Multimodal Machine Learning Approach for Predicting Cognitive Decline in People with Parkinson’s Disease

Multimodal Machine Learning Approach for Predicting Cognitive Decline in People with Parkinson’s Disease | 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 Article Multimodal Machine Learning Approach for Predicting Cognitive Decline in People with Parkinson’s Disease Bohyun Kim, Changhong Youm, Sang-Myung Cheon, Hwayoung Park, Hyejin Choi, and 2 more This is a preprint; it has not been peer reviewed by a journal.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-6695263/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract This study developed machine learning models to predict cognitive decline in individuals with Parkinson’s disease (PD) by integrating clinical characteristics and gait-derived digital biomarkers. Using data from 102 patients diagnosed with PD, we trained least absolute shrinkage and selection operator regression and eXtreme Gradient Boosting models on multimodal features, including clinical characteristics, physical function, lifestyle factors, and gait-derived features. Key predictors included Mini-Mental State Examination scores and gait biomarkers such as stride length of the left foot during preferred speed leftward turning, maximum acceleration of the right ankle during faster speed leftward turning, maximum jerk of the right ankle during forward walking, and the maximum gyroscope at the posterior superior iliac spine during forward walking. Stepwise regression explained 61.7% of the variance in Montreal Cognitive Assessment scores (p < 0.05). A logistic regression classifier using ten selected features achieved 76.5% accuracy and an area under the curve of 0.895 in identifying individuals with cognitive decline. These findings suggest that combining standard cognitive assessments with quantitative gait analysis enhances prediction and classification of cognitive impairment in PD, offering a clinically applicable strategy. Health sciences/Biomarkers Health sciences/Diseases/Neurological disorders/Parkinsons disease Parkinson's disease cognitive decline digital biomarkers multimodal features machine learning wearable sensors Figures Figure 1 Figure 2 Figure 3 INTRODUCTION Parkinson’s disease (PD) is a progressive neurodegenerative disorder marked by both motor and non-motor symptoms, which substantially impair quality of life in affected individuals [ 1 ] . While hallmark motor symptoms include bradykinesia, rigidity, resting tremor, and postural instability, non-motor symptoms—particularly cognitive impairment—can be equally disabling and may even precede the onset of motor symptoms [ 2 ] . Approximately 42.5% of newly diagnosed individuals with PD present with mild cognitive impairment, with 70–80% progressing to dementia over 15–20 years [ 3 , 4 ] . Recent studies have increasingly focused on predicting cognitive decline in PD by integrating clinical evaluations, biomarkers, neuropsychological tests, and longitudinal data [ 5 – 12 ] . These methods promote early identification, timely therapeutic intervention, and enhanced disease management [ 13 ] . Various cognitive assessment scales have been utilized to predict cognitive decline in PD, and several studies have investigated the predictive abilities of these approaches [ 14 , 15 ] . The Montreal Cognitive Assessment (MoCA), though widely adopted for evaluating memory, executive function, and verbal fluency in PD [ 5 , 8 , 9 , 16 , 17 ] , is limited by its intermittent administration, potentially failing to detect subtle or early cognitive changes [ 18 ] . Emerging evidence suggests a link between motor and cognitive symptoms in PD, mediated by overlapping neuropathology in the basal ganglia and prefrontal dopaminergic circuits [ 19 ] . Specifically, gait disturbances, postural instability, and impaired turning performance have been associated with cognitive decline, indicating that movement-based digital metrics could serve as early biomarkers [ 20 , 21 ] . Recent advances in digital healthcare have enabled the use of wearable sensors for non-invasive evaluation of gait impairments in individuals with PD [ 22 – 24 ] . Metrics derived from these sensors—such as stride variability, turning angular velocity, and entropy-based indicators of gait regularity—have proven sensitive in detecting subtle cognitive-motor dysfunctions often overlooked by traditional clinical scales [ 23 , 25 , 26 ] . These devices offer continuous, real-world gait monitoring, providing high-resolution, objective data that address key limitations of conventional assessments [ 27 ] . Previous studies relying exclusively on standard clinical tools have demonstrated poor reproducibility across patient cohorts, highlighting the need to incorporate objective, multimodal data to enhance the prediction and classification of cognitive decline in PD [ 28 , 29 ] . Machine learning (ML) techniques have emerged as effective tools for integrating high-dimensional, heterogeneous multimodal data, enabling the detection of complex motor–cognitive interactions and the systematic selection of predictive features [ 28 , 30 , 31 ] . Recent studies employing ML approaches such as least absolute shrinkage and selection operator (LASSO) regression, Random Forest, and eXtreme Gradient Boosting (XGBoost) have demonstrated superior accuracy in predicting PD symptom severity and progression compared with traditional statistical methods, highlighting their value in developing clinically viable digital biomarkers [ 28 , 30 , 32 ] . However, most existing ML models either insufficiently incorporate sensor-derived gait features or fail to adequately examine their direct relationship with cognitive decline in PD. Moreover, despite their promise as screening tools, current digital biomarker models often lack interpretability and clinical utility [ 33 ] . Thus, developing robust, interpretable, and clinically relevant models that integrate digital biomarkers with conventional clinical assessments is essential for improving the early identification and management of cognitive decline in PD. Despite recent progress, critical gaps remain in incorporating gait features derived from wearable sensors into clinically interpretable models of cognitive impairment in PD. This study aims to enhance the accuracy of predicting and classifying cognitive decline in PD by applying ML techniques to multimodal data, including clinical characteristics, physical function, lifestyle factors, and gait-derived features. Specifically, the objectives are to: Investigate the associations between multimodal features—such as clinical characteristics, physical function, lifestyle factors, and gait-derived features—and cognitive decline in individuals with PD. Develop and evaluate ML models, including LASSO regression and XGBoost, to predict MoCA scores using multimodal data. Perform stepwise multiple linear regression to identify independent predictors of cognitive performance. Assess logistic regression models to classify individuals with and without cognitive impairment based on selected multimodal features. We hypothesize that multimodal features—particularly gait-derived digital biomarkers—will show significant associations with cognitive status in PD. Furthermore, integrating these features with ML techniques is expected to improve MoCA score prediction and the classification of cognitive impairment, offering insights that extend beyond conventional clinical assessments. Results Demographic and clinical characteristics associated with cognitive decline in PD Table 1 presents the demographic and clinical characteristics of the study participants. Individuals were stratified into two groups based on MoCA scores (< 26): PD with normal cognition (PD-NC; n = 60) and PD with cognitive decline (PD-CD; n = 42). The analysis revealed significant demographic and clinical differences between the two groups. Specifically, individuals in the PD-CD group were older and had more severe motor symptoms, as indicated by higher Hoehn and Yahr stages and elevated scores on both the total Unified Parkinson’s Disease Rating Scale (UPDRS) and the motor examination subscale (UPDRS Part III). Additionally, the PD-CD group demonstrated significantly lower scores on cognitive assessments compared with the PD-NC group. Table 1 Demographics and clinical characteristics of the study participants PD-NC (n = 60) PD-CD (n = 42) p -value Mean ± standard deviation Sex (male/female) 24/36 22/20 Age (years) 65.83 ± 7.52 71.24 ± 5.36 < 0.001 a Height (cm) 161.47 ± 8.16 158.79 ± 8.66 0.114 a Body weight (kg) 62.95 ± 11.89 62.82 ± 8.67 0.731 b BMI (kg/m 2 ) 24.01 ± 3.29 24.89 ± 2.67 0.063 b Symptom duration (years) 5.44 ± 3.26 6.08 ± 5.58 0.600 b Treatment duration (years) 4.19 ± 3.18 5.06 ± 5.35 0.303 b L-dopa equivalent dose (mg/day) 542.89 ± 316.50 549.12 ± 250.88 0.514 b Hoehn and Yahr scale (stages) 1.83 ± 0.69 2.14 ± 0.72 0.032 b UPDRS total (scores) 47.78 ± 19.83 64.13 ± 24.40 < 0.001 a UPDRS part I (scores) 9.53 ± 5.33 10.88 ± 5.00 0.180 b UPDRS part II (scores) 11.70 ± 6.38 14.60 ± 8.06 0.071 b UPDRS part III (scores) 24.62 ± 12.64 36.35 ± 16.10 < 0.001 b UPDRS part IV (scores) 1.93 ± 2.73 1.40 ± 2.58 0.212 b MMSE (scores) 28.55 ± 1.32 26.60 ± 2.46 < 0.001 b MoCA (scores) 27.83 ± 1.24 23.00 ± 2.36 < 0.001 b The data are presented as mean ± standard deviation. Statistically significant differences between groups are shown in bold ( p < 0.05). PD-NC, Parkinson’s disease with normal cognition; PD-CD, Parkinson’s disease with cognitive decline; BMI, Body mass index; UPDRS, Unified Parkinson’s Disease Rating Scale; MMSE, Mini-Mental State Examination; MoCA, Montreal Cognitive Assessment. a Independent samples t -test. b Mann–Whitney U test. Performance evaluation of LASSO and XGBoost models for MoCA score prediction The predictive performance of the LASSO regression and XGBoost models was evaluated using cross-validation and an independent test set. Table 2 summarizes the results based on mean absolute error (MAE) and coefficient of determination ( R 2 ) for both models. Table 2 Model performance and feature importance Model CV_MAE CV_ R² Test_MAE Test_ R² LASSO 0.525 0.470 0.613 0.370 XGBoost 0.596 0.294 0.628 0.383 LASSO and XGBoost were used as predictive models. CV_MAE, Mean absolute error during cross-validation; CV_ R² , Coefficient of determination ( R ²) during cross-validation, indicating model fit; Test_MAE, Mean absolute error on the test set; Test_ R² , R ² on the test set, indicating predictive accuracy. LASSO regression achieved a cross-validation MAE (CV_MAE) of 0.525 and a cross-validation R 2 (CV_ R 2 ) of 0.470. In contrast, XGBoost produced a higher CV_MAE of 0.596 and a lower CV_ R 2 of 0.294. For the test set, LASSO recorded a MAE of 0.613 and a R 2 of 0.370, while XGBoost showed a comparable MAE of 0.628 but a marginally higher R 2 of 0.383. These results suggest that although LASSO demonstrated stronger generalization in cross-validation, XGBoost performed slightly better on unseen data, as indicated by its higher Test_ R 2 value. Feature importance was computed using a weighted average approach that integrated the relative contributions of both models. The Korean Mini-Mental State Examination (MMSE) emerged as the strongest predictor (importance = 0.385), followed by motor examination scores (UPDRS Part III, importance = 0.152) and contralateral temporal coordination of the left foot during faster speed rightward turning (TurnFR_CTL; 0.085). Several gait-derived features also contributed substantially to MoCA score prediction, including maximum acceleration of the right ankle during faster speed leftward turning (TurnFL_RANK_MaxAcc; 0.072), maximum jerk of the right ankle during forward walking (FW_RANK_MaxJerk; 0.062), and stride length of the left foot during preferred speed leftward turning (TurnPL_SLL; 0.056) (Fig. 1 ). Regression analysis and network-based feature relationships for MoCA score prediction Stepwise multiple linear regression identified five significant predictors of cognitive performance, as measured by MoCA, yielding an adjusted R 2 = 0.617, p < 0.05. Collectively, these features explained 61.7% of the variance in MoCA scores. The final model included both neuropsychological (MMSE) and gait-derived features, including TurnPL_SLL, TurnFL_RANK_MaxAcc, FW_RANK_MaxJerk, and the maximum gyroscope at the posterior superior iliac spine during forward walking (FW_PSIS_MaxGyr). The model demonstrated strong predictive power, as indicated by a progressive increase in adjusted R 2 across regression iterations. Table 3 reports the corresponding regression coefficients and their statistical significance. Table 3 Stepwise multiple linear regression analysis predicting MoCA scores TurnPL, Preferred speed leftward turning; TurnFL, Faster speed leftward turning; FW, Forward walking; SLL, Stride length of the left foot; RANK, Right ankle; PSIS, Posterior superior iliac spine; MaxAcc, Maximum acceleration; MaxJerk, Maximum jerk; MaxGyr, Maximum gyroscope; SE, Standard error; p < 0.05. Features β (SE) t p -value R² MMSE 0.577(0.065) 8.853 < 0.001 0.617 TurnPL_SLL, 0.225(0.066) 3.403 0.001 TurnFL_RANK_MaxAcc, 0.218(0.068) 3191 0.002 FW_RANK_MaxJerk -0.190(0.068) -2.783 0.006 FW_PSIS_MaxGyr 0.135(0.067) 2.012 0.047 Beyond regression analysis, a network-based visualization revealed that MMSE was the most influential predictor, exhibiting the highest regression coefficient ( β = 0.577) and the strongest connection to MoCA within the feature network, highlighting its central role in cognitive prediction. Additionally, TurnPL_SLL ( β = 0.225) was directly associated with MoCA scores and was closely linked to other gait-derived features in the network (Fig. 2 ). Identification of digital biomarkers based on classification models Ten optimal digital biomarkers for classifying PD-CD group were identified using recursive feature elimination with cross-validation, based on features previously selected by the LASSO and XGBoost models. The final biomarker panel included clinical characteristics —age, MMSE, and UPDRS_III—along quantitative gait-derived features: including TurnFR_CTL, TurnPL_SLL, stride length of the left foot during preferred speed rightward turning (TurnPL_SLR), TurnFL_RANK_MaxAcc, sample entropy of acceleration at the posterior superior iliac spine during faster speed rightward turning (TurnFR_PSIS_SampEnAcc), FW_RANK_MaxJerk, and FW_PSIS_MaxGyr. The logistic regression classifier achieved 76.5% cross-validated accuracy, with an area under the receiver operating characteristic curve (AUC–ROC) of 0.895, indicating strong discriminative performance for distinguishing between individuals with and without cognitive decline (Table 4 , Fig. 3 ). Table 4 Classification performance of PD-CD identification models for MoCA scores based on multimodal features Digital biomarkers Accuracy (%) AUC (Mean) Age MMSE UPDRS_III TurnFR_CTL TurnPL_SLL TurnPL_SLR TurnFL_RANK_MaxAcc TurnFR_PSIS_SampEnAcc FW_RANK_MaxJerk FW_PSIS_MaxGyr 76.5 0.895 PD-CD, Parkinson’s disease with cognitive decline; MoCA, Montreal Cognitive Assessment; AUC, Area under the curve; MMSE, Mini-Mental State Examination; UPDRS, Unified Parkinson’s Disease Rating Scale; TurnFR, Faster speed rightward turning; TurnPL, Preferred speed leftward turning; TurnFL, Faster speed leftward turning; FW, Forward walking; CTL, contralateral temporal coordination of the left foot; SLL, Stride length of the left foot; SLR, Stride length of the right foot; RANK, Right ankle; PSIS, posterior superior iliac spine; MaxAcc, Maximum acceleration; SampEnAcc, Sample entropy of acceleration; Maxjerk, Maximum jerk; MaxGyr, Maximum gyroscope. DISCUSSION This study evaluated the utility of multimodal features—including clinical characteristics, physical function, lifestyle factors, and gait-derived digital biomarkers—for predicting cognitive performance and classifying cognitive status in individuals with PD. By applying ML and regression-based approaches, we identified key predictors of MoCA scores, highlighting the clinical significance of gait-derived features in assessing cognitive decline. Key findings and model performance The LASSO regression and XGBoost models demonstrated moderate-to-strong predictive performance for MoCA scores (LASSO CV_ R² = 0.470, Test_ R ² = 0.370; XGBoost CV_ R² = 0.294, Test_ R² = 