Using machine learning to identify Parkinson’s disease severity subtypes with multimodal data

Using machine learning to identify Parkinson’s disease severity subtypes with multimodal data | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Using machine learning to identify Parkinson’s disease severity subtypes with multimodal data Hwayoung Park, Changhong Youm, Sang-Myung Cheon, Bohyun Kim, Hyejin Choi, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5523724/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 02 Jun, 2025 Read the published version in Journal of NeuroEngineering and Rehabilitation → Version 1 posted 6 You are reading this latest preprint version Abstract Background Classifying and predicting Parkinson's disease (PD) is challenging because of its diverse subtypes based on severity levels. Currently, identifying objective biomarkers associated with disease severity that can distinguish PD subtypes in clinical trials is necessary. This study aims to address the clinical applicability and heterogeneity of PD using PD severity subtypes classification and digital biomarker development by combining objective multimodal data with machine learning (ML) approaches. Methods We analyzed datasets that combine clinical characteristics, physical function and lifestyle data, gait parameters in motion analysis systems, and wearable sensors collected from persons with PD (n = 102) to perform clustering for subtype classification. Results We identified three PD severity subtypes, each exhibiting different patterns of clinical severity, with the severity increasing as it progressed from clusters 1 to 3. We found significant mutual information between all/single modalities and the unified PD rating scale scores, identifying potential modalities with high feature importance using ML. Among all modalities, the principal components of gait parameters derived from wearable sensors were identified as the most associated indicators of PD severity. A model utilizing the first principal component of the left and right ankle achieved perfect classification with an area under the curve of 1.0, accurately distinguishing clinically severe subtypes from mild subtypes of PD. These findings suggest that gait features in both ankles can reflect asymmetry factors associated with PD severity subtypes, which contributes to high classification performance. Conclusions Digital biomarkers obtained from wearable sensors attached bilaterally to body segments demonstrate potential for classifying PD severity subtypes and tracking disease progression. Our findings emphasized the clinical value of sensor-based gait analysis in PD management, which suggested its integration into personalized monitoring systems and therapeutic interventions for persons with PD. Parkinson's disease Severity subtype Multimodal data Machine learning Clustering Digital biomarker Figures Figure 1 Figure 2 Figure 3 Background Classifying Parkinson's disease (PD) into subtypes and predicting disease progression based on severity are essential tasks for understanding the diverse symptoms experienced by patients and providing tailored intervention strategies [ 1 ]. Traditional classification methods rely on specific motor symptoms, such as tremors and postural instability [ 2 ], which hinder the evaluation of the comprehensive range of motor and non-motor PD symptoms. Previous studies employed clinical observations based on age at onset or categorization based on most notable features [ 3 , 4 ] to define PD subtypes based on disease severity levels (e.g., mild, moderate, and severe); inching, moderate, and rapid pace subtypes with PD progression rate; early and late onset; motor and non-motor predominance; and presence of dementia [ 5 – 7 ]. Despite being intuitive, these classifications do not represent the clinical features of PD, which are quantifiable, complex, and interrelated. Recently, data-driven multidimensional approaches have shown promise in overcoming existing limitations [ 1 , 8 ]. These data-driven multidimensional approaches can accurately identify and predict PD subtypes by integrating various data sources and using clustering and machine learning (ML) algorithms [ 1 , 3 , 9 , 10 ]. Hence, they can contribute to early diagnosis, track disease progression, develop personalized treatment strategies, and improve clinical trial efficiency [ 3 ]. The use of digital technologies in healthcare, which includes dedicated wearable devices, artificial intelligence-driven analytical tools, and mobile health applications, has recently seen substantial growth. These smart tools continuously generate objective and rich data, providing valuable insights into disease symptoms and PD management [ 11 – 13 ]. Motor impairment in PD is primarily assessed using the movement disorder society (MDS) unified PD rating scale (UPDRS) [ 14 , 15 ]; however, this assessment requires regular clinical visits, which can be time-consuming, costly, and limited by the availability of specialized neurologists and mobility restrictions [ 16 , 17 ]. Wearable sensors address these issues by offering portability, cost-effectiveness, and the ability to assess spatiotemporal characteristics of gait and balance in laboratories, clinics, and homes [ 11 , 18 , 19 ]. Obtaining clinically meaningful information from sensor data requires employing analytical techniques such as initial feature selection and reduction [ 3 , 18 ]. Although widely used for evaluating motor impairments, gait analysis suffers from subjective assessments and inconsistencies in the gait feature analysis [ 20 – 22 ]. Further, wearable sensor data can be unstructured and noisy [ 20 , 23 ]. Therefore, multimodal data may be required to analyze the combined motor and non-motor symptoms of PD [ 24 , 25 ]. Previous studies on PD subtype classification using only clinical data reported reproducibility issues, underscoring the need to identify subtypes based on objective multimodal data [ 26 ]. Integrating ML algorithms with multimodal data can complement existing clinical assessment scales and provide an improved classification of PD severity subtypes [ 1 , 20 ]. ML algorithms can be trained using time-series signals and features collected from wearable sensors during clinical evaluations with relevant scales serving as labels [ 18 , 27 ]. ML analysis on inertial measurement unit (IMU) data has been used for identifying persons with PD from healthy persons or other Parkinsonian disorders and in detecting the signs of bradykinesia or tremors [ 18 , 19 , 28 – 31 ]. Therefore, combining multimodal data with ML algorithms can aid in disease diagnosis, progression tracking, and treatment strategy development [ 25 ]. Further, this approach can potentially contribute to the understanding of the complex mechanisms of PD and the identification of new digital biomarkers [ 3 ]. Digital biomarkers are being increasingly used; however, they present certain limitations. Although various digital devices can monitor physical activity, physiological signals, speech, sleep patterns, and other variables [ 32 ], data collected from individual wearable devices provide limited information for the classification and prediction of diverse subtypes based on PD severity [ 20 , 33 ]. Further, most studies focused on evaluating motor symptoms during walking or rest using restricted datasets, which makes it challenging to clinically validate high-dimensional features extracted from digital devices [ 34 ]. Although models for classifying and predicting PD subtypes based on derived digital biomarkers are currently limited in performance and may be used for initial screening in clinical settings, they are unsuitable for direct clinical applications [ 20 ]. Therefore, digital biomarkers derived from objective, standardized multimodal data using ML are necessary to ensure clinical efficiency and feasibility, which enables the continuous development of models for tracking and predicting PD. In this study, our goal was addressing the clinical applicability and heterogeneity of PD by classifying PD severity subtypes and developing digital biomarkers for an objective diagnosis. We hypothesized that objectively collected multimodal data, along with existing clinical characteristics, can facilitate the classification of PD severity subtypes and be associated with MDS-UPDRS scores, which is a measure of PD severity. Confirming this hypothesis would help us establish potential classification markers, which can aid in developing representative digital biomarkers for distinguishing PD severity subtypes. Methods We established a data-driven framework combining objective multimodal data with ML and statistical approaches (Fig. 1). This framework included integrating data from demographics, clinical characteristics, physical function and lifestyle (PFL), and principal components (PCs) of gait parameters in motion analysis systems (GP_Motion) and wearable sensors (GP_Sensors). 1) We characterized and preprocessed multimodal data of diverse types collected to identify subtypes from persons with PD. 2) We investigated associations between multimodal data collected from identified PD severity subtypes and existing clinical characteristics. We identified potential modalities with a high feature importance specific to each PD severity subtype with ML approaches. 3) Based on the identified modalities, we established potential classification markers for PD severity subtypes, which led to the development of representative digital biomarkers. Participants We used data from 102 persons diagnosed with PD. The participants were assessed by a movement disorder specialist using the MDS-UPDRS criteria, which evaluate the motor and non-motor symptoms of the participants via clinician-rated and patient-reported measures. Participants were included if they showed mild-to-moderate idiopathic PD, received anti-Parkinsonian medications, and could walk and stand unassisted during clinical tests. However, they were excluded if they had other neurological, orthopedic, or psychiatric disorders. The average age of persons with PD ( n = 102) was 68 years, and 52.9% of them were female. In addition, 49.0% of the patients had Hoehn and Yahr (H&Y) stage 2 disease, an average disease duration of 5.7 years, and a motor score (MDS-UPDRS Part III) of 29.5 (Table 1). [Table ] This study was approved by the Institutional Review Board of the Dong-A University Medical Center (Approval number DAUHIRB-22-089). All participants were informed about the study aims and protocols and signed a written informed consent form before enrollment. The study was conducted according to relevant guidelines and regulations. Our study was registered with the Clinical Research Information Service of the Republic of Korea (KCT0009353). Experimental procedures Each participant visited our laboratory twice during the study period. The participants underwent a multimodal assessment of clinical characteristics, physical function tests, lifestyle questionnaires, and gait measurements using a motion analysis system and wearable sensors. Patients using levodopa were assessed in the “on” medication state after 2–3 h of medication intake. We considered data modalities described below. Detailed descriptions of these measures are provided in Supplementary Material 2 (Table 1). Clinical measurements (12 modalities): Clinical assessments were conducted in collaboration with neurologists and nurses using structured questionnaires and patient observation. Persons with PD were assessed using the H&Y stage and MDS-UPDRS, including the Total and Part I–III scores. MDS-UPDRS Total assesses the severity scales of motor and non-motor PD symptoms; Part I evaluates non-motor experiences of daily living, such as cognitive impairment and mood disorders; Part II assesses the motor experiences of daily living, including speech, handwriting, and hygiene; and Part III measures motor disorder severity, which focuses on motor symptoms such as tremors, rigidity, and bradykinesia. Further, mini-mental state examination (MMSE) and Montreal cognitive assessment (MoCA) were applied for evaluating the global cognitive function. Higher scores on the MDS-UPDRS indicate greater symptom severity, whereas higher scores on the MMSE and MoCA reflect better cognitive function. PFL measurements (11 modalities): The physical function was assessed using grip strength (left and right), five times sit-to-stand test (5 STST), six min walking test (6 MWT), short physical performance battery (SPPB), and mini-balance evaluation systems test (Mini-BEST). Lifestyle scores included the nutrition quotient (NQ), 36-item short form health survey (SF-36), and international physical activity questionnaire (IPAQ) scores. Physical function assessments were conducted using standardized protocols by trained evaluators who guided participants through the tests and assigned scores based on their performance levels. NQ is a validated dietary assessment tool developed by the Korean Ministry of Food and Drug Safety. NQ comprehensively evaluates the nutritional status and dietary quality of adults [35]. SF-36 was used for assessing overall health-related quality of life, which includes total, physical, and mental scores. We clarify that SF-36 was used in this study, not SF-12 (a shorter version with 12 items), to avoid any potential confusion. Participants completed the study under the supervision of trained examiners to ensure consistency and accuracy. GP_Motion (38 modalities; 8 PCs) and GP_Sensors (360 modalities; 35 PCs): Participants performed the following gait tests: 1) forward and 2) backward straight walking at a preferred speed and 3) 360° turning at a preferred and fast speed and turning to the left and right sides. For all tasks, kinematic data were captured using a full-body marker-based motion capture system (Plug-in Gait model, Vicon Motion Systems, Oxford Metrics, UK) and an array of six IMUs (Xsens DOT, Movella Technologies, Enschede, Netherlands). The raw data were collected at 100 Hz from the Vicon system and at 60 Hz from the Xsens DOT sensors. The IMU sensors were attached to six anatomical landmarks: left lateral humeral epicondyle (LELB), right lateral humeral epicondyle (RELB), left lateral malleolus (LANK), right lateral malleolus (RANK), 10th thoracic vertebra (T10), and posterior superior iliac spine (PSIS). A sensor placement was designed for capturing whole-body movement patterns and gait dynamics, which provides both spatial and temporal data from multiple body segments. The IMU sensors were calibrated before each trial according to the manufacturer’s guidelines to ensure the reliability of collected data. The data consistency was ensured by conducting trials under controlled laboratory conditions, which minimizes the effect of external factors. For GP_Motion variables, spatiotemporal gait parameters such as walking speed, stride length, double support phase, and contralateral temporal coordination [36] were extracted from 3D marker data for analysis. These parameters are clinically recognized as important indicators to assess gait dysfunction, which includes gait instability and asymmetry in individuals with PD [36]. GP_Sensors included IMU-derived features such as maximum jerk, angular velocity jerk, mean and maximum acceleration, RMS acceleration and gyroscope measurements, and sample entropy of acceleration and gyroscope data. These extracted features from both motion capture and IMU sensors provided a multimodal dataset to analyze gait abnormalities in PD, which enable a more comprehensive characterization of movement patterns beyond conventional joint kinematic analyses. Data analysis Data analysis involved preprocessing collected multimodal data, clustering, feature selection, model training and selection, and classification based on logistic regression for identifying digital biomarkers. Figure 1 illustrates this process, and the steps are detailed below. Preprocessing The collected data were initially inspected to examine their structure. Preprocessing involved two key tasks. First, k -nearest neighbor imputation was applied to handle missing values. An imputer with five neighbors was initialized and applied to the dataset to fill the missing values. This technique enabled estimating the missing values based on the mean across the nearest neighbors, thereby ensuring that imputed values were informed by similar data points. Second, the data were normalized using z scores to standardize features, which involved scaling the data to have a mean of 0 and standard deviation of 1. Clustering We adopted a multimodal clustering approach. k -means clustering was applied to categorize multimodal data into three clusters. The clustering model was initialized with three clusters (n_clusters = 3) and a random state of 42 for ensuring reproducibility. Subsequently, the model was fitted to the scaled data. Each data point was assigned a cluster label of 1, 2, or 3, which indicates its membership to one of the three clusters. Feature selection Mutual information (MI) was employed for determining multimodal measures most reflective of PD severity, as assessed using MDS-UPDRS [11]. The corresponding nonparametric supervised method was used for estimating the relationship between the MDS-UPDRS parts and multimodal data. MI provided scores that quantify mutual dependencies and amount of shared information between each feature and MDS-UPDRS scores. In our study, MI was calculated using a natural logarithm base (ln). If log base 2 (log 2 ) were used, MI scores would range between 0 and 1. All MI values obtained in our analysis remained within the range of 0 to 1, which indicates that our computed MI scores did not exceed 1 in this study although MI is theoretically unbounded depending on the logarithm base. A higher MI score indicates a stronger dependency between the selected feature and PD severity. The calculation process was repeated 100 times (100-fold validation) with a random partition seed per MDS-UPDRS component to ensure stable and consistent feature rankings. This approach was specifically selected for feature selection instead of using the traditional model evaluation because it helps mitigate random partitioning effects and enhances the robustness of feature importance estimation. In each iteration, 80% of the data were randomly selected for training, while the remaining 20% were used for validation, ensuring different subsets contributed to feature selection. MI scores between each feature and MDS-UPDRS scores were computed in every iteration. The final feature selection score was determined by counting the number of times each feature appeared among top-ranked features and multiplying this count by its average MI score across 100 folds (see Supplementary Material 3). This method ensured that feature selection using MI scores reliably identified multimodal data most relevant to PD severity, which reduces the effect of random sampling effects. Model training and selection Various ML models were evaluated for determining the association between the selected modalities and MDS-UPDRS scores. The data were split into training (80%) and test (20%) sets, which ensures that model evaluation was performed on unseen data. The evaluated models included the random forest regressor and least absolute shrinkage and selection operator (LASSO). Model training was performed using five-fold cross-validation with Pearson’s R 2 as the performance metric. Hyperparameter tuning was performed using the grid search method to identify optimal parameters per model (see Supplementary Material 1). We employed a nested five-fold cross-validation approach to optimize hyperparameters and assess model selection stability. This nested cross-validation framework effectively prevents overfitting and provides a more stable estimate of model generalization performance compared to that of a simple three-way data split. Inner loop (hyperparameter tuning): Five-fold cross-validation was conducted within the training set (80%) for identifying optimal hyperparameters using a grid search method. This ensured that hyperparameter tuning was performed solely within the training set, thereby preventing any data leakage into the final test set. Outer loop (model selection and evaluation): Five-fold nested cross-validation was applied for validating the model selection process, assessing its stability, and minimizing overfitting risks. The final model was trained on the training set and evaluated on the test set for ensuring that it was blinded to the test data during training. Mean absolute error (MAE) and Pearson’s R 2 were calculated to assess the correlation between the predicted and actual MDS-UPDRS scores. In models such as the random forest regressor and LASSO, intrinsic feature importance attributes were explored for identifying modalities that most significantly influenced model predictions. For the random forest regressor, feature importance was derived using the Gini importance (mean decrease in impurity). Importance scores were computed for each feature and analyzed for understanding its contribution to prediction. For LASSO, feature importance was assessed based on the magnitudes of coefficients assigned to each feature by the model. Features with larger absolute coefficients were considered more important. The feature importance scores provided insights into modalities that were the most predictive of the MDS-UPDRS scores. Identification of digital biomarkers based on logistic regression We employed recursive feature elimination with logistic regression for identifying the most relevant digital biomarkers to classify clustering labels. Therefore, we selected features by recursively considering smaller sets, starting with an external estimator that assigned weights to features. Initially, the estimator was trained on the complete feature set, and the importance of each feature was determined either through a coef_ or feature_importances_ attribute. Subsequently, the least important features were pruned from the current set. This process was repeated recursively on the pruned set until the desired number of features was reached, which were the top 40 modalities in this study. These selected modalities served as potential digital biomarkers and were used in subsequent analyses for reducing data dimensionality while retaining representative predictors. A logistic regression model was trained using selected digital biomarkers to predict clustering labels. The data were split into training and test sets containing 80% and 20% of the samples, respectively, using a random state of 2 to ensure reproducibility. The logistic regression model was trained and evaluated on the training and test sets, respectively. Performance metrics such as the receiver operating characteristic (ROC) curve and area under the curve (AUC) were computed. The ROC curve was plotted to visualize the trade-off between sensitivity (true positive rate) and specificity (1 − false positive rate). ROC-AUC was computed for providing the average and standard deviation of each measure of model performance. Statistical analysis Data normality was assessed using the Shapiro–Wilk test. An independent t -test or nonparametric statistics was used for analyzing the mean and standard deviation of the physical and clinical characteristics of all participants. Based on the normality results, appropriate statistical tests were applied for pairwise comparisons between clusters. Independent t -tests were used when data were distributed normally. Mann–Whitney U tests were applied when data did not meet normality assumptions. Multiple comparisons were conducted across overlapping clusters (1 vs. 2, 2 vs. 3, and 1 vs. 3), and therefore, we applied false discovery rate (FDR) correction (Benjamini–Hochberg method) to control for type I errors while maintaining statistical power. Both uncorrected and FDR-adjusted p-values are reported in Table 1. All statistical analyses were performed using SPSS version 22.0 (SPSS, Chicago, IL). The statistical significance level was set to 0.05. Data preprocessing and analysis were conducted using Python (version 3.10) with libraries including Pandas, NumPy, Scikit-learn, and Matplotlib. Results Identification of PD severity subtypes using clustering based on multimodal data Persons with PD were assigned to three subtypes, namely, clusters 1 ( n = 24), 2 ( n = 47), and 3 ( n = 31). We observed statistically significant differences in clinical characteristics after applying FDR correction to account for multiple comparisons. MDS-UPDRS Total, Part I, and Part II showed significant differences between clusters 2 and 3 using independent t -tests, whereas MoCA exhibited significant differences using the Mann–Whitney U test. Similarly, significant differences were found between clusters 1 and 3 in MDS-UPDRS Total, Part I, Part II, and Part III using independent t -tests, whereas MMSE and MoCA exhibited significant differences using the Mann–Whitney U test. We defined clusters 1 to 3 as mild, moderate, and severe subtypes with PD, respectively (Table 1 ). The PD severe subtype showed higher ages and lowest cognitive measures in tests such as MMSE and MoCA compared to those of the other subtypes. In addition, this subtype showed the most severe clinical characteristics, including the H&Y stage, MDS-UPDRS Total, Parts I, II, and III scores. This subtype also demonstrated the lowest scores for physical function measures such as grip strength, 5 STST, SPPB, and Mini-BEST, as well as lifestyle scores, which include the NQ, SF-36 (in total, physical, and mental scores), and IPAQ. Furthermore, they exhibited the lowest straight and turning gait speeds with shorter steps. The related information is summarized in Supplementary Material 2. Feature selection associated with MDS-UPDRS scores We used the MI for the nonparametric supervised estimation of the relationship between the MDS-UPDRS parts and data modalities to identify those most reflective of PD severity. Further, MI measures the dependency or shared information between two variables. A higher MI score indicates a stronger relationship between the feature and the target measure, which implies that the feature is more informative or predictive about the target [ 11 , 37 ]. We determined modalities with the highest association with the scores of MDS-UPDRS Total and Parts I–III (see Supplementary Material 3) and obtained the most prominent domains using the feature selection score (Table 2 ). The modalities were classified into four domains: 1) demographics (4 features), 2) PFL scores (11 features), 3) GP_Motion (8 features), and 4) GP_Sensors (35 features). The most associated UPDRS parts in both all/single modalities were UPDRS Total scores, and the most important domain was GP_Sensors. [Table 2 ] ML regression models for estimating PD severity The most frequently selected and accurate models using five-fold cross-validation were identified for determining the importance of features most associated with MDS-UPDRS scores (Table 3). The selected model on all modalities was the LASSO for MDS-UPDRS Part II, with an average Pearson’s R 2 of 0.41 and MAE of 0.66, which implies the average difference between the predicted and actual MDS-UPDRS scores on the test set. The selected model on single modalities was LASSO for PFL modalities for MDS-UPDRS Part II, with an average Pearson’s R 2 of 0.55 and MAE of 0.54. For GP_Motion modalities, the MDS-UPDRS Part III was suitably predicted using random forest regressor, with an average Pearson’s R 2 of 0.20 and MAE of 0.87. For GP_Sensors modalities, MDS-UPDRS Part II was suitably predicted using a random forest regressor, with an average Pearson’s R 2 of 0.34 and MAE of 0.70. [Table 3] The dominant modalities of top features were identified for exploring the feature importance of the selected model (random forest regressor or LASSO). The top three features of all modalities, including SF-36 (total), Mini-BEST, and 10th thoracic vertebra (T10)_PC2 (time-domain gyroscope) measurements, were important features to estimate the disease severity for MDS-UPDRS Part II. Single PFL modalities, which include SF-36 (total), Mini-BSET, and SF-36 (physical), were important for the MDS-UPDRS Part II. For GP_Motion modalities, Turning_PC6 (contralateral temporal coordination), Backward walking_PC7 (left and right double support phase), and Turning_PC2 (left and right double support phase) in MDS-UPDRS Part III were considered important. For GP_Sensors modalities, important features were T10_PC2 (time-domain gyroscope measurements), right lateral malleolus (RANK)_PC1 (time-domain acceleration and gyroscope measurements), and right lateral humeral epicondyle (RELB)_PC1 (time-domain acceleration and gyroscope measurements) in MDS-UPDRS Part II (Fig. 2 ). Further details on the importance of features contributing to the accuracy of these models and feature descriptions are presented in Supplementary Material 4. Identification of digital biomarkers based on classification models Following the classification of persons with PD according to the three severity subtypes based on multimodal data clustering, we developed PD severity subtypes classification models to identify digital biomarkers for estimating disease severity classes based on MDS-UPDRS scores. The models used all modalities, separate PFL modalities, separate GP_Motion modalities, and separate GP_Sensors modalities. Figure 3 and Table 4 show the ROC curves illustrating the performances of the PD severity subtype classification models for MDS-UPDRS scores using unsupervised clustering based on multimodal data.