Enhancing Patient Rehabilitation Outcomes: Artificial Intelligence-Driven Predictive Modeling for Home Discharge in Neurological and Orthopedic Conditions

Enhancing Patient Rehabilitation Outcomes: Artificial Intelligence-Driven Predictive Modeling for Home Discharge in Neurological and Orthopedic Conditions | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Enhancing Patient Rehabilitation Outcomes: Artificial Intelligence-Driven Predictive Modeling for Home Discharge in Neurological and Orthopedic Conditions Leonardo Buscarini, Paola Romano, Elena Sofia Cocco, Carlo Damiani, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4383785/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 26 May, 2025 Read the published version in Journal of NeuroEngineering and Rehabilitation → Version 1 posted 13 You are reading this latest preprint version Abstract In recent years, the fusion of the medical and computer science domains has gained significant traction in the scientific research landscape. Progress in both fields has enabled the generation of a vast amount of data used for making predictions and identifying interesting clusters and pathways. The Machine Learning model's application in the medical domain is one of the most compelling and challenging topics to explore, bridging the gap between Artificial Intelligence (AI). and healthcare. The combination of AI and medical information offers the possibility to create tools that can benefit both healthcare providers and physicians. This enables the enhancement of rehabilitation therapy and patient care. In the rehabilitation context, this work provides an alternative perspective: prediction of patients’ home discharge upon completing the rehabilitation protocol. Demographic and clinical data were collected on 7282 inpatients from electronic Medical Record, each record was categorized into Neurological Patients (NP, N = 3222) or Orthopedic Patients (OP, N = 4060). To comprehend the most suitable machine learning model, an extensive data preprocessing phase was essential. This included steps such as variable recoding, scaling, and comparing various dataset balancing methods to enhance the model’s performance. Random Forest model was selected after a careful review and comparison of algorithms commonly utilized in the clinical-rehabilitative domain. Following a significant hyperparameter optimization phase through a grid search technique, we achieved an accuracy rate of 97% for OP and slightly lower (94%) for NP. This work points out the increasing importance of AI in medicine, especially in the realm of personalized rehabilitation. The use of such approaches could signify a transformative shift in healthcare. The integration of machine learning not only enhances the precision of treatment but also opens new possibilities for patient-centered care, improving outcomes and quality of care for individuals undergoing rehabilitation. Personalized rehabilitation machine learning classification algorithm random forest unbalance hyperparameters optimization accuracy Figures Figure 1 Figure 2 Figure 3 1. INTRODUCTION In the global population, approximately one in seven individuals is forced to live with disabilities every day. This rate is on the verge of increasing, primarily due to the high incidence of chronic diseases and the demographic shift in Western countries[ 1 ]. For this reason, rehabilitation is considered a priority of the 21st century for healthcare systems, as well as disease treatment and prevention[ 2 ]. From a patient’s perspective, it’s very important to know whether he or she will return to independently perform Activities of Daily Living (ADL) such as dressing, eating, or walking. In recent years, the concept of personalized rehabilitation has incredible development; nowadays, it’s considered a personalized multimodal process aimed at improving patient autonomy[ 3 ]. Prediction in rehabilitation outcomes using AI is a promising area of research. Several studies discuss the potential of AI in rehabilitation, including its use in assisting rehabilitation sessions, evaluating treatment progress, and providing prognosis regarding the risk of complications or treatment success. These advancements in AI have the potential to improve decision-making, develop precision medicine tools, and optimize rehabilitation programs [ 4 – 6 ]. Determining the rehabilitative protocol to use, depending on the patient's medical history and clinical features, is a fundamental purpose. Consequently, recent clinical studies have focused on identifying specific targets that allow us to understand whether the chosen protocol is the most suitable for the patient. Positive outcomes have been achieved using parameters like the modified Barthel Index (mBI) at discharge or the Length of Stay (LOS) [ 7 – 11 ].The leitmotifs of this work are personalized rehabilitation and management of human and non-human resources in healthcare facilities. It integrates clinical information with the remarkable potential of Artificial Intelligence (AI), particularly within its branch known as Machine Learning (ML). The latter is primarily portrayed as the field that allows computers to learn how to make predictions without the need for explicit programming [ 12 – 16 ]. The concept of a learning machine capable of making predictions and drawing conclusions that are difficult to reach with conventional statistical methods is an old idea. However, statistics and ML are not completely distinct; ML algorithms are built upon common statistical methods, and they continue to advance alongside AI development. Statistics focuses on the verification of a hypothesis, whether null (H 0 ) or alternative (H 1 ), to assess how well the data distribution fits other known models, such as the Gaussian one. On the other hand, the aim of ML is prediction through algorithms that attempt to be as generalizable and applicable as possible for new validation datasets[ 17 ]. In this context, our work goal is to provide clinicians and healthcare professionals with a tool capable of estimating the likelihood of patient home discharge at the end of rehabilitation based on individual characteristics. So far, literature provides recent studies that demonstrate the efficacy and the reliability of AI algorithms in the rehabilitation outcomes analysis. Santilli et al. [ 6 ] leveraged a 3-year dataset (4050 patients) comprising Acceptance and Discharge Report of Rehabilitation (ADR-r) data to predict mBI at discharge, exploiting an in-built function for importance ranking. Concerning the Intensive Care Unit (ICU), Safaei et al. [ 18 ] proposed a CatBoost model to predict patient mortality using the discharge status variable and ten other impactful characteristics. Rufo et al.[ 19 ] employed a Light Gradient Boosting Machine (LightGBM) algorithm and other significant classification models, including Random Forest (RF) and eXtreme Gradient Boosting (XGBoost), to predict the condition of diabetes mellitus with reliable results. In recent years, heightened stress on healthcare management, especially during the COVID-19 pandemic, has led to the publication of several works on prediction analysis. Yan et al. [ 20 ] utilized 485 patient blood samples to assess three important biomarkers for quickly predicting mortality risk in COVID-19 individuals. Drawing on the classification and regression ML models employed in these studies, our contribution was to increase the literature knowledge about the impact of rehabilitation therapies on orthopedic and neurological patients in terms of home discharge rate after the treatment period. Our investigation aimed to conduct an intensive and in-depth preprocessing phase on 8468 ADR-r data, addressing the issue of unbalanced distribution in the home discharging target variable. We confirmed previous research findings regarding the importance of certain characteristics in patient autonomy recovery [ 6 ] and provided an accurate clinical model built on a massive input dataset to predict discharging status that could help healthcare management systems. The implemented prototyping tool could be used to explore new and interesting targets for further analysis. 2. MATERIALS AND METHODS The study protocol was approved by the ethical committee of the IRCCS San Raffaele Pisana of Rome on 18/07/2018 (code number 07/18). 2.1 Original dataset and initial cleaning steps The original analysis dataset includes clinical and demographic data of adults admitted to the neurology and orthopedic departments of a rehabilitation hospital in Italy from January 2015 to August 2022. The completion of patients’ ADR-r form [ 21 ] resulted in the collection of data for 10520 individuals, whose information is distributed across 120 initial features. The dataset size reduced to 8468 after the duplicate removal (2052). We anonymized dataset rows by assigning a unique ID to each patient and collected personal information about them, such as gender, age, marital status, date, place of birth, and more. Clinical data, including the primary reason for rehabilitation, any associated medical conditions, impairments, and admission/discharge mBi scores were collected. 2.2 Dataset cleaning and encoding Underlying medical conditions were defined establishing 13 general macro-categories. Through categorization based on Basic Pathology , the dataset has been divided into two subsets: Neurologic Patients (NP), related to individuals whose basic pathology is primarily neurological, and Orthopedic Patients (OP), which pertains to orthopedic basic pathology. An additional macro-category called Not Attributable didn’t belong to any of the identified 13 macro-categories and wasn’t considered in the further analysis. Details are shown in Table 3 . The Basic Pathology consists of alphanumeric codes established by ICD-9-CM, which is the ninth revision of this classification system with clinical modifications [ 21 ]. Most of the used algorithms are unable to interpret a variable with values represented as text strings. This is what happens in the Basic Pathology variable; for this reason, a numeric encoding was necessary to ensure accurate predictions[ 22 ], the one-hot encoding technique, commonly used in ML prediction analyses that involve categorical features, was applied; its application converts text strings into n binary numeric vectors, where n is the macro-categories amount: (i) 1 if the categorical feature corresponds to the vector label, and (ii) 0 in the other cases [ 23 ]. The same encoding method was applied to the variables related to the presence of any patient comorbidity in 17 macro-categories, as shown in Table 2 . Regarding the impairment’s variables, following the ADR-r form the possible values range from 0 to 9. The smaller the number, the lower the degree of impairment shown by the patient. We applied a different encoding technique called label encoding, resulting in modified categorical impairment variables transformed into numerical ones [ 22 ]. The labels used in this work are: (i) 0, (ii) 1, and (iii) 2, which are respectively associated with the absence, presence, and not-evaluability of patient impairment. A fundamental step in the data preprocessing phase was the definition of the target variable. Keeping in mind that our analysis aimed to create a model that allows the prediction of patient-specific home discharge, we defined the Discharge Type Category based on the Discharge Type Feature in the ADR-r form. To this purpose we used four labels to encode the categorization of patient discharge, assigning to each label a specific condition: (i) 1 - the patient is home discharged at the end of rehabilitation, (ii) 2 – the patient is not home discharged at the end of rehabilitation, (iii) Voluntary Discharge – the patient voluntarily leaves the clinic, (iv) Death – the patient deceased during rehabilitation protocol. Finally, we created the final target variable for our analysis, referred to as Home Discharge (HD). Applying one-hot encoding, we assigned the value 1 when the patient falls under label 1 of the Discharge Type Category, indicating the patient is home discharged. On the other hand, 0 is assigned to patients corresponding to the label 2, indicating that they will not home discharge at the end of rehabilitation. Cases related to other labels were not considered in the further analysis (voluntary discharge and death). The mBI is a well-established patient-centered clinical outcome measure that quantifies a patient’s disability, ranging from 0 to 100; this value derives from the sum of scores assigned by the clinician to each of the scale subfields[ 10 ]. It is commonly administrated in rehabilitation settings to evaluate the functional status of patients at admission and discharge[ 24 ]. In particular, it assesses a patient’s independence in performing ADL [ 25 , 26 ]. Label encoding technique