0.383), underscoring the complementary strengths of linear and nonlinear modeling approaches. LASSO exhibited greater consistency across validation folds, while XGBoost showed slightly better generalizability to unseen data, likely owing to its capacity to model nonlinear relationships [ 32 , 34 ] . Feature importance analysis identified the MMSE as the most influential predictor of MoCA scores (importance = 0.385), followed by UPDRS Part III. Although MMSE and MoCA share conceptual overlap, they target distinct cognitive domains: MoCA offers greater sensitivity to mild cognitive impairment, especially in executive and visuospatial function, whereas MMSE is widely used in clinical practice for its simplicity and established validity [ 35 ] . Further analysis revealed a significant drop in model performance when MMSE was excluded (LASSO Test_ R² = -0.078; XGBoost Test_ R² =- 0.282), underscoring its unique predictive value [ 36 , 37 ] . The use of regularization techniques such as LASSO mitigates concerns of multicollinearity, supporting MMSE’s inclusion as both a statistically and clinically relevant predictor [ 38 ] . The UPDRS Part III, which evaluates motor symptom severity, emerged as the second most influential predictor, aligning with previous studies that link motor dysfunction—particularly bradykinesia, postural instability, and axial impairment—to PD-CD [ 39 ] . These findings reinforce the role of overlapping motor–cognitive neural circuits, especially within the basal ganglia and prefrontal cortex [ 2 , 40 , 41 ] . Consequently, axial and postural impairments captured by UPDRS Part III may indicate early cognitive vulnerability, underscoring its utility in integrated motor–cognitive assessments [ 42 ] . Gait-derived digital biomarkers Stepwise multiple linear regression identified four significant gait-derived features, which, alongside the MMSE, collectively accounted for 61.7% of the variance in MoCA scores. These features captured core dimensions of gait control linked to cognitive processes: spatial symmetry (TurnPL_SLL), ankle-level acceleration during directional transitions (TurnFL_RANK_MaxAcc), movement smoothness via jerk (FW_RANK_MaxJerk), and trunk-level rotational dynamics (FW_PSIS_MaxGyr). Distal ankle features—TurnFL_RANK_MaxAcc and FW_RANK_MaxJerk—reflect neuromuscular responsiveness and anticipatory motor planning under dynamic postural demands, underscoring the involvement of executive and attentional processes in coordinating lower-limb control during complex movement transitions [ 43 – 46 ] . Proximal trunk rotation (FW_PSIS_MaxGyr) captures axial stability and cognitive-motor integration, key functions typically disrupted in early cognitive decline [ 47 – 49 ] . Moreover, impairments in gait smoothness, symmetry, and anticipatory postural adjustments—especially during demanding tasks such as turning or dynamic walking—serve as sensitive indicators of early cognitive dysfunction, reinforcing their potential as clinically meaningful digital biomarkers [ 26 , 31 , 44 , 48 , 49 ] . These findings highlight the combined contribution of distal (ankle-level) and proximal (trunk-level) kinematic features in revealing neuromechanical disturbances associated with cognitive impairment in PD. Network analysis and interconnected biomarkers Network-based visualization further illustrated the functional interconnections among key predictive features. The MMSE exhibited strong associations with gait-derived features such as TurnFL_RANK_MaxAcc and FW_RANK_MaxJerk. These associations underscore the integration of cognitive function with dynamic limb control, reflecting the influence of executive and attentional processes on rapid acceleration and smooth transitional movement patterns [ 44 – 46 ] . TurnPL_SLL, a measure of stride length during preferred speed leftward turning, not only had a direct association with MoCA scores but also demonstrated substantial connectivity with trunk rotation metrics (FW_PSIS_MaxGyr), reinforcing its role in spatial coordination and postural control [ 44 , 50 ] . Notably, TurnFL_RANK_MaxAcc emerged as a central node in the predictive network, exhibiting bidirectional connections with both FW_RANK_MaxJerk and FW_PSIS_MaxGyr. This centrality suggests that rapid directional transitions, ankle jerk dynamics, and trunk rotation are interdependent and likely modulated by shared neuro-mechanical processes rooted in executive control mechanisms, commonly affected during early cognitive decline [ 26 , 31 , 46 , 50 , 51 ] . Collectively, the network topology suggests that cognitive status in Parkinson’s disease is underpinned by an integrated system of interconnected motor features rather than isolated predictors, supporting the development of holistic digital biomarker strategies. Classification of cognitive decline The optimized logistic regression model achieved an average classification accuracy of 76.5% (AUC = 0.895) in classifying PD-CD based on ten multimodal predictors. Informative features included clinical variables—age, MMSE, and UPDRS_III—as well as gait-derived digital biomarkers reflecting spatiotemporal coordination (TurnFR_CTL), spatial symmetry (TurnPL_SLL, TurnPL_SLR), ankle dynamics (TurnFL_RANK_MaxAcc, FW_RANK_MaxJerk), and pelvic rotational motion (FW_PSIS_MaxGyr). This integration enabled detailed characterization of cognitive-motor interactions, particularly emphasizing deficits in executive and attentional function as reflected in gait irregularities [ 52 , 53 ] . Incorporating sensor-based features such as TurnFR_PSIS_SampEnAcc and FW_RANK_MaxJerk enhanced model sensitivity, reinforcing the utility of gait irregularity as an early marker of cognitive dysfunction [ 22 , 54 , 55 ] . The interpretability and clinical feasibility of the logistic regression model underscore its potential as a screening tool for early-stage cognitive decline and diagnostic stratification in routine clinical settings. Collectively, these findings validate the clinical value of integrating wearable sensor-derived gait metrics with conventional cognitive assessments to support earlier identification and personalized management of cognitive impairment in PD. Clinical implications and limitations The findings demonstrate that integrating clinical characteristics with gait-derived digital biomarkers enhances the identification of cognitive impairment in individuals with PD. The models developed are both interpretable and computationally efficient, supporting their feasibility for clinical deployment. The incorporation of continuous, non-invasive gait monitoring provides complementary value to traditional cognitive screening tools, enabling earlier diagnosis, individualized therapeutic planning, and longitudinal disease monitoring. Nevertheless, several limitations of this study should be acknowledged. First, the relatively small sample size may limit the generalizability of the findings, underscoring the need for future validation in larger, more diverse, and multicenter cohorts. Second, the cross-sectional study design prevents evaluation of cognitive trajectories over time; longitudinal research is needed to assess the predictive value of multimodal biomarkers in tracking cognitive decline. Third, because data collection occurred exclusively during patients “"ON” medication states, motor and cognitive assessments may not reflect fluctuations across the medication cycle. Future studies should evaluate performance in both “ON” and “OFF” states to enable more comprehensive cognitive–motor profiling. Finally, as gait data were collected under controlled laboratory conditions, the findings may not fully represent everyday variability. Incorporating real-world gait monitoring could enhance ecological validity and strengthen clinical applicability. CONCLUSION This study demonstrates that multimodal features—including clinical characteristics, physical function, lifestyle factors, and gait-derived features—can effectively predict and classify PD-CD. ML models, particularly LASSO regression and XGBoost, identified the MMSE and several gait-derived features—TurnPL_SLL, TurnFL_RANK_MaxAcc, FW_RANK_MaxJerk, and FW_PSIS_MaxGyr —as key predictors. These digital biomarkers, validated through regression analysis, network modeling, and classification performance, reflect clinically relevant cognitive–motor associations. Together, they offer a practical and interpretable foundation for early detection and targeted intervention. Future longitudinal and real-world studies are essential to refine these predictive models and establish their utility in clinical settings. METHODS Participants Data from 102 individuals with a clinician-confirmed diagnosis of idiopathic PD were included, based on the UK Parkinson’s Disease Society Brain Bank Clinical Diagnostic Criteria. Inclusion criteria required participants to have mild-to-moderate idiopathic PD, be on anti-Parkinsonian medication, and be able to stand and walk independently during clinical assessments. Exclusion criteria included any comorbid neurological, orthopedic, or psychiatric disorders. The participants had a mean age of 68.1 years, and 54.9% were female. Among them, 49.0% were classified as Hoehn and Yahr stage 2. The mean disease duration was 5.7 years, and the average UPDRS_III score was 29.4. Based on MoCA scores (< 26), participants were classified into two groups: PD-NC (n = 60) and PD-CD (n = 42) (Table 1 ). The study was approved by the Institutional Review Board of Dong-A University Medical Center (Approval No. DAUHIRB-22-089). All participants were fully informed of the study purpose and procedures and provided written informed consent. Experimental procedures Participants completed two laboratory visits for comprehensive multimodal evaluation, including assessments of demographic and clinical characteristics, physical function, lifestyle factors, and gait parameters. For individuals receiving levodopa treatment, testing was conducted during the ‘ON’ medication state, approximately 2–3 hours after dosing. The study assessed the following categories of multimodal features. Clinical measurements included 22 features, with PD severity evaluated using the Hoehn and Yahr scale and the UPDRS comprising total and Parts I–IV scores. Part I assessed non-motor aspects of daily living, including cognitive impairment and mood disturbances. Part II measured motor-related daily activities, such as speech, handwriting, and hygiene. Part III evaluated motor symptom severity, focusing on tremors, rigidity, and bradykinesia. Part IV captured motor complications related to long-term dopaminergic therapy, such as dyskinesia and motor fluctuations. Clinical history, including symptom and treatment duration and levodopa-equivalent daily dose, was collected to reflect disease progression and pharmacological management. Cognitive function was assessed using the MMSE and the MoCA. The Beck Depression Inventory and Beck Anxiety Inventory were used to measure depressive and anxiety symptoms, respectively. The PD Sleep Scale-2 assessed sleep disturbances, while the Epworth Sleepiness Scale evaluated daytime sleepiness. Fatigue severity was measured using the Fatigue Severity Scale. Fall risk and fear of falling were evaluated based on fall history and the Korean version of the Falls Efficacy Scale. Self-efficacy for exercise was assessed using the Self-Efficacy for Exercise Scale. The New Freezing of Gait Questionnaire measured the severity of freezing episodes. The Non-Motor Symptoms Scale provided a comprehensive evaluation of a range of non-motor symptoms. Lastly, the PD Questionnaire-39 assessed health-related quality of life. The physical function and lifestyle assessments comprised 10 features. Grip strength was measured bilaterally, along with performance-based tests including the Five-Times Sit-to-Stand Test, Six-Minute Walk Test, Short Physical Performance Battery, and Mini-Balance Evaluation Systems Test. Lifestyle-related measures included the Nutrition Quotient, 36-item Short-Form Health Survey, which provided total, physical, and mental health scores, and the International Physical Activity Questionnaire. Gait-derived features obtained from motion capture and wearable sensors comprised 38 and 360 features, respectively. Participants completed three walking tasks: (1) forward and (2) backward straight walking at a self-selected speed, and (3) 360° turning to the left and right at both preferred and faster speeds. Kinematic data were collected using a full-body marker-based system (Plug-in Gait model, Vicon Motion Systems, Oxford Metrics, UK) and six inertial measurement units (Xsens DOT, Movella Technologies, Enschede, Netherlands). The sampling rates were 100 Hz for the Vicon system and 60 Hz for the Xsens DOT. Motion capture–derived features included walking speed, stride length, double-support phase, and contralateral temporal coordination. In contrast, wearable sensor-derived features encompassed maximum jerk and angular velocity jerk, mean and maximum acceleration, mean and maximum gyroscope values, root mean square acceleration and gyroscope signals, and sample entropy of acceleration and gyroscope data (Table 5 ). Table 5 Gait features derived from wearable sensors Feature Mathematical expression Maximum jerk ( \(\:\text{m}/{s}^{2}\) ) \(\:\frac{dAcceleration}{dt}\) Maximum angular velocity jerk ( \(\:\text{r}\text{a}\text{d}/{s}^{2}\) ) \(\:\frac{dAngular\:Velocity}{dt}\) Mean acceleration ( \(\:\text{m}/{s}^{2}\) ) \(\:mean\left(\sqrt{{acc}_{x}^{2}+{acc}_{y}^{2}+{acc}_{z}^{2}}\right)\) Mean gyroscopes ( \(\:\text{r}\text{a}\text{d}/\text{s}\) ) \(\:mean\left(\sqrt{{gyr}_{x}^{2}+{gyr}_{y}^{2}+{gyr}_{z}^{2}}\right)\) Maximum acceleration ( \(\:\text{m}/{s}^{2}\) ) \(\:max\left(\sqrt{{acc}_{x}^{2}+{acc}_{y}^{2}+{acc}_{z}^{2}}\right)\) Maximum gyroscopes ( \(\:\text{r}\text{a}\text{d}/\text{s}\) ) \(\:max\left(\sqrt{{gyr}_{x}^{2}+{gyr}_{y}^{2}+{gyr}_{z}^{2}}\right)\) RMS acceleration ( \(\:\text{m}/{s}^{2}\) ) \(\:\sqrt{mean({acc}_{x}^{2}+{acc}_{y}^{2}+{acc}_{z}^{2})}\) RMS gyroscopes ( \(\:\text{r}\text{a}\text{d}/\text{s}\) ) \(\:\sqrt{mean({gyr}_{x}^{2}+{gyr}_{y}^{2}+{gyr}_{z}^{2})}\) Sample entropy acceleration \(\:-\text{log}\left(\frac{Aacc}{Bacc}\right)\) Sample entropy gyroscopes \(\:-\text{log}\left(\frac{Agyr}{Bgyr}\right)\) RMS, root mean square; acc, acceleration; gyr, gyroscope; A: Total number of similar vector pairs of length m + 1 (i.e., m + 1 consecutive data points) that remain within a defined tolerance r; B, total number of similar vector pairs of length m (i.e., m consecutive data points) within the same tolerance r. Preprocessing The dataset was first reviewed to understand its structure and identify inconsistencies. Preprocessing consisted of two main steps. First, missing values were addressed using k-nearest neighbors’ imputation. A k-nearest neighbor imputer with five neighbors estimated missing values based on the mean of the five most similar data points. Second, z-score normalization was applied to standardize all features, transforming them to have a mean of zero and a standard deviation of one. This ensured uniform feature scaling and improved comparability across variables. Feature selection and predictive modeling with LASSO and XGBoost: Performance evaluation and importance analysis An initial feature selection step was performed using correlation analysis between all multimodal features and MoCA scores. To accommodate different data distributions, Pearson correlation was used for parametric variables, and Spearman correlation for non-parametric variables. This dual approach allowed a comprehensive assessment of both linear and monotonic relationships between features and MoCA scores. Only variables showing statistically significant correlations ( p < 0.05) were retained for subsequent predictive modeling. MoCA scores were converted into a one-dimensional array. The dataset was then split into training (80%) and testing (20%) subsets using stratified random sampling to preserve the distribution of cognitive function scores. LASSO regression was used to identify the most relevant features for predicting cognitive function in individuals with PD. LASSO is a regularization technique that performs feature selection by shrinking the coefficients of less informative predictors to zero. A grid search with five-fold cross-validation was applied to