[Table 4 ] Discussion We processed the multimodal data collected from persons with PD to classify PD severity subtypes and identified digital biomarkers associated with the widely used PD severity assessment tools, MDS-UPDRS parts, using ML algorithms. Our analysis provided the following insights: Multimodal data, including demographics, PFL, and PCs of GP_Motion and GP_Sensors, enabled clustering persons with PD into three PD severity subtypes (mild, moderate, and severe). Our multimodal data could be associated with MDS-UPDRS Total scores and the most important domain was found to be the PCs of GP_Sensors. The LASSO model, based on all and PFL modalities, achieved the highest feature importance in MDS-UPDRS Part II. The model yielded average Pearson’s R ² values of 0.41 for all modalities and 0.55 for PFL modalities, with MAE values of 0.66 and 0.54, respectively. The random forest regressor model based on GP_Motion and GP_Sensors modalities achieved the highest feature importance in MDS-UPDRS Parts III and II, respectively. The average Pearson’s R ² values were 0.20 for GP_Motion and 0.34 for GP_Sensors, with MAE values of 0.87 and 0.70, respectively. Digital biomarkers were derived from high-importance multimodal data to develop classification models for PD severity subtypes. A model with LANK_PC1 (forward walking and turning) and RANK_PC1 (forward and backward walking and turning) among all modalities (100.0%, AUC: 0.99) accurately distinguished clinically severe subtypes from mild subtypes with PD. In addition, the PFL, GP_Motion, and GP_Sensors modalities accurately distinguished severe subtypes from mild subtypes with PD with accuracies of 0.75, 0.83, and 1.00, and AUCs of 0.74, 0.72, and 0.91, respectively. Identification of PD severity subtypes using clustering based on multimodal data In this study, persons with PD were divided into three PD severity subtypes using clustering. The key clinical characteristics showed statistically significant differences between PD severity subtypes in terms of the comprehensive disease severity score (MDS-UPDRS Total), motor experiences of daily living (MDS-UPDRS Part II), and cognitive measures (MoCA). Patients in the severe subtype, which included persons with PD with the most severe symptoms and lowest cognitive measures, had the lowest PFL scores (see Supplementary Material 1). This classification aligns with clinical scales such as MDS-UPDRS scores and provides additional quantitative insights into motor and non-motor dysfunction in persons with PD, which can be used for guiding personalized rehabilitation strategies [ 5 ]. For example, mild persons with PD exhibit relatively preserved motor function, which suggests that early exercise interventions and proactive rehabilitation can help slow disease progression [ 3 , 38 ]. Moderate persons with PD experience gait and balance impairments, emphasizing the need for targeted gait training and fall prevention programs [ 38 ]. Severe persons with PD present significant motor dysfunction and postural instability, indicating that assistive devices or intensive physiotherapy may be necessary for maintaining functional mobility [ 11 ]. Unlike conventional H&Y stages, which rely on clinical observation, our clustering model incorporates multimodal features such as sensor-derived, kinematic, PFL, and clinical data to provide a more data-driven stratification of PD severity. This finding highlights the heterogeneity of PD and underscores the need for personalized interventions based on severity subtype classification [ 1 , 25 ]. Feature selection associated with MDS-UPDRS scores We used the MI algorithm for identifying multimodal data that can reflect PD severity. Thus far, numerous studies validated and demonstrated the benefits of using objective and highly correlated features for PD severity classification and disease progression monitoring [ 11 , 38 , 39 ], indicating their potential as clinical support tools [ 40 ]. Similarly, our findings reveal associations between multimodal data and MDS-UPDRS parts, with particularly high Total scores. Therefore, multimodal datasets can be included for comprehensively evaluating PD signs and symptom severity, which highlights the value of analyzing multimodal data for comprehensively understanding PD severity [ 20 , 25 ]. Interestingly, GP_sensors outperformed GP_motion in PD severity classification, which can be attributed to the ability of IMU sensors to capture continuous movement dynamics, acceleration-based features, and entropy-based measures, which are highly relevant to PD-related gait abnormalities [ 41 , 42 ]. Further, IMU-derived features provide richer information on postural instability and movement variability, which are key factors in PD motor dysfunction [ 43 ]. Moreover, these sensor-based features exhibited the highest associations with multimodal data and were significantly correlated with MDS-UPDRS Total and Part II scores, reflecting their relevance to motor experiences of daily living in persons with PD [ 11 , 44 ]. Wearable sensor-based measurements offer a distinct advantage over laboratory-based assessments because they can capture real-world behavioral contexts and fluctuations in motor symptoms [ 11 ]. Unlike clinical gait tests, which are controlled by external instructions, wearable devices enable long-term monitoring in naturalistic settings, thereby enabling data collection across different behavioral contexts, such as motor fluctuations and on/off medication states [ 11 , 45 ]. This ability to continuously track motor function over time is valuable in PD management, where symptom variability is a key challenge [ 46 ]. Therefore, validating sensor-based measurements must involve not only clinical assessments but also real-world functional outcomes, considering temporal changes and individualized therapeutic responses [ 46 ]. Future studies should explore how combining IMU and motion capture features could enhance classification performance and provide a more comprehensive understanding of PD gait pathology [ 11 , 46 , 47 ]. ML regression models for estimating PD severity We used ML regression models and five-fold cross-validation to estimate PD severity. The LASSO and random forest regressor models were used based on all and single modalities to identify the importance of various features for predicting MDS-UPDRS parts. ML has been widely applied to learn patterns from diverse data and estimate PD severity [ 3 , 18 , 19 , 23 , 28 , 48 ]. Random forest regressors can account for collinearity in high-dimensional datasets and achieve an adequate discriminative performance [ 28 , 49 ]. Therefore, the most relevant features could be identified without excessive influence during feature selection and importance processing despite numerous features derived from wearable sensors in multimodal data [ 11 ]. Modalities related to MDS-UPDRS Part II included the total scores of SF-36 and Mini-BEST from the PFL modalities and T10_PC2 and LANK_PC3 from the GP_Sensors modalities in our study. Further, the single modalities for PFL modalities related to MDS-UPDRS Part II included the total and physical scores of SF-36 and Mini-BEST. Persons with PD perceive that slow movements (bradykinesia), tremors, postural instability, and gait disturbances are the most challenging symptoms in daily living that substantially reduce their quality of life [ 50 ]. Our findings suggest that the measures most closely associated with the severity of daily living motor experiences were the quality of life, posture and balance measures, and gyroscope measurements from the body’s central axis and distal segments during gait tasks [ 3 , 23 ]. Therefore, the application of specific multimodal data, clinical evaluation scales, and ML algorithms for motor and non-motor symptom evaluation can support precision in therapeutic interventions [ 18 ]. Identification of digital biomarkers based on classification models We used the ROC-AUC to evaluate our classification models based on modalities with high feature importance for identifying digital biomarkers to classify PD severity subtypes. Movement disorder specialists use MDS-UPDRS as a screening tool for evaluating PD severity, monitoring disease progression, and assessing treatments and interventions [ 3 , 14 , 15 ]. However, clinical assessments require experience, expertise, and time [ 16 , 17 ]. Inaccurate clinical assessments of persons with PD can lead to the suboptimal characterization of patients and clinical processes, which can affect diagnosis and treatment [ 3 ]. We systematically evaluated meaningful multimodal data collected from persons with PD, assessed their accuracy in relation to MDS-UPDRS parts, and identified digital biomarkers that contributed to accurately classifying PD severity subtypes. GP_Sensors modalities such as LANK_PC1, T10_PC2, and RANK_PC1 from multimodal data showed the best classification performance between clusters based on PD severity. These features were derived from wearable sensor measurements and closely associated with MDS-UPDRS Total and Part III scores, which underscores the importance of gait measurements and standard clinical scales in identifying PD severity subtypes [ 11 , 18 ]. As PD progresses from its early stages, gait disturbances, among the major motor symptoms, are likely to be objective and highly sensitive biomarkers [ 11 ]. UPDRS scores have been predicted using gait time-series measurements, demonstrating promising results in terms of the MAE and RMS error [ 18 , 20 ]. In this study, accelerometer and gyroscope measurements from wearable sensors attached to the ankles and back were used for describing the posture and gait patterns. A previous study reported that reduced gait performance and greater postural sway were associated with higher PD severity [ 32 ]. Quantitative gait measurements and their characteristics suggest that they can be used as digital biomarkers to realize enhanced PD severity classification [ 32 , 51 ]. Considering individual PFL modalities, grip strength, 6 MWT, and Mini-BEST tests effectively distinguished PD severity subtypes in our study, thereby highlighting how such measures reflect the decline in motor function with PD progression. Grip strength is not only a measure of upper limb strength but also an indicator of PD progression from the mild to moderate stages [ 52 ]. Reduced grip strength can adversely affect activities of daily living and increase the risk of falls while performing tasks such as opening door handles or refrigerator doors. This issue can be attributed to the weakening of the upper and lower limbs in persons with PD [ 53 ]. In our study, the 6 MWT, which was related to MDS-UPDRS Total, was valuable for assessing PD severity because it measured the distance walked by the patient as fast as possible in 6 min [ 54 ]. In addition, the Mini-BEST score was among the top features, and it measured the ability of a patient for balancing or performing in daily life activities, also reflecting the fall risk [ 55 ]. Gait parameters showed that specific spatiotemporal features could effectively classify PD severity subtypes in our study, supporting the use of wearable sensors in clinical assessments [ 17 – 19 , 22 ]. However, the spectrum of disease severity has not been considered, and several variables are difficult to interpret clinically [ 56 , 57 ]. Further, using only continuous data collected from a single wearable sensor may be insufficient for identifying persons with PD from healthy controls or classifying PD severity subtypes [ 22 ]. Rehman et al. attempted to differentiate persons with PD from healthy controls using more than 100 gait variables and ML approaches using the GAITRite walkway. They identified six spatiotemporal variables that achieved an accuracy of 73–97% [ 58 ]. In our previous study, we distinguished persons with PD from healthy controls with a 98.0% accuracy using five spatiotemporal variables during the same 360° turning task [ 36 ]. Persons with PD often exhibit considerable clinical asymmetry, which can cause gait disturbances and freezing of gait [ 59 ]. Asymmetry appears in the early PD stages and can persist with an increase in disease severity [ 60 ]. Our results confirm that gait features in both ankles such as LANK_PC1 and RANK_PC1 can indicate asymmetry factors based on the PD severity subtype, which promotes a high classification performance. Therefore, we suggest that features obtained by bilaterally attaching wearable sensors to body segments can be used and monitored to classify PD severity subtypes. Although this study represents a step forward in our efforts to classify PD severity subtypes and predict PD severity, further investigation is required. In fact, our analysis has the following limitations: Although clustering is data-driven, the outcomes depend on the appropriate selection of variables and clustering algorithms. Further, PD severity subtypes must be validated considering independent cohorts covering similar domains to confirm the clinical applicability of the study because an unsupervised approach to subtyping was employed in the study. All participants were diagnosed with idiopathic PD based on criteria established by a neurologist. However, we cannot completely exclude the possibility that some participants had other underlying pathologies that could have affected our results. The sample size was too small for ML analysis. However, our approach highlights the importance of incorporating multimodal features for comprehensively analyzing PD severity subtypes. Data were collected exclusively in a controlled laboratory setting. Such a controlled environment may not fully represent real-world variability and complexity, potentially limiting the generalizability of our findings. Data were collected only in the “on” medication state. Continuous monitoring during the transition periods between the “on” and “off” medication states is necessary for capturing the full spectrum of motor symptom fluctuations. Conclusions This study demonstrated the efficacy of multimodal data and advanced ML algorithms for identifying PD severity subtypes and estimating disease severity. Our approach provided a comprehensive framework that integrates demographics, physical function, lifestyle measures, and gait parameters to understand PD heterogeneity. Our findings highlighted the potential of bilaterally attached wearable sensors as digital biomarkers to classify PD severity subtypes and track disease progression. This underscores the clinical value of sensor-based gait analysis in PD management, which supports its integration into personalized monitoring systems and therapeutic interventions [ 18 , 31 , 32 ]. Future research should validate these findings in larger independent cohorts and explore the clinical integration of digital biomarkers for real-time disease monitoring and therapy adjustments. Wearable sensors and mobile health applications can enhance personalized PD management by enabling continuous assessment and tailored rehabilitation strategies [ 36 ]. Abbreviations PD Parkinson's disease ML machine learning MDS Movement disorder society UPDRS Unified Parkinson's disease rating scale IMU inertial measurement unit PFL Physical function and lifestyle PCs principal components GP_Motion Gait parameter PCs in motion analysis system GP_Sensors gait parameter PCs in wearable sensors MoCA Montreal cognitive assessment MMSE Mini-mental state examination H&Y Hoehn and Yahr 5STST five times sit-to-stand test Mini-BEST Mini-balance evaluation systems test SF-36 36-item short form health survey IPAQ International physical activity questionnaire MI Mutual information LASSO Least absolute shrinkage and selection operator MAE mean absolute error T10 10th thoracic vertebra RANK right lateral malleolus RELB right lateral humeral epicondyle ROC receiver operating characteristic PSIS posterior superior iliac spine LANK left lateral malleolus AUC area under the curve LELB left lateral humeral epicondyle 6MWT 6 min walking test. Declarations Ethics approval and consent to participate All procedures involving human participants were performed in accordance with the ethical standards of the institutional and/or national research committee and observing the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. The study and its additional files were approved by the Institutional Review Board of Dong-A University Hospital (approval number DAUHIRB-22-089) (see Supplementary Material 5). All patients provided written informed consent prior to data collection. The study was registered with the Clinical Research Information Service of the Republic of Korea (KCT0009353). Consent for publication Not applicable. Competing interests The authors declare no conflicts of interest. Funding This work was supported by grants from the National Research Foundation of Korea (NRF) funded by the Korean government (MSIT) (No. 2022R1A2C100933711; Changhong Youm), the Basic Science Research Program through the NRF funded by the Ministry of Education (No. 2022R1A6A3A0108756411; Hwayoung Park), and the Ministry of Education of the Republic of Korea and the NRF (No. 2024S1A5B5A16021673; Hwayoung Park). This study received no specific grants from funding agencies in the public, commercial, or non-profit sectors. The funding sources had no role in the study design; collection, analysis, and interpretation of the data; or in writing the manuscript. Author Contribution H.P., C.Y., and S.C. conceived and designed the study. H.P., S.C., and B.K. recruited the participants. H.P., C.Y., S.C., B.K., H.C., J.H., and M.K. performed data acquisition. H.P. and C.Y. analyzed and interpreted the data. H.P., C.Y., and S.C. drafted the manuscript. All authors read and approved the final version of the manuscript. Acknowledgement The authors thank all participants who contributed to this study. This work was supported by the Dong-A University research fund. The authors also thank Editage (www.editage.co.kr) for English language editing. Data Availability Datasets supporting the findings of this study are available from the corresponding author upon request. References Birkenbihl C, Ahmad A, Massat NJ, Raschka T, Avbersek A, Downey P, et al. Artificial intelligence-based clustering and characterization of Parkinson’s disease trajectories. Sci Rep. 2023;13:2897. https://doi.org/10.1038/s41598-023-30038-8 . Fereshtehnejad SM, Postuma RB. Subtypes of Parkinson’s disease: What do they tell us about disease progression? Curr Neurol Neurosci Rep. 2017;17:34. https://doi.org/10.1007/s11910-017-0738-x . Dadu A, Satone V, Kaur R, Hashemi SH, Leonard H, Iwaki H, et al. Identification and prediction of Parkinson’s disease subtypes and progression using machine learning in two cohorts. npj Parkinsons Dis. 2022;8:172. https://doi.org/10.1038/s41531-022-00439-z . Stebbins GT, Goetz CG, Burn DJ, Jankovic J, Khoo TK, Tilley BC. How to identify tremor dominant and postural instability/gait difficulty groups with the movement disorder society unified Parkinson’s disease rating scale: Comparison with the unified Parkinson’s disease rating scale. Mov Disord. 2013;28:668–70. https://doi.org/10.1002/mds.25383 . Su C, Hou Y, Xu J, Xu Z, Zhou M, Ke A, et al. Identification of Parkinson’s disease PACE subtypes and repurposing treatments through integrative analyses of multimodal data. npj Digit Med. 2024;7:184. https://doi.org/10.1038/s41746-024-01175-9 . Zetusky WJ, Jankovic J, Pirozzolo FJ. The heterogeneity of Parkinson’s disease: Clinical and prognostic implications. Neurology. 1985;35:522. https://doi.org/10.1212/wnl.35.4.522 . Jankovic J, McDermott M, Carter J, Gauthier S, Goetz C, Golbe L, et al. Variable expression of Parkinson’s disease: A baseline analysis of the DAT ATOP cohort. Neurology. 1990;40:1529. https://doi.org/10.1212/wnl.40.10.1529 . Faghri F, Brunn F, Dadu A, PARALS consortium ERRALS consortium, Zucchi E et al. Identifying and predicting amyotrophic lateral sclerosis clinical subgroups: A population-based machine-learning study. Lancet Digit Health. 2022;4:e359–69. https://doi.org/10.1016/S2589-7500(21)00274-0 Van Rooden SM, Heiser WJ, Kok JN, Verbaan D, van Hilten JJ, Marinus J. The identification of Parkinson’s disease subtypes using cluster analysis: A systematic review. Mov Disord. 2010;25:969–78. https://doi.org/10.1002/mds.23116 . Fereshtehnejad SM, Zeighami Y, Dagher A, Postuma RB. Clinical criteria for subtyping Parkinson’s disease: Biomarkers and longitudinal progression. Brain. 2017;140:1959–76. https://doi.org/10.1093/brain/awx118 . Mirelman A, Volkov J, Salomon A, Gazit E, Nieuwboer A, Rochester L, et al. Digital mobility measures: A window into real-world severity and progression of Parkinson’s disease. Mov Disord. 2024;39:328–38. https://doi.org/10.1002/mds.29689 . Espay AJ, Bonato P, Nahab FB, Maetzler W, Dean JM, Klucken J, et al. Technology in Parkinson’s disease: Challenges and opportunities. Mov Disord. 2016;31:1272–82. https://doi.org/10.1002/mds.26642 . Houts CR, Patrick-Lake B, Clay I, Wirth RJ. The path forward for digital measures: Suppressing the desire to compare apples and pineapples. Digit Biomark. 2020;4(Suppl 1):3–12. https://doi.org/10.1159/000511586 . Goetz CG, Tilley BC, Shaftman SR, Stebbins GT, Fahn S, Martinez-Martin P, et al. Movement Disorder Society-sponsored revision of the Unified Parkinson’s disease Rating Scale (MDS‐UPDRS): Scale presentation and clinimetric testing results. Mov Disord. 2008;23:2129–70. https://doi.org/10.1002/mds.22340 . Stephenson D, Badawy R, Mathur S, Tome M, Rochester L. Digital progression biomarkers as novel endpoints in clinical trials: A multistakeholder perspective. J Parkinsons Dis. 2021;11:S103–9. https://doi.org/10.3233/JPD-202428 . Post B, Merkus MP, de Bie RMA, de Haan RJ, Speelman JD. Unified Parkinson’s disease rating scale motor examination: Are ratings of nurses, residents in neurology, and movement disorders specialists interchangeable? Mov Disord. 