divided the collected continuous variable into classes to better perform the model training. Based on previous studies [ 27 , 28 ], we created six classes corresponding to the degree of dependency exhibited by the patient (Table 1 ). The mBI henceforth will be referred to as “Categorized mBI” in the analysis. Table 1 Categories and labels of mBI at admission and discharge. ADR-r value Label Criteria Level of dependence mBI on admission or mBI at discharge 1 mBI Є [0, 24] Complete 2 mBI Є [ 25 , 49 ] Serious 3 mBI Є [50, 74] Moderate 4 mBI Є [75, 90] Mild 5 mBI Є [91, 99] Minimal 6 mBI = 100 Indipendent Before the final cleaning step, the dataset consisted of 8468 patients (rows) and the new 50 categorical and non-categorical variables (columns) relevant to our study to schedule the demographic and clinical characteristics of patients admitted to the clinic. Table 2 provides an overview of the encoded input features used for prediction models. Table 2 Summary of input and target variables for ML models with corresponding encoded labels in round brackets. Variable Name Variable Type Value Range Variable Group Patient ID Q [1-8468] NP/OP Gender C [female (0), male (1)] NP/OP Stroke C [A (0), P (1)] NP Parkinson C [A (0), P (1)] NP Multiple Sclerosis C [A (0), P (1)] NP Brain Tumors C [A (0), P (1)] NP Post mild/moderate trauma C [A (0), P (1)] NP Other neurological pathologies C [A (0), P (1)] NP Non-traumatic myeloradiculopathies C [A (0), P (1)] NP Hip arthroplasty C [A (0), P (1)] OP Knee arthroplasty C [A (0), P (1)] OP Femur osteosynthesis C [A (0), P (1)] OP Amputation C [A (0), P (1)] OP Spinal pathologies C [A (0), P (1)] OP Other orthopedic pathologies C [A (0), P (1)] OP Other types of comorbidities C [A (0), P (1)] NP/OP Hematological C [A (0), P (1)] NP/OP Dysmetabolic C [A (0), P (1)] NP/OP Hepatic C [A (0), P (1)] NP/OP Neurological C [A (0), P (1)] NP/OP Respiratory C [A (0), P (1)] NP/OP Diabetics C [A (0), P (1)] NP/OP Hypertension C [A (0), P (1)] NP/OP Heart diseases C [A (0), P (1)] NP/OP Cardiac arrhythmias C [A (0), P (1)] NP/OP Dyslipidemic C [A (0), P (1)] NP/OP Tumors C [A (0), P (1)] NP/OP Rheumatology/Orthopedics C [A (0), P (1)] NP/OP Circulatory complications C [A (0), P (1)] NP/OP Intestinal complications C [A (0), P (1)] NP/OP Kidney and urinary tract complications C [A (0), P (1)] NP/OP Different types of comorbidities C [A (0), P (1)] NP/OP Cognitive impairment C [A (0), P (1), NE (2)] NP/OP Behavior impairment C [A (0), P (1), NE (2)] NP/OP Communication/language impairment C [A (0), P (1), NE (2)] NP/OP Sensory impairment C [A (0), P (1), NE (2)] NP/OP Manipulation impairment C [A (0), P (1), NE (2)] NP/OP Balance impairment C [A (0), P (1), NE (2)] NP/OP Locomotion impairment C [A (0), P (1), NE (2)] NP/OP Cardiovascular impairment C [A (0), P (1), NE (2)] NP/OP Respiratory system impairment C [A (0), P (1), NE (2)] NP/OP Ulcers C [A (0), P (1)] NP/OP Sphincter control impairment C [A (0), P (1), NE (2)] NP/OP Urinary system impairment C [A (0), P (1), NE (2)] NP/OP Nutrition impairment C [A (0), P (1), NE (2)] NP/OP Categorized mBI admission C [1 (CD) − 6 (I)] NP/OP Age Q [18–97] NP/OP Target variable: Home discharge C [NHD (0), HD (1)] NP/OP Q: quantitative variables; C: categorical variables; A: absence of pathology/impairment; P: presence of pathology/impairment; NE: not-evaluability; CD: complete dependence; I: independence; NHD: not home discharge; HD: home discharge. 2.3 Analysis dataset To address the presence of missing values and the dataset integrity, we didn’t consider patients for further analysis when: the admission and/or discharge mBI was not recorded in the ADR-r form; Discharge Type Category variable returned Voluntary Discharge or Death; ICD-9-CM code in ADR-r form wasn’t registered; patient pathology was not attributable into identified categories for the Basic Pathology . Figure 1 provides an overview of the analysis dataset building. The definitive analysis dataset consists of 7282 patients (rows): 3222 were admitted for a neurological basic pathology and 4060 for orthopedic conditions. Table 3 shows substantial dataset details for NP and OP groups. Table 3 Demographic insights on model input patient. Demographic Data NP OP Total Sample size 3222 4060 7282 (44% NP; 56% OP) Mean age ( \(\pm\) St. Dev.) 73 ( \(\pm\) 13) 74 ( \(\pm\) 11) 74 ( \(\pm\) 12) Median mBI at admission ( \(\pm\) St. Dev.) 30 ( \(\pm\) 13) 40 \((\pm\) 9) 37 ( \(\pm\) 12) Median mBI at discharge ( \(\pm\) St. Dev.) 73 ( \(\pm\) 26) 91 ( \(\pm\) 18) 86 ( \(\pm\) 23) Median mBI change ( \(\pm\) St. Dev.) 41 ( \(\pm\) 20) 48 ( \(\pm\) 14) 46 ( \(\pm\) 17) Gender: Male (%) 1715 (53%) 1399 (34%) 3114 (43%) Gender: Female (%) 1507 (47%) 2661 (66%) 4168 (57%) Home discharge ratio (%) 2742 (85%) 3800 (94%) 6542 (90%) 2.4 Supervised Learning Algorithms In this study, we tested the supervised learning models. The goal was to implement a predictive function that can map the features exhibited at the time of a patient's admission (X) to a target variable (y). This is achieved using a training set consisting of tuples formed by the output variable and the i-th patient features[ 29 ]. Generally, the supervised ML domain is employed to solve classification and regression problems [ 30 ]. When the target variable is categorical, we refer to it as classification; on the other hand, a quantitative target variable is used in regression algorithms. The principal aim was to predict label 1 of the home discharge variable relied on the individual’s characteristics observed at admission. We trained and tested five of the most widely used classification algorithms in the rehabilitation field: some of them relied on decision trees, such as Random Forest (RF) and eXtreme Gradient Boosting (XGBoost), as well as others like Gradient Boosting (GB), Light Gradient Boosting Machine (LightGBM), and Categorical Boosting (CatBoost). The dataset was partitioned into NP and OP, which were further divided into training (70%) and test (30%) sets. The preprocessing phase, discussed in sections 2.5 and 2.6, as well as the prediction step, was carried out using Python 3.10.12 and proper libraries, including SciKit-Learn (for model learning)[ 31 ], Imbalanced-Learn (for dataset imbalance issues), Matplotlib (for graphical visualization), SciPy (for statistical tests), Numpy (for mathematical operations)[ 32 ] and Pandas (for data manipulation and analysis). We improved the efficiency of the model and, consequently, prediction accuracy through an optimization phase of hyper-parameters using a grid search technique. Hyper-parameters are specific variables of models that can be configured during the training step; their tuning (HPT) influences model performance[ 33 ]. We chose to use the grid search because it exhaustively investigates all hyperparameter combinations, returning their best values. However, it’s worth noting that this search process is computationally and temporally demanding[ 34 ]. During the splitting step, there is a risk of making a significant mistake. It is common to train the model with a dataset that may not allow it to take full advantage of the maximum available information. This can potentially hinder the achievement of the most accurate prediction. Randomly, a patient may be assigned to either the training or the test set; however, if we are in the latter condition, their information will not be used for model learning. This approach can lead to a high risk of overfitting: a model exhibits high accuracy in predictions on the training set but performs poorly on the test set. A solution to address this issue is the k-fold Cross-Validation (CV) method. It involves dividing the dataset into k folds, where k is a positive integer: k-1 folds are used for model training, and one-fold is reserved for testing. In this analysis, we choose k = 10, resulting in 10 distinct subsets, each of which was used as the test set in each iteration. The accuracy of the final model is obtained by averaging the accuracies of the 10 iterations [ 34 ]. This process also makes it possible to provide approximate model validation when it is not possible to give input to a dataset never considered in the training and testing phase in a tight time frame. The combination of k-CV and HPT enables the optimal training and fine-tuning of models used in our work. 2.5 Feature Scaling The feature scaling step has a significant impact on the quality of predictions. Its application ensures that variables distributed in large value ranges do not dominate features distributed in smaller intervals [ 35 ], to guarantee the latter mentioned be examined by the model [ 22 ]. There are two main techniques for feature scaling: Standard Scaling and Min-Max Scaling. In this work, we used the Min-Max Scaling technique, which involves applying a formula to the variables’ values to be scaled: \({X}_{norm}= \frac{X- {X}_{min}}{{X}_{max}- {X}_{min}}\) 1) where X is the original value of the variable to be scaled, X min is the minimum value, X max is the maximum value and X norm is the new value obtained after the application of this scaling method. This technique ensures that the distribution range of the scaled variable is reduced to [0, + 1] if it shows only positive values or to [-1, + 1] if both positive and negative values are present. This technique is powerful when Standard Scaling is not suitable, such as when the variable to be scaled doesn’t fit a Gaussian curve[ 36 ], as in the case of the variable Age in our study. Section 3.1 shows the results of non-normality tests applied to this variable. 2.6 Balancing Methods The last preprocessing step involves the use of balancing techniques to address the unbalanced distribution of the target variable: 0 (15%) vs 1 (85%) for the NP set; 0 (6%) vs 1 (94%) for the OP set. This study shows a binary classification analysis; in this case, it’s common to observe a high unbalance of target variable distribution. This condition adversely affects the predictive performance of the model[ 22 ]. Not solving this problem results in an algorithm that tends to accurately predict the most common classes, at the expense of the rare ones [ 37 , 38 ]. Proposed solutions in the scientific landscape are mainly based on resampling techniques. Usually, in ML analysis we observe the low number of minority classes rather than the majority class abundance. For this reason, oversampling (synthesis) is preferred to undersampling (reduction)[ 39 ]. This study compared the main methodologies of both approaches, including Random Over Sampling (ROS)[ 38 ], Synthetic Minority Oversampling TEchnique (SMOTE)[ 40 ], ADAptive SYNthetic (ADASYN)[ 41 ], as well as Random Under Sampling (RUS)[ 42 ] and Cluster Centroids (CC) [ 43 ]. In addition, the SMOTE-Tomek technique, which combines both methodologies simultaneously, was also considered [ 44 ]. 2.7 Evaluation Metrics Once the model is implemented, we must establish the effectiveness of its performance. Since this is a classification task, the main evaluation metrics used to compare predictions are Accuracy, Precision, Recall and F1-score[ 45 ]. The possible range of these values is [0, 1]: the closer to the right extreme, the better the model's ability to correctly predict instances belonging to both Home discharge classes. 3. RESULTS Figure 1 is a graphical representation of ML input datasets construction, along with their absolute and percentage frequencies. Duplicate entries, absence of admission and/or discharge mBI scores, and other criteria were considered in constructing the analysis dataset. 3.1 Non-normality verification We needed to conduct some statistical checks to determine the appropriate technique for scaling the Age quantitative variable. These checks were performed to confirm statistically the non-normal distribution, justifying the subsequent application of the Min-Max Scaling technique. In addition to visual inspection (Fig. 2 ) and a comparison of nominal percentiles (Fig. 3 ), two statistical tests were also applied for a quantitative assessment. The Shapiro-Wilk test is properly employed for small sample sizes but remains effective for larger datasets. The Kolmogorov-Smirnov test is primarily applicable to larger datasets. In both checks, we chose a significance level (alpha) of 0.05. Table 4 shows the results for the Age variable in NP and OP. For this reason, we accepted the alternative hypothesis: Age variable for both datasets doesn’t fit a normal distribution. Table 4 Statistical confirmation of the non-normal distribution of the Age variable in NP and OP datasets. Test p-value NP p-value OP Shapiro-Wilk 7.76E-35 1.08E-34 Kolmogorov-Smirnov 4.34E-34 8.68E-30 Quantitative assessments justify the application of the Min-Max Scaler to this variable, helping to avoid problems that could lead to less accurate predictions. 