identify the optimal regularization parameter ( α ), based on the highest R 2 score. The α search space included the values {0.01, 0.1, 1, 10}. After determining the optimal α , the LASSO model was retrained using the features with non-zero coefficients. The final dataset used for subsequent modeling included only these selected features. Two predictive models were developed to estimate MoCA scores: LASSO regression—a linear model with L1 regularization that enhances interpretability and performs feature selection—and XGBoost regression, a gradient-boosting method optimized for capturing nonlinear relationships and feature interactions. For the XGBoost model, hyperparameter tuning was conducted using grid search with five-fold cross-validation. The search space included the number of estimators {100, 200, 300}, maximum tree depth {3, 6, 9}, and learning rate {0.01, 0.1, 0.2}. Model performance was assessed using MAE and R² on the cross-validation and test sets. Each model was implemented within a data-processing pipeline, incorporating feature scaling via StandardScaler prior to model training. Model performance was evaluated on the test dataset, and key metrics were recorded. Final predictions from both models were stored for downstream analysis. Feature importance was assessed using multiple approaches: for LASSO, importance was derived from the absolute values of the regression coefficients; for XGBoost, feature importance was computed using the model’s built-in importance attribute. To generate a unified importance ranking, a weighted averaging method was applied to combine scores from both models. Weights were proportional to each model’s R² score on the test set, ensuring that the final ranking reflected relative predictive performance [ 56 , 57 ] . Stepwise linear regression was conducted to further examine the relationship between selected features and MoCA scores. This method iteratively added or removed predictors based on statistical significance, retaining only those that significantly contributed to the final model. Through this analysis, the explanatory contributions of selected features to MoCA score prediction were validated, and their relative importance in cognitive function variability in PD was assessed. Optimizing digital biomarker identification using recursive feature selection and logistic regression Recursive feature elimination with cross-validation was implemented to refine the model and enhance predictive performance, using logistic regression as the base estimator. This method iteratively removed less informative features while optimizing model performance based on AUC–ROC. The procedure identified the optimal number of features, retaining only the most informative variables. The final selected features included age, MMSE, UPDRS_III, and several gait-related features, which were subsequently used for model training. A logistic regression model was trained using these selected biomarkers to classify cognitive status groups. The dataset was split into training (80%) and test (20%) subsets, with a fixed random seed applied to ensure reproducibility. To evaluate generalizability, 10-fold cross-validation was conducted. In each fold, the model was trained on nine subsets and validated on the remaining one, ensuring that each subset served as the validation set exactly once. Model performance was assessed using the ROC curve and AUC metrics. The ROC curve illustrated the trade-off between sensitivity and specificity, while AUC scores quantified overall classification performance. This approach effectively identified digital biomarkers and optimized feature selection for predictive modeling. Statistical analysis Data normality was assessed using the Shapiro–Wilk test. Depending on the distribution, either an independent t-test or a non-parametric test was used to compare physical and clinical characteristics between groups. Statistical analyses were performed using SPSS version 21.0 (IBM Corp., Armonk, NY, USA), with significance set at p < 0.05. Additionally, data preprocessing and analysis were conducted in Python (version 3.8.8) using standard libraries, including pandas, NumPy, scikit-learn, and matplotlib. Abbreviations PD Parkinson's disease MoCA Montreal cognitive assessment ML Machine learning LASSO Least absolute shrinkage and selection operator,XGBoost,eXtreme Gradient Boosting PD-NC PD with normal cognition PD-CD PD with cognitive decline UPDRS Unified parkinson's disease rating scale MAE Mean absolute error R 2 Coefficient of determination CV Cross-validation MMSE Mini-mental state examination TurnFR_CTL Contralateral temporal coordination of the left foot during faster speed rightward turning TurnFL_RANK_MaxAcc Maximum acceleration of the right ankle during faster speed leftward turning FW_RANK_MaxJerk Maximum jerk of the right ankle during forward walking TurnPL_SLL Stride length of the left foot during preferred speed leftward turning FW_PSIS_MaxGyr Maximum gyroscope at the posterior superior iliac spine during forward walking TurnPL_SLR Stride length of the left foot during preferred speed rightward turning TurnFR_PSIS_SampEnAcc Sample entropy of acceleration at the posterior superior iliac spine during faster speed rightward turning AUC‒ROC area under the receiver operating characteristic curve. Declarations ACKNOWLEDGEMENTS 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. AUTHOR CONTRIBUTIONS B.K., C.Y., and S.C. conceived and designed the study. B.K., S.C., and H.P. recruited the participants. B.K., C.Y., S.C., H.P., H.C., J.H., and M.K. conducted data acquisition. B.K. and C.Y. analyzed and interpreted the data. B.K., C.Y., and S.C. drafted the manuscript. All authors reviewed and approved the final version of the manuscript. DATA AVAILABILITY STATEMENT The datasets supporting the findings of this study are available from the corresponding author upon reasonable request. COMPETING INTERESTS STATEMENT The authors declare that they have no competing interests. ETHICS APPROVAL AND CONSENT TO PARTICIPATE All procedures involving human participants were conducted in accordance with the ethical standards of the institutional and/or national research committee and the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. The study protocol and supplementary materials were approved by the Institutional Review Board of Dong-A University Hospital (Approval No. DAUHIRB-22-089). All participants provided written informed consent prior to data collection. 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, data collection, analysis, interpretation, or manuscript writing. CODE AVAILABILITY No open-source code is available for this study. The code for training and testing the machine learning models was written in Python 3.8.8 using PyTorch 1.13.1 and Torchvision 0.14.1. Data management and feature processing scripts were written in Python 3.8.8 using pandas 1.2.4 and NumPy 1.20.1. The code used for the analysis can be obtained from the corresponding author upon request. References Leroi, I., McDonald, K., Pantula, H. & Harbishettar, V. Cognitive impairment in Parkinson disease: impact on quality of life, disability, and caregiver burden. J. Geriatr. Psychiatry Neurol. 25 , 208–214 (2012). Sasikumar, S. & Strafella, A. P. Imaging mild cognitive impairment and dementia in Parkinson's disease. Front. Neurol. 11 , 47 (2020). Aarsland, D. et al. Mild cognitive impairment in Parkinson disease: a multicenter pooled analysis. Neurology 75 , 1062–1069 (2010). Palavra, N. C., Naismith, S. L. & Lewis, S. J. Mild cognitive impairment in Parkinson’s disease: a review of current concepts. Neurol. Res. Int. 576091 (2013). (2013). Salmanpour, M. R. et al. Optimized machine learning methods for prediction of cognitive outcome in Parkinson's disease. Comput. Biol. Med. 111 , 103347 (2019). Harvey, J. et al. Machine learning-based prediction of cognitive outcomes in de novo Parkinson’s disease. npj Parkinsons Dis. 8 , 150 (2022). Booth, S., Park, K. W., Lee, C. S. & Ko, J. H. Predicting cognitive decline in Parkinson’s disease using FDG-PET–based supervised learning. J. Clin. Invest. 132 , e157074 (2022). Gramotnev, G., Gramotnev, D. K. & Gramotnev, A. Parkinson’s disease prognostic scores for progression of cognitive decline. Sci. Rep. 9 , 17485 (2019). Hayete, B. et al. A Bayesian mathematical model of motor and cognitive outcomes in Parkinson’s disease. PLOS One . 12 , e0178982 (2017). Hobson, P. & Meara, J. Mild cognitive impairment in Parkinson's disease and its progression onto dementia: a 16-year outcome evaluation of the Denbighshire cohort. Int. J. Geriatr. Psychiatry . 30 , 1048–1055 (2015). Pigott, K. et al. Longitudinal study of normal cognition in Parkinson disease. Neurology 85 , 1276–1282 (2015). Wood, K. L. et al. Different PD-MCI criteria and risk of dementia in Parkinson’s disease: 4-year longitudinal study. npj Parkinsons Dis. 2 , 15027 (2016). Gorji, A. & Fathi Jouzdani, A. Machine learning for predicting cognitive decline within five years in Parkinson’s disease: comparing cognitive assessment scales with DAT SPECT and clinical biomarkers. PLOS One . 19 , e0304355 (2024). Kim, H. M. et al. Prediction of cognitive progression in Parkinson's disease using three cognitive screening measures. Clin. Park Relat. Disord . 1 , 91–97 (2019). Almgren, H. et al. Machine learning-based prediction of longitudinal cognitive decline in early Parkinson’s disease using multimodal features. Sci. Rep. 13 , 13193 (2023). Camargo, C. H. et al. Comparison of the use of screening tools for evaluating cognitive impairment in patients with Parkinson's disease. Dement. Neuropsychol. 10 , 344–350 (2016). Nasreddine, Z. S. et al. The Montreal Cognitive Assessment, MoCA: a brief screening tool for mild cognitive impairment. J. Am. Geriatr. Soc. 53 , 695–699 (2005). Goldman, J. G. et al. Cognitive impairment in Parkinson’s disease: a report from a multidisciplinary symposium on unmet needs and future directions to maintain cognitive health. npj Parkinsons Dis. 4, 19 (2018). Redgrave, P. et al. Goal-directed and habitual control in the basal ganglia: implications for Parkinson's disease. Nat. Rev. Neurosci. 11 , 760–772 (2010). Barbosa, A. F. et al. Gait, posture and cognition in Parkinson's disease. Dement. Neuropsychol. 10 , 280–286 (2016). Pieruccini-Faria, F. et al. Gait variability across neurodegenerative and cognitive disorders: results from the Canadian Consortium of Neurodegeneration in Aging (CCNA) and the Gait and Brain Study. Alzheimers Dement. 17 , 1317–1328 (2021). 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) . 20 , 2631 (2020). Shah, V. V. et al. Digital biomarkers of mobility in Parkinson’s disease during daily living. J. Parkinsons Dis. 10 , 1099–1111 (2020). Zhang, W. et al. Wearable sensor-based quantitative gait analysis in Parkinson’s disease patients with different motor subtypes. npj Digit. Med. 7 , 169 (2024). Del Din, S., Godfrey, A., Mazzà, C., Lord, S. & Rochester, L. Free-living monitoring of Parkinson's disease: lessons from the field. Mov. Disord . 31 , 1293–1313 (2016). Morris, R. et al. Cognition as a mediator for gait and balance impairments in GBA-related Parkinson’s disease. npj Parkinsons Dis. 8 , 78 (2022). Sun, Y. M., Wang, Z. Y., Liang, Y. Y., Hao, C. W. & Shi, C. H. Digital biomarkers for precision diagnosis and monitoring in Parkinson’s disease. npj Digit. Med. 7 , 218 (2024). Albrecht, F. et al. Unraveling Parkinson's disease heterogeneity using subtypes based on multimodal data. Parkinsonism Relat. Disord . 102 , 19–29 (2022). Inguanzo, A. et al. MRI subtypes in Parkinson’s disease across diverse populations and clustering approaches. npj Parkinsons Dis. 10 , 159 (2024). Amboni, M. et al. Machine learning can predict mild cognitive impairment in Parkinson's disease. Front. Neurol. 13 , 1010147 (2022). Russo, M. et al. Kinematic and kinetic gait features associated with mild cognitive impairment in Parkinson’s disease. IEEE Trans. Neural Syst. Rehabil Eng. 32 , 2676–2687 (2024). Ghosh, D., Pal, S., Lutz, M. & Luo, S. Ensemble survival analysis for preclinical cognitive decline prediction in Alzheimer's disease using longitudinal biomarkers. arXiv Preprint arXiv :250316645 (2025). Janssen Daalen, J. M. et al. Digital biomarkers for non-motor symptoms in Parkinson’s disease: the state of the art. npj Digit. Med. 7 , 186 (2024). Madakkatel, I., Zhou, A., McDonnell, M. D. & Hyppönen, E. Combining machine learning and conventional statistical approaches for risk factor discovery in a large cohort study. Sci. Rep. 11 , 22997 (2021). Hoops, S. et al. Validity of the MoCA and MMSE in the detection of MCI and dementia in Parkinson disease. Neurology 73 , 1738–1745 (2009). Chou, K. L., Lenhart, A., Koeppe, R. A. & Bohnen, N. I. Abnormal MoCA and normal range MMSE scores in Parkinson disease without dementia: cognitive and neurochemical correlates. Parkinsonism Relat. Disord . 20 , 1076–1080 (2014). Fiorenzato, E. et al. Optimal MMSE and MoCA cutoffs for cognitive diagnoses in Parkinson's disease: A data-driven decision tree model. J. Neurol. Sci. 466 , 123283 (2024). Liu, J. et al. Lasso-based machine learning algorithm for predicting postoperative lung complications in elderly: a single-center retrospective study from China. Clin. Interv Aging . 18 , 597–606 (2023). Liampas, I. et al. Motor signs and incident dementia with Lewy bodies in older adults with mild cognitive impairment. J. Am. Geriatr. Soc. 73 , 50–62 (2025). Di Tella, S. et al. Cognitive Reserve proxies can modulate motor and non-motor basal ganglia circuits in early Parkinson’s disease. Brain Imaging Behav. 18 , 220–230 (2024). Sosnik, R., Fahoum, F., Katzir, Z., Mirelman, A. & Maidan, I. Key shifts in frontoparietal network activity in Parkinson’s disease. npj Parkinsons Dis. 11 , 2 (2025). Wilson, H. et al. Predict cognitive decline with clinical markers in Parkinson’s disease (PRECODE-1). J. Neural Transm (Vienna) . 127 , 51–59 (2020). Penko, A. L., Streicher, M. C., Dey, T., Rosenfeldt, A. B. & Alberts, J. L. Parkinson’s gait kinematics deteriorates across multiple cognitive domains under dual-task paradigms. Clin. Neurol. Neurosurg. 197 , 106083 (2020). Morris, R. et al. Cognitive associations with comprehensive gait and static balance measures in Parkinson's disease. Parkinsonism Relat. Disord . 69 , 104–110 (2019). Sarasso, E. et al. Dual-task clinical and functional MRI correlates in Parkinson's disease with postural instability and gait disorders. Parkinsonism Relat. Disord . 91 , 88–95 (2021). Caronni, A. et al. In Parkinson’s disease, dual-tasking reduces gait smoothness during the straight-walking and turning-while-walking phases of the Timed Up and Go test. BMC Sports Sci. Med. Rehabil . 17 , 42 (2025). Wu, Y. et al. Non-motor symptoms and quality of life in tremor dominant vs postural instability gait disorder Parkinson′s disease patients. Acta Neurol. Scand. 133 , 330–337 (2016). Artusi, C. A. et al. Pisa syndrome in Parkinson's disease is associated with specific cognitive alterations. Front. Neurol. 10 , 577 (2019). Tait, P. et al. Neuroimaging and cognitive correlates of postural control in Parkinson’s disease: a systematic review. J. Neuroeng. Rehabil . 22 , 24 (2025). Pal, G. et al. Global cognitive function and processing speed are associated with gait and balance dysfunction in Parkinson’s disease. J. Neuroeng. Rehabil . 13 , 94 (2016). Burnfield, M. Gait analysis: normal and pathological function. J. Sports Sci. Med. 9 , 353 (2010). Gago, M. F. et al. How do cognitive and axial motor signs correlate in Parkinson’s disease? A 6-year prospective study. J. Neurol. 256 , 1655–1662 (2009). Miri, A. L. et al. A biomechanical analysis of turning during gait in individuals with different subtypes of Parkinson's disease. Clin. Biomech. (Bristol) . 112 , 106166 (2024). Pantall, A. et al. Postural dynamics are associated with cognitive decline in Parkinson's disease. Front. Neurol. 9 , 1044 (2018). Morris, R. et al. Gait rather than cognition predicts decline in specific cognitive domains in early Parkinson’s disease. J. Gerontol. Biol. Sci. Med. Sci. 72 , 1656–1662 (2017). Wang, J., Xu, J., Zhao, C., Peng, Y. & Wang, H. An ensemble feature selection method for high-dimensional data based on sort aggregation. Syst. Sci. Control Eng. 7 , 32–39 (2019). Abhisheka, B., Biswas, S. K. & Purkayastha, B. Infusing weighted average ensemble diversity for advanced breast cancer detection. Int. J. Imaging Syst. Technol. 