2005;20:1577–84. https://doi.org/10.1002/mds.20640 . Mirelman A, Hillel I, Rochester L, Del Din S, Bloem BR, Avanzino L, et al. Tossing and turning in bed: Nocturnal movements in Parkinson’s disease. Mov Disord. 2020;35:959–68. https://doi.org/10.1002/mds.28006 . Sotirakis C, Su Z, Brzezicki MA, Conway N, Tarassenko L, FitzGerald JJ, et al. Identification of motor progression in Parkinson’s disease using wearable sensors and machine learning. npj Parkinsons Dis. 2023;9:142. https://doi.org/10.1038/s41531-023-00581-2 . Zadka A, Rabin N, Gazit E, Mirelman A, Nieuwboer A, Rochester L, et al. A wearable sensor and machine learning estimate step length in older adults and patients with neurological disorders. npj Digit Med. 2024;7:142. https://doi.org/10.1038/s41746-024-01136-2 . Faiem N, Asuroglu T, Acici K, Kallonen A, Van Gils M. Assessment of Parkinson’s disease severity using gait data: A deep learning-based multimodal approach. Nordic Conf Digit Health Wirel Solutions 29–48; 2024. Vásquez-Correa JC, Arias-Vergara T, Orozco-Arroyave JR, Eskofier B, Klucken J, Noth E. Multimodal assessment of Parkinson’s disease: A deep learning approach. IEEE J Biomed Health Inf. 2019;23:1618–30. https://doi.org/10.1109/JBHI.2018.2866873 . Godi M, Arcolin I, Giardini M, Corna S, Schieppati M. A pathophysiological model of gait captures the details of the impairment of pace/rhythm, variability and asymmetry in Parkinsonian patients at distinct stages of the disease. Sci Rep. 2021;11:21143. https://doi.org/10.1038/s41598-021-00543-9 . Chandrabhatla AS, Pomeraniec IJ, Ksendzovsky A. Co-evolution of machine learning and digital technologies to improve monitoring of Parkinson’s disease motor symptoms. npj Digit Med. 2022;5:32. https://doi.org/10.1038/s41746-022-00568-y . Nalls MA, McLean CY, Rick J, Eberly S, Hutten SJ, Gwinn K, et al. Diagnosis of Parkinson’s disease on the basis of clinical and genetic classification: A population-based modelling study. Lancet Neurol. 2015;14:1002–9. https://doi.org/10.1016/S1474-4422(15)00178-7 . Albrecht F, Poulakis K, Freidle M, Johansson H, Ekman U, Volpe G, et al. Unraveling Parkinson’s disease heterogeneity using subtypes based on multimodal data. Parkinsonism Relat Disord. 2022;102:19–29. https://doi.org/10.1016/j.parkreldis.2022.07.014 . Mestre TA, Eberly S, Tanner C, Grimes D, Lang AE, Oakes D, et al. Reproducibility of data-driven Parkinson’s disease subtypes for clinical research. Parkinsonism Relat Disord. 2018;56:102–6. https://doi.org/10.1016/j.parkreldis.2018.07.009 . Rehman RZU, Rochester L, Yarnall AJ, Del Din S. Predicting the progression of Parkinson’s disease MDS-UPDRS-III motor severity score from gait data using deep learning 43rd Ann. Int Conf IEEE Eng Med Biol Soc (EMBC) 249–52; 2021. De Vos M, Prince J, Buchanan T, FitzGerald JJ, Antoniades CA. Discriminating progressive supranuclear palsy from Parkinson’s disease using wearable technology and machine learning. Gait Posture. 2020;77:257–63. https://doi.org/10.1016/j.gaitpost.2020.02.007 . Mancini M, Weiss A, Herman T, Hausdorff JM. Turn around freezing: Community-living turning behavior in people with Parkinson’s disease. Front Neurol. 2018;9:18. https://doi.org/10.3389/fneur.2018.00018 . Del Din S, Elshehabi M, Galna B, Hobert MA, Warmerdam E, Suenkel U, et al. Gait analysis with wearables predicts conversion to Parkinson disease. Ann Neurol. 2019;86:357–67. https://doi.org/10.1002/ana.25548 . Lonini L, Dai A, Shawen N, Simuni T, Poon C, Shimanovich L, et al. Wearable sensors for Parkinson’s disease: Which data are worth collecting for training symptom detection models. npj Digit Med. 2018;1:64. https://doi.org/10.1038/s41746-018-0071-z . Adams JL, Dinesh K, Snyder CW, Xiong M, Tarolli CG, Sharma S, et al. A real-world study of wearable sensors in Parkinson’s disease. npj Parkinsons Dis. 2021;7:106. https://doi.org/10.1038/s41531-021-00248-w . Scherbaum R, Moewius A, Oppermann J, Geritz J, Hansen C, Gold R, et al. Parkinson’s disease multimodal complex treatment improves gait performance: An exploratory wearable digital device-supported study. J Neurol. 2022;269:6067–85. https://doi.org/10.1007/s00415-022-11257-x . Mahadevan N, Demanuele C, Zhang H, Volfson D, Ho B, Erb MK, et al. Development of digital biomarkers for resting tremor and bradykinesia using a wrist-worn wearable device. npj Digit Med. 2020;3:5. https://doi.org/10.1038/s41746-019-0217-7 . Lee JS, Kim HY, Hwang JY, Kwon S, Chung HR, Kwak TK et al. Development of nutrition quotient for Korean adults: Item selection and validation of factor structure. J Nutr Health. 2018:51:340–56. https://doi.org/10.4163/jnh.2018.51.4.340 Park H, Shin S, Youm C, Cheon SM, Lee M, Noh B. Classification of Parkinson’s disease with freezing of gait based on 360 turning analysis using 36 kinematic features. J Neuroeng Rehabil. 2021;18:1–18. https://doi.org/10.1186/s12984-021-00975-4 35 . Peng H, Long F, Ding C. Feature selection based on mutual information criteria of max-dependency, max-relevance, and min-redundancy. IEEE Trans Pattern Anal Mach Intell. 2005;27:1226–38. https://doi.org/10.1109/TPAMI.2005.159 . Debelle H, Packer E, Beales E, Bailey HGB, Mc Ardle R, Brown P, et al. Feasibility and usability of a digital health technology system to monitor mobility and assess medication adherence in mild-to-moderate Parkinson’s disease. Front Neurol. 2023;14:1111260. https://doi.org/10.3389/fneur.2023.1111260 . Del Din S, Kirk C, Yarnall AJ, Rochester L, Hausdorff JM. Body-worn sensors for remote monitoring of Parkinson’s disease motor symptoms: Vision, state of the art, and challenges ahead. J Parkinsons Dis. 2021;11:S35–47. https://doi.org/10.3233/JPD-202471 . Sundgren M, Andréasson M, Svenningsson P, Noori RM, Johansson A. Does information from the Parkinson KinetiGraph™ (PKG) influence the neurologist’s treatment decisions?—An observational study in routine clinical care of people with Parkinson’s disease. J Pers Med. 2021;11:519. https://doi.org/10.3390/jpm11060519 . Castiglia SF, Trabassi D, Conte C, Ranavolo A, Coppola G, Sebastianelli G, et al. Multiscale entropy algorithms to analyze complexity and variability of trunk accelerations time series in subjects with Parkinson’s disease. Sens (Basel). 2023;23:4983. https://doi.org/10.3390/s23104983 . Coates L, Shi J, Rochester L, Del Din S, Pantall A. Entropy of real-world gait in Parkinson’s disease determined from wearable sensors as a digital marker of altered ambulatory behavior. Sens (Basel). 2020;20:2631. https://doi.org/10.3390/s20092631 . Mancini M, Horak FB, Zampieri C, Carlson-Kuhta P, Nutt JG, Chiari L. Trunk accelerometry reveals postural instability in untreated Parkinson’s disease. Parkinsonism Relat Disord. 2011;17:557–62. https://doi.org/10.1016/j.parkreldis.2011.05.010 . World Health Organization. Towards a common language for functioning, disability, and health: ICF. Int Classif Functioning Disabil Health; 2002. Giannouli E, Bock O, Mellone S, Zijlstra W. Mobility in old age: Capacity is not performance. BioMed Res Int. 2016;2016:3261567. https://doi.org/10.1155/2016/3261567 . Guo Y, Yang J, Liu Y, Chen X, Yang GZ. Detection and assessment of Parkinson's disease based on gait analysis: A survey. Front Aging Neurosci. 2022;14:916971. https://doi.org/10.3389/fnagi.2022.916971 . Mammen JR, Speck RM, Stebbins GM, Müller MLTM, Yang PT, Campbell M, et al. Mapping relevance of digital measures to meaningful symptoms and impacts in early Parkinson’s disease. J Parkinsons Dis. 2023;13:589–607. https://doi.org/10.3233/JPD-225122 . Zhang J. Mining imaging and clinical data with machine learning approaches for the diagnosis and early detection of Parkinson’s disease. npj Parkinsons Dis. 2022;8:13. https://doi.org/10.1038/s41531-021-00266-8 . Wahid F, Begg RK, Hass CJ, Halgamuge S, Ackland DC. Classification of Parkinson’s disease gait using spatial-temporal gait features. IEEE J Biomed Health Inf. 2015;19:1794–802. https://doi.org/10.1109/JBHI.2015.2450232 . Port RJ, Rumsby M, Brown G, Harrison IF, Amjad A, Bale CJ. People with Parkinson’s disease: What symptoms do they most want to improve and how does this change with disease duration? J Parkinsons Dis. 2021;11:715–24. https://doi.org/10.3233/JPD-202346 . Elshehabi M, Maier KS, Hasmann SE, Nussbaum S, Herbst H, Heger T, et al. Limited effect of dopaminergic medication on straight walking and turning in early-to-moderate Parkinson’s disease during single and dual tasking. Front Aging Neurosci. 2016;8:4. https://doi.org/10.3389/fnagi.2016.00004 . Panyakaew P, Duangjino K, Kerddonfag A, Ploensin T, Piromsopa K, Kongkamol C, et al. Exploring the complex phenotypes of impaired finger dexterity in mild-to-moderate stage Parkinson’s disease: A time-series analysis. J Parkinsons Dis. 2023;13:975–88. https://doi.org/10.3233/JPD-230029 . Gamborg M, Hvid LG, Thrue C, Johansson S, Franzén E, Dalgas U, et al. Muscle strength and power in people with Parkinson disease: A systematic review and meta-analysis. J Neurol Phys Ther. 2023;47:3–15. https://doi.org/10.1097/NPT.0000000000000421 . Bailo G, Saibene FL, Bandini V, Arcuri P, Salvatore A, Meloni M, et al. Characterization of walking in mild Parkinson’s disease: Reliability, validity and discriminant ability of the six-minute walk test instrumented with a single inertial sensor. Sens (Basel). 2024;24:662. https://doi.org/10.3390/s24020662 . Egger M, Finsterhölzl M, Buetikofer A, Wippenbeck F, Müller F, Jahn K, et al. Balance function in critical illness survivors and evaluation of psychometric properties of the Mini-BESTest. Sci Rep. 2024;14:12089. https://doi.org/10.1038/s41598-024-61745-5 . Rehman RZU, Buckley C, Micó-Amigo ME, Kirk C, Dunne-Willows M, Mazzà C, et al. Accelerometry-based digital gait characteristics for classification of Parkinson’s disease: What counts? IEEE open J Eng Med Biol. 2020;1:65–73. https://doi.org/10.1109/OJEMB.2020.2966295 . Alam MN, Garg A, Munia TTK, Fazel-Rezai R, Tavakolian K. Vertical ground reaction force marker for Parkinson’s disease. PLoS ONE. 2017;12:e0175951. https://doi.org/10.1371/journal.pone.0175951 . Rehman RZU, Del Din S, Guan Y, Yarnall AJ, Shi JQ, Rochester L. Selecting clinically relevant gait characteristics for classification of early Parkinson’s disease: A comprehensive machine learning approach. Sci Rep. 2019;9:17269. https://doi.org/10.1038/s41598-019-53656-7 . Djaldetti R, Ziv I, Melamed E. The mystery of motor asymmetry in Parkinson’s disease. Lancet Neurol. 2006;5:796–802. https://doi.org/10.1016/S1474-4422(06)70549-X . Monje MHG, Sánchez-Ferro Á, Pineda-Pardo JA, Vela-Desojo L, Alonso-Frech F, Obeso JA. Motor onset topography and progression in Parkinson’s disease: The upper limb is first. Mov Disord. 2021;36:905–15. https://doi.org/10.1002/mds.28462 . Tables Table 1. Demographics and clinical characteristics of the study participants. Persons with PD ( n = 102) Individuals according to the PD severity subtype p -value adjusted p -value Mild ( n = 24; 23.5%) Moderate ( n = 47; 46.1%) Severe ( n = 31; 30.4%) 1 vs. 2 2 vs. 3 1 vs. 3 Sex (male/female) 48/54 16/8 19/28 13/18 – – – Age (years) 68.06 ± 7.20 64.83 ± 7.25 68.00 ± 5.98 70.65 ± 8.01 0.054 a 0.081 a 0.099 a 0.014 a 0.007 a 0.011 a Height (cm) 160.36 ± 8.43 164.07 ± 7.82 159.90 ± 6.10 158.20 ± 10.91 0.016 a 0.047 a 0.434 a 0.118 a 0.024 a 0.010 a Body weight (kg) 62.90 ± 10.64 64.81 ± 11.87 61.98 ± 9.62 62.81 ± 11.26 0.282 a 0.211 a 0.737 a 0.170 a 0.528 a 0.158 a BMI (kg/m 2 ) 24.37 ± 3.07 23.89 ± 2.76 24.17 ± 2.90 25.07 ± 3.50 0.698 a 0.161 a 0.239 a 0.080 a 0.168 a 0.056 a Disease duration (years) 5.67 ± 4.39 5.69 ± 3.58 5.71 ± 5.22 5.59 ± 3.65 0.618 b 0.181 b 0.567 b 0.142 b 0.805 b 0.186 b Treatment duration (years) 4.77 ± 4.25 4.72 ± 3.81 4.78 ± 5.10 4.81 ± 3.12 0.662 b 0.181 b 0.244 b 0.073 b 0.532 b 0.145 b L-dopa equivalent dose (mg/day) 545.46 ± 289.96 608.60 ± 405.14 490.82 ± 207.53 579.40 ± 286.55 0.191 a 0.191 a 0.144 a 0.054 a 0.766 a 0.192 a H&Y stage I 28 (27.5%) 12 (50.0%) 11 (23.4%) 5 (16.1%) – – – H&Y stage II 50 (49.0%) 8 (33.3%) 25 (53.2%) 17 (54.8%) – – – H&Y stage III 24 (23.5%) 4 (16.7%) 11 (23.4%) 9 (29.0%) – – – MDS-UPDRS Total (scores) 54.51 ± 23.17 47.83 ± 24.80 51.22 ± 20.53 64.68 ± 23.00 0.542 a 0.181 a 0.009 a 0.014 a 0.012 a 0.007 a MDS-UPDRS Part I (scores) 10.09 ± 5.22 8.50 ± 4.68 9.72 ± 6.26 11.87 ± 5.19 0.340 a 0.170 a 0.080 a 0.048 a 0.016 a 0.008 a MDS-UPDRS Part II (scores) 12.89 ± 7.22 10.83 ± 7.38 11.26 ± 5.90 16.97 ± 7.48 0.495 a 0.212 a 0.001 a 0.003 a 0.004 a 0.012 a MDS-UPDRS Part III (scores) 29.45 ± 15.23 26.25 ± 14.14 27.84 ± 14.68 34.35 ± 16.13 0.663 a 0.166 a 0.069 a 0.052 a 0.057 a 0.021 a MMSE (scores) 27.75 ± 2.10 28.46 ± 1.32 27.94 ± 1.77 26.90 ± 2.74 0.305 b 0.183 b 0.138 b 0.059 b 0.008 b 0.006 b MoCA (scores) 25.84 ± 2.98 26.79 ± 2.25 26.15 ± 2.92 24.65 ± 3.25 0.495 b 0.186 b 0.017 b 0.017 b 0.008 b 0.008 b The data are presented as mean ± standard deviation, with significant differences between groups (1, 2, and 3 indicate mild, moderate, and severe subtypes, respectively) indicated in bold (Uncorrected p and adjusted p < 0.05). False discovery rate (FDR) correction using the Benjamini–Hochberg method was applied to adjust for multiple comparisons. PD, Parkinson’s disease; BMI, body mass index; L-dopa, levodopa; H&Y, Hoehn and Yahr; MDS-UPDRS, movement disorder society-unified Parkinson’s disease rating scale; MMSE, mini-mental state examination; and MoCA, Montreal cognitive assessment. a Independent samples t -test result b Mann–Whitney U test result Table 2. Selected modalities and category domains by the MI algorithm for the relationship between all/single modalities and MDS-UPDRS parts. Modality Selected domains (Number of features) Feature selection score MDS-UPDRS scores All modalities UPDRS Total GP_Sensors (17) GP_Motion (5) PFL (10) Demographics (3) 2.74 1.70 1.22 0.13 UPDRS Part I GP_Sensors (14) GP_Motion (5) PFL (7) Demographics (3) 1.68 0.94 0.74 0.15 UPDRS Part II GP_Sensors (21) GP_Motion (4) PFL (7) Demographics (2) 1.97 0.87 0.69 0.07 UPDRS Part III GP_Sensors (21) GP_Motion (5) PFL (10) Demographics (3) 2.42 1.39 1.17 0.30 PFL modalities UPDRS Total (10) 1.11 UPDRS Part I (7) 0.54 UPDRS Part II (11) 0.73 UPDRS Part III (9) 0.81 GP_Motion modalities UPDRS Total (5) 1.59 UPDRS Part I (5) 0.69 UPDRS Part II (4) 0.95 UPDRS Part III (5) 1.01 GP_Sensors modalities UPDRS Total (16) 2.63 UPDRS Part I (14) 1.4 UPDRS Part II (22) 2.08 UPDRS Part III (20) 1.97 PFL, Physical function and lifestyle; GP_Motion, Gait parameters' principal components (PCs) in motion analysis system; GP_Sensors, Gait parameters' PCs in wearable sensors; and MDS-UPDRS, Movement disorder society-unified Parkinson’s disease rating scale. Table 3. ML regression models selected using cross-validation, grid search, and model evaluation based on multiple modalities. Modality Selected model MAE R 2 MDS-UPDRS scores All modalities UPDRS Total Random forest regressor 0.69 0.35 UPDRS Part I LASSO 0.79 0.13 UPDRS Part II LASSO 0.66 0.41 UPDRS Part III Random forest regressor 0.68 0.16 PFL modalities UPDRS Total Random forest regressor 0.54 0.51 UPDRS Part I LASSO 0.78 0.15 UPDRS Part II LASSO 0.54 0.55 UPDRS Part III Random forest regressor 0.65 0.25 GP_Motion modalities UPDRS Total Random forest regressor 0.94 0.04 UPDRS Part I LASSO 0.88 0.01 UPDRS Part II Random forest regressor 0.94 0.14 UPDRS Part III Random forest regressor 0.87 0.20 GP_Sensors modalities UPDRS Total Random forest regressor 0.81 0.18 UPDRS Part I Random forest regressor 0.87 0.01 UPDRS Part II Random forest regressor 0.70 0.34 UPDRS Part III Random forest regressor 0.76 0.02 The results indicate the average model performance on the test set. Model selection and training were performed using five-fold cross-validation to obtain MAE and the coefficient of determination R 2 as performance metrics. MAE measures the average absolute difference between actual and predicted values. R 2 indicates the model fitting to the variability of the dependent variable. ML, machine learning; MDS-UPDRS, movement disorder society-unified Parkinson’s disease rating scale; MAE, mean absolute error; PFL, physical function and lifestyle; and GP, gait parameter. Table 4. Performance of the PD severity subtype classification models for MDS-UPDRS scores based on multimodal data Severity subtypes Modality Digital biomarkers MDS-UPDRS scores Accuracy (%) AUC (Mean) 1 vs. 2 All modalities RELB_PC1 (Turning: Time-domain acceleration and gyroscope measurements, and max angular velocity jerk) RANK_PC3 (Forward walking and turning: Maximum and sample entropy acceleration and jerk) PSIS_PC1 (Turning: Time-domain acceleration and gyroscope measurements) LANK_PC1 (Forward walking and turning: RMS and mean acceleration and gyroscope) Turning _PC6 (Contralateral temporal coordination) RANK_PC1 (Forward & backward walking, and turning: Time-domain acceleration and gyroscope) UPDRS Part III 93.3 0.73 2 vs. 3 T10_PC2 (Forward walking and turning: Time-domain gyroscope) RELB_PC1 PSIS_PC6 (Forward walking: Time-domain gyroscope) PSIS_PC1 LELB_PC3 (Forward & backward walking: Time-domain acceleration, gyroscope, and angular velocity jerk) LELB_PC1 (Forward walking and turning: Gyroscope and angular velocity jerk) LANK_PC2 (Forward & backward walking, and turning: RMS and mean acceleration and gyroscope) IPAQ UPDRS Total 60.0 0.76 1 vs. 3 LANK_PC1 RANK_PC1 UPDRS Part III 100.0 0.99 1 vs. 2 PFL modalities Grip strength_L and R UPDRS Part III 66.7 0.74 2 vs. 3 6MWT UPDRS Total 73.0 0.65 1 vs. 3 Grip strength_R Mini-BEST SF-36 (total) IPAQ 5STST SF-36 (physical) UPDRS Total 75.0 0.74 1 vs. 2 GP_Motion modalities Backward walking_PC4 (Walking speed and stride length) Turning_PC6 Turning_PC2 ( Left and right double support phase) UPDRS Part III 73.3 0.67 2 vs. 3 Turning_PC8 (Left and right contralateral temporal coordination) UPDRS Part III 73.0 0.64 1 vs. 3 Turning_PC1 (Walking speed and stride length) Forward walking_PC3 (Walking speed, stride length, and double support phase) Backward walking_PC4 Turning_PC2 UPDRS Part II 83.0 0.72 1 vs. 2 GP_Sensors modalities LELB_PC4 (Turning: RMS and mean acceleration) RELB_PC1 LANK_PC1 RANK_PC3 PSIS_PC1 PSIS_PC2 (Forward walking and turning: Maximum and sample entropy acceleration and jerk) UPDRS Part III 86.7 0.72 2 vs. 3 LELB_PC1 LELB_PC2 (Forward walking and turning: Maximum acceleration, jerk, and angular velocity jerk) LELB_PC3 RELB_PC1 LANK_PC1 RANK_PC1 PSIS_PC1 T10_PC2 UPDRS Part II 80.0 0.83 1 vs. 3 RANK_PC1 LANK_PC1 PSIS_PC1 LELB_PC2 LELB_PC3 UPDRS Part II 100.0 0.91 Time-domain indicated the root mean square (RMS), mean, and max. PD, Parkinson's disease; MDS-UPDRS, movement disorder society-unified Parkinson’s disease rating scale; AUC, area under the curve; PFL, physical function and lifestyle; GP, gait parameter; PC, principal components; RELB, right lateral humeral epicondyle; RANK, right lateral malleolus; PSIS, posterior superior iliac spine; LANK, left lateral malleolus; T10, 10th thoracic vertebra; LELB, left lateral humeral epicondyle; IPAQ, international physical activity questionnaire; L, left; R, right; 6MWT, 6 min walking test; Mini-BEST, mini-balance evaluation systems test; SF-36, 36-item short form health survey; and 5STST, five times sit-to-stand test. Additional Declarations No competing interests reported. Supplementary Files SupplementaryMaterial1R1.docx SupplementaryMaterial2R1.docx SupplementaryMaterial3R1.docx SupplementaryMaterial4.pdf SupplementaryMaterial5.pdf Cite Share Download PDF Status: Published Journal Publication published 02 Jun, 2025 Read the published version in Journal of NeuroEngineering and Rehabilitation → Version 1 posted Editorial decision: Accepted 13 May, 2025 Reviews received at journal 06 May, 2025 Reviewers agreed at journal 18 Apr, 2025 Reviewers invited by journal 17 Apr, 2025 Submission checks completed at journal 10 Apr, 2025 First submitted to journal 09 Apr, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5523724","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":444724508,"identity":"d70ff352-fb84-4634-a1ed-943020f87442","order_by":0,"name":"Hwayoung Park","email":"","orcid":"","institution":"Dong-A University","correspondingAuthor":false,"prefix":"","firstName":"Hwayoung","middleName":"","lastName":"Park","suffix":""},{"id":444724509,"identity":"eb689cbe-4378-457f-9e69-7db820c9430b","order_by":1,"name":"Changhong Youm","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAt0lEQVRIiWNgGAWjYDACCTBpw9gA4SYQrSWNdC2HSdDCP7vH8DFvznnZ/hkJjB9+MKTlE7bkzhljY95tt41n3EhgluxhyLFsIKTFQCLHTBqoJbHhRgKDNANDhQFBW6BaziXOB9rymxQtBxI33EhgA9qSQ1iLxI20YsO525KNN5552GbZY5BGWAv/jOSND95us5Oddzz58I0fFcmEtTAwcMAUgaKGGA0MDOwPiFI2CkbBKBgFIxgAAMSZOXu0+WMMAAAAAElFTkSuQmCC","orcid":"","institution":"Dong-A University","correspondingAuthor":true,"prefix":"","firstName":"Changhong","middleName":"","lastName":"Youm","suffix":""},{"id":444724510,"identity":"a39e28a3-e136-4793-aac4-3fb27143e7f0","order_by":2,"name":"Sang-Myung Cheon","email":"","orcid":"","institution":"Dong-A University","correspondingAuthor":false,"prefix":"","firstName":"Sang-Myung","middleName":"","lastName":"Cheon","suffix":""},{"id":444724511,"identity":"89a14790-6ff0-4785-925e-d288f03ef152","order_by":3,"name":"Bohyun Kim","email":"","orcid":"","institution":"Dong-A University","correspondingAuthor":false,"prefix":"","firstName":"Bohyun","middleName":"","lastName":"Kim","suffix":""},{"id":444724512,"identity":"7b04291f-0189-4e1c-8302-b384af1c1e15","order_by":4,"name":"Hyejin Choi","email":"","orcid":"","institution":"Dong-A University","correspondingAuthor":false,"prefix":"","firstName":"Hyejin","middleName":"","lastName":"Choi","suffix":""},{"id":444724513,"identity":"31de6b93-c11e-4037-b346-991eb9143a04","order_by":5,"name":"Juseon Hwang","email":"","orcid":"","institution":"Dong-A University","correspondingAuthor":false,"prefix":"","firstName":"Juseon","middleName":"","lastName":"Hwang","suffix":""},{"id":444724514,"identity":"8967c369-4ec8-412d-8aea-a03cee7fece9","order_by":6,"name":"Minsoo Kim","email":"","orcid":"","institution":"Dong-A University","correspondingAuthor":false,"prefix":"","firstName":"Minsoo","middleName":"","lastName":"Kim","suffix":""}],"badges":[],"createdAt":"2024-11-26 02:23:15","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5523724/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5523724/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s12984-025-01648-2","type":"published","date":"2025-06-02T15:57:24+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":81036883,"identity":"cddb8549-d99c-4d03-9fb9-ddda80a4432b","added_by":"auto","created_at":"2025-04-21 12:32:18","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":278335,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eWorkflow to classify PD severity subtypes and develop digital biomarkers.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5523724/v1/f8eab8260e2bebd7971b01b6.jpg"},{"id":81036889,"identity":"b461cde7-4da5-4121-ab2c-b0db98326890","added_by":"auto","created_at":"2025-04-21 12:32:18","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":110461,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eHeatmap of feature importance based on multimodal data with demographics, PFL, and gait parameters. \u003c/strong\u003eThe relative importance score of each feature was determined using feature importances from the random forest or LASSO model. The random forest attribute “feature_importances_” enabled determining the contribution of each feature toward reducing the impurity of nodes in the forest. Higher values indicated more important features. The LASSO attribute “coef_” represented the coefficients of the linear model, where larger coefficients (absolute values) indicated more important features. PFL, physical function and lifestyle; GP_Motion, gait parameters' principal components (PCs) in the motion analysis system; GP_Sensors, gait parameters' PCs in wearable sensors; and UPDRS, unified Parkinson’s disease rating scale.\u003c/p\u003e","description":"","filename":"Picture2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5523724/v1/4da6bafc551659a3113eeb84.jpg"},{"id":81036887,"identity":"5015683f-d8c7-42c8-a698-8da0875ff048","added_by":"auto","created_at":"2025-04-21 12:32:18","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":400612,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eIdentified digital biomarkers associated with MDS-UPDRS scores for distinguishing three PD severity subtypes. \u003c/strong\u003eRepresentative variables were selected based on logistic regression results. The AUC for each ROC curve was calculated, and the mean and standard deviation of the AUCs were obtained.\u003c/p\u003e","description":"","filename":"Picture3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5523724/v1/112e2c8e66691e0551997e38.jpg"},{"id":84242715,"identity":"931266d3-7bef-42f0-8001-5b16f2c38b1e","added_by":"auto","created_at":"2025-06-09 16:11:47","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2803683,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5523724/v1/0276df68-1987-4b98-a998-efce281cae07.pdf"},{"id":81037951,"identity":"9a14bf17-e5a7-4cc5-9d82-9c4ef7e6491d","added_by":"auto","created_at":"2025-04-21 12:48:18","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":40049,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterial1R1.docx","url":"https://assets-eu.researchsquare.com/files/rs-5523724/v1/fe48752242bf718a888f3b23.docx"},{"id":81037577,"identity":"06b5ab21-c909-46ee-b581-dd7c9ce33085","added_by":"auto","created_at":"2025-04-21 12:40:18","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":43174,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterial2R1.docx","url":"https://assets-eu.researchsquare.com/files/rs-5523724/v1/24491788666cf517672717d4.docx"},{"id":81037583,"identity":"48b5321a-6635-4e29-a2be-fb4cd408b334","added_by":"auto","created_at":"2025-04-21 12:40:18","extension":"docx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":86347,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterial3R1.docx","url":"https://assets-eu.researchsquare.com/files/rs-5523724/v1/e55d5a5d95af41c7c3892905.docx"},{"id":81037952,"identity":"3546919e-be95-4e15-b9c7-2ebcd10bfdec","added_by":"auto","created_at":"2025-04-21 12:48:18","extension":"pdf","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":75462,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterial4.