3.2 Best combination between model and balancing techniques We used accuracy values to select the optimal combination of balancing techniques and classifiers for achieving accurate predictions. Accuracy is a key parameter in classification analyses. Results for NP and OP are shown in Table 5 . Table 5 Accuracy values for different combinations of balancing techniques and classifier predictions in the NP and OP datasets. Accuracy ROS RUS SMOTE SMOTE Tomek ADASYN CC NP OP NP OP NP OP NP OP NP OP NP OP GB 0.7224 0.7474 0.6285 0.6603 0.7898 0.8114 0.7858 0.8121 0.7721 0.8152 0.7326 0.8333 RF 0.9502 0.9750 0.6701 0.6282 0.8706 0.8908 0.8588 0.9067 0.8571 0.8966 0.7396 0.8526 XGBoost 0.9064 0.9364 0.6528 0.5769 0.8767 0.9456 0.8856 0.9464 0.8870 0.9430 0.7500 0.8654 LightGBM 0.8554 0.8904 0.6424 0.6154 0.8755 0.9386 0.8606 0.9432 0.8589 0.9359 0.7500 0.8462 CatBoost 0.8518 0.9075 0.6701 0.6090 0.8645 0.9364 0.8667 0.9415 0.8511 0.9341 0.7431 0.8526 The highest accuracy is achieved by applying ROS (Random OverSampling) as the balancing technique and the Random Forest Classifier as the ML model. This is true for both neurological and orthopedic patients. 3.3 Feature Importance Ranking Using built-in functions of various models in the Python SK-Learn library, it’s possible to determine a feature importance ranking in predicting the outcome. As already shown in Table 5 , the model with the highest accuracy is the RF classifier. Table 6 shows the ranking obtained with this model for NP and OP, respectively. Table 6 Top 20 most critical features identified through Random Forest (RF) feature ranking in the NP and OP datasets. Rank NP OP 1 Age Age 2 Categorized mBI on admission Urinary system condition 3 Balance impairment Knee arthroplasty 4 Cognitive impairment Balance impairment 5 Nutrition impairment Hypertension 6 Sensory impairment Different types of comorbitidies 7 Gender Hip arthroplasty 8 Manipulation impairment Gender 9 Neurological Rheumatology/Orthopedics 10 Hypertension Amputation 11 Sphincter control impairment Ulcers 12 Communication/language impairment Diabetes 13 Urinary system condition Dysmetabolic 14 Ulcers Other orthopedic pathologies 15 Diabetes Dyslipidemic 16 Different types of comorbidities Categorized mBI on admission 17 Behavior impairment Other neurological pathologies 18 Cardiovascular impairment Cardiovascular impairment 19 Heart diseases Heart diseases 20 Cardiac arrhythmias Tumors Both RF rankings confirm that Age holds the greatest influence in forecasting patient discharge destination at the end of the rehabilitation program, simultaneously validating the substantial importance of cognitive impairment. Urinary and cardiac systems hold a place among the top twenty most influential features, along with hypertension diagnosis. In addition, by logical expectations, substantial weight is also associated with balance impairment, ranking third for NP and fourth for OP. The mBI proves to be a significant variable for NP; its importance decreases among orthopedic ones but still ranks among the top 20 most influential predictors. Finally, it can be observed how the presence of comorbidities influences the prediction of the outcome in both subsets [ 46 ] and how the gender isn’t a top five important feature. 3.4 Hyperparameter Optimization This step consists of the optimization of the classifier hyperparameters. We conducted a grid search on the most common values of RF classifier specific parameters, such as n_estimators, max_depth, min_samples_split, min_samples_leaf, and max_features. The aim is to find the best combination of those hyper-parameters to optimize the model performance. Table 7 shows the best values found for NP and OP sets. Table 7 Best combination of RF hyperparameters in NP and OP implemented model. Best values HPT NP OP n estimators 200 100 max depth 20 20 min samples split 2 5 min samples leaf 1 1 max features sqrt auto 3.5 Model evaluation Table 8 shows the values obtained from the model on the respective test sets of NP and OP, after training it with the remaining 70% of both data sets. A 10-fold cross-validation was employed to provide an approximate validation of models and to mitigate overfitting, and hyperparameters were optimized. The table evaluation metrics were used to analyze model performance in predicting the outcome of our study, both in terms of positive (patient home discharge) and negative instances (patient not home discharge). In both cases, values were calculated using SK-learn specific functions (accuracy_score, f1_score, precision_score, recall_score). They often get close to 1, showing that the model performs very well in predicting home discharge, both for NP and OP. Table 8 Model Evaluation Metrics in NP and OP dataset. Model Evaluation Metrics NP OP Accuracy 93.8% 96.7% Precision 94.4% 96.9% Recall 93.9% 96.8% f1-score 93.7% 96.7% 4. DISCUSSION This study aimed at advancing the patient-centered approach to rehabilitation by integrating medical knowledge with supervised learning methods to predict probability of patients' home discharge considering modified Barthel Index and ADR-r’s data. Utilizing a substantial dataset for robust model training and employing machine learning techniques represent significant strengths of our analysis. However, a considerable challenge arose due to the unbalanced distribution of the target variable, which we addressed through extensive data preprocessing. This phase significantly contributed to the excellent performance evaluation metrics observed in the implemented classifier, detailed in Table 8 . Within the field of rehabilitation, selecting appropriate protocols for each patient remains a formidable challenge due to the lack of uniformity in their application among individuals showing similar demographic and clinical conditions. Despite this challenge, our work aims to simplify the selection process, narrowing down potentially applicable protocols to promote a more consistent approach to rehabilitation program choice and enhance overall personalization. A common limitation of ML work is the absence of a model validation set. While both training and testing steps are crucial for verifying model learning and performance in supervised ML analysis, a complete model evaluation requires the use of an unseen input dataset. In the context of our work, this limitation is planned to be addressed by collecting a substantial number of new ADR-r forms related to clinic-admitted patients over the next few years. These will be utilized as unseen data for model validation, thereby enhancing the actual functionality of the implemented tool. Despite that, an approximate model validation is obtained performing a 10-fold Cross Validation. Moving to the specific findings of our analysis, we aim to provide a predictive model for patient home discharge at the end of the rehabilitation period. The dataset was divided into two subsets based on the nature of the basic pathology encountered: Neurological (NP) and Orthopedic (OP). Our analysis identified the top 20 characteristics with the most significant impact on outcomes in both subsets, detailed in Table 6 . The results align with previous studies, confirming Age as the leading predictive feature for both subsets, consistent with recent pioneering work targeting mBI at the end of the rehabilitation program [ 6 , 47 – 50 ]. Notably, patient independence (mBI) has a more substantial impact on NP predictions than on OP, highlighting the lower autonomy levels in performing ADLs among individuals undergoing rehabilitation for basic neurological pathologies. Despite its importance decreases among orthopedic ones, it still ranks among the top 20 most influential predictors, thus confirming the findings in Masaru Uragami's study on readmission of patients with hip fracture[ 50 ]. In the case of OP, the most influential variable ranking reveals that the level of urinary system impairment is the second most important feature in predicting home discharge. This suggests that elderly patients facing urinary system issues are likely to have a lower probability of home discharge after rehabilitation, as specific mBI sub-items relate to intestinal and urinary continence, as well as toilet use. The consistent identification of age as the leading predictive feature for both NP and OP subsets reinforces the importance of age in assessing patient outcomes and discharge disposition [ 6 , 47 ]. Moreover, the observation that patient independence exerts a more substantial impact on predictions for NP compared to OP highlights the complexity of rehabilitation needs among individuals with neurological conditions. The significant influence of urinary system condition as the second most important feature in predicting discharge destination among OP underscores the importance of comprehensive assessment and targeted interventions to address patient-specific needs [ 48 , 49 ]. Elderly patients with urinary system issues may require multidisciplinary care and support to facilitate their transition to home settings post-rehabilitation. These implications emphasize the necessity for tailored interventions aimed at improving patient outcomes and reducing healthcare utilization post-rehabilitation. Future research may focus on exploring additional factors contributing to discharge destination after rehabilitation, refining predictive models, and implementing targeted interventions to optimize patient care and reduce healthcare costs. The extensive data pre-processing phase, including the use of the ROS technique and hyperparameter tuning of the RF classifier, enhanced our predictive model performance. Its evaluation metrics, such as accuracy, precision, recall, and F1 score, yielded promising results for both OP and NP sets (Table 8 ). The differences in performance are rooted in clinical principles, where neurological conditions impact a wider range of human systems and functions than orthopedic pathologies. The greater diversity of instances among NP patients results in a reduced model accuracy for the NP classifier (93.8%) compared to OP (96.7%), highlighting the intricate and multifaceted impact of neurological conditions on patient health. This underscores the challenge faced by AI engineers in predicting rehabilitation outcomes for individual neurologic patients. 5. CONCLUSIONS This study introduces a medical tool leveraging ADR-r data to predict patient clinic home discharge post-rehabilitation, facilitating the personalization of treatment plans. The RF output could be used to align rehabilitation programs with patient needs, enabling the full recovery of ADLs. Additionally, it provides timely alerts to family members and caregivers, optimizing home care management. Successful home discharge predictions could be linked to in-home health assistance, while the other case predictions support preparing patients, physically and mentally, for a not home discharge. The model could also be used for clinical resource management, improving bed allocation, medical equipment, and staffing. Future hopes spin around gathering new ADR-r forms to validate more robustly the implemented tool, which will persist in being refined and enhanced along with the significant advancements in AI and ML. This ongoing process will simplify the exploration of new clusters, pathways, and targets, thereby improving the personalization of rehabilitation. The successful integration of machine learning techniques into clinical practice offers promising avenues for enhancing patient-centered care and improving rehabilitation outcomes. By accurately predicting patient’s home discharge, healthcare providers can tailor treatment plans to individual patient needs, optimize resource allocation, and provide timely support to patients and their caregivers. This study represents a significant step forward in employing artificial intelligence and machine learning to optimize patient care and improve healthcare delivery. Abbreviations OP Orthopedic Patients; NP Neurologic Patients; ML Machine Learning; AI Artificial Intelligence; ADL Activities of Daily Living ; mBI Modified Barthel Index; RF Random Forest; XGBoost eXtreme Gradient Boosting; GB Gradient Boosting; CatBoost Categorical Boosting; LightGBM Light Gradient Boosting Model; ADR-r Acceptance and Discharge Report of rehabilitation; ICD-9-CM International Classification of Diseases, revision number 9 with Clinical Modifications; HPT Hyper-Parameter Optimization; k-CV k-fold Cross Validation; ROS Random Over Sampling; SMOTE Synthetic Minority Oversampling TEchnique; ADASYN ADAptive SYNthetic; RUS Random Under Sampling; CC Cluster Centroids; HPT Hyper-Parameter Tuning. Declarations AUTHORSHIP MF has made substantial contributions to conception and design of the study. SP and CD carried out the data collection. FI, PR and LB designed the algorithm for data analysis. SP and FI participated in the study design and coordination. LB, PR, ESC performed the statistical analysis. LB has elaborated the original draft of the manuscript. MF, FI, SP, and PR and ESC participated in the manuscript revisions. MF gave the final approval of the version. All authors contributed to the article and approved the submitted version. Funding statement This study has been funded by the Italian Ministry of Health (Ricerca Corrente). Conflict of interest disclosure The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. AVAILABILTY OF DATA AND MATERIALS The dataset supporting the conclusion of this article is available in Zenodo repository with this DOI: 10.5281/zenodo.10991206 and this link: https://zenodo.org/records/10991206 ETHICS APPROVAL STATEMENT The study protocol was approved by the ethical committee of the IRCCS San Raffaele Pisana of Rome on 18/07/2018 (code number 07/18). PATIENT CONSENT STATEMENT Each patient has received and signed informed consent. PERMISSION TO REPRODUCE MATERIAL FROM OTHER SOURCES No data have been reproduced. References Organization WH. WHO global disability action plan 2014-2021: Better health for all people with disability [Internet]. World Health Organization; 2015 [cited 2024 Mar 26]. Available from: https://apps.who.int/iris/bitstream/handle/10665/199544/9789241509619_eng.pdf Gimigliano F, Negrini S. The World Health Organization" rehabilitation 2030: a call for action". European Journal of Physical and Rehabilitation Medicine. 