34 , e23146 (2024). Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted 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-6695263","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":470596980,"identity":"50f46555-9d86-4361-a737-658a71a3082d","order_by":0,"name":"Bohyun Kim","email":"","orcid":"","institution":"Dong-A University","correspondingAuthor":false,"prefix":"","firstName":"Bohyun","middleName":"","lastName":"Kim","suffix":""},{"id":470596981,"identity":"59018fa4-8626-417d-9531-ca2aa162b1b8","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":470596982,"identity":"639dea8d-4c31-4b68-a8de-c5bfe88469c8","order_by":2,"name":"Sang-Myung Cheon","email":"","orcid":"","institution":"Dong-A University","correspondingAuthor":false,"prefix":"","firstName":"Sang-Myung","middleName":"","lastName":"Cheon","suffix":""},{"id":470596983,"identity":"d567e695-0e6b-4f9b-bf7a-c491e826c63f","order_by":3,"name":"Hwayoung Park","email":"","orcid":"","institution":"Dong-A University","correspondingAuthor":false,"prefix":"","firstName":"Hwayoung","middleName":"","lastName":"Park","suffix":""},{"id":470596984,"identity":"c879c681-aa3f-41fa-974d-51e6536c13e0","order_by":4,"name":"Hyejin Choi","email":"","orcid":"","institution":"Dong-A University","correspondingAuthor":false,"prefix":"","firstName":"Hyejin","middleName":"","lastName":"Choi","suffix":""},{"id":470596985,"identity":"2f5fd5a4-c992-4072-a419-43bca72877e9","order_by":5,"name":"Juseon Hwang","email":"","orcid":"","institution":"Dong-A University","correspondingAuthor":false,"prefix":"","firstName":"Juseon","middleName":"","lastName":"Hwang","suffix":""},{"id":470596986,"identity":"16b4a9be-7bee-4d50-993b-1734eee25564","order_by":6,"name":"Minsoo Kim","email":"","orcid":"","institution":"Dong-A University","correspondingAuthor":false,"prefix":"","firstName":"Minsoo","middleName":"","lastName":"Kim","suffix":""}],"badges":[],"createdAt":"2025-05-19 05:38:04","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6695263/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6695263/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":84682517,"identity":"27f51fb4-6183-401e-866a-026bf4a2efc6","added_by":"auto","created_at":"2025-06-16 08:37:34","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":348800,"visible":true,"origin":"","legend":"\u003cp\u003eRelative feature importance derived from LASSO regression and XGBoost models for MoCA score prediction. MoCA, Montreal Cognitive Assessment; MMSE, Mini-Mental State Examination; UPDRS, Unified Parkinson’s Disease Rating Scale; TurnFR, Faster speed rightward turning; TurnFL, Faster speed leftward turning; FW, Forward walking; TurnPL, Preferred speed leftward turning; CTL, Contralateral temporal coordination of the left foot; RANK, Right ankle; SLL, Stride length of the left foot; PSIS, Posterior superior iliac spine; MaxAcc, Maximum acceleration; MaxJerk, Maximum jerk; MaxGyr, Maximum gyroscope; SampEnAcc, Ssample entropy of acceleration.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-6695263/v1/8143dedd53ecf4667a054f03.png"},{"id":84682508,"identity":"49670c45-6f3f-42f9-869c-aa3a9e98e09a","added_by":"auto","created_at":"2025-06-16 08:37:34","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":108484,"visible":true,"origin":"","legend":"\u003cp\u003eFeature types and predictive weights in the MoCA regression network. Node colors indicate variable types (red: dependent variable; blue: cognitive and gait predictors), while node size reflects the magnitude of standardized regression coefficients. Edge thickness represents the strength of partial correlations, estimated using graphical LASSO. MoCA, Montreal Cognitive Assessment; MMSE, Mini-Mental State Examination; TurnPL, Preferred speed leftward turning; TurnFL, Faster speed leftward turning; FW, Forward walking; SLL, Stride length of the left foot; PSIS, Posterior superior iliac spine; RANK, Right ankle; MaxGyr, Maximum gyroscope; MaxAcc, Maximum acceleration; MaxJerk, Maximum jerk.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-6695263/v1/fb7b16f3e729c19ef15ef2cc.png"},{"id":84682516,"identity":"a60cbb38-ad94-470c-9ad9-21190ea51214","added_by":"auto","created_at":"2025-06-16 08:37:34","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":321503,"visible":true,"origin":"","legend":"\u003cp\u003eROC curves for digital biomarkers identified to distinguish PD-CD. a) ROC curve derived from clinical and cognitive features (age, MMSE, and UPDRS_III). b, c) ROC curves based on gait-derived features, including spatiotemporal and sensor-based metrics. MMSE, Mini-Mental State Examination; UPDRS, Unified Parkinson’s Disease Rating Scale; TurnFR, Faster speed rightward turning; TurnPL, Preferred speed leftward turning; TurnFL, Faster speed leftward turning; FW, Forward walking; CTL, Contralateral temporal coordination of the left foot; SLL, Stride length of the left foot; SLR, Stride length of the right foot; RANK, Right ankle; PSIS, Posterior superior iliac spine; MaxAcc, Maximum acceleration; SampEnAcc, Ssample entropy of acceleration; MaxJerk, Maximum jerk; MaxGyr, Maximum gyroscope.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-6695263/v1/c3a35e643b237f606ddd523d.png"},{"id":89266781,"identity":"a39f234c-6aa1-4617-acb7-712507f3d1b8","added_by":"auto","created_at":"2025-08-18 08:17:11","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1947816,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6695263/v1/c6b9e4a2-1b48-45aa-a121-8f9ec3388294.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Multimodal Machine Learning Approach for Predicting Cognitive Decline in People with Parkinson’s Disease","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eParkinson\u0026rsquo;s disease (PD) is a progressive neurodegenerative disorder marked by both motor and non-motor symptoms, which substantially impair quality of life in affected individuals \u003csup\u003e[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]\u003c/sup\u003e. While hallmark motor symptoms include bradykinesia, rigidity, resting tremor, and postural instability, non-motor symptoms\u0026mdash;particularly cognitive impairment\u0026mdash;can be equally disabling and may even precede the onset of motor symptoms \u003csup\u003e[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]\u003c/sup\u003e. Approximately 42.5% of newly diagnosed individuals with PD present with mild cognitive impairment, with 70\u0026ndash;80% progressing to dementia over 15\u0026ndash;20 years \u003csup\u003e[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]\u003c/sup\u003e. Recent studies have increasingly focused on predicting cognitive decline in PD by integrating clinical evaluations, biomarkers, neuropsychological tests, and longitudinal data \u003csup\u003e[\u003cspan additionalcitationids=\"CR6 CR7 CR8 CR9 CR10 CR11\" citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]\u003c/sup\u003e. These methods promote early identification, timely therapeutic intervention, and enhanced disease management \u003csup\u003e[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eVarious cognitive assessment scales have been utilized to predict cognitive decline in PD, and several studies have investigated the predictive abilities of these approaches \u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003e. The Montreal Cognitive Assessment (MoCA), though widely adopted for evaluating memory, executive function, and verbal fluency in PD \u003csup\u003e[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]\u003c/sup\u003e, is limited by its intermittent administration, potentially failing to detect subtle or early cognitive changes \u003csup\u003e[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]\u003c/sup\u003e. Emerging evidence suggests a link between motor and cognitive symptoms in PD, mediated by overlapping neuropathology in the basal ganglia and prefrontal dopaminergic circuits \u003csup\u003e[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]\u003c/sup\u003e. Specifically, gait disturbances, postural instability, and impaired turning performance have been associated with cognitive decline, indicating that movement-based digital metrics could serve as early biomarkers \u003csup\u003e[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eRecent advances in digital healthcare have enabled the use of wearable sensors for non-invasive evaluation of gait impairments in individuals with PD \u003csup\u003e[\u003cspan additionalcitationids=\"CR23\" citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]\u003c/sup\u003e. Metrics derived from these sensors\u0026mdash;such as stride variability, turning angular velocity, and entropy-based indicators of gait regularity\u0026mdash;have proven sensitive in detecting subtle cognitive-motor dysfunctions often overlooked by traditional clinical scales \u003csup\u003e[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]\u003c/sup\u003e. These devices offer continuous, real-world gait monitoring, providing high-resolution, objective data that address key limitations of conventional assessments \u003csup\u003e[\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]\u003c/sup\u003e. Previous studies relying exclusively on standard clinical tools have demonstrated poor reproducibility across patient cohorts, highlighting the need to incorporate objective, multimodal data to enhance the prediction and classification of cognitive decline in PD \u003csup\u003e[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eMachine learning (ML) techniques have emerged as effective tools for integrating high-dimensional, heterogeneous multimodal data, enabling the detection of complex motor\u0026ndash;cognitive interactions and the systematic selection of predictive features \u003csup\u003e[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]\u003c/sup\u003e. Recent studies employing ML approaches such as least absolute shrinkage and selection operator (LASSO) regression, Random Forest, and eXtreme Gradient Boosting (XGBoost) have demonstrated superior accuracy in predicting PD symptom severity and progression compared with traditional statistical methods, highlighting their value in developing clinically viable digital biomarkers \u003csup\u003e[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/sup\u003e. However, most existing ML models either insufficiently incorporate sensor-derived gait features or fail to adequately examine their direct relationship with cognitive decline in PD. Moreover, despite their promise as screening tools, current digital biomarker models often lack interpretability and clinical utility \u003csup\u003e[\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/sup\u003e. Thus, developing robust, interpretable, and clinically relevant models that integrate digital biomarkers with conventional clinical assessments is essential for improving the early identification and management of cognitive decline in PD.\u003c/p\u003e \u003cp\u003eDespite recent progress, critical gaps remain in incorporating gait features derived from wearable sensors into clinically interpretable models of cognitive impairment in PD. This study aims to enhance the accuracy of predicting and classifying cognitive decline in PD by applying ML techniques to multimodal data, including clinical characteristics, physical function, lifestyle factors, and gait-derived features. Specifically, the objectives are to:\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eInvestigate the associations between multimodal features\u0026mdash;such as clinical characteristics, physical function, lifestyle factors, and gait-derived features\u0026mdash;and cognitive decline in individuals with PD.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eDevelop and evaluate ML models, including LASSO regression and XGBoost, to predict MoCA scores using multimodal data.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003ePerform stepwise multiple linear regression to identify independent predictors of cognitive performance.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eAssess logistic regression models to classify individuals with and without cognitive impairment based on selected multimodal features.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003cp\u003eWe hypothesize that multimodal features\u0026mdash;particularly gait-derived digital biomarkers\u0026mdash;will show significant associations with cognitive status in PD. Furthermore, integrating these features with ML techniques is expected to improve MoCA score prediction and the classification of cognitive impairment, offering insights that extend beyond conventional clinical assessments.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eDemographic and clinical characteristics associated with cognitive decline in PD\u003c/h2\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e presents the demographic and clinical characteristics of the study participants. Individuals were stratified into two groups based on MoCA scores (\u0026lt;\u0026thinsp;26): PD with normal cognition (PD-NC; n\u0026thinsp;=\u0026thinsp;60) and PD with cognitive decline (PD-CD; n\u0026thinsp;=\u0026thinsp;42). The analysis revealed significant demographic and clinical differences between the two groups. Specifically, individuals in the PD-CD group were older and had more severe motor symptoms, as indicated by higher Hoehn and Yahr stages and elevated scores on both the total Unified Parkinson\u0026rsquo;s Disease Rating Scale (UPDRS) and the motor examination subscale (UPDRS Part III). Additionally, the PD-CD group demonstrated significantly lower scores on cognitive assessments compared with the PD-NC group.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDemographics and clinical characteristics of the study participants\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePD-NC\u003c/p\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;60)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePD-CD\u003c/p\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;42)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e\u003cem\u003ep\u003c/em\u003e-value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eMean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSex (male/female)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e24/36\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e22/20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge (years)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e65.83\u0026thinsp;\u0026plusmn;\u0026thinsp;7.52\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e71.24\u0026thinsp;\u0026plusmn;\u0026thinsp;5.36\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHeight (cm)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e161.47\u0026thinsp;\u0026plusmn;\u0026thinsp;8.16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e158.79\u0026thinsp;\u0026plusmn;\u0026thinsp;8.66\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.114\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBody weight (kg)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e62.95\u0026thinsp;\u0026plusmn;\u0026thinsp;11.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e62.82\u0026thinsp;\u0026plusmn;\u0026thinsp;8.67\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.731\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBMI (kg/m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e24.01\u0026thinsp;\u0026plusmn;\u0026thinsp;3.29\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e24.89\u0026thinsp;\u0026plusmn;\u0026thinsp;2.67\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.063\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSymptom duration (years)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e5.44\u0026thinsp;\u0026plusmn;\u0026thinsp;3.26\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e6.08\u0026thinsp;\u0026plusmn;\u0026thinsp;5.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.600\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTreatment duration (years)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4.19\u0026thinsp;\u0026plusmn;\u0026thinsp;3.18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e5.06\u0026thinsp;\u0026plusmn;\u0026thinsp;5.35\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.303\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eL-dopa equivalent dose (mg/day)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e542.89\u0026thinsp;\u0026plusmn;\u0026thinsp;316.50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e549.12\u0026thinsp;\u0026plusmn;\u0026thinsp;250.