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5523724/v1/8e8dfa0accf8cd2789da9a06.pdf"},{"id":81037953,"identity":"8de5a9b7-fdda-49ad-907a-2920e9a49d56","added_by":"auto","created_at":"2025-04-21 12:48:18","extension":"pdf","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":950963,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterial5.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5523724/v1/ac162ba3650eefcf04d2187a.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Using machine learning to identify Parkinson’s disease severity subtypes with multimodal data","fulltext":[{"header":"Background","content":"\u003cp\u003eClassifying Parkinson's disease (PD) into subtypes and predicting disease progression based on severity are essential tasks for understanding the diverse symptoms experienced by patients and providing tailored intervention strategies [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Traditional classification methods rely on specific motor symptoms, such as tremors and postural instability [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e], which hinder the evaluation of the comprehensive range of motor and non-motor PD symptoms. Previous studies employed clinical observations based on age at onset or categorization based on most notable features [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e] to define PD subtypes based on disease severity levels (e.g., mild, moderate, and severe); inching, moderate, and rapid pace subtypes with PD progression rate; early and late onset; motor and non-motor predominance; and presence of dementia [\u003cspan additionalcitationids=\"CR6\" citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Despite being intuitive, these classifications do not represent the clinical features of PD, which are quantifiable, complex, and interrelated. Recently, data-driven multidimensional approaches have shown promise in overcoming existing limitations [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. These data-driven multidimensional approaches can accurately identify and predict PD subtypes by integrating various data sources and using clustering and machine learning (ML) algorithms [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Hence, they can contribute to early diagnosis, track disease progression, develop personalized treatment strategies, and improve clinical trial efficiency [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe use of digital technologies in healthcare, which includes dedicated wearable devices, artificial intelligence-driven analytical tools, and mobile health applications, has recently seen substantial growth. These smart tools continuously generate objective and rich data, providing valuable insights into disease symptoms and PD management [\u003cspan additionalcitationids=\"CR12\" citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Motor impairment in PD is primarily assessed using the movement disorder society (MDS) unified PD rating scale (UPDRS) [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]; however, this assessment requires regular clinical visits, which can be time-consuming, costly, and limited by the availability of specialized neurologists and mobility restrictions [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Wearable sensors address these issues by offering portability, cost-effectiveness, and the ability to assess spatiotemporal characteristics of gait and balance in laboratories, clinics, and homes [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. Obtaining clinically meaningful information from sensor data requires employing analytical techniques such as initial feature selection and reduction [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAlthough widely used for evaluating motor impairments, gait analysis suffers from subjective assessments and inconsistencies in the gait feature analysis [\u003cspan additionalcitationids=\"CR21\" citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Further, wearable sensor data can be unstructured and noisy [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Therefore, multimodal data may be required to analyze the combined motor and non-motor symptoms of PD [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Previous studies on PD subtype classification using only clinical data reported reproducibility issues, underscoring the need to identify subtypes based on objective multimodal data [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. Integrating ML algorithms with multimodal data can complement existing clinical assessment scales and provide an improved classification of PD severity subtypes [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eML algorithms can be trained using time-series signals and features collected from wearable sensors during clinical evaluations with relevant scales serving as labels [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. ML analysis on inertial measurement unit (IMU) data has been used for identifying persons with PD from healthy persons or other Parkinsonian disorders and in detecting the signs of bradykinesia or tremors [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan additionalcitationids=\"CR29 CR30\" citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. Therefore, combining multimodal data with ML algorithms can aid in disease diagnosis, progression tracking, and treatment strategy development [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Further, this approach can potentially contribute to the understanding of the complex mechanisms of PD and the identification of new digital biomarkers [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eDigital biomarkers are being increasingly used; however, they present certain limitations. Although various digital devices can monitor physical activity, physiological signals, speech, sleep patterns, and other variables [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e], data collected from individual wearable devices provide limited information for the classification and prediction of diverse subtypes based on PD severity [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. Further, most studies focused on evaluating motor symptoms during walking or rest using restricted datasets, which makes it challenging to clinically validate high-dimensional features extracted from digital devices [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. Although models for classifying and predicting PD subtypes based on derived digital biomarkers are currently limited in performance and may be used for initial screening in clinical settings, they are unsuitable for direct clinical applications [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Therefore, digital biomarkers derived from objective, standardized multimodal data using ML are necessary to ensure clinical efficiency and feasibility, which enables the continuous development of models for tracking and predicting PD.\u003c/p\u003e \u003cp\u003eIn this study, our goal was addressing the clinical applicability and heterogeneity of PD by classifying PD severity subtypes and developing digital biomarkers for an objective diagnosis. We hypothesized that objectively collected multimodal data, along with existing clinical characteristics, can facilitate the classification of PD severity subtypes and be associated with MDS-UPDRS scores, which is a measure of PD severity. Confirming this hypothesis would help us establish potential classification markers, which can aid in developing representative digital biomarkers for distinguishing PD severity subtypes.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003eWe established a data-driven framework combining objective multimodal data with ML and statistical approaches (Fig. 1). This framework included integrating data from demographics, clinical characteristics, physical function and lifestyle (PFL), and principal components (PCs) of gait parameters in motion analysis systems (GP_Motion) and wearable sensors (GP_Sensors).\u003c/p\u003e\n\u003cp\u003e1) We characterized and preprocessed multimodal data of diverse types collected to identify subtypes from persons with PD.\u003c/p\u003e\n\u003cp\u003e2) We investigated associations between multimodal data collected from identified PD severity subtypes and existing clinical characteristics. We identified potential modalities with a high feature importance specific to each PD severity subtype with ML approaches.\u003c/p\u003e\n\u003cp\u003e3) Based on the identified modalities, we established potential classification markers for PD severity subtypes, which led to the development of representative digital biomarkers.\u003c/p\u003e\n\u003cdiv id=\"Sec3\"\u003e\n \u003ch2\u003eParticipants\u003c/h2\u003e\n \u003cp\u003eWe used data from 102 persons diagnosed with PD. The participants were assessed by a movement disorder specialist using the MDS-UPDRS criteria, which evaluate the motor and non-motor symptoms of the participants via clinician-rated and patient-reported measures. Participants were included if they showed mild-to-moderate idiopathic PD, received anti-Parkinsonian medications, and could walk and stand unassisted during clinical tests. However, they were excluded if they had other neurological, orthopedic, or psychiatric disorders. The average age of persons with PD (\u003cem\u003en\u003c/em\u003e = 102) was 68 years, and 52.9% of them were female. In addition, 49.0% of the patients had Hoehn and Yahr (H\u0026amp;Y) stage 2 disease, an average disease duration of 5.7 years, and a motor score (MDS-UPDRS Part III) of 29.5 (Table 1).\u003c/p\u003e\n\u003c/div\u003e\n\u003cp\u003e[Table\u0026nbsp;]\u003c/p\u003e\n\u003cp\u003eThis study was approved by the Institutional Review Board of the Dong-A University Medical Center (Approval number DAUHIRB-22-089). All participants were informed about the study aims and protocols and signed a written informed consent form before enrollment. The study was conducted according to relevant guidelines and regulations. Our study was registered with the Clinical Research Information Service of the Republic of Korea (KCT0009353).\u003c/p\u003e\n\u003ch3\u003eExperimental procedures\u003c/h3\u003e\n\u003cp\u003eEach participant visited our laboratory twice during the study period. The participants underwent a multimodal assessment of clinical characteristics, physical function tests, lifestyle questionnaires, and gait measurements using a motion analysis system and wearable sensors. Patients using levodopa were assessed in the “on” medication state after 2–3 h of medication intake.\u003c/p\u003e\n\u003cp\u003eWe considered data modalities described below. Detailed descriptions of these measures are provided in Supplementary Material 2 (Table 1).\u003c/p\u003e\n\u003cp\u003eClinical measurements (12 modalities): Clinical assessments were conducted in collaboration with neurologists and nurses using structured questionnaires and patient observation. Persons with PD were assessed using the H\u0026amp;Y stage and MDS-UPDRS, including the Total and Part I–III scores. MDS-UPDRS Total assesses the severity scales of motor and non-motor PD symptoms; Part I evaluates non-motor experiences of daily living, such as cognitive impairment and mood disorders; Part II assesses the motor experiences of daily living, including speech, handwriting, and hygiene; and Part III measures motor disorder severity, which focuses on motor symptoms such as tremors, rigidity, and bradykinesia. Further, mini-mental state examination (MMSE) and Montreal cognitive assessment (MoCA) were applied for evaluating the global cognitive function. Higher scores on the MDS-UPDRS indicate greater symptom severity, whereas higher scores on the MMSE and MoCA reflect better cognitive function.\u003c/p\u003e\n\u003cp\u003ePFL measurements (11 modalities): The physical function was assessed using grip strength (left and right), five times sit-to-stand test (5 STST), six min walking test (6 MWT), short physical performance battery (SPPB), and mini-balance evaluation systems test (Mini-BEST). Lifestyle scores included the nutrition quotient (NQ), 36-item short form health survey (SF-36), and international physical activity questionnaire (IPAQ) scores. Physical function assessments were conducted using standardized protocols by trained evaluators who guided participants through the tests and assigned scores based on their performance levels. NQ is a validated dietary assessment tool developed by the Korean Ministry of Food and Drug Safety. NQ comprehensively evaluates the nutritional status and dietary quality of adults [35]. SF-36 was used for assessing overall health-related quality of life, which includes total, physical, and mental scores. We clarify that SF-36 was used in this study, not SF-12 (a shorter version with 12 items), to avoid any potential confusion. Participants completed the study under the supervision of trained examiners to ensure consistency and accuracy.\u003c/p\u003e\n\u003cp\u003eGP_Motion (38 modalities; 8 PCs) and GP_Sensors (360 modalities; 35 PCs): Participants performed the following gait tests: 1) forward and 2) backward straight walking at a preferred speed and 3) 360° turning at a preferred and fast speed and turning to the left and right sides. For all tasks, kinematic data were captured using a full-body marker-based motion capture system (Plug-in Gait model, Vicon Motion Systems, Oxford Metrics, UK) and an array of six IMUs (Xsens DOT, Movella Technologies, Enschede, Netherlands). The raw data were collected at 100 Hz from the Vicon system and at 60 Hz from the Xsens DOT sensors. The IMU sensors were attached to six anatomical landmarks: left lateral humeral epicondyle (LELB), right lateral humeral epicondyle (RELB), left lateral malleolus (LANK), right lateral malleolus (RANK), 10th thoracic vertebra (T10), and posterior superior iliac spine (PSIS). A sensor placement was designed for capturing whole-body movement patterns and gait dynamics, which provides both spatial and temporal data from multiple body segments. The IMU sensors were calibrated before each trial according to the manufacturer’s guidelines to ensure the reliability of collected data. The data consistency was ensured by conducting trials under controlled laboratory conditions, which minimizes the effect of external factors.\u003c/p\u003e\n\u003cp\u003eFor GP_Motion variables, spatiotemporal gait parameters such as walking speed, stride length, double support phase, and contralateral temporal coordination [36] were extracted from 3D marker data for analysis. These parameters are clinically recognized as important indicators to assess gait dysfunction, which includes gait instability and asymmetry in individuals with PD [36]. GP_Sensors included IMU-derived features such as maximum jerk, angular velocity jerk, mean and maximum acceleration, RMS acceleration and gyroscope measurements, and sample entropy of acceleration and gyroscope data. These extracted features from both motion capture and IMU sensors provided a multimodal dataset to analyze gait abnormalities in PD, which enable a more comprehensive characterization of movement patterns beyond conventional joint kinematic analyses.\u003c/p\u003e\n\u003cdiv id=\"Sec6\"\u003e\n \u003ch2\u003eData analysis\u003c/h2\u003e\n \u003cp\u003eData analysis involved preprocessing collected multimodal data, clustering, feature selection, model training and selection, and classification based on logistic regression for identifying digital biomarkers. Figure 1 illustrates this process, and the steps are detailed below.\u003c/p\u003e\n\u003c/div\u003e\n\u003ch3\u003ePreprocessing\u003c/h3\u003e\n\u003cp\u003eThe collected data were initially inspected to examine their structure. Preprocessing involved two key tasks. First, \u003cem\u003ek\u003c/em\u003e-nearest neighbor imputation was applied to handle missing values. An imputer with five neighbors was initialized and applied to the dataset to fill the missing values. This technique enabled estimating the missing values based on the mean across the nearest neighbors, thereby ensuring that imputed values were informed by similar data points. Second, the data were normalized using \u003cem\u003ez\u003c/em\u003e scores to standardize features, which involved scaling the data to have a mean of 0 and standard deviation of 1.\u003c/p\u003e\n\u003cdiv id=\"Sec8\"\u003e\n \u003ch2\u003eClustering\u003c/h2\u003e\n \u003cp\u003eWe adopted a multimodal clustering approach. \u003cem\u003ek\u003c/em\u003e-means clustering was applied to categorize multimodal data into three clusters. The clustering model was initialized with three clusters (n_clusters = 3) and a random state of 42 for ensuring reproducibility. Subsequently, the model was fitted to the scaled data. Each data point was assigned a cluster label of 1, 2, or 3, which indicates its membership to one of the three clusters.\u003c/p\u003e\n\u003c/div\u003e\n\u003ch3\u003eFeature selection\u003c/h3\u003e\n\u003cp\u003eMutual information (MI) was employed for determining multimodal measures most reflective of PD severity, as assessed using MDS-UPDRS [11]. The corresponding nonparametric supervised method was used for estimating the relationship between the MDS-UPDRS parts and multimodal data. MI provided scores that quantify mutual dependencies and amount of shared information between each feature and MDS-UPDRS scores. In our study, MI was calculated using a natural logarithm base (ln). If log base 2 (log\u003csub\u003e2\u003c/sub\u003e) were used, MI scores would range between 0 and 1. All MI values obtained in our analysis remained within the range of 0 to 1, which indicates that our computed MI scores did not exceed 1 in this study although MI is theoretically unbounded depending on the logarithm base. A higher MI score indicates a stronger dependency between the selected feature and PD severity.\u003c/p\u003e\n\u003cp\u003eThe calculation process was repeated 100 times (100-fold validation) with a random partition seed per MDS-UPDRS component to ensure stable and consistent feature rankings. This approach was specifically selected for feature selection instead of using the traditional model evaluation because it helps mitigate random partitioning effects and enhances the robustness of feature importance estimation. In each iteration, 80% of the data were randomly selected for training, while the remaining 20% were used for validation, ensuring different subsets contributed to feature selection. MI scores between each feature and MDS-UPDRS scores were computed in every iteration. The final feature selection score was determined by counting the number of times each feature appeared among top-ranked features and multiplying this count by its average MI score across 100 folds (see Supplementary Material 3). This method ensured that feature selection using MI scores reliably identified multimodal data most relevant to PD severity, which reduces the effect of random sampling effects.\u003c/p\u003e\n\u003ch3\u003eModel training and selection\u003c/h3\u003e\n\u003cp\u003eVarious ML models were evaluated for determining the association between the selected modalities and MDS-UPDRS scores. The data were split into training (80%) and test (20%) sets, which ensures that model evaluation was performed on unseen data. The evaluated models included the random forest regressor and least absolute shrinkage and selection operator (LASSO).\u003c/p\u003e\n\u003cp\u003eModel training was performed using five-fold cross-validation with Pearson’s \u003cem\u003eR\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e as the performance metric. Hyperparameter tuning was performed using the grid search method to identify optimal parameters per model (see Supplementary Material 1). We employed a nested five-fold cross-validation approach to optimize hyperparameters and assess model selection stability. This nested cross-validation framework effectively prevents overfitting and provides a more stable estimate of model generalization performance compared to that of a simple three-way data split.\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\n \u003cp\u003eInner loop (hyperparameter tuning): Five-fold cross-validation was conducted within the training set (80%) for identifying optimal hyperparameters using a grid search method. This ensured that hyperparameter tuning was performed solely within the training set, thereby preventing any data leakage into the final test set.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eOuter loop (model selection and evaluation): Five-fold nested cross-validation was applied for validating the model selection process, assessing its stability, and minimizing overfitting risks.\u003c/p\u003e\n \u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003eThe final model was trained on the training set and evaluated on the test set for ensuring that it was blinded to the test data during training. Mean absolute error (MAE) and Pearson’s \u003cem\u003eR\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e were calculated to assess the correlation between the predicted and actual MDS-UPDRS scores.\u003c/p\u003e\n\u003cp\u003eIn models such as the random forest regressor and LASSO, intrinsic feature importance attributes were explored for identifying modalities that most significantly influenced model predictions. For the random forest regressor, feature importance was derived using the Gini importance (mean decrease in impurity). Importance scores were computed for each feature and analyzed for understanding its contribution to prediction. For LASSO, feature importance was assessed based on the magnitudes of coefficients assigned to each feature by the model. Features with larger absolute coefficients were considered more important. The feature importance scores provided insights into modalities that were the most predictive of the MDS-UPDRS scores.\u003c/p\u003e\n\u003cdiv id=\"Sec11\"\u003e\n \u003ch2\u003eIdentification of digital biomarkers based on logistic regression\u003c/h2\u003e\n \u003cp\u003eWe employed recursive feature elimination with logistic regression for identifying the most relevant digital biomarkers to classify clustering labels. Therefore, we selected features by recursively considering smaller sets, starting with an external estimator that assigned weights to features. Initially, the estimator was trained on the complete feature set, and the importance of each feature was determined either through a coef_ or feature_importances_ attribute. Subsequently, the least important features were pruned from the current set. This process was repeated recursively on the pruned set until the desired number of features was reached, which were the top 40 modalities in this study. These selected modalities served as potential digital biomarkers and were used in subsequent analyses for reducing data dimensionality while retaining representative predictors.\u003c/p\u003e\n \u003cp\u003eA logistic regression model was trained using selected digital biomarkers to predict clustering labels. The data were split into training and test sets containing 80% and 20% of the samples, respectively, using a random state of 2 to ensure reproducibility. The logistic regression model was trained and evaluated on the training and test sets, respectively. Performance metrics such as the receiver operating characteristic (ROC) curve and area under the curve (AUC) were computed. The ROC curve was plotted to visualize the trade-off between sensitivity (true positive rate) and specificity (1 − false positive rate). ROC-AUC was computed for providing the average and standard deviation of each measure of model performance.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec12\"\u003e\n \u003ch2\u003eStatistical analysis\u003c/h2\u003e\n \u003cp\u003eData normality was assessed using the Shapiro–Wilk test. An independent \u003cem\u003et\u003c/em\u003e-test or nonparametric statistics was used for analyzing the mean and standard deviation of the physical and clinical characteristics of all participants. Based on the normality results, appropriate statistical tests were applied for pairwise comparisons between clusters. Independent \u003cem\u003et\u003c/em\u003e-tests were used when data were distributed normally. Mann–Whitney U tests were applied when data did not meet normality assumptions. Multiple comparisons were conducted across overlapping clusters (1 vs. 2, 2 vs. 3, and 1 vs. 3), and therefore, we applied false discovery rate (FDR) correction (Benjamini–Hochberg method) to control for type I errors while maintaining statistical power. Both uncorrected and FDR-adjusted p-values are reported in Table 1.\u003c/p\u003e\n \u003cp\u003eAll statistical analyses were performed using SPSS version 22.0 (SPSS, Chicago, IL). The statistical significance level was set to 0.05. Data preprocessing and analysis were conducted using Python (version 3.10) with libraries including Pandas, NumPy, Scikit-learn, and Matplotlib.