2017;53:155–68. Negrini S, Meyer T, Arienti C, Kiekens C, Pollock A, Selb M, et al. The 3rd Cochrane Rehabilitation Methodology Meeting:" Rehabilitation definition for scientific research purposes". European journal of physical and rehabilitation medicine. 2020;56:658–60. Rahgozar P. AIM in Rehabilitation. Artificial Intelligence in Medicine [Internet]. Springer; 2022 [cited 2024 Mar 27]. p. 1809–17. Available from: https://link.springer.com/content/pdf/10.1007/978-3-030-64573-1_177.pdf Mennella C, Maniscalco U, De Pietro G, Esposito M. The Role of Artificial Intelligence in Future Rehabilitation Services: A Systematic Literature Review. IEEE Access. 2023;11:11024–43. Santilli V, Mangone M, Diko A, Alviti F, Bernetti A, Agostini F, et al. The Use of Machine Learning for Inferencing the Effectiveness of a Rehabilitation Program for Orthopedic and Neurological Patients. International Journal of Environmental Research and Public Health. 2023;20:5575. Koh GC-H, Chen C, Cheong A, Choo TB, Pui CK, Phoon FN, et al. Trade-offs between effectiveness and efficiency in stroke rehabilitation. International Journal of Stroke. 2012;7:606–14. Chow P, Chen C, Cheong A, Fong NP, Chan KM, Tan BY, et al. Factors and trade-offs with rehabilitation effectiveness and efficiency in newly disabled older persons. Archives of physical medicine and rehabilitation. 2014;95:1510-1520. e4. Damiani C, Mangone M, Paoloni M, Goffredo M, Franceschini M, Servidio M, et al. Trade-Offs with rehabilitation Effectiveness (REs) and Efficiency (REy) in a sample of Italian disabled persons in a in post-acuity rehabilitation unit. Ann Ig. 2020;32:327–35. Collin C, Wade DT, Davies S, Horne V. The Barthel ADL Index: a reliability study. International disability studies. 1988;10:61–3. Hsueh I-P, Lin J-H, Jeng J-S, Hsieh C-L. Comparison of the psychometric characteristics of the functional independence measure, 5 item Barthel index, and 10 item Barthel index in patients with stroke. Journal of Neurology, Neurosurgery & Psychiatry. 2002;73:188–90. Krittanawong C, Zhang H, Wang Z, Aydar M, Kitai T. Artificial intelligence in precision cardiovascular medicine. Journal of the American College of Cardiology. 2017;69:2657–64. Rech J, Althoff K-D. Artificial intelligence and software engineering: Status and future trends. KI. 2004;18:5–11. Rajkomar A, Dean J, Kohane I. Machine learning in medicine. New England Journal of Medicine. 2019;380:1347–58. Saria S, Butte A, Sheikh A. Better medicine through machine learning: What’s real, and what’s artificial? PLoS medicine. Public Library of Science San Francisco, CA USA; 2018. p. e1002721. Mitchell TM. Machine Learning. New York: McGraw-Hill; 1997. Handelman GS, Kok HK, Chandra RV, Razavi AH, Lee MJ, Asadi H. eD octor: machine learning and the future of medicine. Journal of internal medicine. 2018;284:603–19. Safaei N, Safaei B, Seyedekrami S, Talafidaryani M, Masoud A, Wang S, et al. E-CatBoost: An efficient machine learning framework for predicting ICU mortality using the eICU Collaborative Research Database. Plos one. 2022;17:e0262895. Rufo DD, Debelee TG, Ibenthal A, Negera WG. Diagnosis of diabetes mellitus using gradient boosting machine (LightGBM). Diagnostics. 2021;11:1714. Yan L, Zhang H-T, Goncalves J, Xiao Y, Wang M, Guo Y, et al. An interpretable mortality prediction model for COVID-19 patients. Nature machine intelligence. 2020;2:283–8. Seccia R, Boresta M, Fusco F, Tronci E, Di Gemma E, Palagi L, et al. Data of patients undergoing rehabilitation programs. Data in brief. 2020;30:105419. Al-Shehari T, Alsowail RA. An insider data leakage detection using one-hot encoding, synthetic minority oversampling and machine learning techniques. Entropy. 2021;23:1258. El Affendi MA, Al Rajhi KHS. Text encoding for deep learning neural networks: A reversible base 64 (Tetrasexagesimal) Integer Transformation (RIT64) alternative to one hot encoding with applications to Arabic morphology. 2018 sixth international conference on digital information, networking, and wireless communications (DINWC). IEEE; 2018. p. 70–4. Pournajaf S, Pellicciari L, Proietti S, Agostini F, Gabbani D, Goffredo M, et al. Which items of the modified Barthel Index can predict functional independence at discharge from inpatient rehabilitation? A secondary analysis retrospective cohort study. International Journal of Rehabilitation Research. 2023;46:230–7. Liu F, Tsang RC, Zhou J, Zhou M, Zha F, Long J, et al. Relationship of Barthel Index and its Short Form with the Modified Rankin Scale in acute stroke patients. Journal of stroke and cerebrovascular diseases. 2020;29:105033. Ocagli H, Cella N, Stivanello L, Degan M, Canova C. The Barthel index as an indicator of hospital outcomes: A retrospective cross‐sectional study with healthcare data from older people. Journal of Advanced Nursing. 2021;77:1751–61. Shah S. Modified Barthel index or Barthel index (expanded). Compendium of Quality of Life Instruments, Part II Chichester, UK: Wiley and Sons. 1998; Sulter G, Steen C, De Keyser J. Use of the Barthel index and modified Rankin scale in acute stroke trials. Stroke. 1999;30:1538–41. Witten IH, Frank E. Data mining: practical machine learning tools and techniques with Java implementations. Acm Sigmod Record. 2002;31:76–7. Kotsiantis SB, Zaharakis ID, Pintelas PE. Machine learning: a review of classification and combining techniques. Artif Intell Rev. 2006;26:159–90. Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O, et al. Scikit-learn: Machine learning in Python. the Journal of machine Learning research. 2011;12:2825–30. Von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. The Lancet. 2007;370:1453–7. Jalaeian Zaferani E, Teshnehlab M, Khodadadian A, Heitzinger C, Vali M, Noii N, et al. Hyper-parameter optimization of stacked asymmetric auto-encoders for automatic personality traits perception. Sensors. 2022;22:6206. Younis EM, Zaki SM, Kanjo E, Houssein EH. Evaluating ensemble learning methods for multi-modal emotion recognition using sensor data fusion. Sensors. 2022;22:5611. Tangadpalliwar SR, Vishwakarma S, Nimbalkar R, Garg P. ChemSuite: A package for chemoinformatics calculations and machine learning. Chemical Biology & Drug Design. 2019;93:960–4. Cao XH, Stojkovic I, Obradovic Z. A robust data scaling algorithm to improve classification accuracies in biomedical data. BMC Bioinformatics. 2016;17:359. Fernández A, García S, Luengo J, Bernadó-Mansilla E, Herrera F. Genetics-based machine learning for rule induction: state of the art, taxonomy, and comparative study. IEEE Transactions on Evolutionary Computation. 2010;14:913–41. Haixiang G, Yijing L, Shang J, Mingyun G, Yuanyue H, Bing G. Learning from class-imbalanced data: Review of methods and applications. Expert systems with applications. 2017;73:220–39. Laurikkala J. Improving identification of difficult small classes by balancing class distribution. Artificial Intelligence in Medicine: 8th Conference on Artificial Intelligence in Medicine in Europe, AIME 2001 Cascais, Portugal, July 1–4, 2001, Proceedings 8. Springer; 2001. p. 63–6. Duan F, Zhang S, Yan Y, Cai Z. An oversampling method of unbalanced data for mechanical fault diagnosis based on MeanRadius-SMOTE. Sensors. 2022;22:5166. Haibo He, Yang Bai, Garcia EA, Shutao Li. ADASYN: Adaptive synthetic sampling approach for imbalanced learning. 2008 IEEE International Joint Conference on Neural Networks (IEEE World Congress on Computational Intelligence) [Internet]. Hong Kong, China: IEEE; 2008 [cited 2024 Mar 27]. p. 1322–8. Available from: http://ieeexplore.ieee.org/document/4633969/ Wu Y, Fang Y. Stroke prediction with machine learning methods among older Chinese. International journal of environmental research and public health. 2020;17:1828. Ali Z, Hayat MF, Shaukat K, Alam TM, Hameed IA, Luo S, et al. A proposed framework for early prediction of Schistosomiasis. Diagnostics. 2022;12:3138. Swana EF, Doorsamy W, Bokoro P. Tomek link and SMOTE approaches for machine fault classification with an imbalanced dataset. Sensors. 2022;22:3246. Kirk D, Catal C, Tekinerdogan B. Precision nutrition: A systematic literature review. Computers in Biology and Medicine. 2021;133:104365. Shah AA, Devana SK, Lee C, Bugarin A, Lord EL, Shamie AN, et al. Machine learning driven identification of novel patient factors for prediction of major complications after posterior cervical spinal fusion. Eur Spine J. 2022;31:1952–9. Adhiya J, Barghi B, Azadeh-Fard N. Predicting the risk of hospital readmissions using a machine learning approach: a case study on patients undergoing skin procedures. Front Artif Intell. 2024;6:1213378. Lin Y, Wu JY, Lin K, Hu YH, Kong GL. [Prediction of intensive care unit readmission for critically ill patients based on ensemble learning]. Beijing Da Xue Xue Bao Yi Xue Ban. 2021;53:566–72. Zhu Y, Venugopalan J, Zhang Z, Chanani NK, Maher KO, Wang MD. Domain Adaptation Using Convolutional Autoencoder and Gradient Boosting for Adverse Events Prediction in the Intensive Care Unit. Front Artif Intell. 2022;5:640926. Uragami M, Matsushita K, Shibata Y, Takata S, Karasugi T, Sueyoshi T, et al. A machine learning-based scoring system and ten factors associated with hip fracture occurrence in the elderly. Bone. 2023;176:116865. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 26 May, 2025 Read the published version in Journal of NeuroEngineering and Rehabilitation → Version 1 posted Editorial decision: Revision requested 17 Mar, 2025 Reviews received at journal 16 Mar, 2025 Reviews received at journal 02 Mar, 2025 Reviewers agreed at journal 02 Mar, 2025 Reviewers agreed at journal 26 Feb, 2025 Reviewers agreed at journal 16 Dec, 2024 Reviews received at journal 09 Sep, 2024 Reviewers agreed at journal 20 Aug, 2024 Reviewers agreed at journal 24 May, 2024 Reviewers invited by journal 08 May, 2024 Submission checks completed at journal 08 May, 2024 Editor assigned by journal 08 May, 2024 First submitted to journal 07 May, 2024 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-4383785","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":302025524,"identity":"aac26edf-ee56-448a-adc5-350ef8cd00ea","order_by":0,"name":"Leonardo Buscarini","email":"","orcid":"","institution":"IRCCS San Raffaele Roma","correspondingAuthor":false,"prefix":"","firstName":"Leonardo","middleName":"","lastName":"Buscarini","suffix":""},{"id":302025525,"identity":"5cf3fcd8-b957-409d-ad22-5c8b5317e3db","order_by":1,"name":"Paola Romano","email":"data:image/png;base64,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","orcid":"","institution":"IRCCS San Raffaele Roma","correspondingAuthor":true,"prefix":"","firstName":"Paola","middleName":"","lastName":"Romano","suffix":""},{"id":302025526,"identity":"5fb58901-da73-47fd-a139-6b733e2d6155","order_by":2,"name":"Elena Sofia Cocco","email":"","orcid":"","institution":"IRCCS San Raffaele Roma","correspondingAuthor":false,"prefix":"","firstName":"Elena","middleName":"Sofia","lastName":"Cocco","suffix":""},{"id":302025527,"identity":"8325ad30-e2e4-4ab2-9ffa-2b9782bf5e01","order_by":3,"name":"Carlo Damiani","email":"","orcid":"","institution":"IRCCS San Raffaele Roma","correspondingAuthor":false,"prefix":"","firstName":"Carlo","middleName":"","lastName":"Damiani","suffix":""},{"id":302025528,"identity":"b43d0f2b-8634-4b0f-a2b9-0f6e2f7fd284","order_by":4,"name":"Sanaz Pournajaf","email":"","orcid":"","institution":"IRCCS San Raffaele Roma","correspondingAuthor":false,"prefix":"","firstName":"Sanaz","middleName":"","lastName":"Pournajaf","suffix":""},{"id":302025529,"identity":"6818e3a5-ecf1-4efc-aaf3-25a0d92296a3","order_by":5,"name":"Marco Franceschini","email":"","orcid":"","institution":"IRCCS San Raffaele Roma","correspondingAuthor":false,"prefix":"","firstName":"Marco","middleName":"","lastName":"Franceschini","suffix":""},{"id":302025530,"identity":"521cdf75-5a0a-4784-8c35-d09f820537ef","order_by":6,"name":"Francesco Infarinato","email":"","orcid":"","institution":"IRCCS San Raffaele Roma","correspondingAuthor":false,"prefix":"","firstName":"Francesco","middleName":"","lastName":"Infarinato","suffix":""}],"badges":[],"createdAt":"2024-05-07 14:48:11","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4383785/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4383785/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s12984-025-01654-4","type":"published","date":"2025-05-26T15:57:27+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":56682178,"identity":"7b6c4a3c-f742-4a3c-af26-fe7a191e06ad","added_by":"auto","created_at":"2024-05-17 18:24:49","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":22859,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart detailing the data cleaning and splitting process.