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.514\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHoehn and Yahr scale (stages)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.83\u0026thinsp;\u0026plusmn;\u0026thinsp;0.69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2.14\u0026thinsp;\u0026plusmn;\u0026thinsp;0.72\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e0.032\u003c/b\u003e\u003csup\u003e\u003cb\u003eb\u003c/b\u003e\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUPDRS total (scores)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e47.78\u0026thinsp;\u0026plusmn;\u0026thinsp;19.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e64.13\u0026thinsp;\u0026plusmn;\u0026thinsp;24.40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUPDRS part I (scores)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e9.53\u0026thinsp;\u0026plusmn;\u0026thinsp;5.33\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e10.88\u0026thinsp;\u0026plusmn;\u0026thinsp;5.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.180\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUPDRS part II (scores)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e11.70\u0026thinsp;\u0026plusmn;\u0026thinsp;6.38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e14.60\u0026thinsp;\u0026plusmn;\u0026thinsp;8.06\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.071\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUPDRS part III (scores)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e24.62\u0026thinsp;\u0026plusmn;\u0026thinsp;12.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e36.35\u0026thinsp;\u0026plusmn;\u0026thinsp;16.10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003csup\u003e\u003cb\u003eb\u003c/b\u003e\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUPDRS part IV (scores)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.93\u0026thinsp;\u0026plusmn;\u0026thinsp;2.73\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.40\u0026thinsp;\u0026plusmn;\u0026thinsp;2.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.212\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMMSE (scores)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e28.55\u0026thinsp;\u0026plusmn;\u0026thinsp;1.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e26.60\u0026thinsp;\u0026plusmn;\u0026thinsp;2.46\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003csup\u003e\u003cb\u003eb\u003c/b\u003e\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMoCA (scores)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e27.83\u0026thinsp;\u0026plusmn;\u0026thinsp;1.24\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e23.00\u0026thinsp;\u0026plusmn;\u0026thinsp;2.36\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003csup\u003e\u003cb\u003eb\u003c/b\u003e\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"4\" nameend=\"c4\" namest=\"c1\"\u003e \u003cp\u003eThe data are presented as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation. Statistically significant differences between groups are shown in bold (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05).\u003c/p\u003e \u003cp\u003ePD-NC, Parkinson\u0026rsquo;s disease with normal cognition; PD-CD, Parkinson\u0026rsquo;s disease with cognitive decline; BMI, Body mass index; UPDRS, Unified Parkinson\u0026rsquo;s Disease Rating Scale; MMSE, Mini-Mental State Examination; MoCA, Montreal Cognitive Assessment.\u003c/p\u003e \u003cp\u003e\u003csup\u003ea\u003c/sup\u003e Independent samples \u003cem\u003et\u003c/em\u003e-test.\u003c/p\u003e \u003cp\u003e\u003csup\u003eb\u003c/sup\u003e Mann\u0026ndash;Whitney \u003cem\u003eU\u003c/em\u003e test.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003ePerformance evaluation of LASSO and XGBoost models for MoCA score prediction\u003c/h3\u003e\n\u003cp\u003eThe predictive performance of the LASSO regression and XGBoost models was evaluated using cross-validation and an independent test set. Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e summarizes the results based on mean absolute error (MAE) and coefficient of determination (\u003cem\u003eR\u003c/em\u003e\u003csup\u003e\u003cem\u003e2\u003c/em\u003e\u003c/sup\u003e) for both models.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eModel performance and feature importance\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCV_MAE\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCV_\u003cem\u003eR\u0026sup2;\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTest_MAE\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eTest_ \u003cem\u003eR\u0026sup2;\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLASSO\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.525\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.470\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.613\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.370\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eXGBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.596\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.294\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.628\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.383\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003eLASSO and XGBoost were used as predictive models. CV_MAE, Mean absolute error during cross-validation; CV_\u003cem\u003eR\u0026sup2;\u003c/em\u003e, Coefficient of determination (\u003cem\u003eR\u003c/em\u003e\u0026sup2;) during cross-validation, indicating model fit; Test_MAE, Mean absolute error on the test set; Test_\u003cem\u003eR\u0026sup2;\u003c/em\u003e, \u003cem\u003eR\u003c/em\u003e\u0026sup2; on the test set, indicating predictive accuracy.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eLASSO regression achieved a cross-validation MAE (CV_MAE) of 0.525 and a cross-validation \u003cem\u003eR\u003c/em\u003e\u003csup\u003e\u003cem\u003e2\u003c/em\u003e\u003c/sup\u003e (CV_\u003cem\u003eR\u003c/em\u003e\u003csup\u003e\u003cem\u003e2\u003c/em\u003e\u003c/sup\u003e) of 0.470. In contrast, XGBoost produced a higher CV_MAE of 0.596 and a lower CV_\u003cem\u003eR\u003c/em\u003e\u003csup\u003e\u003cem\u003e2\u003c/em\u003e\u003c/sup\u003e of 0.294. For the test set, LASSO recorded a MAE of 0.613 and a \u003cem\u003eR\u003c/em\u003e\u003csup\u003e\u003cem\u003e2\u003c/em\u003e\u003c/sup\u003e of 0.370, while XGBoost showed a comparable MAE of 0.628 but a marginally higher \u003cem\u003eR\u003c/em\u003e\u003csup\u003e\u003cem\u003e2\u003c/em\u003e\u003c/sup\u003e of 0.383. These results suggest that although LASSO demonstrated stronger generalization in cross-validation, XGBoost performed slightly better on unseen data, as indicated by its higher Test_\u003cem\u003eR\u003c/em\u003e\u003csup\u003e\u003cem\u003e2\u003c/em\u003e\u003c/sup\u003e value. Feature importance was computed using a weighted average approach that integrated the relative contributions of both models. The Korean Mini-Mental State Examination (MMSE) emerged as the strongest predictor (importance\u0026thinsp;=\u0026thinsp;0.385), followed by motor examination scores (UPDRS Part III, importance\u0026thinsp;=\u0026thinsp;0.152) and contralateral temporal coordination of the left foot during faster speed rightward turning (TurnFR_CTL; 0.085). Several gait-derived features also contributed substantially to MoCA score prediction, including maximum acceleration of the right ankle during faster speed leftward turning (TurnFL_RANK_MaxAcc; 0.072), maximum jerk of the right ankle during forward walking (FW_RANK_MaxJerk; 0.062), and stride length of the left foot during preferred speed leftward turning (TurnPL_SLL; 0.056) (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e\n\u003ch3\u003eRegression analysis and network-based feature relationships for MoCA score prediction\u003c/h3\u003e\n\u003cp\u003eStepwise multiple linear regression identified five significant predictors of cognitive performance, as measured by MoCA, yielding an adjusted \u003cem\u003eR\u003c/em\u003e\u003csup\u003e\u003cem\u003e2\u003c/em\u003e\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.617, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05. Collectively, these features explained 61.7% of the variance in MoCA scores. The final model included both neuropsychological (MMSE) and gait-derived features, including TurnPL_SLL, TurnFL_RANK_MaxAcc, FW_RANK_MaxJerk, and the maximum gyroscope at the posterior superior iliac spine during forward walking (FW_PSIS_MaxGyr). The model demonstrated strong predictive power, as indicated by a progressive increase in adjusted \u003cem\u003eR\u003c/em\u003e\u003csup\u003e\u003cem\u003e2\u003c/em\u003e\u003c/sup\u003e across regression iterations. Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e reports the corresponding regression coefficients and their statistical significance.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eStepwise multiple linear regression analysis predicting MoCA scores TurnPL, Preferred speed leftward turning; TurnFL, Faster speed leftward turning; FW, Forward walking; SLL, Stride length of the left foot; RANK, Right ankle; PSIS, Posterior superior iliac spine; MaxAcc, Maximum acceleration; MaxJerk, Maximum jerk; MaxGyr, Maximum gyroscope; SE, Standard error; \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFeatures\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cem\u003eβ\u003c/em\u003e (SE)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003et\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cem\u003ep\u003c/em\u003e-value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cem\u003eR\u0026sup2;\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMMSE\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.577(0.065)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e8.853\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\" morerows=\"4\" rowspan=\"5\"\u003e \u003cp\u003e0.617\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTurnPL_SLL,\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.225(0.066)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3.403\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTurnFL_RANK_MaxAcc,\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.218(0.068)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3191\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.002\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFW_RANK_MaxJerk\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-0.190(0.068)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-2.783\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.006\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFW_PSIS_MaxGyr\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.135(0.067)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2.012\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.047\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eBeyond regression analysis, a network-based visualization revealed that MMSE was the most influential predictor, exhibiting the highest regression coefficient (\u003cem\u003eβ\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.577) and the strongest connection to MoCA within the feature network, highlighting its central role in cognitive prediction. Additionally, TurnPL_SLL (\u003cem\u003eβ\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.225) was directly associated with MoCA scores and was closely linked to other gait-derived features in the network (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e\n\u003ch3\u003eIdentification of digital biomarkers based on classification models\u003c/h3\u003e\n\u003cp\u003eTen optimal digital biomarkers for classifying PD-CD group were identified using recursive feature elimination with cross-validation, based on features previously selected by the LASSO and XGBoost models. The final biomarker panel included clinical characteristics \u0026mdash;age, MMSE, and UPDRS_III\u0026mdash;along quantitative gait-derived features: including TurnFR_CTL, TurnPL_SLL, stride length of the left foot during preferred speed rightward turning (TurnPL_SLR), TurnFL_RANK_MaxAcc, sample entropy of acceleration at the posterior superior iliac spine during faster speed rightward turning (TurnFR_PSIS_SampEnAcc), FW_RANK_MaxJerk, and FW_PSIS_MaxGyr.\u003c/p\u003e \u003cp\u003eThe logistic regression classifier achieved 76.5% cross-validated accuracy, with an area under the receiver operating characteristic curve (AUC\u0026ndash;ROC) of 0.895, indicating strong discriminative performance for distinguishing between individuals with and without cognitive decline (Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eClassification performance of PD-CD identification models for MoCA scores based on multimodal features\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDigital biomarkers\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAUC (Mean)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge\u003c/p\u003e \u003cp\u003eMMSE\u003c/p\u003e \u003cp\u003eUPDRS_III\u003c/p\u003e \u003cp\u003eTurnFR_CTL\u003c/p\u003e \u003cp\u003eTurnPL_SLL\u003c/p\u003e \u003cp\u003eTurnPL_SLR\u003c/p\u003e \u003cp\u003eTurnFL_RANK_MaxAcc\u003c/p\u003e \u003cp\u003eTurnFR_PSIS_SampEnAcc\u003c/p\u003e \u003cp\u003eFW_RANK_MaxJerk\u003c/p\u003e \u003cp\u003eFW_PSIS_MaxGyr\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e76.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.895\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003ePD-CD, Parkinson\u0026rsquo;s disease with cognitive decline; MoCA, Montreal Cognitive Assessment; AUC, Area under the curve; MMSE, Mini-Mental State Examination; UPDRS, Unified Parkinson\u0026rsquo;s Disease Rating Scale; TurnFR, Faster speed rightward turning; TurnPL, Preferred speed leftward turning; TurnFL, Faster speed leftward turning; FW, Forward walking; CTL, contralateral temporal coordination of the left foot; SLL, Stride length of the left foot; SLR, Stride length of the right foot; RANK, Right ankle; PSIS, posterior superior iliac spine; MaxAcc, Maximum acceleration; SampEnAcc, Sample entropy of acceleration; Maxjerk, Maximum jerk; MaxGyr, Maximum gyroscope.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eThis study evaluated the utility of multimodal features\u0026mdash;including clinical characteristics, physical function, lifestyle factors, and gait-derived digital biomarkers\u0026mdash;for predicting cognitive performance and classifying cognitive status in individuals with PD. By applying ML and regression-based approaches, we identified key predictors of MoCA scores, highlighting the clinical significance of gait-derived features in assessing cognitive decline.