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\n \u003ch2\u003eIdentification of PD severity subtypes using clustering based on multimodal data\u003c/h2\u003e\n \u003cp\u003ePersons with PD were assigned to three subtypes, namely, clusters 1 (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;24), 2 (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;47), and 3 (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;31). We observed statistically significant differences in clinical characteristics after applying FDR correction to account for multiple comparisons. MDS-UPDRS Total, Part I, and Part II showed significant differences between clusters 2 and 3 using independent \u003cem\u003et\u003c/em\u003e-tests, whereas MoCA exhibited significant differences using the Mann\u0026ndash;Whitney U test. Similarly, significant differences were found between clusters 1 and 3 in MDS-UPDRS Total, Part I, Part II, and Part III using independent \u003cem\u003et\u003c/em\u003e-tests, whereas MMSE and MoCA exhibited significant differences using the Mann\u0026ndash;Whitney U test. We defined clusters 1 to 3 as mild, moderate, and severe subtypes with PD, respectively (Table \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003eThe PD severe subtype showed higher ages and lowest cognitive measures in tests such as MMSE and MoCA compared to those of the other subtypes. In addition, this subtype showed the most severe clinical characteristics, including the H\u0026amp;Y stage, MDS-UPDRS Total, Parts I, II, and III scores. This subtype also demonstrated the lowest scores for physical function measures such as grip strength, 5 STST, SPPB, and Mini-BEST, as well as lifestyle scores, which include the NQ, SF-36 (in total, physical, and mental scores), and IPAQ. Furthermore, they exhibited the lowest straight and turning gait speeds with shorter steps. The related information is summarized in Supplementary Material 2.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\n \u003ch2\u003eFeature selection associated with MDS-UPDRS scores\u003c/h2\u003e\n \u003cp\u003eWe used the MI for the nonparametric supervised estimation of the relationship between the MDS-UPDRS parts and data modalities to identify those most reflective of PD severity. Further, MI measures the dependency or shared information between two variables. A higher MI score indicates a stronger relationship between the feature and the target measure, which implies that the feature is more informative or predictive about the target [\u003cspan class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e37\u003c/span\u003e]. We determined modalities with the highest association with the scores of MDS-UPDRS Total and Parts I\u0026ndash;III (see Supplementary Material 3) and obtained the most prominent domains using the feature selection score (Table \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e). The modalities were classified into four domains: 1) demographics (4 features), 2) PFL scores (11 features), 3) GP_Motion (8 features), and 4) GP_Sensors (35 features). The most associated UPDRS parts in both all/single modalities were UPDRS Total scores, and the most important domain was GP_Sensors.\u003c/p\u003e\n \u003cp\u003e[Table \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e]\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\n \u003cdiv id=\"Sec17\" class=\"Section3\"\u003e\n \u003ch2\u003eML regression models for estimating PD severity\u003c/h2\u003e\n \u003cp\u003eThe most frequently selected and accurate models using five-fold cross-validation were identified for determining the importance of features most associated with MDS-UPDRS scores (Table 3). The selected model on all modalities was the LASSO for MDS-UPDRS Part II, with an average Pearson\u0026rsquo;s \u003cem\u003eR\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e of 0.41 and MAE of 0.66, which implies the average difference between the predicted and actual MDS-UPDRS scores on the test set. The selected model on single modalities was LASSO for PFL modalities for MDS-UPDRS Part II, with an average Pearson\u0026rsquo;s \u003cem\u003eR\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e of 0.55 and MAE of 0.54. For GP_Motion modalities, the MDS-UPDRS Part III was suitably predicted using random forest regressor, with an average Pearson\u0026rsquo;s \u003cem\u003eR\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e of 0.20 and MAE of 0.87. For GP_Sensors modalities, MDS-UPDRS Part II was suitably predicted using a random forest regressor, with an average Pearson\u0026rsquo;s \u003cem\u003eR\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e of 0.34 and MAE of 0.70.\u003c/p\u003e[Table 3]\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec18\" class=\"Section2\"\u003e\n \u003cp\u003eThe dominant modalities of top features were identified for exploring the feature importance of the selected model (random forest regressor or LASSO). The top three features of all modalities, including SF-36 (total), Mini-BEST, and 10th thoracic vertebra (T10)_PC2 (time-domain gyroscope) measurements, were important features to estimate the disease severity for MDS-UPDRS Part II. Single PFL modalities, which include SF-36 (total), Mini-BSET, and SF-36 (physical), were important for the MDS-UPDRS Part II. For GP_Motion modalities, Turning_PC6 (contralateral temporal coordination), Backward walking_PC7 (left and right double support phase), and Turning_PC2 (left and right double support phase) in MDS-UPDRS Part III were considered important. For GP_Sensors modalities, important features were T10_PC2 (time-domain gyroscope measurements), right lateral malleolus (RANK)_PC1 (time-domain acceleration and gyroscope measurements), and right lateral humeral epicondyle (RELB)_PC1 (time-domain acceleration and gyroscope measurements) in MDS-UPDRS Part II (Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e). Further details on the importance of features contributing to the accuracy of these models and feature descriptions are presented in Supplementary Material 4.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec19\" class=\"Section2\"\u003e\n \u003ch2\u003eIdentification of digital biomarkers based on classification models\u003c/h2\u003e\n \u003cp\u003eFollowing the classification of persons with PD according to the three severity subtypes based on multimodal data clustering, we developed PD severity subtypes classification models to identify digital biomarkers for estimating disease severity classes based on MDS-UPDRS scores. The models used all modalities, separate PFL modalities, separate GP_Motion modalities, and separate GP_Sensors modalities. Figure \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e and Table \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e show the ROC curves illustrating the performances of the PD severity subtype classification models for MDS-UPDRS scores using unsupervised clustering based on multimodal data.[Table \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e]\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec20\" class=\"Section2\"\u003e\u003cbr\u003e\u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eWe processed the multimodal data collected from persons with PD to classify PD severity subtypes and identified digital biomarkers associated with the widely used PD severity assessment tools, MDS-UPDRS parts, using ML algorithms. Our analysis provided the following insights:\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eMultimodal data, including demographics, PFL, and PCs of GP_Motion and GP_Sensors, enabled clustering persons with PD into three PD severity subtypes (mild, moderate, and severe).\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eOur multimodal data could be associated with MDS-UPDRS Total scores and the most important domain was found to be the PCs of GP_Sensors.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eThe LASSO model, based on all and PFL modalities, achieved the highest feature importance in MDS-UPDRS Part II. The model yielded average Pearson\u0026rsquo;s \u003cem\u003eR\u003c/em\u003e\u0026sup2; values of 0.41 for all modalities and 0.55 for PFL modalities, with MAE values of 0.66 and 0.54, respectively. The random forest regressor model based on GP_Motion and GP_Sensors modalities achieved the highest feature importance in MDS-UPDRS Parts III and II, respectively. The average Pearson\u0026rsquo;s \u003cem\u003eR\u003c/em\u003e\u0026sup2; values were 0.20 for GP_Motion and 0.34 for GP_Sensors, with MAE values of 0.87 and 0.70, respectively.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eDigital biomarkers were derived from high-importance multimodal data to develop classification models for PD severity subtypes. A model with LANK_PC1 (forward walking and turning) and RANK_PC1 (forward and backward walking and turning) among all modalities (100.0%, AUC: 0.99) accurately distinguished clinically severe subtypes from mild subtypes with PD. In addition, the PFL, GP_Motion, and GP_Sensors modalities accurately distinguished severe subtypes from mild subtypes with PD with accuracies of 0.75, 0.83, and 1.00, and AUCs of 0.74, 0.72, and 0.91, respectively.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003eIdentification of PD severity subtypes using clustering based on multimodal data\u003c/h2\u003e \u003cp\u003eIn this study, persons with PD were divided into three PD severity subtypes using clustering. The key clinical characteristics showed statistically significant differences between PD severity subtypes in terms of the comprehensive disease severity score (MDS-UPDRS Total), motor experiences of daily living (MDS-UPDRS Part II), and cognitive measures (MoCA). Patients in the severe subtype, which included persons with PD with the most severe symptoms and lowest cognitive measures, had the lowest PFL scores (see Supplementary Material 1). This classification aligns with clinical scales such as MDS-UPDRS scores and provides additional quantitative insights into motor and non-motor dysfunction in persons with PD, which can be used for guiding personalized rehabilitation strategies [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. For example, mild persons with PD exhibit relatively preserved motor function, which suggests that early exercise interventions and proactive rehabilitation can help slow disease progression [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. Moderate persons with PD experience gait and balance impairments, emphasizing the need for targeted gait training and fall prevention programs [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. Severe persons with PD present significant motor dysfunction and postural instability, indicating that assistive devices or intensive physiotherapy may be necessary for maintaining functional mobility [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Unlike conventional H\u0026amp;Y stages, which rely on clinical observation, our clustering model incorporates multimodal features such as sensor-derived, kinematic, PFL, and clinical data to provide a more data-driven stratification of PD severity. This finding highlights the heterogeneity of PD and underscores the need for personalized interventions based on severity subtype classification [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e].\u003c/p\u003e \u003cdiv id=\"Sec23\" class=\"Section3\"\u003e \u003ch2\u003eFeature selection associated with MDS-UPDRS scores\u003c/h2\u003e \u003cp\u003eWe used the MI algorithm for identifying multimodal data that can reflect PD severity. Thus far, numerous studies validated and demonstrated the benefits of using objective and highly correlated features for PD severity classification and disease progression monitoring [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e], indicating their potential as clinical support tools [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]. Similarly, our findings reveal associations between multimodal data and MDS-UPDRS parts, with particularly high Total scores. Therefore, multimodal datasets can be included for comprehensively evaluating PD signs and symptom severity, which highlights the value of analyzing multimodal data for comprehensively understanding PD severity [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Interestingly, GP_sensors outperformed GP_motion in PD severity classification, which can be attributed to the ability of IMU sensors to capture continuous movement dynamics, acceleration-based features, and entropy-based measures, which are highly relevant to PD-related gait abnormalities [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]. Further, IMU-derived features provide richer information on postural instability and movement variability, which are key factors in PD motor dysfunction [\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]. Moreover, these sensor-based features exhibited the highest associations with multimodal data and were significantly correlated with MDS-UPDRS Total and Part II scores, reflecting their relevance to motor experiences of daily living in persons with PD [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eWearable sensor-based measurements offer a distinct advantage over laboratory-based assessments because they can capture real-world behavioral contexts and fluctuations in motor symptoms [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Unlike clinical gait tests, which are controlled by external instructions, wearable devices enable long-term monitoring in naturalistic settings, thereby enabling data collection across different behavioral contexts, such as motor fluctuations and on/off medication states [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e]. This ability to continuously track motor function over time is valuable in PD management, where symptom variability is a key challenge [\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e]. Therefore, validating sensor-based measurements must involve not only clinical assessments but also real-world functional outcomes, considering temporal changes and individualized therapeutic responses [\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e]. Future studies should explore how combining IMU and motion capture features could enhance classification performance and provide a more comprehensive understanding of PD gait pathology [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e, \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec24\" class=\"Section2\"\u003e \u003ch2\u003eML regression models for estimating PD severity\u003c/h2\u003e \u003cp\u003eWe used ML regression models and five-fold cross-validation to estimate PD severity. The LASSO and random forest regressor models were used based on all and single modalities to identify the importance of various features for predicting MDS-UPDRS parts. ML has been widely applied to learn patterns from diverse data and estimate PD severity [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e]. Random forest regressors can account for collinearity in high-dimensional datasets and achieve an adequate discriminative performance [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e]. Therefore, the most relevant features could be identified without excessive influence during feature selection and importance processing despite numerous features derived from wearable sensors in multimodal data [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Modalities related to MDS-UPDRS Part II included the total scores of SF-36 and Mini-BEST from the PFL modalities and T10_PC2 and LANK_PC3 from the GP_Sensors modalities in our study. Further, the single modalities for PFL modalities related to MDS-UPDRS Part II included the total and physical scores of SF-36 and Mini-BEST. Persons with PD perceive that slow movements (bradykinesia), tremors, postural instability, and gait disturbances are the most challenging symptoms in daily living that substantially reduce their quality of life [\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e]. Our findings suggest that the measures most closely associated with the severity of daily living motor experiences were the quality of life, posture and balance measures, and gyroscope measurements from the body\u0026rsquo;s central axis and distal segments during gait tasks [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Therefore, the application of specific multimodal data, clinical evaluation scales, and ML algorithms for motor and non-motor symptom evaluation can support precision in therapeutic interventions [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e].\u003c/p\u003e \u003cdiv id=\"Sec25\" class=\"Section3\"\u003e \u003ch2\u003eIdentification of digital biomarkers based on classification models\u003c/h2\u003e \u003cp\u003eWe used the ROC-AUC to evaluate our classification models based on modalities with high feature importance for identifying digital biomarkers to classify PD severity subtypes. Movement disorder specialists use MDS-UPDRS as a screening tool for evaluating PD severity, monitoring disease progression, and assessing treatments and interventions [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. However, clinical assessments require experience, expertise, and time [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Inaccurate clinical assessments of persons with PD can lead to the suboptimal characterization of patients and clinical processes, which can affect diagnosis and treatment [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. We systematically evaluated meaningful multimodal data collected from persons with PD, assessed their accuracy in relation to MDS-UPDRS parts, and identified digital biomarkers that contributed to accurately classifying PD severity subtypes. GP_Sensors modalities such as LANK_PC1, T10_PC2, and RANK_PC1 from multimodal data showed the best classification performance between clusters based on PD severity. These features were derived from wearable sensor measurements and closely associated with MDS-UPDRS Total and Part III scores, which underscores the importance of gait measurements and standard clinical scales in identifying PD severity subtypes [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. As PD progresses from its early stages, gait disturbances, among the major motor symptoms, are likely to be objective and highly sensitive biomarkers [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. UPDRS scores have been predicted using gait time-series measurements, demonstrating promising results in terms of the MAE and RMS error [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. In this study, accelerometer and gyroscope measurements from wearable sensors attached to the ankles and back were used for describing the posture and gait patterns. A previous study reported that reduced gait performance and greater postural sway were associated with higher PD severity [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. Quantitative gait measurements and their characteristics suggest that they can be used as digital biomarkers to realize enhanced PD severity classification [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e, \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eConsidering individual PFL modalities, grip strength, 6 MWT, and Mini-BEST tests effectively distinguished PD severity subtypes in our study, thereby highlighting how such measures reflect the decline in motor function with PD progression. Grip strength is not only a measure of upper limb strength but also an indicator of PD progression from the mild to moderate stages [\u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e]. Reduced grip strength can adversely affect activities of daily living and increase the risk of falls while performing tasks such as opening door handles or refrigerator doors. This issue can be attributed to the weakening of the upper and lower limbs in persons with PD [\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e]. In our study, the 6 MWT, which was related to MDS-UPDRS Total, was valuable for assessing PD severity because it measured the distance walked by the patient as fast as possible in 6 min [\u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e]. In addition, the Mini-BEST score was among the top features, and it measured the ability of a patient for balancing or performing in daily life activities, also reflecting the fall risk [\u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e55\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eGait parameters showed that specific spatiotemporal features could effectively classify PD severity subtypes in our study, supporting the use of wearable sensors in clinical assessments [\u003cspan additionalcitationids=\"CR18\" citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. However, the spectrum of disease severity has not been considered, and several variables are difficult to interpret clinically [\u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e56\u003c/span\u003e, \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e57\u003c/span\u003e]. Further, using only continuous data collected from a single wearable sensor may be insufficient for identifying persons with PD from healthy controls or classifying PD severity subtypes [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Rehman et al. attempted to differentiate persons with PD from healthy controls using more than 100 gait variables and ML approaches using the GAITRite walkway. They identified six spatiotemporal variables that achieved an accuracy of 73\u0026ndash;97% [\u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e58\u003c/span\u003e]. In our previous study, we distinguished persons with PD from healthy controls with a 98.0% accuracy using five spatiotemporal variables during the same 360\u0026deg; turning task [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]. Persons with PD often exhibit considerable clinical asymmetry, which can cause gait disturbances and freezing of gait [\u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e59\u003c/span\u003e]. Asymmetry appears in the early PD stages and can persist with an increase in disease severity [\u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e60\u003c/span\u003e]. Our results confirm that gait features in both ankles such as LANK_PC1 and RANK_PC1 can indicate asymmetry factors based on the PD severity subtype, which promotes a high classification performance. Therefore, we suggest that features obtained by bilaterally attaching wearable sensors to body segments can be used and monitored to classify PD severity subtypes.\u003c/p\u003e \u003cp\u003eAlthough this study represents a step forward in our efforts to classify PD severity subtypes and predict PD severity, further investigation is required. In fact, our analysis has the following limitations:\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eAlthough clustering is data-driven, the outcomes depend on the appropriate selection of variables and clustering algorithms. Further, PD severity subtypes must be validated considering independent cohorts covering similar domains to confirm the clinical applicability of the study because an unsupervised approach to subtyping was employed in the study.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eAll participants were diagnosed with idiopathic PD based on criteria established by a neurologist. However, we cannot completely exclude the possibility that some participants had other underlying pathologies that could have affected our results.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eThe sample size was too small for ML analysis. However, our approach highlights the importance of incorporating multimodal features for comprehensively analyzing PD severity subtypes.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eData were collected exclusively in a controlled laboratory setting. Such a controlled environment may not fully represent real-world variability and complexity, potentially limiting the generalizability of our findings.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eData were collected only in the \u0026ldquo;on\u0026rdquo; medication state. Continuous monitoring during the transition periods between the \u0026ldquo;on\u0026rdquo; and \u0026ldquo;off\u0026rdquo; medication states is necessary for capturing the full spectrum of motor symptom fluctuations.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Conclusions","content":"\u003cp\u003eThis study demonstrated the efficacy of multimodal data and advanced ML algorithms for identifying PD severity subtypes and estimating disease severity. Our approach provided a comprehensive framework that integrates demographics, physical function, lifestyle measures, and gait parameters to understand PD heterogeneity. Our findings highlighted the potential of bilaterally attached wearable sensors as digital biomarkers to classify PD severity subtypes and track disease progression. This underscores the clinical value of sensor-based gait analysis in PD management, which supports its integration into personalized monitoring systems and therapeutic interventions [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eFuture research should validate these findings in larger independent cohorts and explore the clinical integration of digital biomarkers for real-time disease monitoring and therapy adjustments. Wearable sensors and mobile health applications can enhance personalized PD management by enabling continuous assessment and tailored rehabilitation strategies [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e].