\u003c/p\u003e","description":"","filename":"image1.png","url":"https://assets-eu.researchsquare.com/files/rs-4383785/v1/4d10d07fa19541cf3edf67ff.png"},{"id":56682177,"identity":"168413b1-7bfc-4915-bfa2-67505c6fd26e","added_by":"auto","created_at":"2024-05-17 18:24:49","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":138303,"visible":true,"origin":"","legend":"\u003cp\u003eClinical examination of the non-normal Age distribution in NP and OP datasets. The red line shows the ideal Gaussian distribution, while the blue histograms depict the Age distribution in NP (left) and OP sets (right). Both exhibit a leptokurtic distribution: the histograms are asymmetric, leaning towards the right extreme, and reach higher levels, deviating from the normal symmetric red curve.\u003c/p\u003e","description":"","filename":"image2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4383785/v1/318ad64f34c31c74124d2368.jpg"},{"id":56682179,"identity":"4be4acd4-d599-4a64-91e2-58a761657758","added_by":"auto","created_at":"2024-05-17 18:24:50","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":156823,"visible":true,"origin":"","legend":"\u003cp\u003eGraphical representation of nominal percentiles to assess the non-normality of the Age distribution in NP (left) and OP (right) datasets. The green line corresponds to the normal percentiles of the Age variable, while the red dashed line represents the Gaussian distribution. The noticeable gap between the lines in each graph serves as a visual confirmation of the non-normal Age distribution.\u003c/p\u003e","description":"","filename":"image3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4383785/v1/5a5cc1064f83bc0925cba449.jpg"},{"id":83783548,"identity":"c52bad20-15be-42f9-a863-816677cb158d","added_by":"auto","created_at":"2025-06-02 16:11:37","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1490005,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4383785/v1/ca518d67-70b2-4734-9914-4c6b9ded5829.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Enhancing Patient Rehabilitation Outcomes: Artificial Intelligence-Driven Predictive Modeling for Home Discharge in Neurological and Orthopedic Conditions","fulltext":[{"header":"1. INTRODUCTION","content":"\u003cp\u003eIn the global population, approximately one in seven individuals is forced to live with disabilities every day. This rate is on the verge of increasing, primarily due to the high incidence of chronic diseases and the demographic shift in Western countries[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. For this reason, rehabilitation is considered a priority of the 21st century for healthcare systems, as well as disease treatment and prevention[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. From a patient\u0026rsquo;s perspective, it\u0026rsquo;s very important to know whether he or she will return to independently perform Activities of Daily Living (ADL) such as dressing, eating, or walking. In recent years, the concept of personalized rehabilitation has incredible development; nowadays, it\u0026rsquo;s considered a personalized multimodal process aimed at improving patient autonomy[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e].\u003c/p\u003e \u003cp\u003ePrediction in rehabilitation outcomes using AI is a promising area of research. Several studies discuss the potential of AI in rehabilitation, including its use in assisting rehabilitation sessions, evaluating treatment progress, and providing prognosis regarding the risk of complications or treatment success. These advancements in AI have the potential to improve decision-making, develop precision medicine tools, and optimize rehabilitation programs [\u003cspan additionalcitationids=\"CR5\" citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Determining the rehabilitative protocol to use, depending on the patient's medical history and clinical features, is a fundamental purpose. Consequently, recent clinical studies have focused on identifying specific targets that allow us to understand whether the chosen protocol is the most suitable for the patient. Positive outcomes have been achieved using parameters like the modified Barthel Index (mBI) at discharge or the Length of Stay (LOS) [\u003cspan additionalcitationids=\"CR8 CR9 CR10\" citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e].The leitmotifs of this work are personalized rehabilitation and management of human and non-human resources in healthcare facilities. It integrates clinical information with the remarkable potential of Artificial Intelligence (AI), particularly within its branch known as Machine Learning (ML). The latter is primarily portrayed as the field that allows computers to learn how to make predictions without the need for explicit programming [\u003cspan additionalcitationids=\"CR13 CR14 CR15\" citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. The concept of a learning machine capable of making predictions and drawing conclusions that are difficult to reach with conventional statistical methods is an old idea. However, statistics and ML are not completely distinct; ML algorithms are built upon common statistical methods, and they continue to advance alongside AI development. Statistics focuses on the verification of a hypothesis, whether null (H\u003csub\u003e0\u003c/sub\u003e) or alternative (H\u003csub\u003e1\u003c/sub\u003e), to assess how well the data distribution fits other known models, such as the Gaussian one. On the other hand, the aim of ML is prediction through algorithms that attempt to be as generalizable and applicable as possible for new validation datasets[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn this context, our work goal is to provide clinicians and healthcare professionals with a tool capable of estimating the likelihood of patient home discharge at the end of rehabilitation based on individual characteristics. So far, literature provides recent studies that demonstrate the efficacy and the reliability of AI algorithms in the rehabilitation outcomes analysis. Santilli et al. [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e] leveraged a 3-year dataset (4050 patients) comprising Acceptance and Discharge Report of Rehabilitation (ADR-r) data to predict mBI at discharge, exploiting an in-built function for importance ranking. Concerning the Intensive Care Unit (ICU), Safaei et al. [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e] proposed a CatBoost model to predict patient mortality using the discharge status variable and ten other impactful characteristics. Rufo et al.[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e] employed a Light Gradient Boosting Machine (LightGBM) algorithm and other significant classification models, including Random Forest (RF) and eXtreme Gradient Boosting (XGBoost), to predict the condition of diabetes mellitus with reliable results. In recent years, heightened stress on healthcare management, especially during the COVID-19 pandemic, has led to the publication of several works on prediction analysis. Yan et al. [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e] utilized 485 patient blood samples to assess three important biomarkers for quickly predicting mortality risk in COVID-19 individuals. Drawing on the classification and regression ML models employed in these studies, our contribution was to increase the literature knowledge about the impact of rehabilitation therapies on orthopedic and neurological patients in terms of home discharge rate after the treatment period. Our investigation aimed to conduct an intensive and in-depth preprocessing phase on 8468 ADR-r data, addressing the issue of unbalanced distribution in the home discharging target variable. We confirmed previous research findings regarding the importance of certain characteristics in patient autonomy recovery [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e] and provided an accurate clinical model built on a massive input dataset to predict discharging status that could help healthcare management systems. The implemented prototyping tool could be used to explore new and interesting targets for further analysis.\u003c/p\u003e"},{"header":"2. MATERIALS AND METHODS","content":"\u003cp\u003eThe study protocol was approved by the ethical committee of the IRCCS San Raffaele Pisana of Rome on 18/07/2018 (code number 07/18).\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Original dataset and initial cleaning steps\u003c/h2\u003e \u003cp\u003eThe original analysis dataset includes clinical and demographic data of adults admitted to the neurology and orthopedic departments of a rehabilitation hospital in Italy from January 2015 to August 2022. The completion of patients\u0026rsquo; ADR-r form [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e] resulted in the collection of data for 10520 individuals, whose information is distributed across 120 initial features. The dataset size reduced to 8468 after the duplicate removal (2052). We anonymized dataset rows by assigning a unique ID to each patient and collected personal information about them, such as gender, age, marital status, date, place of birth, and more. Clinical data, including the primary reason for rehabilitation, any associated medical conditions, impairments, and admission/discharge mBi scores were collected.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Dataset cleaning and encoding\u003c/h2\u003e \u003cp\u003eUnderlying medical conditions were defined establishing 13 general macro-categories. Through categorization based on \u003cem\u003eBasic Pathology\u003c/em\u003e, the dataset has been divided into two subsets: Neurologic Patients (NP), related to individuals whose basic pathology is primarily neurological, and Orthopedic Patients (OP), which pertains to orthopedic basic pathology. An additional macro-category called \u003cem\u003eNot Attributable\u003c/em\u003e didn\u0026rsquo;t belong to any of the identified 13 macro-categories and wasn\u0026rsquo;t considered in the further analysis. Details are shown in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eThe \u003cem\u003eBasic Pathology\u003c/em\u003e consists of alphanumeric codes established by ICD-9-CM, which is the ninth revision of this classification system with clinical modifications [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eMost of the used algorithms are unable to interpret a variable with values represented as text strings. This is what happens in the \u003cem\u003eBasic Pathology\u003c/em\u003e variable; for this reason, a numeric encoding was necessary to ensure accurate predictions[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e], the one-hot encoding technique, commonly used in ML prediction analyses that involve categorical features, was applied; its application converts text strings into n binary numeric vectors, where n is the macro-categories amount: (i) 1 if the categorical feature corresponds to the vector label, and (ii) 0 in the other cases [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe same encoding method was applied to the variables related to the presence of any patient comorbidity in 17 macro-categories, as shown in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eRegarding the impairment\u0026rsquo;s variables, following the ADR-r form the possible values range from 0 to 9. The smaller the number, the lower the degree of impairment shown by the patient. We applied a different encoding technique called label encoding, resulting in modified categorical impairment variables transformed into numerical ones [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. The labels used in this work are: (i) 0, (ii) 1, and (iii) 2, which are respectively associated with the absence, presence, and not-evaluability of patient impairment.