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eKey findings and model performance\u003c/h2\u003e \u003cp\u003eThe LASSO regression and XGBoost models demonstrated moderate-to-strong predictive performance for MoCA scores (LASSO CV_\u003cem\u003eR\u0026sup2;\u003c/em\u003e = 0.470, Test_ \u003cem\u003eR\u003c/em\u003e\u0026sup2; = 0.370; XGBoost CV_\u003cem\u003eR\u0026sup2;\u003c/em\u003e = 0.294, Test_\u003cem\u003eR\u0026sup2;\u003c/em\u003e = 0.383), underscoring the complementary strengths of linear and nonlinear modeling approaches. LASSO exhibited greater consistency across validation folds, while XGBoost showed slightly better generalizability to unseen data, likely owing to its capacity to model nonlinear relationships \u003csup\u003e[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]\u003c/sup\u003e. Feature importance analysis identified the MMSE as the most influential predictor of MoCA scores (importance\u0026thinsp;=\u0026thinsp;0.385), followed by UPDRS Part III. Although MMSE and MoCA share conceptual overlap, they target distinct cognitive domains: MoCA offers greater sensitivity to mild cognitive impairment, especially in executive and visuospatial function, whereas MMSE is widely used in clinical practice for its simplicity and established validity \u003csup\u003e[\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]\u003c/sup\u003e. Further analysis revealed a significant drop in model performance when MMSE was excluded (LASSO Test_\u003cem\u003eR\u0026sup2;\u003c/em\u003e = -0.078; XGBoost Test_\u003cem\u003eR\u0026sup2;\u003c/em\u003e =- 0.282), underscoring its unique predictive value \u003csup\u003e[\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e, \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]\u003c/sup\u003e. The use of regularization techniques such as LASSO mitigates concerns of multicollinearity, supporting MMSE\u0026rsquo;s inclusion as both a statistically and clinically relevant predictor \u003csup\u003e[\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThe UPDRS Part III, which evaluates motor symptom severity, emerged as the second most influential predictor, aligning with previous studies that link motor dysfunction\u0026mdash;particularly bradykinesia, postural instability, and axial impairment\u0026mdash;to PD-CD \u003csup\u003e[\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]\u003c/sup\u003e. These findings reinforce the role of overlapping motor\u0026ndash;cognitive neural circuits, especially within the basal ganglia and prefrontal cortex \u003csup\u003e[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]\u003c/sup\u003e. Consequently, axial and postural impairments captured by UPDRS Part III may indicate early cognitive vulnerability, underscoring its utility in integrated motor\u0026ndash;cognitive assessments \u003csup\u003e[\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eGait-derived digital biomarkers\u003c/h3\u003e\n\u003cp\u003eStepwise multiple linear regression identified four significant gait-derived features, which, alongside the MMSE, collectively accounted for 61.7% of the variance in MoCA scores. These features captured core dimensions of gait control linked to cognitive processes: spatial symmetry (TurnPL_SLL), ankle-level acceleration during directional transitions (TurnFL_RANK_MaxAcc), movement smoothness via jerk (FW_RANK_MaxJerk), and trunk-level rotational dynamics (FW_PSIS_MaxGyr). Distal ankle features\u0026mdash;TurnFL_RANK_MaxAcc and FW_RANK_MaxJerk\u0026mdash;reflect neuromuscular responsiveness and anticipatory motor planning under dynamic postural demands, underscoring the involvement of executive and attentional processes in coordinating lower-limb control during complex movement transitions \u003csup\u003e[\u003cspan additionalcitationids=\"CR44 CR45\" citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e]\u003c/sup\u003e. Proximal trunk rotation (FW_PSIS_MaxGyr) captures axial stability and cognitive-motor integration, key functions typically disrupted in early cognitive decline \u003csup\u003e[\u003cspan additionalcitationids=\"CR48\" citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e]\u003c/sup\u003e. Moreover, impairments in gait smoothness, symmetry, and anticipatory postural adjustments\u0026mdash;especially during demanding tasks such as turning or dynamic walking\u0026mdash;serve as sensitive indicators of early cognitive dysfunction, reinforcing their potential as clinically meaningful digital biomarkers \u003csup\u003e[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e, \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e, \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e]\u003c/sup\u003e. These findings highlight the combined contribution of distal (ankle-level) and proximal (trunk-level) kinematic features in revealing neuromechanical disturbances associated with cognitive impairment in PD.\u003c/p\u003e\n\u003ch3\u003eNetwork analysis and interconnected biomarkers\u003c/h3\u003e\n\u003cp\u003eNetwork-based visualization further illustrated the functional interconnections among key predictive features. The MMSE exhibited strong associations with gait-derived features such as TurnFL_RANK_MaxAcc and FW_RANK_MaxJerk. These associations underscore the integration of cognitive function with dynamic limb control, reflecting the influence of executive and attentional processes on rapid acceleration and smooth transitional movement patterns \u003csup\u003e[\u003cspan additionalcitationids=\"CR45\" citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e]\u003c/sup\u003e. TurnPL_SLL, a measure of stride length during preferred speed leftward turning, not only had a direct association with MoCA scores but also demonstrated substantial connectivity with trunk rotation metrics (FW_PSIS_MaxGyr), reinforcing its role in spatial coordination and postural control \u003csup\u003e[\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e, \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e]\u003c/sup\u003e. Notably, TurnFL_RANK_MaxAcc emerged as a central node in the predictive network, exhibiting bidirectional connections with both FW_RANK_MaxJerk and FW_PSIS_MaxGyr. This centrality suggests that rapid directional transitions, ankle jerk dynamics, and trunk rotation are interdependent and likely modulated by shared neuro-mechanical processes rooted in executive control mechanisms, commonly affected during early cognitive decline \u003csup\u003e[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e, \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e, \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e]\u003c/sup\u003e. Collectively, the network topology suggests that cognitive status in Parkinson\u0026rsquo;s disease is underpinned by an integrated system of interconnected motor features rather than isolated predictors, supporting the development of holistic digital biomarker strategies.\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eClassification of cognitive decline\u003c/h2\u003e \u003cp\u003eThe optimized logistic regression model achieved an average classification accuracy of 76.5% (AUC\u0026thinsp;=\u0026thinsp;0.895) in classifying PD-CD based on ten multimodal predictors. Informative features included clinical variables\u0026mdash;age, MMSE, and UPDRS_III\u0026mdash;as well as gait-derived digital biomarkers reflecting spatiotemporal coordination (TurnFR_CTL), spatial symmetry (TurnPL_SLL, TurnPL_SLR), ankle dynamics (TurnFL_RANK_MaxAcc, FW_RANK_MaxJerk), and pelvic rotational motion (FW_PSIS_MaxGyr). This integration enabled detailed characterization of cognitive-motor interactions, particularly emphasizing deficits in executive and attentional function as reflected in gait irregularities \u003csup\u003e[\u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e, \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e]\u003c/sup\u003e. Incorporating sensor-based features such as TurnFR_PSIS_SampEnAcc and FW_RANK_MaxJerk enhanced model sensitivity, reinforcing the utility of gait irregularity as an early marker of cognitive dysfunction \u003csup\u003e[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e, \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e55\u003c/span\u003e]\u003c/sup\u003e. The interpretability and clinical feasibility of the logistic regression model underscore its potential as a screening tool for early-stage cognitive decline and diagnostic stratification in routine clinical settings. Collectively, these findings validate the clinical value of integrating wearable sensor-derived gait metrics with conventional cognitive assessments to support earlier identification and personalized management of cognitive impairment in PD.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eClinical implications and limitations\u003c/h2\u003e \u003cp\u003eThe findings demonstrate that integrating clinical characteristics with gait-derived digital biomarkers enhances the identification of cognitive impairment in individuals with PD. The models developed are both interpretable and computationally efficient, supporting their feasibility for clinical deployment. The incorporation of continuous, non-invasive gait monitoring provides complementary value to traditional cognitive screening tools, enabling earlier diagnosis, individualized therapeutic planning, and longitudinal disease monitoring.\u003c/p\u003e \u003cp\u003eNevertheless, several limitations of this study should be acknowledged. First, the relatively small sample size may limit the generalizability of the findings, underscoring the need for future validation in larger, more diverse, and multicenter cohorts. Second, the cross-sectional study design prevents evaluation of cognitive trajectories over time; longitudinal research is needed to assess the predictive value of multimodal biomarkers in tracking cognitive decline. Third, because data collection occurred exclusively during patients \u0026ldquo;\"ON\u0026rdquo; medication states, motor and cognitive assessments may not reflect fluctuations across the medication cycle. Future studies should evaluate performance in both \u0026ldquo;ON\u0026rdquo; and \u0026ldquo;OFF\u0026rdquo; states to enable more comprehensive cognitive\u0026ndash;motor profiling. Finally, as gait data were collected under controlled laboratory conditions, the findings may not fully represent everyday variability. Incorporating real-world gait monitoring could enhance ecological validity and strengthen clinical applicability.\u003c/p\u003e \u003c/div\u003e"},{"header":"CONCLUSION","content":"\u003cp\u003eThis study demonstrates that multimodal features\u0026mdash;including clinical characteristics, physical function, lifestyle factors, and gait-derived features\u0026mdash;can effectively predict and classify PD-CD. ML models, particularly LASSO regression and XGBoost, identified the MMSE and several gait-derived features\u0026mdash;TurnPL_SLL, TurnFL_RANK_MaxAcc, FW_RANK_MaxJerk, and FW_PSIS_MaxGyr \u0026mdash;as key predictors. These digital biomarkers, validated through regression analysis, network modeling, and classification performance, reflect clinically relevant cognitive\u0026ndash;motor associations. Together, they offer a practical and interpretable foundation for early detection and targeted intervention. Future longitudinal and real-world studies are essential to refine these predictive models and establish their utility in clinical settings.\u003c/p\u003e"},{"header":"METHODS","content":" \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003cdiv id=\"Sec15\" class=\"Section3\"\u003e \u003ch2\u003eParticipants\u003c/h2\u003e \u003cp\u003eData from 102 individuals with a clinician-confirmed diagnosis of idiopathic PD were included, based on the UK Parkinson\u0026rsquo;s Disease Society Brain Bank Clinical Diagnostic Criteria. Inclusion criteria required participants to have mild-to-moderate idiopathic PD, be on anti-Parkinsonian medication, and be able to stand and walk independently during clinical assessments. Exclusion criteria included any comorbid neurological, orthopedic, or psychiatric disorders. The participants had a mean age of 68.1 years, and 54.9% were female. Among them, 49.0% were classified as Hoehn and Yahr stage 2. The mean disease duration was 5.7 years, and the average UPDRS_III score was 29.4. Based on MoCA scores (\u0026lt;\u0026thinsp;26), participants were classified into two groups: PD-NC (n\u0026thinsp;=\u0026thinsp;60) and PD-CD (n\u0026thinsp;=\u0026thinsp;42) (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e The study was approved by the Institutional Review Board of Dong-A University Medical Center (Approval No. DAUHIRB-22-089). All participants were fully informed of the study purpose and procedures and provided written informed consent.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eExperimental procedures\u003c/h2\u003e \u003cp\u003eParticipants completed two laboratory visits for comprehensive multimodal evaluation, including assessments of demographic and clinical characteristics, physical function, lifestyle factors, and gait parameters. For individuals receiving levodopa treatment, testing was conducted during the \u0026lsquo;ON\u0026rsquo; medication state, approximately 2\u0026ndash;3 hours after dosing.\u003c/p\u003e \u003cp\u003eThe study assessed the following categories of multimodal features. Clinical measurements included 22 features, with PD severity evaluated using the Hoehn and Yahr scale and the UPDRS comprising total and Parts I\u0026ndash;IV scores. Part I assessed non-motor aspects of daily living, including cognitive impairment and mood disturbances. Part II measured motor-related daily activities, such as speech, handwriting, and hygiene. Part III evaluated motor symptom severity, focusing on tremors, rigidity, and bradykinesia. Part IV captured motor complications related to long-term dopaminergic therapy, such as dyskinesia and motor fluctuations. Clinical history, including symptom and treatment duration and levodopa-equivalent daily dose, was collected to reflect disease progression and pharmacological management. Cognitive function was assessed using the MMSE and the MoCA. The Beck Depression Inventory and Beck Anxiety Inventory were used to measure depressive and anxiety symptoms, respectively. The PD Sleep Scale-2 assessed sleep disturbances, while the Epworth Sleepiness Scale evaluated daytime sleepiness. Fatigue severity was measured using the Fatigue Severity Scale. Fall risk and fear of falling were evaluated based on fall history and the Korean version of the Falls Efficacy Scale. Self-efficacy for exercise was assessed using the Self-Efficacy for Exercise Scale. The New Freezing of Gait Questionnaire measured the severity of freezing episodes. The Non-Motor Symptoms Scale provided a comprehensive evaluation of a range of non-motor symptoms. Lastly, the PD Questionnaire-39 assessed health-related quality of life.\u003c/p\u003e \u003cp\u003eThe physical function and lifestyle assessments comprised 10 features. Grip strength was measured bilaterally, along with performance-based tests including the Five-Times Sit-to-Stand Test, Six-Minute Walk Test, Short Physical Performance Battery, and Mini-Balance Evaluation Systems Test. Lifestyle-related measures included the Nutrition Quotient, 36-item Short-Form Health Survey, which provided total, physical, and mental health scores, and the International Physical Activity Questionnaire.\u003c/p\u003e \u003cp\u003eGait-derived features obtained from motion capture and wearable sensors comprised 38 and 360 features, respectively. Participants completed three walking tasks: (1) forward and (2) backward straight walking at a self-selected speed, and (3) 360\u0026deg; turning to the left and right at both preferred and faster speeds. Kinematic data were collected using a full-body marker-based system (Plug-in Gait model, Vicon Motion Systems, Oxford Metrics, UK) and six inertial measurement units (Xsens DOT, Movella Technologies, Enschede, Netherlands). The sampling rates were 100 Hz for the Vicon system and 60 Hz for the Xsens DOT. Motion capture\u0026ndash;derived features included walking speed, stride length, double-support phase, and contralateral temporal coordination. In contrast, wearable sensor-derived features encompassed maximum jerk and angular velocity jerk, mean and maximum acceleration, mean and maximum gyroscope values, root mean square acceleration and gyroscope signals, and sample entropy of acceleration and gyroscope data (Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eGait features derived from wearable sensors\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFeature\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMathematical expression\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMaximum jerk (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\text{m}/{s}^{2}\\)\u003c/span\u003e\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\frac{dAcceleration}{dt}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMaximum angular velocity jerk (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\text{r}\\text{a}\\text{d}/{s}^{2}\\)\u003c/span\u003e\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\frac{dAngular\\:Velocity}{dt}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMean acceleration (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\text{m}/{s}^{2}\\)\u003c/span\u003e\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:mean\\left(\\sqrt{{acc}_{x}^{2}+{acc}_{y}^{2}+{acc}_{z}^{2}}\\right)\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMean gyroscopes (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\text{r}\\text{a}\\text{d}/\\text{s}\\)\u003c/span\u003e\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:mean\\left(\\sqrt{{gyr}_{x}^{2}+{gyr}_{y}^{2}+{gyr}_{z}^{2}}\\right)\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMaximum acceleration (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\text{m}/{s}^{2}\\)\u003c/span\u003e\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:max\\left(\\sqrt{{acc}_{x}^{2}+{acc}_{y}^{2}+{acc}_{z}^{2}}\\right)\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMaximum gyroscopes (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\text{r}\\text{a}\\text{d}/\\text{s}\\)\u003c/span\u003e\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:max\\left(\\sqrt{{gyr}_{x}^{2}+{gyr}_{y}^{2}+{gyr}_{z}^{2}}\\right)\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRMS acceleration (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\text{m}/{s}^{2}\\)\u003c/span\u003e\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\sqrt{mean({acc}_{x}^{2}+{acc}_{y}^{2}+{acc}_{z}^{2})}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRMS gyroscopes (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\text{r}\\text{a}\\text{d}/\\text{s}\\)\u003c/span\u003e\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\sqrt{mean({gyr}_{x}^{2}+{gyr}_{y}^{2}+{gyr}_{z}^{2})}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSample entropy acceleration\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:-\\text{log}\\left(\\frac{Aacc}{Bacc}\\right)\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSample entropy gyroscopes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:-\\text{log}\\left(\\frac{Agyr}{Bgyr}\\right)\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"2\"\u003eRMS, root mean square; acc, acceleration; gyr, gyroscope; A: Total number of similar vector pairs of length m\u0026thinsp;+\u0026thinsp;1 (i.e., m\u0026thinsp;+\u0026thinsp;1 consecutive data points) that remain within a defined tolerance r; B, total number of similar vector pairs of length m (i.e., m consecutive data points) within the same tolerance r.