\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePD\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eParkinson's disease\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eML\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003emachine learning\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMDS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMovement disorder society\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eUPDRS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eUnified Parkinson's disease rating scale\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eIMU\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003einertial measurement unit\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePFL\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePhysical function and lifestyle\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePCs\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eprincipal components\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eGP_Motion\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eGait parameter PCs in motion analysis system\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eGP_Sensors\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003egait parameter PCs in wearable sensors\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMoCA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMontreal cognitive assessment\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMMSE\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMini-mental state examination\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eH\u0026amp;Y\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eHoehn and Yahr\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e5STST\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003efive times sit-to-stand test\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMini-BEST\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMini-balance evaluation systems test\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eSF-36\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003e36-item short form health survey\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eIPAQ\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eInternational physical activity questionnaire\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMutual information\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eLASSO\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eLeast absolute shrinkage and selection operator\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMAE\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003emean absolute error\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eT10\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003e10th thoracic vertebra\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eRANK\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eright lateral malleolus\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eRELB\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eright lateral humeral epicondyle\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eROC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ereceiver operating characteristic\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePSIS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eposterior superior iliac spine\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eLANK\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eleft lateral malleolus\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eAUC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003earea under the curve\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eLELB\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eleft lateral humeral epicondyle\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e6MWT\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003e6 min walking test.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Declarations","content":" \u003cp\u003e \u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e \u003cp\u003e All procedures involving human participants were performed in accordance with the ethical standards of the institutional and/or national research committee and observing the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. The study and its additional files were approved by the Institutional Review Board of Dong-A University Hospital (approval number DAUHIRB-22-089) (see Supplementary Material 5). All patients provided written informed consent prior to data collection. The study was registered with the Clinical Research Information Service of the Republic of Korea (KCT0009353).\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eConsent for publication\u003c/strong\u003e \u003cp\u003eNot applicable.\u003c/p\u003e \u003c/p\u003e\u003cp\u003e \u003ch2\u003eCompeting interests\u003c/h2\u003e \u003cp\u003eThe authors declare no conflicts of interest.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThis work was supported by grants from the National Research Foundation of Korea (NRF) funded by the Korean government (MSIT) (No. 2022R1A2C100933711; Changhong Youm), the Basic Science Research Program through the NRF funded by the Ministry of Education (No. 2022R1A6A3A0108756411; Hwayoung Park), and the Ministry of Education of the Republic of Korea and the NRF (No. 2024S1A5B5A16021673; Hwayoung Park). This study received no specific grants from funding agencies in the public, commercial, or non-profit sectors. The funding sources had no role in the study design; collection, analysis, and interpretation of the data; or in writing the manuscript.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eH.P., C.Y., and S.C. conceived and designed the study. H.P., S.C., and B.K. recruited the participants. H.P., C.Y., S.C., B.K., H.C., J.H., and M.K. performed data acquisition. H.P. and C.Y. analyzed and interpreted the data. H.P., C.Y., and S.C. drafted the manuscript. All authors read and approved the final version of the manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eThe authors thank all participants who contributed to this study. This work was supported by the Dong-A University research fund. The authors also thank Editage (www.editage.co.kr) for English language editing.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eDatasets supporting the findings of this study are available from the corresponding author upon request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eBirkenbihl C, Ahmad A, Massat NJ, Raschka T, Avbersek A, Downey P, et al. Artificial intelligence-based clustering and characterization of Parkinson\u0026rsquo;s disease trajectories. Sci Rep. 2023;13:2897. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41598-023-30038-8\u003c/span\u003e\u003cspan address=\"10.1038/s41598-023-30038-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFereshtehnejad SM, Postuma RB. Subtypes of Parkinson\u0026rsquo;s disease: What do they tell us about disease progression? Curr Neurol Neurosci Rep. 2017;17:34. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11910-017-0738-x\u003c/span\u003e\u003cspan address=\"10.1007/s11910-017-0738-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDadu A, Satone V, Kaur R, Hashemi SH, Leonard H, Iwaki H, et al. Identification and prediction of Parkinson\u0026rsquo;s disease subtypes and progression using machine learning in two cohorts. npj Parkinsons Dis. 2022;8:172. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41531-022-00439-z\u003c/span\u003e\u003cspan address=\"10.1038/s41531-022-00439-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStebbins GT, Goetz CG, Burn DJ, Jankovic J, Khoo TK, Tilley BC. How to identify tremor dominant and postural instability/gait difficulty groups with the movement disorder society unified Parkinson\u0026rsquo;s disease rating scale: Comparison with the unified Parkinson\u0026rsquo;s disease rating scale. Mov Disord. 2013;28:668\u0026ndash;70. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/mds.25383\u003c/span\u003e\u003cspan address=\"10.1002/mds.25383\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSu C, Hou Y, Xu J, Xu Z, Zhou M, Ke A, et al. Identification of Parkinson\u0026rsquo;s disease PACE subtypes and repurposing treatments through integrative analyses of multimodal data. npj Digit Med. 2024;7:184. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41746-024-01175-9\u003c/span\u003e\u003cspan address=\"10.1038/s41746-024-01175-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZetusky WJ, Jankovic J, Pirozzolo FJ. The heterogeneity of Parkinson\u0026rsquo;s disease: Clinical and prognostic implications. Neurology. 1985;35:522. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1212/wnl.35.4.522\u003c/span\u003e\u003cspan address=\"10.1212/wnl.35.4.522\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJankovic J, McDermott M, Carter J, Gauthier S, Goetz C, Golbe L, et al. Variable expression of Parkinson\u0026rsquo;s disease: A baseline analysis of the DAT ATOP cohort. Neurology. 1990;40:1529. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1212/wnl.40.10.1529\u003c/span\u003e\u003cspan address=\"10.1212/wnl.40.10.1529\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFaghri F, Brunn F, Dadu A, PARALS consortium ERRALS consortium, Zucchi E et al. Identifying and predicting amyotrophic lateral sclerosis clinical subgroups: A population-based machine-learning study. Lancet Digit Health. 2022;4:e359\u0026ndash;69. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S2589-7500(21)00274-0\u003c/span\u003e\u003cspan address=\"10.1016/S2589-7500(21)00274-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVan Rooden SM, Heiser WJ, Kok JN, Verbaan D, van Hilten JJ, Marinus J. The identification of Parkinson\u0026rsquo;s disease subtypes using cluster analysis: A systematic review. Mov Disord. 2010;25:969\u0026ndash;78. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/mds.23116\u003c/span\u003e\u003cspan address=\"10.1002/mds.23116\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFereshtehnejad SM, Zeighami Y, Dagher A, Postuma RB. Clinical criteria for subtyping Parkinson\u0026rsquo;s disease: Biomarkers and longitudinal progression. Brain. 2017;140:1959\u0026ndash;76. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/brain/awx118\u003c/span\u003e\u003cspan address=\"10.1093/brain/awx118\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMirelman A, Volkov J, Salomon A, Gazit E, Nieuwboer A, Rochester L, et al. Digital mobility measures: A window into real-world severity and progression of Parkinson\u0026rsquo;s disease. Mov Disord. 2024;39:328\u0026ndash;38. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/mds.29689\u003c/span\u003e\u003cspan address=\"10.1002/mds.29689\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEspay AJ, Bonato P, Nahab FB, Maetzler W, Dean JM, Klucken J, et al. Technology in Parkinson\u0026rsquo;s disease: Challenges and opportunities. Mov Disord. 2016;31:1272\u0026ndash;82. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/mds.26642\u003c/span\u003e\u003cspan address=\"10.1002/mds.26642\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHouts CR, Patrick-Lake B, Clay I, Wirth RJ. The path forward for digital measures: Suppressing the desire to compare apples and pineapples. Digit Biomark. 2020;4(Suppl 1):3\u0026ndash;12. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1159/000511586\u003c/span\u003e\u003cspan address=\"10.1159/000511586\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGoetz CG, Tilley BC, Shaftman SR, Stebbins GT, Fahn S, Martinez-Martin P, et al. Movement Disorder Society-sponsored revision of the Unified Parkinson\u0026rsquo;s disease Rating Scale (MDS‐UPDRS): Scale presentation and clinimetric testing results. Mov Disord. 2008;23:2129\u0026ndash;70. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/mds.22340\u003c/span\u003e\u003cspan address=\"10.1002/mds.22340\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStephenson D, Badawy R, Mathur S, Tome M, Rochester L. Digital progression biomarkers as novel endpoints in clinical trials: A multistakeholder perspective. J Parkinsons Dis. 2021;11:S103\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3233/JPD-202428\u003c/span\u003e\u003cspan address=\"10.3233/JPD-202428\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePost B, Merkus MP, de Bie RMA, de Haan RJ, Speelman JD. Unified Parkinson\u0026rsquo;s disease rating scale motor examination: Are ratings of nurses, residents in neurology, and movement disorders specialists interchangeable? Mov Disord. 2005;20:1577\u0026ndash;84. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/mds.20640\u003c/span\u003e\u003cspan address=\"10.1002/mds.20640\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMirelman A, Hillel I, Rochester L, Del Din S, Bloem BR, Avanzino L, et al. Tossing and turning in bed: Nocturnal movements in Parkinson\u0026rsquo;s disease. Mov Disord. 2020;35:959\u0026ndash;68. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/mds.28006\u003c/span\u003e\u003cspan address=\"10.1002/mds.28006\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSotirakis C, Su Z, Brzezicki MA, Conway N, Tarassenko L, FitzGerald JJ, et al. Identification of motor progression in Parkinson\u0026rsquo;s disease using wearable sensors and machine learning. npj Parkinsons Dis. 2023;9:142. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41531-023-00581-2\u003c/span\u003e\u003cspan address=\"10.1038/s41531-023-00581-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZadka A, Rabin N, Gazit E, Mirelman A, Nieuwboer A, Rochester L, et al. A wearable sensor and machine learning estimate step length in older adults and patients with neurological disorders. npj Digit Med. 2024;7:142. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41746-024-01136-2\u003c/span\u003e\u003cspan address=\"10.1038/s41746-024-01136-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFaiem N, Asuroglu T, Acici K, Kallonen A, Van Gils M. Assessment of Parkinson\u0026rsquo;s disease severity using gait data: A deep learning-based multimodal approach. Nordic Conf Digit Health Wirel Solutions 29\u0026ndash;48; 2024.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eV\u0026aacute;squez-Correa JC, Arias-Vergara T, Orozco-Arroyave JR, Eskofier B, Klucken J, Noth E. Multimodal assessment of Parkinson\u0026rsquo;s disease: A deep learning approach. IEEE J Biomed Health Inf. 2019;23:1618\u0026ndash;30. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/JBHI.2018.2866873\u003c/span\u003e\u003cspan address=\"10.1109/JBHI.2018.2866873\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGodi M, Arcolin I, Giardini M, Corna S, Schieppati M. A pathophysiological model of gait captures the details of the impairment of pace/rhythm, variability and asymmetry in Parkinsonian patients at distinct stages of the disease. Sci Rep. 2021;11:21143. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41598-021-00543-9\u003c/span\u003e\u003cspan address=\"10.1038/s41598-021-00543-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChandrabhatla AS, Pomeraniec IJ, Ksendzovsky A. Co-evolution of machine learning and digital technologies to improve monitoring of Parkinson\u0026rsquo;s disease motor symptoms. npj Digit Med. 2022;5:32. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41746-022-00568-y\u003c/span\u003e\u003cspan address=\"10.1038/s41746-022-00568-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNalls MA, McLean CY, Rick J, Eberly S, Hutten SJ, Gwinn K, et al. Diagnosis of Parkinson\u0026rsquo;s disease on the basis of clinical and genetic classification: A population-based modelling study. Lancet Neurol. 2015;14:1002\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S1474-4422(15)00178-7\u003c/span\u003e\u003cspan address=\"10.1016/S1474-4422(15)00178-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlbrecht F, Poulakis K, Freidle M, Johansson H, Ekman U, Volpe G, et al. Unraveling Parkinson\u0026rsquo;s disease heterogeneity using subtypes based on multimodal data. Parkinsonism Relat Disord. 2022;102:19\u0026ndash;29. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.parkreldis.2022.07.014\u003c/span\u003e\u003cspan address=\"10.1016/j.parkreldis.2022.07.014\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMestre TA, Eberly S, Tanner C, Grimes D, Lang AE, Oakes D, et al. Reproducibility of data-driven Parkinson\u0026rsquo;s disease subtypes for clinical research. Parkinsonism Relat Disord. 2018;56:102\u0026ndash;6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.parkreldis.2018.07.009\u003c/span\u003e\u003cspan address=\"10.1016/j.parkreldis.2018.07.009\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRehman RZU, Rochester L, Yarnall AJ, Del Din S. Predicting the progression of Parkinson\u0026rsquo;s disease MDS-UPDRS-III motor severity score from gait data using deep learning 43rd Ann. Int Conf IEEE Eng Med Biol Soc (EMBC) 249\u0026ndash;52; 2021.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDe Vos M, Prince J, Buchanan T, FitzGerald JJ, Antoniades CA. Discriminating progressive supranuclear palsy from Parkinson\u0026rsquo;s disease using wearable technology and machine learning. Gait Posture. 2020;77:257\u0026ndash;63. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.gaitpost.2020.02.007\u003c/span\u003e\u003cspan address=\"10.1016/j.gaitpost.2020.02.007\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMancini M, Weiss A, Herman T, Hausdorff JM. Turn around freezing: Community-living turning behavior in people with Parkinson\u0026rsquo;s disease. Front Neurol. 2018;9:18. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fneur.2018.00018\u003c/span\u003e\u003cspan address=\"10.3389/fneur.2018.00018\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDel Din S, Elshehabi M, Galna B, Hobert MA, Warmerdam E, Suenkel U, et al. Gait analysis with wearables predicts conversion to Parkinson disease. Ann Neurol. 2019;86:357\u0026ndash;67. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/ana.25548\u003c/span\u003e\u003cspan address=\"10.1002/ana.25548\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLonini L, Dai A, Shawen N, Simuni T, Poon C, Shimanovich L, et al. Wearable sensors for Parkinson\u0026rsquo;s disease: Which data are worth collecting for training symptom detection models. npj Digit Med. 2018;1:64. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41746-018-0071-z\u003c/span\u003e\u003cspan address=\"10.1038/s41746-018-0071-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAdams JL, Dinesh K, Snyder CW, Xiong M, Tarolli CG, Sharma S, et al. A real-world study of wearable sensors in Parkinson\u0026rsquo;s disease. npj Parkinsons Dis. 2021;7:106. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41531-021-00248-w\u003c/span\u003e\u003cspan address=\"10.1038/s41531-021-00248-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eScherbaum R, Moewius A, Oppermann J, Geritz J, Hansen C, Gold R, et al. Parkinson\u0026rsquo;s disease multimodal complex treatment improves gait performance: An exploratory wearable digital device-supported study. J Neurol. 2022;269:6067\u0026ndash;85. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00415-022-11257-x\u003c/span\u003e\u003cspan address=\"10.1007/s00415-022-11257-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMahadevan N, Demanuele C, Zhang H, Volfson D, Ho B, Erb MK, et al. Development of digital biomarkers for resting tremor and bradykinesia using a wrist-worn wearable device. npj Digit Med. 2020;3:5. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41746-019-0217-7\u003c/span\u003e\u003cspan address=\"10.1038/s41746-019-0217-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee JS, Kim HY, Hwang JY, Kwon S, Chung HR, Kwak TK et al. Development of nutrition quotient for Korean adults: Item selection and validation of factor structure. J Nutr Health. 2018:51:340\u0026ndash;56. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.4163/jnh.2018.51.4.340\u003c/span\u003e\u003cspan address=\"10.4163/jnh.2018.51.4.340\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePark H, Shin S, Youm C, Cheon SM, Lee M, Noh B. Classification of Parkinson\u0026rsquo;s disease with freezing of gait based on 360 turning analysis using 36 kinematic features. J Neuroeng Rehabil. 2021;18:1\u0026ndash;18. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12984-021-00975-4 35\u003c/span\u003e\u003cspan address=\"10.1186/s12984-021-00975-4 35\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePeng H, Long F, Ding C. Feature selection based on mutual information criteria of max-dependency, max-relevance, and min-redundancy. IEEE Trans Pattern Anal Mach Intell. 2005;27:1226\u0026ndash;38. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/TPAMI.2005.159\u003c/span\u003e\u003cspan address=\"10.1109/TPAMI.2005.159\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDebelle H, Packer E, Beales E, Bailey HGB, Mc Ardle R, Brown P, et al. Feasibility and usability of a digital health technology system to monitor mobility and assess medication adherence in mild-to-moderate Parkinson\u0026rsquo;s disease. Front Neurol. 2023;14:1111260. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fneur.2023.1111260\u003c/span\u003e\u003cspan address=\"10.3389/fneur.2023.1111260\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDel Din S, Kirk C, Yarnall AJ, Rochester L, Hausdorff JM. Body-worn sensors for remote monitoring of Parkinson\u0026rsquo;s disease motor symptoms: Vision, state of the art, and challenges ahead. J Parkinsons Dis. 2021;11:S35\u0026ndash;47. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3233/JPD-202471\u003c/span\u003e\u003cspan address=\"10.3233/JPD-202471\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSundgren M, Andr\u0026eacute;asson M, Svenningsson P, Noori RM, Johansson A. Does information from the Parkinson KinetiGraph\u0026trade; (PKG) influence the neurologist\u0026rsquo;s treatment decisions?\u0026mdash;An observational study in routine clinical care of people with Parkinson\u0026rsquo;s disease. J Pers Med. 2021;11:519. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/jpm11060519\u003c/span\u003e\u003cspan address=\"10.3390/jpm11060519\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCastiglia SF, Trabassi D, Conte C, Ranavolo A, Coppola G, Sebastianelli G, et al. Multiscale entropy algorithms to analyze complexity and variability of trunk accelerations time series in subjects with Parkinson\u0026rsquo;s disease. Sens (Basel). 2023;23:4983. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/s23104983\u003c/span\u003e\u003cspan address=\"10.3390/s23104983\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCoates L, Shi J, Rochester L, Del Din S, Pantall A. Entropy of real-world gait in Parkinson\u0026rsquo;s disease determined from wearable sensors as a digital marker of altered ambulatory behavior. Sens (Basel). 2020;20:2631. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/s20092631\u003c/span\u003e\u003cspan address=\"10.3390/s20092631\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMancini M, Horak FB, Zampieri C, Carlson-Kuhta P, Nutt JG, Chiari L. Trunk accelerometry reveals postural instability in untreated Parkinson\u0026rsquo;s disease. Parkinsonism Relat Disord. 2011;17:557\u0026ndash;62. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.parkreldis.2011.05.010\u003c/span\u003e\u003cspan address=\"10.1016/j.parkreldis.2011.05.010\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWorld Health Organization. Towards a common language for functioning, disability, and health: ICF. Int Classif Functioning Disabil Health; 2002.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGiannouli E, Bock O, Mellone S, Zijlstra W. Mobility in old age: Capacity is not performance. BioMed Res Int. 2016;2016:3261567. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1155/2016/3261567\u003c/span\u003e\u003cspan address=\"10.1155/2016/3261567\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuo Y, Yang J, Liu Y, Chen X, Yang GZ. Detection and assessment of Parkinson's disease based on gait analysis: A survey. Front Aging Neurosci. 2022;14:916971. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fnagi.2022.916971\u003c/span\u003e\u003cspan address=\"10.3389/fnagi.2022.916971\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMammen JR, Speck RM, Stebbins GM, M\u0026uuml;ller MLTM, Yang PT, Campbell M, et al. Mapping relevance of digital measures to meaningful symptoms and impacts in early Parkinson\u0026rsquo;s disease. J Parkinsons Dis. 2023;13:589\u0026ndash;607. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3233/JPD-225122\u003c/span\u003e\u003cspan address=\"10.3233/JPD-225122\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang J. Mining imaging and clinical data with machine learning approaches for the diagnosis and early detection of Parkinson\u0026rsquo;s disease. npj Parkinsons Dis. 2022;8:13. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41531-021-00266-8\u003c/span\u003e\u003cspan address=\"10.1038/s41531-021-00266-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWahid F, Begg RK, Hass CJ, Halgamuge S, Ackland DC. Classification of Parkinson\u0026rsquo;s disease gait using spatial-temporal gait features. IEEE J Biomed Health Inf. 2015;19:1794\u0026ndash;802. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/JBHI.2015.2450232\u003c/span\u003e\u003cspan address=\"10.1109/JBHI.2015.2450232\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePort RJ, Rumsby M, Brown G, Harrison IF, Amjad A, Bale CJ. People with Parkinson\u0026rsquo;s disease: What symptoms do they most want to improve and how does this change with disease duration? J Parkinsons Dis. 2021;11:715\u0026ndash;24. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3233/JPD-202346\u003c/span\u003e\u003cspan address=\"10.3233/JPD-202346\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eElshehabi M, Maier KS, Hasmann SE, Nussbaum S, Herbst H, Heger T, et al. Limited effect of dopaminergic medication on straight walking and turning in early-to-moderate Parkinson\u0026rsquo;s disease during single and dual tasking. Front Aging Neurosci. 