\u003c/p\u003e \u003cp\u003eA fundamental step in the data preprocessing phase was the definition of the target variable. Keeping in mind that our analysis aimed to create a model that allows the prediction of patient-specific home discharge, we defined the Discharge Type Category based on the Discharge Type Feature in the ADR-r form. To this purpose we used four labels to encode the categorization of patient discharge, assigning to each label a specific condition: (i) 1 - the patient is home discharged at the end of rehabilitation, (ii) 2 \u0026ndash; the patient is not home discharged at the end of rehabilitation, (iii) Voluntary Discharge \u0026ndash; the patient voluntarily leaves the clinic, (iv) Death \u0026ndash; the patient deceased during rehabilitation protocol.\u003c/p\u003e \u003cp\u003eFinally, we created the final target variable for our analysis, referred to as Home Discharge (HD). Applying one-hot encoding, we assigned the value 1 when the patient falls under label 1 of the Discharge Type Category, indicating the patient is home discharged. On the other hand, 0 is assigned to patients corresponding to the label 2, indicating that they will not home discharge at the end of rehabilitation. Cases related to other labels were not considered in the further analysis (voluntary discharge and death).\u003c/p\u003e \u003cp\u003eThe mBI is a well-established patient-centered clinical outcome measure that quantifies a patient\u0026rsquo;s disability, ranging from 0 to 100; this value derives from the sum of scores assigned by the clinician to each of the scale subfields[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. It is commonly administrated in rehabilitation settings to evaluate the functional status of patients at admission and discharge[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. In particular, it assesses a patient\u0026rsquo;s independence in performing ADL [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. Label encoding technique divided the collected continuous variable into classes to better perform the model training. Based on previous studies [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e], we created six classes corresponding to the degree of dependency exhibited by the patient (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). The mBI henceforth will be referred to as \u0026ldquo;Categorized mBI\u0026rdquo; in the analysis.\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\u003eCategories and labels of mBI at admission and discharge.\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=\"char\" char=\".\" 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\"\u003e \u003cp\u003eADR-r value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLabel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCriteria\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLevel of dependence\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"5\" rowspan=\"6\"\u003e \u003cp\u003emBI on admission or mBI at discharge\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003emBI Є [0, 24]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eComplete\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003emBI Є [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSerious\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003emBI Є [50, 74]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eModerate\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003emBI Є [75, 90]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMild\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003emBI Є [91, 99]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMinimal\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003emBI\u0026thinsp;=\u0026thinsp;100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eIndipendent\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\u003eBefore the final cleaning step, the dataset consisted of 8468 patients (rows) and the new 50 categorical and non-categorical variables (columns) relevant to our study to schedule the demographic and clinical characteristics of patients admitted to the clinic. Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e provides an overview of the encoded input features used for prediction 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\u003eSummary of input and target variables for ML models with corresponding encoded labels in round brackets.\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\"\u003e \u003cp\u003eVariable Name\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVariable Type\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eValue Range\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eVariable Group\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePatient ID\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eQ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[1-8468]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGender\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[female (0), male (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStroke\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eParkinson\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMultiple Sclerosis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBrain Tumors\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePost mild/moderate trauma\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOther neurological pathologies\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNon-traumatic myeloradiculopathies\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHip arthroplasty\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eOP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eKnee arthroplasty\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eOP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFemur osteosynthesis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eOP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAmputation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eOP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSpinal pathologies\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eOP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOther orthopedic pathologies\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eOP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOther types of comorbidities\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHematological\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDysmetabolic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHepatic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNeurological\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRespiratory\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDiabetics\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHypertension\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHeart diseases\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCardiac arrhythmias\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDyslipidemic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTumors\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRheumatology/Orthopedics\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCirculatory complications\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIntestinal complications\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eKidney and urinary tract complications\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDifferent types of comorbidities\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCognitive impairment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1), NE (2)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBehavior impairment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1), NE (2)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCommunication/language impairment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1), NE (2)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSensory impairment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1), NE (2)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eManipulation impairment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1), NE (2)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBalance impairment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1), NE (2)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLocomotion impairment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1), NE (2)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCardiovascular impairment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1), NE (2)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRespiratory system impairment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1), NE (2)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUlcers\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSphincter control impairment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1), NE (2)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUrinary system impairment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1), NE (2)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNutrition impairment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[A (0), P (1), NE (2)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCategorized mBI admission\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[1 (CD) \u0026minus;\u0026thinsp;6 (I)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eQ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[18\u0026ndash;97]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTarget variable: Home discharge\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[NHD (0), HD (1)]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNP/OP\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\u003eQ: quantitative variables; C: categorical variables; A: absence of pathology/impairment; P: presence of pathology/impairment; NE: not-evaluability; CD: complete dependence; I: independence; NHD: not home discharge; HD: home discharge.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Analysis dataset\u003c/h2\u003e \u003cp\u003eTo address the presence of missing values and the dataset integrity, we didn\u0026rsquo;t consider patients for further analysis when: the admission and/or discharge mBI was not recorded in the ADR-r form; Discharge Type Category variable returned Voluntary Discharge or Death; ICD-9-CM code in ADR-r form wasn\u0026rsquo;t registered; patient pathology was not attributable into identified categories for the \u003cem\u003eBasic Pathology\u003c/em\u003e. Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e provides an overview of the analysis dataset building.\u003c/p\u003e \u003cp\u003eThe definitive analysis dataset consists of 7282 patients (rows): 3222 were admitted for a neurological basic pathology and 4060 for orthopedic conditions. Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e shows substantial dataset details for NP and OP groups.\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\u003eDemographic insights on model input patient.\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\"\u003e \u003cp\u003eDemographic Data\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNP\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eOP\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTotal\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSample size\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3222\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4060\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e7282\u003c/p\u003e \u003cp\u003e(44% NP;\u003c/p\u003e \u003cp\u003e56% OP)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMean age (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e St. Dev.)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e73 (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e 13)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e74 (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e 11)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e74 (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e 12)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMedian mBI at admission (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e St. Dev.)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e30 (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e 13)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e40 \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\((\\pm\\)\u003c/span\u003e\u003c/span\u003e 9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e37 (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e 12)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMedian mBI at discharge (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e St. Dev.)