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003ePreprocessing\u003c/h2\u003e \u003cp\u003eThe dataset was first reviewed to understand its structure and identify inconsistencies. Preprocessing consisted of two main steps. First, missing values were addressed using k-nearest neighbors\u0026rsquo; imputation. A k-nearest neighbor imputer with five neighbors estimated missing values based on the mean of the five most similar data points. Second, z-score normalization was applied to standardize all features, transforming them to have a mean of zero and a standard deviation of one. This ensured uniform feature scaling and improved comparability across variables.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003eFeature selection and predictive modeling with LASSO and XGBoost: Performance evaluation and importance analysis\u003c/h2\u003e \u003cp\u003eAn initial feature selection step was performed using correlation analysis between all multimodal features and MoCA scores. To accommodate different data distributions, Pearson correlation was used for parametric variables, and Spearman correlation for non-parametric variables. This dual approach allowed a comprehensive assessment of both linear and monotonic relationships between features and MoCA scores. Only variables showing statistically significant correlations (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05) were retained for subsequent predictive modeling.\u003c/p\u003e \u003cp\u003eMoCA scores were converted into a one-dimensional array. The dataset was then split into training (80%) and testing (20%) subsets using stratified random sampling to preserve the distribution of cognitive function scores. LASSO regression was used to identify the most relevant features for predicting cognitive function in individuals with PD. LASSO is a regularization technique that performs feature selection by shrinking the coefficients of less informative predictors to zero. A grid search with five-fold cross-validation was applied to identify the optimal regularization parameter (\u003cem\u003eα\u003c/em\u003e), based on the highest \u003cem\u003eR\u003c/em\u003e\u003csup\u003e\u003cem\u003e2\u003c/em\u003e\u003c/sup\u003e score. The \u003cem\u003eα\u003c/em\u003e search space included the values {0.01, 0.1, 1, 10}. After determining the optimal \u003cem\u003eα\u003c/em\u003e, the LASSO model was retrained using the features with non-zero coefficients. The final dataset used for subsequent modeling included only these selected features.\u003c/p\u003e \u003cp\u003eTwo predictive models were developed to estimate MoCA scores: LASSO regression\u0026mdash;a linear model with L1 regularization that enhances interpretability and performs feature selection\u0026mdash;and XGBoost regression, a gradient-boosting method optimized for capturing nonlinear relationships and feature interactions. For the XGBoost model, hyperparameter tuning was conducted using grid search with five-fold cross-validation. The search space included the number of estimators {100, 200, 300}, maximum tree depth {3, 6, 9}, and learning rate {0.01, 0.1, 0.2}. Model performance was assessed using MAE and \u003cem\u003eR\u0026sup2;\u003c/em\u003e on the cross-validation and test sets. Each model was implemented within a data-processing pipeline, incorporating feature scaling via StandardScaler prior to model training. Model performance was evaluated on the test dataset, and key metrics were recorded. Final predictions from both models were stored for downstream analysis. Feature importance was assessed using multiple approaches: for LASSO, importance was derived from the absolute values of the regression coefficients; for XGBoost, feature importance was computed using the model\u0026rsquo;s built-in importance attribute. To generate a unified importance ranking, a weighted averaging method was applied to combine scores from both models. Weights were proportional to each model\u0026rsquo;s \u003cem\u003eR\u0026sup2;\u003c/em\u003e score on the test set, ensuring that the final ranking reflected relative predictive performance \u003csup\u003e[\u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e56\u003c/span\u003e, \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e57\u003c/span\u003e]\u003c/sup\u003e. Stepwise linear regression was conducted to further examine the relationship between selected features and MoCA scores. This method iteratively added or removed predictors based on statistical significance, retaining only those that significantly contributed to the final model. Through this analysis, the explanatory contributions of selected features to MoCA score prediction were validated, and their relative importance in cognitive function variability in PD was assessed.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003eOptimizing digital biomarker identification using recursive feature selection and logistic regression\u003c/h2\u003e \u003cp\u003eRecursive feature elimination with cross-validation was implemented to refine the model and enhance predictive performance, using logistic regression as the base estimator. This method iteratively removed less informative features while optimizing model performance based on AUC\u0026ndash;ROC. The procedure identified the optimal number of features, retaining only the most informative variables. The final selected features included age, MMSE, UPDRS_III, and several gait-related features, which were subsequently used for model training.\u003c/p\u003e \u003cp\u003eA logistic regression model was trained using these selected biomarkers to classify cognitive status groups. The dataset was split into training (80%) and test (20%) subsets, with a fixed random seed applied to ensure reproducibility. To evaluate generalizability, 10-fold cross-validation was conducted. In each fold, the model was trained on nine subsets and validated on the remaining one, ensuring that each subset served as the validation set exactly once.\u003c/p\u003e \u003cp\u003eModel performance was assessed using the ROC curve and AUC metrics. The ROC curve illustrated the trade-off between sensitivity and specificity, while AUC scores quantified overall classification performance. This approach effectively identified digital biomarkers and optimized feature selection for predictive modeling.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003eStatistical analysis\u003c/h2\u003e \u003cp\u003eData normality was assessed using the Shapiro\u0026ndash;Wilk test. Depending on the distribution, either an independent t-test or a non-parametric test was used to compare physical and clinical characteristics between groups. Statistical analyses were performed using SPSS version 21.0 (IBM Corp., Armonk, NY, USA), with significance set at \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05. Additionally, data preprocessing and analysis were conducted in Python (version 3.8.8) using standard libraries, including pandas, NumPy, scikit-learn, and matplotlib.\u003c/p\u003e \u003c/div\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\"\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\"\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\"\u003eLASSO\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eLeast absolute shrinkage and selection operator,XGBoost,eXtreme Gradient Boosting\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePD-NC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePD with normal cognition\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePD-CD\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePD with cognitive decline\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eUPDRS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e\u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eUnified parkinson's disease rating scale\u003c/div\u003e \u003cdiv class=\"Description\"\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\"\u003e\u003cem\u003eR\u003c/em\u003e\u003csup\u003e\u003cem\u003e2\u003c/em\u003e\u003c/sup\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eCoefficient of determination\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCV\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eCross-validation\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\"\u003eTurnFR_CTL\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eContralateral temporal coordination of the left foot during faster speed rightward turning\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTurnFL_RANK_MaxAcc\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMaximum acceleration of the right ankle during faster speed leftward turning\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eFW_RANK_MaxJerk\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMaximum jerk of the right ankle during forward walking\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTurnPL_SLL\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eStride length of the left foot during preferred speed leftward turning\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eFW_PSIS_MaxGyr\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMaximum gyroscope at the posterior superior iliac spine during forward walking\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTurnPL_SLR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eStride length of the left foot during preferred speed rightward turning\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTurnFR_PSIS_SampEnAcc\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eSample entropy of acceleration at the posterior superior iliac spine during faster speed rightward turning\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eAUC‒ROC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003earea under the receiver operating characteristic curve.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eACKNOWLEDGEMENTS\u003c/strong\u003e\u003c/p\u003e\n\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\n\u003cp\u003e\u003cstrong\u003eAUTHOR CONTRIBUTIONS\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eB.K., C.Y., and S.C. conceived and designed the study. B.K., S.C., and H.P. recruited the participants. B.K., C.Y., S.C., H.P., H.C., J.H., and M.K. conducted data acquisition. B.K. and C.Y. analyzed and interpreted the data. B.K., C.Y., and S.C. drafted the manuscript. All authors reviewed and approved the final version of the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDATA AVAILABILITY STATEMENT\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets supporting the findings of this study are available from the corresponding author upon reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCOMPETING INTERESTS\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eSTATEMENT\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eETHICS APPROVAL AND CONSENT TO PARTICIPATE\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll procedures involving human participants were conducted in accordance with the ethical standards of the institutional and/or national research committee and the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. The study protocol and supplementary materials were approved by the Institutional Review Board of Dong-A University Hospital (Approval No. DAUHIRB-22-089). All participants provided written informed consent prior to data collection.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFUNDING\u003c/strong\u003e\u003c/p\u003e\n\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, data collection, analysis, interpretation, or manuscript writing.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCODE AVAILABILITY\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNo open-source code is available for this study. The code for training and testing the machine learning models was written in Python 3.8.8 using PyTorch 1.13.1 and Torchvision 0.14.1. Data management and feature processing scripts were written in Python 3.8.8 using pandas 1.2.4 and NumPy 1.20.1. The code used for the analysis can be obtained from the corresponding author upon request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eLeroi, I., McDonald, K., Pantula, H. \u0026amp; Harbishettar, V. Cognitive impairment in Parkinson disease: impact on quality of life, disability, and caregiver burden. \u003cem\u003eJ. Geriatr. Psychiatry Neurol.\u003c/em\u003e \u003cb\u003e25\u003c/b\u003e, 208\u0026ndash;214 (2012).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSasikumar, S. \u0026amp; Strafella, A. P. Imaging mild cognitive impairment and dementia in Parkinson's disease. \u003cem\u003eFront. Neurol.\u003c/em\u003e \u003cb\u003e11\u003c/b\u003e, 47 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAarsland, D. et al. Mild cognitive impairment in Parkinson disease: a multicenter pooled analysis. \u003cem\u003eNeurology\u003c/em\u003e \u003cb\u003e75\u003c/b\u003e, 1062\u0026ndash;1069 (2010).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePalavra, N. C., Naismith, S. L. \u0026amp; Lewis, S. J. Mild cognitive impairment in Parkinson\u0026rsquo;s disease: a review of current concepts. \u003cem\u003eNeurol. Res. Int.\u003c/em\u003e 576091 (2013). (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSalmanpour, M. R. et al. Optimized machine learning methods for prediction of cognitive outcome in Parkinson's disease. \u003cem\u003eComput. Biol. Med.\u003c/em\u003e \u003cb\u003e111\u003c/b\u003e, 103347 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHarvey, J. et al. Machine learning-based prediction of cognitive outcomes in de novo Parkinson\u0026rsquo;s disease. \u003cem\u003enpj Parkinsons Dis.\u003c/em\u003e \u003cb\u003e8\u003c/b\u003e, 150 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBooth, S., Park, K. W., Lee, C. S. \u0026amp; Ko, J. H. Predicting cognitive decline in Parkinson\u0026rsquo;s disease using FDG-PET\u0026ndash;based supervised learning. \u003cem\u003eJ. Clin. Invest.\u003c/em\u003e \u003cb\u003e132\u003c/b\u003e, e157074 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGramotnev, G., Gramotnev, D. K. \u0026amp; Gramotnev, A. Parkinson\u0026rsquo;s disease prognostic scores for progression of cognitive decline. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e9\u003c/b\u003e, 17485 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHayete, B. et al. A Bayesian mathematical model of motor and cognitive outcomes in Parkinson\u0026rsquo;s disease. \u003cem\u003ePLOS One\u003c/em\u003e. \u003cb\u003e12\u003c/b\u003e, e0178982 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHobson, P. \u0026amp; Meara, J. Mild cognitive impairment in Parkinson's disease and its progression onto dementia: a 16-year outcome evaluation of the Denbighshire cohort. \u003cem\u003eInt. J. Geriatr. Psychiatry\u003c/em\u003e. \u003cb\u003e30\u003c/b\u003e, 1048\u0026ndash;1055 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePigott, K. et al. Longitudinal study of normal cognition in Parkinson disease. \u003cem\u003eNeurology\u003c/em\u003e \u003cb\u003e85\u003c/b\u003e, 1276\u0026ndash;1282 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWood, K. L. et al. Different PD-MCI criteria and risk of dementia in Parkinson\u0026rsquo;s disease: 4-year longitudinal study. \u003cem\u003enpj Parkinsons Dis.\u003c/em\u003e \u003cb\u003e2\u003c/b\u003e, 15027 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGorji, A. \u0026amp; Fathi Jouzdani, A. Machine learning for predicting cognitive decline within five years in Parkinson\u0026rsquo;s disease: comparing cognitive assessment scales with DAT SPECT and clinical biomarkers. \u003cem\u003ePLOS One\u003c/em\u003e. \u003cb\u003e19\u003c/b\u003e, e0304355 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, H. M. et al. Prediction of cognitive progression in Parkinson's disease using three cognitive screening measures. \u003cem\u003eClin. Park Relat. Disord\u003c/em\u003e. \u003cb\u003e1\u003c/b\u003e, 91\u0026ndash;97 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlmgren, H. et al. Machine learning-based prediction of longitudinal cognitive decline in early Parkinson\u0026rsquo;s disease using multimodal features. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e, 13193 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCamargo, C. H. et al. Comparison of the use of screening tools for evaluating cognitive impairment in patients with Parkinson's disease. \u003cem\u003eDement. Neuropsychol.\u003c/em\u003e \u003cb\u003e10\u003c/b\u003e, 344\u0026ndash;350 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNasreddine, Z. S. et al. The Montreal Cognitive Assessment, MoCA: a brief screening tool for mild cognitive impairment. \u003cem\u003eJ. Am. Geriatr. Soc.\u003c/em\u003e \u003cb\u003e53\u003c/b\u003e, 695\u0026ndash;699 (2005).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGoldman, J. G. et al. Cognitive impairment in Parkinson\u0026rsquo;s disease: a report from a multidisciplinary symposium on unmet needs and future directions to maintain cognitive health. \u003cem\u003enpj Parkinsons Dis.\u003c/em\u003e 4, 19 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRedgrave, P. et al. Goal-directed and habitual control in the basal ganglia: implications for Parkinson's disease. \u003cem\u003eNat. Rev. Neurosci.\u003c/em\u003e \u003cb\u003e11\u003c/b\u003e, 760\u0026ndash;772 (2010).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBarbosa, A. F. et al. Gait, posture and cognition in Parkinson's disease. \u003cem\u003eDement. Neuropsychol.\u003c/em\u003e \u003cb\u003e10\u003c/b\u003e, 280\u0026ndash;286 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePieruccini-Faria, F. et al. Gait variability across neurodegenerative and cognitive disorders: results from the Canadian Consortium of Neurodegeneration in Aging (CCNA) and the Gait and Brain Study. \u003cem\u003eAlzheimers Dement.\u003c/em\u003e \u003cb\u003e17\u003c/b\u003e, 1317\u0026ndash;1328 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCoates, L., Shi, J., Rochester, L., Del Din, S. \u0026amp; Pantall, A. Entropy of real-world gait in Parkinson\u0026rsquo;s disease determined from wearable sensors as a digital marker of altered ambulatory behavior. \u003cem\u003eSens. (Basel)\u003c/em\u003e. \u003cb\u003e20\u003c/b\u003e, 2631 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShah, V. V. et al. Digital biomarkers of mobility in Parkinson\u0026rsquo;s disease during daily living. \u003cem\u003eJ. Parkinsons Dis.\u003c/em\u003e \u003cb\u003e10\u003c/b\u003e, 1099\u0026ndash;1111 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, W. et al. Wearable sensor-based quantitative gait analysis in Parkinson\u0026rsquo;s disease patients with different motor subtypes. \u003cem\u003enpj Digit. Med.\u003c/em\u003e \u003cb\u003e7\u003c/b\u003e, 169 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDel Din, S., Godfrey, A., Mazz\u0026agrave;, C., Lord, S. \u0026amp; Rochester, L. Free-living monitoring of Parkinson's disease: lessons from the field. \u003cem\u003eMov. Disord\u003c/em\u003e. \u003cb\u003e31\u003c/b\u003e, 1293\u0026ndash;1313 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMorris, R. et al. Cognition as a mediator for gait and balance impairments in GBA-related Parkinson\u0026rsquo;s disease. \u003cem\u003enpj Parkinsons Dis.\u003c/em\u003e \u003cb\u003e8\u003c/b\u003e, 78 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSun, Y. M., Wang, Z. Y., Liang, Y. Y., Hao, C. W. \u0026amp; Shi, C. H. Digital biomarkers for precision diagnosis and monitoring in Parkinson\u0026rsquo;s disease. \u003cem\u003enpj Digit. Med.\u003c/em\u003e \u003cb\u003e7\u003c/b\u003e, 218 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlbrecht, F. et al. Unraveling Parkinson's disease heterogeneity using subtypes based on multimodal data. \u003cem\u003eParkinsonism Relat. Disord\u003c/em\u003e. \u003cb\u003e102\u003c/b\u003e, 19\u0026ndash;29 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eInguanzo, A. et al. MRI subtypes in Parkinson\u0026rsquo;s disease across diverse populations and clustering approaches. \u003cem\u003enpj Parkinsons Dis.\u003c/em\u003e \u003cb\u003e10\u003c/b\u003e, 159 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAmboni, M. et al. Machine learning can predict mild cognitive impairment in Parkinson's disease. \u003cem\u003eFront. Neurol.\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e, 1010147 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRusso, M. et al. Kinematic and kinetic gait features associated with mild cognitive impairment in Parkinson\u0026rsquo;s disease. \u003cem\u003eIEEE Trans. Neural Syst. Rehabil Eng.\u003c/em\u003e \u003cb\u003e32\u003c/b\u003e, 2676\u0026ndash;2687 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGhosh, D., Pal, S., Lutz, M. \u0026amp; Luo, S. Ensemble survival analysis for preclinical cognitive decline prediction in Alzheimer's disease using longitudinal biomarkers. \u003cem\u003earXiv Preprint arXiv\u003c/em\u003e :250316645 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJanssen Daalen, J. M. et al. Digital biomarkers for non-motor symptoms in Parkinson\u0026rsquo;s disease: the state of the art. \u003cem\u003enpj Digit. Med.\u003c/em\u003e \u003cb\u003e7\u003c/b\u003e, 186 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMadakkatel, I., Zhou, A., McDonnell, M. D. \u0026amp; Hypp\u0026ouml;nen, E. Combining machine learning and conventional statistical approaches for risk factor discovery in a large cohort study. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e11\u003c/b\u003e, 22997 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHoops, S. et al. Validity of the MoCA and MMSE in the detection of MCI and dementia in Parkinson disease. \u003cem\u003eNeurology\u003c/em\u003e \u003cb\u003e73\u003c/b\u003e, 1738\u0026ndash;1745 (2009).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChou, K. L., Lenhart, A., Koeppe, R. A. \u0026amp; Bohnen, N. I. Abnormal MoCA and normal range MMSE scores in Parkinson disease without dementia: cognitive and neurochemical correlates. \u003cem\u003eParkinsonism Relat. Disord\u003c/em\u003e. \u003cb\u003e20\u003c/b\u003e, 1076\u0026ndash;1080 (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFiorenzato, E. et al. Optimal MMSE and MoCA cutoffs for cognitive diagnoses in Parkinson's disease: A data-driven decision tree model. \u003cem\u003eJ. Neurol. Sci.\u003c/em\u003e \u003cb\u003e466\u003c/b\u003e, 123283 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, J. et al. Lasso-based machine learning algorithm for predicting postoperative lung complications in elderly: a single-center retrospective study from China. \u003cem\u003eClin. Interv Aging\u003c/em\u003e. \u003cb\u003e18\u003c/b\u003e, 597\u0026ndash;606 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiampas, I. et al. Motor signs and incident dementia with Lewy bodies in older adults with mild cognitive impairment. \u003cem\u003eJ. Am. Geriatr. Soc.\u003c/em\u003e \u003cb\u003e73\u003c/b\u003e, 50\u0026ndash;62 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDi Tella, S. et al. Cognitive Reserve proxies can modulate motor and non-motor basal ganglia circuits in early Parkinson\u0026rsquo;s disease. \u003cem\u003eBrain Imaging Behav.\u003c/em\u003e \u003cb\u003e18\u003c/b\u003e, 220\u0026ndash;230 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSosnik, R., Fahoum, F., Katzir, Z., Mirelman, A. \u0026amp; Maidan, I. Key shifts in frontoparietal network activity in Parkinson\u0026rsquo;s disease. \u003cem\u003enpj Parkinsons Dis.\u003c/em\u003e \u003cb\u003e11\u003c/b\u003e, 2 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWilson, H. et al. Predict cognitive decline with clinical markers in Parkinson\u0026rsquo;s disease (PRECODE-1). \u003cem\u003eJ. Neural Transm (Vienna)\u003c/em\u003e. \u003cb\u003e127\u003c/b\u003e, 51\u0026ndash;59 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePenko, A. L., Streicher, M. C., Dey, T., Rosenfeldt, A. B. \u0026amp; Alberts, J. L. Parkinson\u0026rsquo;s gait kinematics deteriorates across multiple cognitive domains under dual-task paradigms. \u003cem\u003eClin. Neurol. Neurosurg.\u003c/em\u003e \u003cb\u003e197\u003c/b\u003e, 106083 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMorris, R. et al. Cognitive associations with comprehensive gait and static balance measures in Parkinson's disease. \u003cem\u003eParkinsonism Relat. Disord\u003c/em\u003e. \u003cb\u003e69\u003c/b\u003e, 104\u0026ndash;110 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSarasso, E. et al. Dual-task clinical and functional MRI correlates in Parkinson's disease with postural instability and gait disorders. \u003cem\u003eParkinsonism Relat. Disord\u003c/em\u003e. \u003cb\u003e91\u003c/b\u003e, 88\u0026ndash;95 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCaronni, A. et al. In Parkinson\u0026rsquo;s disease, dual-tasking reduces gait smoothness during the straight-walking and turning-while-walking phases of the Timed Up and Go test. \u003cem\u003eBMC Sports Sci. Med. Rehabil\u003c/em\u003e. \u003cb\u003e17\u003c/b\u003e, 42 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu, Y. et al. Non-motor symptoms and quality of life in tremor dominant vs postural instability gait disorder Parkinson\u0026prime;s disease patients. \u003cem\u003eActa Neurol. Scand.\u003c/em\u003e \u003cb\u003e133\u003c/b\u003e, 330\u0026ndash;337 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eArtusi, C. A. et al. Pisa syndrome in Parkinson's disease is associated with specific cognitive alterations. \u003cem\u003eFront. Neurol.\u003c/em\u003e \u003cb\u003e10\u003c/b\u003e, 577 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTait, P. et al. Neuroimaging and cognitive correlates of postural control in Parkinson\u0026rsquo;s disease: a systematic review. \u003cem\u003eJ. Neuroeng. Rehabil\u003c/em\u003e. \u003cb\u003e22\u003c/b\u003e, 24 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePal, G. et al. Global cognitive function and processing speed are associated with gait and balance dysfunction in Parkinson\u0026rsquo;s disease. \u003cem\u003eJ. Neuroeng. Rehabil\u003c/em\u003e. \u003cb\u003e13\u003c/b\u003e, 94 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBurnfield, M. Gait analysis: normal and pathological function. \u003cem\u003eJ. Sports Sci. Med.\u003c/em\u003e \u003cb\u003e9\u003c/b\u003e, 353 (2010).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGago, M. F. et al. How do cognitive and axial motor signs correlate in Parkinson\u0026rsquo;s disease? A 6-year prospective study. \u003cem\u003eJ. Neurol.\u003c/em\u003e \u003cb\u003e256\u003c/b\u003e, 1655\u0026ndash;1662 (2009).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMiri, A. L. et al. A biomechanical analysis of turning during gait in individuals with different subtypes of Parkinson's disease. \u003cem\u003eClin. Biomech. (Bristol)\u003c/em\u003e. \u003cb\u003e112\u003c/b\u003e, 106166 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePantall, A. et al. Postural dynamics are associated with cognitive decline in Parkinson's disease. \u003cem\u003eFront. Neurol.\u003c/em\u003e \u003cb\u003e9\u003c/b\u003e, 1044 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMorris, R. et al. Gait rather than cognition predicts decline in specific cognitive domains in early Parkinson\u0026rsquo;s disease. \u003cem\u003eJ. Gerontol. Biol. Sci. Med. Sci.\u003c/em\u003e \u003cb\u003e72\u003c/b\u003e, 1656\u0026ndash;1662 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang, J., Xu, J., Zhao, C., Peng, Y. \u0026amp; Wang, H. An ensemble feature selection method for high-dimensional data based on sort aggregation. \u003cem\u003eSyst. Sci. Control Eng.\u003c/em\u003e \u003cb\u003e7\u003c/b\u003e, 32\u0026ndash;39 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAbhisheka, B., Biswas, S. K. \u0026amp; Purkayastha, B. Infusing weighted average ensemble diversity for advanced breast cancer detection. \u003cem\u003eInt. J. Imaging Syst. Technol.\u003c/em\u003e \u003cb\u003e34\u003c/b\u003e, e23146 (2024).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Parkinson's disease, cognitive decline, digital biomarkers, multimodal features, machine learning, wearable sensors","lastPublishedDoi":"10.21203/rs.3.rs-6695263/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6695263/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis study developed machine learning models to predict cognitive decline in individuals with Parkinson\u0026rsquo;s disease (PD) by integrating clinical characteristics and gait-derived digital biomarkers. Using data from 102 patients diagnosed with PD, we trained least absolute shrinkage and selection operator regression and eXtreme Gradient Boosting models on multimodal features, including clinical characteristics, physical function, lifestyle factors, and gait-derived features. Key predictors included Mini-Mental State Examination scores and gait biomarkers such as stride length of the left foot during preferred speed leftward turning, maximum acceleration of the right ankle during faster speed leftward turning, maximum jerk of the right ankle during forward walking, and the maximum gyroscope at the posterior superior iliac spine during forward walking. Stepwise regression explained 61.7% of the variance in Montreal Cognitive Assessment scores (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05). A logistic regression classifier using ten selected features achieved 76.5% accuracy and an area under the curve of 0.895 in identifying individuals with cognitive decline. These findings suggest that combining standard cognitive assessments with quantitative gait analysis enhances prediction and classification of cognitive impairment in PD, offering a clinically applicable strategy.\u003c/p\u003e","manuscriptTitle":"Multimodal Machine Learning Approach for Predicting Cognitive Decline in People with Parkinson’s Disease","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-06-16 08:37:30","doi":"10.21203/rs.3.rs-6695263/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"38fa7293-76e0-424b-8ec8-49b4201e4631","owner":[],"postedDate":"June 16th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":49976818,"name":"Health sciences/Biomarkers"},{"id":49976819,"name":"Health sciences/Diseases/Neurological disorders/Parkinsons disease"}],"tags":[],"updatedAt":"2025-08-18T08:09:02+00:00","versionOfRecord":[],"versionCreatedAt":"2025-06-16 08:37:30","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6695263","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6695263","identity":"rs-6695263","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}