2016;8:4. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fnagi.2016.00004\u003c/span\u003e\u003cspan address=\"10.3389/fnagi.2016.00004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePanyakaew P, Duangjino K, Kerddonfag A, Ploensin T, Piromsopa K, Kongkamol C, et al. Exploring the complex phenotypes of impaired finger dexterity in mild-to-moderate stage Parkinson\u0026rsquo;s disease: A time-series analysis. J Parkinsons Dis. 2023;13:975\u0026ndash;88. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3233/JPD-230029\u003c/span\u003e\u003cspan address=\"10.3233/JPD-230029\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGamborg M, Hvid LG, Thrue C, Johansson S, Franz\u0026eacute;n E, Dalgas U, et al. Muscle strength and power in people with Parkinson disease: A systematic review and meta-analysis. J Neurol Phys Ther. 2023;47:3\u0026ndash;15. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1097/NPT.0000000000000421\u003c/span\u003e\u003cspan address=\"10.1097/NPT.0000000000000421\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBailo G, Saibene FL, Bandini V, Arcuri P, Salvatore A, Meloni M, et al. Characterization of walking in mild Parkinson\u0026rsquo;s disease: Reliability, validity and discriminant ability of the six-minute walk test instrumented with a single inertial sensor. Sens (Basel). 2024;24:662. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/s24020662\u003c/span\u003e\u003cspan address=\"10.3390/s24020662\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEgger M, Finsterh\u0026ouml;lzl M, Buetikofer A, Wippenbeck F, M\u0026uuml;ller F, Jahn K, et al. Balance function in critical illness survivors and evaluation of psychometric properties of the Mini-BESTest. Sci Rep. 2024;14:12089. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41598-024-61745-5\u003c/span\u003e\u003cspan address=\"10.1038/s41598-024-61745-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRehman RZU, Buckley C, Mic\u0026oacute;-Amigo ME, Kirk C, Dunne-Willows M, Mazz\u0026agrave; C, et al. Accelerometry-based digital gait characteristics for classification of Parkinson\u0026rsquo;s disease: What counts? IEEE open J Eng Med Biol. 2020;1:65\u0026ndash;73. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/OJEMB.2020.2966295\u003c/span\u003e\u003cspan address=\"10.1109/OJEMB.2020.2966295\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlam MN, Garg A, Munia TTK, Fazel-Rezai R, Tavakolian K. Vertical ground reaction force marker for Parkinson\u0026rsquo;s disease. PLoS ONE. 2017;12:e0175951. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1371/journal.pone.0175951\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0175951\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRehman RZU, Del Din S, Guan Y, Yarnall AJ, Shi JQ, Rochester L. Selecting clinically relevant gait characteristics for classification of early Parkinson\u0026rsquo;s disease: A comprehensive machine learning approach. Sci Rep. 2019;9:17269. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41598-019-53656-7\u003c/span\u003e\u003cspan address=\"10.1038/s41598-019-53656-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDjaldetti R, Ziv I, Melamed E. The mystery of motor asymmetry in Parkinson\u0026rsquo;s disease. Lancet Neurol. 2006;5:796\u0026ndash;802. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S1474-4422(06)70549-X\u003c/span\u003e\u003cspan address=\"10.1016/S1474-4422(06)70549-X\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMonje MHG, S\u0026aacute;nchez-Ferro \u0026Aacute;, Pineda-Pardo JA, Vela-Desojo L, Alonso-Frech F, Obeso JA. Motor onset topography and progression in Parkinson\u0026rsquo;s disease: The upper limb is first. Mov Disord. 2021;36:905\u0026ndash;15. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/mds.28462\u003c/span\u003e\u003cspan address=\"10.1002/mds.28462\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003ctable width=\"586\"\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"8\" width=\"585\"\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1. Demographics and clinical characteristics of the study participants.\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"2\" width=\"95\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"2\" width=\"85\"\u003e\n\u003cp\u003e\u003cstrong\u003ePersons with PD (\u003cem\u003en\u0026nbsp;\u003c/em\u003e=\u0026nbsp;102)\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"3\" width=\"255\"\u003e\n\u003cp\u003e\u003cstrong\u003eIndividuals according to the PD severity subtype\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"3\" width=\"151\"\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003ep\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e-value\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eadjusted \u003cem\u003ep\u003c/em\u003e-value\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e\u003cstrong\u003eMild\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(\u003cem\u003en\u0026nbsp;\u003c/em\u003e=\u0026nbsp;24; 23.5%)\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e\u003cstrong\u003eModerate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(\u003cem\u003en\u0026nbsp;\u003c/em\u003e=\u0026nbsp;47; 46.1%)\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e\u003cstrong\u003eSevere\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(\u003cem\u003en\u0026nbsp;\u003c/em\u003e=\u0026nbsp;31; 30.4%)\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u003cstrong\u003e1 vs. 2\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"38\"\u003e\n\u003cp\u003e\u003cstrong\u003e2 vs. 3\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u003cstrong\u003e1 vs. 3\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003eSex (male/female)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e48/54\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e16/8\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e19/28\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e13/18\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u0026ndash;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"38\"\u003e\n\u003cp\u003e\u0026ndash;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u0026ndash;\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003eAge (years)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e68.06\u0026nbsp;\u0026plusmn;\u0026nbsp;7.20\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e64.83\u0026nbsp;\u0026plusmn;\u0026nbsp;7.25\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e68.00\u0026nbsp;\u0026plusmn;\u0026nbsp;5.98\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e70.65\u0026nbsp;\u0026plusmn;\u0026nbsp;8.01\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e0.054\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.081\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"38\"\u003e\n\u003cp\u003e0.099\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.014\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.007\u003c/strong\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e0.011\u003c/strong\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003eHeight (cm)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e160.36\u0026nbsp;\u0026plusmn;\u0026nbsp;8.43\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e164.07\u0026nbsp;\u0026plusmn;\u0026nbsp;7.82\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e159.90\u0026nbsp;\u0026plusmn;\u0026nbsp;6.10\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e158.20\u0026nbsp;\u0026plusmn;\u0026nbsp;10.91\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.016\u003c/strong\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e0.047\u003c/strong\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"38\"\u003e\n\u003cp\u003e0.434\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.118\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.024\u003c/strong\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e0.010\u003c/strong\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003eBody weight (kg)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e62.90\u0026nbsp;\u0026plusmn;\u0026nbsp;10.64\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e64.81\u0026nbsp;\u0026plusmn;\u0026nbsp;11.87\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e61.98\u0026nbsp;\u0026plusmn;\u0026nbsp;9.62\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e62.81\u0026nbsp;\u0026plusmn;\u0026nbsp;11.26\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e0.282\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.211\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"38\"\u003e\n\u003cp\u003e0.737\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.170\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e0.528\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.158\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003eBMI (kg/m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e24.37\u0026nbsp;\u0026plusmn;\u0026nbsp;3.07\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e23.89\u0026nbsp;\u0026plusmn;\u0026nbsp;2.76\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e24.17\u0026nbsp;\u0026plusmn;\u0026nbsp;2.90\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e25.07\u0026nbsp;\u0026plusmn;\u0026nbsp;3.50\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e0.698\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.161\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"38\"\u003e\n\u003cp\u003e0.239\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.080\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e0.168\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.056\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003eDisease duration (years)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e5.67\u0026nbsp;\u0026plusmn;\u0026nbsp;4.39\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e5.69\u0026nbsp;\u0026plusmn;\u0026nbsp;3.58\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e5.71\u0026nbsp;\u0026plusmn;\u0026nbsp;5.22\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e5.59\u0026nbsp;\u0026plusmn;\u0026nbsp;3.65\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e0.618\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.181\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"38\"\u003e\n\u003cp\u003e0.567\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.142\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e0.805\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.186\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003eTreatment duration (years)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e4.77\u0026nbsp;\u0026plusmn;\u0026nbsp;4.25\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e4.72\u0026nbsp;\u0026plusmn;\u0026nbsp;3.81\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e4.78\u0026nbsp;\u0026plusmn;\u0026nbsp;5.10\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e4.81\u0026nbsp;\u0026plusmn;\u0026nbsp;3.12\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e0.662\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.181\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"38\"\u003e\n\u003cp\u003e0.244\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.073\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e0.532\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.145\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003eL-dopa equivalent dose (mg/day)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e545.46\u0026nbsp;\u0026plusmn;\u0026nbsp;289.96\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e608.60\u0026nbsp;\u0026plusmn;\u0026nbsp;405.14\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e490.82\u0026nbsp;\u0026plusmn;\u0026nbsp;207.53\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e579.40\u0026nbsp;\u0026plusmn;\u0026nbsp;286.55\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e0.191\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.191\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"38\"\u003e\n\u003cp\u003e0.144\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.054\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e0.766\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.192\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003eH\u0026amp;Y stage I\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e28 (27.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e12 (50.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e11 (23.4%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e5 (16.1%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u0026ndash;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"38\"\u003e\n\u003cp\u003e\u0026ndash;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u0026ndash;\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003eH\u0026amp;Y stage II\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e50 (49.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e8 (33.3%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e25 (53.2%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e17 (54.8%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u0026ndash;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"38\"\u003e\n\u003cp\u003e\u0026ndash;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u0026ndash;\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003eH\u0026amp;Y stage III\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e24 (23.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e4 (16.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e11 (23.4%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e9 (29.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u0026ndash;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"38\"\u003e\n\u003cp\u003e\u0026ndash;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u0026ndash;\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003eMDS-UPDRS Total (scores)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e54.51\u0026nbsp;\u0026plusmn;\u0026nbsp;23.17\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e47.83\u0026nbsp;\u0026plusmn;\u0026nbsp;24.80\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e51.22\u0026nbsp;\u0026plusmn;\u0026nbsp;20.53\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e64.68\u0026nbsp;\u0026plusmn;\u0026nbsp;23.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e0.542\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.181\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"38\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.009\u003c/strong\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e0.014\u003c/strong\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.012\u003c/strong\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e0.007\u003c/strong\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003eMDS-UPDRS Part I (scores)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e10.09\u0026nbsp;\u0026plusmn;\u0026nbsp;5.22\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e8.50\u0026nbsp;\u0026plusmn;\u0026nbsp;4.68\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e9.72\u0026nbsp;\u0026plusmn;\u0026nbsp;6.26\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e11.87\u0026nbsp;\u0026plusmn;\u0026nbsp;5.19\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e0.340\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.170\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"38\"\u003e\n\u003cp\u003e0.080\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e0.048\u003c/strong\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.016\u003c/strong\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e0.008\u003c/strong\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003eMDS-UPDRS Part II (scores)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e12.89\u0026nbsp;\u0026plusmn;\u0026nbsp;7.22\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e10.83\u0026nbsp;\u0026plusmn;\u0026nbsp;7.38\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e11.26\u0026nbsp;\u0026plusmn;\u0026nbsp;5.90\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e16.97\u0026nbsp;\u0026plusmn;\u0026nbsp;7.48\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e0.495\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.212\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"38\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.001\u003c/strong\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e0.003\u003c/strong\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.004\u003c/strong\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e0.012\u003c/strong\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003eMDS-UPDRS Part III (scores)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e29.45\u0026nbsp;\u0026plusmn;\u0026nbsp;15.23\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e26.25\u0026nbsp;\u0026plusmn;\u0026nbsp;14.14\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e27.84\u0026nbsp;\u0026plusmn;\u0026nbsp;14.68\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e34.35\u0026nbsp;\u0026plusmn;\u0026nbsp;16.13\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e0.663\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.166\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"38\"\u003e\n\u003cp\u003e0.069\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.052\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e0.057\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e0.021\u003c/strong\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003eMMSE (scores)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e27.75\u0026nbsp;\u0026plusmn;\u0026nbsp;2.10\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e28.46\u0026nbsp;\u0026plusmn;\u0026nbsp;1.32\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e27.94\u0026nbsp;\u0026plusmn;\u0026nbsp;1.77\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e26.90\u0026nbsp;\u0026plusmn;\u0026nbsp;2.74\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e0.305\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.183\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"38\"\u003e\n\u003cp\u003e0.138\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.059\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.008\u003c/strong\u003e\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e0.006\u003c/strong\u003e\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003eMoCA (scores)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e25.84\u0026nbsp;\u0026plusmn;\u0026nbsp;2.98\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e26.79\u0026nbsp;\u0026plusmn;\u0026nbsp;2.25\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e26.15\u0026nbsp;\u0026plusmn;\u0026nbsp;2.92\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e24.65\u0026nbsp;\u0026plusmn;\u0026nbsp;3.25\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e0.495\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e0.186\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"38\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.017\u003c/strong\u003e\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e0.017\u003c/strong\u003e\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.008\u003c/strong\u003e\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e0.008\u003c/strong\u003e\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"8\" width=\"586\"\u003e\n\u003cp\u003eThe data are presented as mean \u0026plusmn; standard deviation, with significant differences between groups (1, 2, and 3 indicate mild, moderate, and severe subtypes, respectively) indicated in bold (Uncorrected\u003cem\u003e p\u003c/em\u003e and adjusted \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.05). False discovery rate (FDR) correction using the Benjamini\u0026ndash;Hochberg method was applied to adjust for multiple comparisons.\u003c/p\u003e\n\u003cp\u003ePD, Parkinson\u0026rsquo;s disease; BMI, body mass index; L-dopa, levodopa; H\u0026amp;Y, Hoehn and Yahr; MDS-UPDRS, movement disorder society-unified Parkinson\u0026rsquo;s disease rating scale; MMSE, mini-mental state examination; and MoCA, Montreal cognitive assessment.\u003c/p\u003e\n\u003cp\u003e\u003csup\u003ea\u003c/sup\u003e Independent samples \u003cem\u003et\u003c/em\u003e-test result\u003c/p\u003e\n\u003cp\u003e\u003csup\u003eb\u003c/sup\u003e Mann\u0026ndash;Whitney \u003cem\u003eU\u003c/em\u003e test result\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003ctable\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"3\" width=\"514\"\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2. Selected modalities and category domains by the MI algorithm for the relationship between all/single modalities and MDS-UPDRS parts.\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"105\"\u003e\n\u003cp\u003e\u003cstrong\u003eModality\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"2\" width=\"262\"\u003e\n\u003cp\u003e\u003cstrong\u003eSelected domains\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(Number of features)\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"2\" width=\"200\"\u003e\n\u003cp\u003e\u003cstrong\u003eFeature selection score\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"105\"\u003e\n\u003cp\u003e\u003cstrong\u003eMDS-UPDRS scores \u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"3\" width=\"567\"\u003e\n\u003cp\u003e\u003cstrong\u003eAll modalities\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"105\"\u003e\n\u003cp\u003eUPDRS Total\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"262\"\u003e\n\u003cp\u003eGP_Sensors (17)\u003c/p\u003e\n\u003cp\u003eGP_Motion (5)\u003c/p\u003e\n\u003cp\u003ePFL (10)\u003c/p\u003e\n\u003cp\u003eDemographics (3)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"200\"\u003e\n\u003cp\u003e2.74\u003c/p\u003e\n\u003cp\u003e1.70\u003c/p\u003e\n\u003cp\u003e1.22\u003c/p\u003e\n\u003cp\u003e0.13\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"105\"\u003e\n\u003cp\u003eUPDRS Part I\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"262\"\u003e\n\u003cp\u003eGP_Sensors (14)\u003c/p\u003e\n\u003cp\u003eGP_Motion (5)\u003c/p\u003e\n\u003cp\u003ePFL (7)\u003c/p\u003e\n\u003cp\u003eDemographics (3)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"200\"\u003e\n\u003cp\u003e1.68\u003c/p\u003e\n\u003cp\u003e0.94\u003c/p\u003e\n\u003cp\u003e0.74\u003c/p\u003e\n\u003cp\u003e0.15\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"105\"\u003e\n\u003cp\u003eUPDRS Part II\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"262\"\u003e\n\u003cp\u003eGP_Sensors (21)\u003c/p\u003e\n\u003cp\u003eGP_Motion (4)\u003c/p\u003e\n\u003cp\u003ePFL (7)\u003c/p\u003e\n\u003cp\u003eDemographics (2)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"200\"\u003e\n\u003cp\u003e1.97\u003c/p\u003e\n\u003cp\u003e0.87\u003c/p\u003e\n\u003cp\u003e0.69\u003c/p\u003e\n\u003cp\u003e0.07\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"105\"\u003e\n\u003cp\u003eUPDRS Part III\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"262\"\u003e\n\u003cp\u003eGP_Sensors (21)\u003c/p\u003e\n\u003cp\u003eGP_Motion (5)\u003c/p\u003e\n\u003cp\u003ePFL (10)\u003c/p\u003e\n\u003cp\u003eDemographics (3)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"200\"\u003e\n\u003cp\u003e2.42\u003c/p\u003e\n\u003cp\u003e1.39\u003c/p\u003e\n\u003cp\u003e1.17\u003c/p\u003e\n\u003cp\u003e0.30\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"3\" width=\"567\"\u003e\n\u003cp\u003e\u003cstrong\u003ePFL modalities\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"105\"\u003e\n\u003cp\u003eUPDRS Total\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"262\"\u003e\n\u003cp\u003e(10)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"200\"\u003e\n\u003cp\u003e1.11\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"105\"\u003e\n\u003cp\u003eUPDRS Part I\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"262\"\u003e\n\u003cp\u003e(7)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"200\"\u003e\n\u003cp\u003e0.54\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"105\"\u003e\n\u003cp\u003eUPDRS Part II\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"262\"\u003e\n\u003cp\u003e(11)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"200\"\u003e\n\u003cp\u003e0.73\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"105\"\u003e\n\u003cp\u003eUPDRS Part III\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"262\"\u003e\n\u003cp\u003e(9)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"200\"\u003e\n\u003cp\u003e0.81\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"3\" width=\"567\"\u003e\n\u003cp\u003e\u003cstrong\u003eGP_Motion modalities\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"105\"\u003e\n\u003cp\u003eUPDRS Total\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"262\"\u003e\n\u003cp\u003e(5)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"200\"\u003e\n\u003cp\u003e1.59\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"105\"\u003e\n\u003cp\u003eUPDRS Part I\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"262\"\u003e\n\u003cp\u003e(5)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"200\"\u003e\n\u003cp\u003e0.69\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"105\"\u003e\n\u003cp\u003eUPDRS Part II\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"262\"\u003e\n\u003cp\u003e(4)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"200\"\u003e\n\u003cp\u003e0.95\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"105\"\u003e\n\u003cp\u003eUPDRS Part III\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"262\"\u003e\n\u003cp\u003e(5)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"200\"\u003e\n\u003cp\u003e1.01\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"3\" width=\"567\"\u003e\n\u003cp\u003e\u003cstrong\u003eGP_Sensors modalities\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"105\"\u003e\n\u003cp\u003eUPDRS Total\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"262\"\u003e\n\u003cp\u003e(16)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"200\"\u003e\n\u003cp\u003e2.63\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"105\"\u003e\n\u003cp\u003eUPDRS Part I\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"262\"\u003e\n\u003cp\u003e(14)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"200\"\u003e\n\u003cp\u003e1.4\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"105\"\u003e\n\u003cp\u003eUPDRS Part II\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"262\"\u003e\n\u003cp\u003e(22)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"200\"\u003e\n\u003cp\u003e2.08\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"105\"\u003e\n\u003cp\u003eUPDRS Part III\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"262\"\u003e\n\u003cp\u003e(20)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"200\"\u003e\n\u003cp\u003e1.97\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"3\" width=\"567\"\u003e\n\u003cp\u003ePFL, Physical function and lifestyle; GP_Motion, Gait parameters' principal components (PCs) in motion analysis system; GP_Sensors, Gait parameters' PCs in wearable sensors; and MDS-UPDRS, Movement disorder society-unified Parkinson\u0026rsquo;s disease rating scale.