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e73 (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e 26)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e91 (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e 18)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e86 (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e 23)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMedian mBI change\u003c/p\u003e \u003cp\u003e(\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e St. Dev.)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e41 (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e 20)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e48 (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e 14)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e46 (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e 17)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGender: Male (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1715 (53%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1399 (34%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3114 (43%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGender: Female (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1507 (47%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2661 (66%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4168 (57%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHome discharge ratio (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2742 (85%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3800 (94%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e6542 (90%)\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 \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Supervised Learning Algorithms\u003c/h2\u003e \u003cp\u003eIn this study, we tested the supervised learning models. The goal was to implement a predictive function that can map the features exhibited at the time of a patient's admission (X) to a target variable (y). This is achieved using a training set consisting of tuples formed by the output variable and the i-th patient features[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. Generally, the supervised ML domain is employed to solve classification and regression problems [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. When the target variable is categorical, we refer to it as classification; on the other hand, a quantitative target variable is used in regression algorithms. The principal aim was to predict label 1 of the home discharge variable relied on the individual\u0026rsquo;s characteristics observed at admission. We trained and tested five of the most widely used classification algorithms in the rehabilitation field: some of them relied on decision trees, such as Random Forest (RF) and eXtreme Gradient Boosting (XGBoost), as well as others like Gradient Boosting (GB), Light Gradient Boosting Machine (LightGBM), and Categorical Boosting (CatBoost). The dataset was partitioned into NP and OP, which were further divided into training (70%) and test (30%) sets. The preprocessing phase, discussed in sections 2.5 and 2.6, as well as the prediction step, was carried out using Python 3.10.12 and proper libraries, including SciKit-Learn (for model learning)[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e], Imbalanced-Learn (for dataset imbalance issues), Matplotlib (for graphical visualization), SciPy (for statistical tests), Numpy (for mathematical operations)[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e] and Pandas (for data manipulation and analysis). We improved the efficiency of the model and, consequently, prediction accuracy through an optimization phase of hyper-parameters using a grid search technique. Hyper-parameters are specific variables of models that can be configured during the training step; their tuning (HPT) influences model performance[\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. We chose to use the grid search because it exhaustively investigates all hyperparameter combinations, returning their best values. However, it\u0026rsquo;s worth noting that this search process is computationally and temporally demanding[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eDuring the splitting step, there is a risk of making a significant mistake. It is common to train the model with a dataset that may not allow it to take full advantage of the maximum available information. This can potentially hinder the achievement of the most accurate prediction. Randomly, a patient may be assigned to either the training or the test set; however, if we are in the latter condition, their information will not be used for model learning. This approach can lead to a high risk of overfitting: a model exhibits high accuracy in predictions on the training set but performs poorly on the test set. A solution to address this issue is the k-fold Cross-Validation (CV) method. It involves dividing the dataset into k folds, where k is a positive integer: k-1 folds are used for model training, and one-fold is reserved for testing. In this analysis, we choose k\u0026thinsp;=\u0026thinsp;10, resulting in 10 distinct subsets, each of which was used as the test set in each iteration. The accuracy of the final model is obtained by averaging the accuracies of the 10 iterations [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. This process also makes it possible to provide approximate model validation when it is not possible to give input to a dataset never considered in the training and testing phase in a tight time frame. The combination of k-CV and HPT enables the optimal training and fine-tuning of models used in our work.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.5 Feature Scaling\u003c/h2\u003e \u003cp\u003eThe feature scaling step has a significant impact on the quality of predictions. Its application ensures that variables distributed in large value ranges do not dominate features distributed in smaller intervals [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e], to guarantee the latter mentioned be examined by the model [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. There are two main techniques for feature scaling: Standard Scaling and Min-Max Scaling.\u003c/p\u003e \u003cp\u003eIn this work, we used the Min-Max Scaling technique, which involves applying a formula to the variables\u0026rsquo; values to be scaled:\u003c/p\u003e \u003cp\u003e \u003cspan class=\"InlineEquation\"\u003e \u003cspan class=\"mathinline\"\u003e\\({X}_{norm}= \\frac{X- {X}_{min}}{{X}_{max}- {X}_{min}}\\)\u003c/span\u003e \u003c/span\u003e 1)\u003c/p\u003e \u003cp\u003ewhere X is the original value of the variable to be scaled, X\u003csub\u003emin\u003c/sub\u003e is the minimum value, X\u003csub\u003emax\u003c/sub\u003e is the maximum value and X\u003csub\u003enorm\u003c/sub\u003e is the new value obtained after the application of this scaling method. This technique ensures that the distribution range of the scaled variable is reduced to [0, +\u0026thinsp;1] if it shows only positive values or to [-1, +\u0026thinsp;1] if both positive and negative values are present. This technique is powerful when Standard Scaling is not suitable, such as when the variable to be scaled doesn\u0026rsquo;t fit a Gaussian curve[\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e], as in the case of the variable Age in our study. Section 3.1 shows the results of non-normality tests applied to this variable.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e2.6 Balancing Methods\u003c/h2\u003e \u003cp\u003eThe last preprocessing step involves the use of balancing techniques to address the unbalanced distribution of the target variable: 0 (15%) vs 1 (85%) for the NP set; 0 (6%) vs 1 (94%) for the OP set. This study shows a binary classification analysis; in this case, it\u0026rsquo;s common to observe a high unbalance of target variable distribution. This condition adversely affects the predictive performance of the model[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Not solving this problem results in an algorithm that tends to accurately predict the most common classes, at the expense of the rare ones [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. Proposed solutions in the scientific landscape are mainly based on resampling techniques. Usually, in ML analysis we observe the low number of minority classes rather than the majority class abundance. For this reason, oversampling (synthesis) is preferred to undersampling (reduction)[\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]. This study compared the main methodologies of both approaches, including Random Over Sampling (ROS)[\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e], Synthetic Minority Oversampling TEchnique (SMOTE)[\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e], ADAptive SYNthetic (ADASYN)[\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e], as well as Random Under Sampling (RUS)[\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e] and Cluster Centroids (CC) [\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]. In addition, the SMOTE-Tomek technique, which combines both methodologies simultaneously, was also considered [\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e2.7 Evaluation Metrics\u003c/h2\u003e \u003cp\u003eOnce the model is implemented, we must establish the effectiveness of its performance. Since this is a classification task, the main evaluation metrics used to compare predictions are Accuracy, Precision, Recall and F1-score[\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e]. The possible range of these values is [0, 1]: the closer to the right extreme, the better the model's ability to correctly predict instances belonging to both \u003cem\u003eHome discharge\u003c/em\u003e classes.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. RESULTS","content":"\u003cp\u003eFigure \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e is a graphical representation of ML input datasets construction, along with their absolute and percentage frequencies. Duplicate entries, absence of admission and/or discharge mBI scores, and other criteria were considered in constructing the analysis dataset.\u003c/p\u003e\n\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\n\u003ch2\u003e3.1 Non-normality verification\u003c/h2\u003e\n\u003cp\u003eWe needed to conduct some statistical checks to determine the appropriate technique for scaling the \u003cem\u003eAge\u003c/em\u003e quantitative variable. These checks were performed to confirm statistically the non-normal distribution, justifying the subsequent application of the Min-Max Scaling technique. In addition to visual inspection (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e) and a comparison of nominal percentiles (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e), two statistical tests were also applied for a quantitative assessment.\u003c/p\u003e\n\u003cp\u003eThe Shapiro-Wilk test is properly employed for small sample sizes but remains effective for larger datasets. The Kolmogorov-Smirnov test is primarily applicable to larger datasets. In both checks, we chose a significance level (alpha) of 0.05. Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e shows the results for the \u003cem\u003eAge\u003c/em\u003e variable in NP and OP. For this reason, we accepted the alternative hypothesis: \u003cem\u003eAge\u003c/em\u003e variable for both datasets doesn\u0026rsquo;t fit a normal distribution.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n\u003ctable id=\"Tab4\" border=\"1\"\u003e\u003ccaption\u003e\n\u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\n\u003cdiv class=\"CaptionContent\"\u003e\n\u003cp\u003eStatistical confirmation of the non-normal distribution of the \u003cem\u003eAge\u003c/em\u003e variable in NP and OP datasets.\u003c/p\u003e\n\u003c/div\u003e\n\u003c/caption\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eTest\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003ep-value NP\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003ep-value OP\u003c/p\u003e\n\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eShapiro-Wilk\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e7.76E-35\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.08E-34\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eKolmogorov-Smirnov\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e4.34E-34\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e8.68E-30\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eQuantitative assessments justify the application of the Min-Max Scaler to this variable, helping to avoid problems that could lead to less accurate predictions.