\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003ctable width=\"558\"\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd width=\"1\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"4\" width=\"557\"\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3. ML regression models selected using cross-validation, grid search, and model evaluation based on multiple modalities.\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" width=\"180\"\u003e\n\u003cp\u003e\u003cstrong\u003eModality\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"2\" width=\"198\"\u003e\n\u003cp\u003e\u003cstrong\u003eSelected model\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"2\" width=\"95\"\u003e\n\u003cp\u003e\u003cstrong\u003eMAE\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"2\" width=\"85\"\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eR\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e\u003csup\u003e2\u003c/sup\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" width=\"180\"\u003e\n\u003cp\u003e\u003cstrong\u003eMDS-UPDRS scores \u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"5\" width=\"558\"\u003e\n\u003cp\u003e\u003cstrong\u003eAll modalities\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" width=\"180\"\u003e\n\u003cp\u003eUPDRS Total\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"198\"\u003e\n\u003cp\u003eRandom forest regressor\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003e0.69\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e0.35\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" width=\"180\"\u003e\n\u003cp\u003eUPDRS Part I\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"198\"\u003e\n\u003cp\u003eLASSO\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003e0.79\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e0.13\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" width=\"180\"\u003e\n\u003cp\u003eUPDRS Part II\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"198\"\u003e\n\u003cp\u003eLASSO\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.66\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.41\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" width=\"180\"\u003e\n\u003cp\u003eUPDRS Part III\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"198\"\u003e\n\u003cp\u003eRandom forest regressor\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003e0.68\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e0.16\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"5\" width=\"558\"\u003e\n\u003cp\u003e\u003cstrong\u003ePFL modalities\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" width=\"180\"\u003e\n\u003cp\u003eUPDRS Total\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"198\"\u003e\n\u003cp\u003eRandom forest regressor\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003e0.54\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e0.51\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" width=\"180\"\u003e\n\u003cp\u003eUPDRS Part I\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"198\"\u003e\n\u003cp\u003eLASSO\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003e0.78\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e0.15\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" width=\"180\"\u003e\n\u003cp\u003eUPDRS Part II\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"198\"\u003e\n\u003cp\u003eLASSO\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.54\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.55\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" width=\"180\"\u003e\n\u003cp\u003eUPDRS Part III\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"198\"\u003e\n\u003cp\u003eRandom forest regressor\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003e0.65\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e0.25\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"5\" width=\"558\"\u003e\n\u003cp\u003e\u003cstrong\u003eGP_Motion modalities\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" width=\"180\"\u003e\n\u003cp\u003eUPDRS Total\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"198\"\u003e\n\u003cp\u003eRandom forest regressor\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003e0.94\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e0.04\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" width=\"180\"\u003e\n\u003cp\u003eUPDRS Part I\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"198\"\u003e\n\u003cp\u003eLASSO\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003e0.88\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e0.01\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" width=\"180\"\u003e\n\u003cp\u003eUPDRS Part II\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"198\"\u003e\n\u003cp\u003eRandom forest regressor\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003e0.94\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e0.14\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" width=\"180\"\u003e\n\u003cp\u003eUPDRS Part III\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"198\"\u003e\n\u003cp\u003eRandom forest regressor\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.87\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.20\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"5\" width=\"558\"\u003e\n\u003cp\u003e\u003cstrong\u003eGP_Sensors modalities\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" width=\"180\"\u003e\n\u003cp\u003eUPDRS Total\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"198\"\u003e\n\u003cp\u003eRandom forest regressor\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003e0.81\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e0.18\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" width=\"180\"\u003e\n\u003cp\u003eUPDRS Part I\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"198\"\u003e\n\u003cp\u003eRandom forest regressor\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003e0.87\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e0.01\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" width=\"180\"\u003e\n\u003cp\u003eUPDRS Part II\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"198\"\u003e\n\u003cp\u003eRandom forest regressor\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.70\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.34\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" width=\"180\"\u003e\n\u003cp\u003eUPDRS Part III\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"198\"\u003e\n\u003cp\u003eRandom forest regressor\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"95\"\u003e\n\u003cp\u003e0.76\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e0.02\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"5\" width=\"558\"\u003e\n\u003cp\u003eThe results indicate the average model performance on the test set. Model selection and training were performed using five-fold cross-validation to obtain MAE and the coefficient of determination \u003cem\u003eR\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e as performance metrics. MAE measures the average absolute difference between actual and predicted values. \u003cem\u003eR\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e indicates the model fitting to the variability of the dependent variable.\u003c/p\u003e\n\u003cp\u003eML, machine learning; MDS-UPDRS, movement disorder society-unified Parkinson\u0026rsquo;s disease rating scale; MAE, mean absolute error; PFL, physical function and lifestyle; and GP, gait parameter.\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003ctable width=\"586\"\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"6\" width=\"586\"\u003e\n\u003cp\u003e\u003cstrong\u003eTable 4. Performance of the PD severity subtype classification models for MDS-UPDRS scores based on multimodal data\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u003cstrong\u003eSeverity subtypes\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e\u003cstrong\u003eModality\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"236\"\u003e\n\u003cp\u003e\u003cstrong\u003eDigital biomarkers\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003e\u003cstrong\u003eMDS-UPDRS scores\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e\u003cstrong\u003eAccuracy\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(%)\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e\u003cstrong\u003eAUC\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(Mean)\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u003cstrong\u003e1 vs. 2\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"3\" width=\"85\"\u003e\n\u003cp\u003e\u003cstrong\u003eAll modalities\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"236\"\u003e\n\u003cp\u003e\u003cstrong\u003eRELB_PC1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(Turning: Time-domain acceleration and gyroscope measurements, and max angular velocity jerk)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRANK_PC3\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(Forward walking and turning: Maximum and sample entropy acceleration and jerk)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePSIS_PC1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(Turning: Time-domain acceleration and gyroscope measurements)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLANK_PC1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(Forward walking and turning: RMS and mean acceleration and gyroscope)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTurning _PC6\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(Contralateral temporal coordination)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRANK_PC1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(Forward \u0026amp; backward walking, and turning: Time-domain acceleration and gyroscope)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003eUPDRS Part III\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e93.3\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e0.73\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u003cstrong\u003e2 vs. 3\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"236\"\u003e\n\u003cp\u003e\u003cstrong\u003eT10_PC2\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(Forward walking and turning: Time-domain gyroscope)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRELB_PC1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePSIS_PC6\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(Forward walking: Time-domain gyroscope)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePSIS_PC1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLELB_PC3\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(Forward \u0026amp; backward walking: Time-domain acceleration, gyroscope, and angular velocity jerk)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLELB_PC1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(Forward walking and turning: Gyroscope and angular velocity jerk)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLANK_PC2\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(Forward \u0026amp; backward walking, and turning: RMS and mean acceleration and gyroscope)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eIPAQ\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003eUPDRS Total\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e60.0\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e0.76\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u003cstrong\u003e1 vs. 3\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"236\"\u003e\n\u003cp\u003e\u003cstrong\u003eLANK_PC1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRANK_PC1\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003eUPDRS Part III\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e100.0\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e0.99\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u003cstrong\u003e1 vs. 2\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"3\" width=\"85\"\u003e\n\u003cp\u003e\u003cstrong\u003ePFL modalities\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"236\"\u003e\n\u003cp\u003e\u003cstrong\u003eGrip strength_L and R\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003eUPDRS Part III\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e66.7\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e0.74\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u003cstrong\u003e2 vs. 3\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"236\"\u003e\n\u003cp\u003e\u003cstrong\u003e6MWT\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003eUPDRS Total\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e73.0\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e0.65\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u003cstrong\u003e1 vs. 3\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"236\"\u003e\n\u003cp\u003e\u003cstrong\u003eGrip strength_R\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMini-BEST\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSF-36 (total)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eIPAQ\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e5STST\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSF-36 (physical)\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003eUPDRS Total\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e75.0\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e0.74\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u003cstrong\u003e1 vs. 2\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"3\" width=\"85\"\u003e\n\u003cp\u003e\u003cstrong\u003eGP_Motion modalities\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"236\"\u003e\n\u003cp\u003e\u003cstrong\u003eBackward walking_PC4\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(Walking speed and stride length)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTurning_PC6\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTurning_PC2\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(\u003c/strong\u003eLeft and right double support phase)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003eUPDRS Part III\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e73.3\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e0.67\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u003cstrong\u003e2 vs. 3\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"236\"\u003e\n\u003cp\u003e\u003cstrong\u003eTurning_PC8\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(Left and right contralateral temporal coordination)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003eUPDRS Part III\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e73.0\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e0.64\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u003cstrong\u003e1 vs. 3\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"236\"\u003e\n\u003cp\u003e\u003cstrong\u003eTurning_PC1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(Walking speed and stride length)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eForward walking_PC3\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(Walking speed, stride length, and double support phase)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eBackward walking_PC4\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTurning_PC2\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003eUPDRS Part II\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e83.0\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e0.72\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u003cstrong\u003e1 vs. 2\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"3\" width=\"85\"\u003e\n\u003cp\u003e\u003cstrong\u003eGP_Sensors modalities\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"236\"\u003e\n\u003cp\u003e\u003cstrong\u003eLELB_PC4\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(Turning: RMS and mean acceleration)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRELB_PC1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLANK_PC1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRANK_PC3\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePSIS_PC1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePSIS_PC2\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(Forward walking and turning: Maximum and sample entropy acceleration and jerk)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003eUPDRS Part III\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e86.7\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e0.72\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u003cstrong\u003e2 vs. 3\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"236\"\u003e\n\u003cp\u003e\u003cstrong\u003eLELB_PC1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLELB_PC2\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(Forward walking and turning: Maximum acceleration, jerk, and angular velocity jerk)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLELB_PC3\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRELB_PC1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLANK_PC1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRANK_PC1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePSIS_PC1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eT10_PC2\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003eUPDRS Part II\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e80.0\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e0.83\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"57\"\u003e\n\u003cp\u003e\u003cstrong\u003e1 vs. 3\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"236\"\u003e\n\u003cp\u003e\u003cstrong\u003eRANK_PC1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLANK_PC1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePSIS_PC1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLELB_PC2\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLELB_PC3\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"85\"\u003e\n\u003cp\u003eUPDRS Part II\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e100.0\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"61\"\u003e\n\u003cp\u003e0.91\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"6\" width=\"586\"\u003e\n\u003cp\u003eTime-domain indicated the root mean square (RMS), mean, and max.\u003c/p\u003e\n\u003cp\u003ePD, Parkinson's disease; MDS-UPDRS, movement disorder society-unified Parkinson\u0026rsquo;s disease rating scale; AUC, area under the curve; PFL, physical function and lifestyle; GP, gait parameter; PC, principal components; RELB, right lateral humeral epicondyle; RANK, right lateral malleolus; PSIS, posterior superior iliac spine; LANK, left lateral malleolus; T10, 10th thoracic vertebra; LELB, left lateral humeral epicondyle; IPAQ, international physical activity questionnaire; L, left; R, right; 6MWT, 6 min walking test; Mini-BEST, mini-balance evaluation systems test; SF-36, 36-item short form health survey; and 5STST, five times sit-to-stand test.\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"journal-of-neuroengineering-and-rehabilitation","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"jner","sideBox":"Learn more about [Journal of NeuroEngineering and Rehabilitation](http://jneuroengrehab.biomedcentral.com/)","snPcode":"12984","submissionUrl":"https://submission.nature.com/new-submission/12984/3","title":"Journal of NeuroEngineering and Rehabilitation","twitterHandle":"@BioMedCentral","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Parkinson's disease, Severity subtype, Multimodal data, Machine learning, Clustering, Digital biomarker","lastPublishedDoi":"10.21203/rs.3.rs-5523724/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5523724/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eClassifying and predicting Parkinson's disease (PD) is challenging because of its diverse subtypes based on severity levels. Currently, identifying objective biomarkers associated with disease severity that can distinguish PD subtypes in clinical trials is necessary. This study aims to address the clinical applicability and heterogeneity of PD using PD severity subtypes classification and digital biomarker development by combining objective multimodal data with machine learning (ML) approaches.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eWe analyzed datasets that combine clinical characteristics, physical function and lifestyle data, gait parameters in motion analysis systems, and wearable sensors collected from persons with PD (n\u0026thinsp;=\u0026thinsp;102) to perform clustering for subtype classification.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eWe identified three PD severity subtypes, each exhibiting different patterns of clinical severity, with the severity increasing as it progressed from clusters 1 to 3. We found significant mutual information between all/single modalities and the unified PD rating scale scores, identifying potential modalities with high feature importance using ML. Among all modalities, the principal components of gait parameters derived from wearable sensors were identified as the most associated indicators of PD severity. A model utilizing the first principal component of the left and right ankle achieved perfect classification with an area under the curve of 1.0, accurately distinguishing clinically severe subtypes from mild subtypes of PD. These findings suggest that gait features in both ankles can reflect asymmetry factors associated with PD severity subtypes, which contributes to high classification performance.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eDigital biomarkers obtained from wearable sensors attached bilaterally to body segments demonstrate potential for classifying PD severity subtypes and tracking disease progression. Our findings emphasized the clinical value of sensor-based gait analysis in PD management, which suggested its integration into personalized monitoring systems and therapeutic interventions for persons with PD.\u003c/p\u003e","manuscriptTitle":"Using machine learning to identify Parkinson’s disease severity subtypes with multimodal data","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-04-21 12:32:13","doi":"10.21203/rs.3.rs-5523724/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Accepted","date":"2025-05-13T05:54:07+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-05-06T08:48:21+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"316573332336423845453782815196724393324","date":"2025-04-18T07:33:00+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-04-17T20:04:50+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-04-10T12:05:50+00:00","index":"","fulltext":""},{"type":"submitted","content":"Journal of NeuroEngineering and Rehabilitation","date":"2025-04-09T11:21:38+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"journal-of-neuroengineering-and-rehabilitation","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"jner","sideBox":"Learn more about [Journal of NeuroEngineering and Rehabilitation](http://jneuroengrehab.biomedcentral.com/)","snPcode":"12984","submissionUrl":"https://submission.nature.com/new-submission/12984/3","title":"Journal of NeuroEngineering and Rehabilitation","twitterHandle":"@BioMedCentral","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"38fa7293-76e0-424b-8ec8-49b4201e4631","owner":[],"postedDate":"April 21st, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-06-09T16:06:27+00:00","versionOfRecord":{"articleIdentity":"rs-5523724","link":"https://doi.org/10.1186/s12984-025-01648-2","journal":{"identity":"journal-of-neuroengineering-and-rehabilitation","isVorOnly":false,"title":"Journal of NeuroEngineering and Rehabilitation"},"publishedOn":"2025-06-02 15:57:24","publishedOnDateReadable":"June 2nd, 2025"},"versionCreatedAt":"2025-04-21 12:32:13","video":"","vorDoi":"10.1186/s12984-025-01648-2","vorDoiUrl":"https://doi.org/10.1186/s12984-025-01648-2","workflowStages":[]},"version":"v1","identity":"rs-5523724","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5523724","identity":"rs-5523724","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}