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\n\u003ch2\u003e3.2 Best combination between model and balancing techniques\u003c/h2\u003e\n\u003cp\u003eWe used accuracy values to select the optimal combination of balancing techniques and classifiers for achieving accurate predictions. Accuracy is a key parameter in classification analyses. Results for NP and OP are shown in Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n\u003ctable id=\"Tab5\" border=\"1\"\u003e\u003ccaption\u003e\n\u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e\n\u003cdiv class=\"CaptionContent\"\u003e\n\u003cp\u003eAccuracy values for different combinations of balancing techniques and classifier predictions in the NP and OP datasets.\u003c/p\u003e\n\u003c/div\u003e\n\u003c/caption\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth colspan=\"13\" align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eAccuracy\u003c/em\u003e\u003c/p\u003e\n\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"2\" align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eROS\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eRUS\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eSMOTE\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eSMOTE Tomek\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eADASYN\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eCC\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eNP\u003c/em\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eOP\u003c/em\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eNP\u003c/em\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eOP\u003c/em\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eNP\u003c/em\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eOP\u003c/em\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eNP\u003c/em\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eOP\u003c/em\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eNP\u003c/em\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eOP\u003c/em\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eNP\u003c/em\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eOP\u003c/em\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eGB\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.7224\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.7474\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.6285\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.6603\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.7898\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8114\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.7858\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8121\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.7721\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8152\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.7326\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8333\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eRF\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.9502\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003e0.9750\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.6701\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.6282\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8706\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8908\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8588\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.9067\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8571\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8966\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.7396\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8526\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eXGBoost\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.9064\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.9364\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.6528\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.5769\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8767\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.9456\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8856\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.9464\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8870\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.9430\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.7500\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8654\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eLightGBM\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8554\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8904\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.6424\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.6154\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8755\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.9386\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8606\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.9432\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8589\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.9359\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.7500\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8462\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eCatBoost\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8518\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.9075\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.6701\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.6090\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8645\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.9364\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8667\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.9415\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8511\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.9341\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.7431\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.8526\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eThe highest accuracy is achieved by applying ROS (Random OverSampling) as the balancing technique and the Random Forest Classifier as the ML model. This is true for both neurological and orthopedic patients.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\n\u003ch2\u003e3.3 Feature Importance Ranking\u003c/h2\u003e\n\u003cp\u003eUsing built-in functions of various models in the Python SK-Learn library, it\u0026rsquo;s possible to determine a feature importance ranking in predicting the outcome. As already shown in Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e, the model with the highest accuracy is the RF classifier. Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003e shows the ranking obtained with this model for NP and OP, respectively.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n\u003ctable id=\"Tab6\" border=\"1\"\u003e\u003ccaption\u003e\n\u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e\n\u003cdiv class=\"CaptionContent\"\u003e\n\u003cp\u003eTop 20 most critical features identified through Random Forest (RF) feature ranking in the NP and OP datasets.\u003c/p\u003e\n\u003c/div\u003e\n\u003c/caption\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eRank\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eNP\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eOP\u003c/p\u003e\n\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eAge\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eAge\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eCategorized mBI on admission\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eUrinary system condition\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e3\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eBalance impairment\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eKnee arthroplasty\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e4\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eCognitive impairment\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eBalance impairment\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e5\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNutrition impairment\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eHypertension\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eSensory impairment\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eDifferent types of comorbitidies\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e7\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eGender\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eHip arthroplasty\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e8\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eManipulation impairment\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eGender\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e9\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNeurological\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eRheumatology/Orthopedics\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e10\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eHypertension\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eAmputation\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e11\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eSphincter control impairment\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eUlcers\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e12\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eCommunication/language impairment\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eDiabetes\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e13\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eUrinary system condition\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eDysmetabolic\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e14\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eUlcers\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eOther orthopedic pathologies\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e15\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eDiabetes\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eDyslipidemic\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e16\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eDifferent types of comorbidities\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eCategorized mBI on admission\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e17\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eBehavior impairment\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eOther neurological pathologies\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e18\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eCardiovascular impairment\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eCardiovascular impairment\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e19\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eHeart diseases\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eHeart diseases\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e20\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eCardiac arrhythmias\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eTumors\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eBoth RF rankings confirm that Age holds the greatest influence in forecasting patient discharge destination at the end of the rehabilitation program, simultaneously validating the substantial importance of cognitive impairment. Urinary and cardiac systems hold a place among the top twenty most influential features, along with hypertension diagnosis. In addition, by logical expectations, substantial weight is also associated with balance impairment, ranking third for NP and fourth for OP. The mBI proves to be a significant variable for NP; its importance decreases among orthopedic ones but still ranks among the top 20 most influential predictors. Finally, it can be observed how the presence of comorbidities influences the prediction of the outcome in both subsets [\u003cspan class=\"CitationRef\"\u003e46\u003c/span\u003e] and how the gender isn\u0026rsquo;t a top five important feature.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\n\u003ch2\u003e3.4 Hyperparameter Optimization\u003c/h2\u003e\n\u003cp\u003eThis step consists of the optimization of the classifier hyperparameters. We conducted a grid search on the most common values of RF classifier specific parameters, such as n_estimators, max_depth, min_samples_split, min_samples_leaf, and max_features. The aim is to find the best combination of those hyper-parameters to optimize the model performance. Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e shows the best values found for NP and OP sets.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n\u003ctable id=\"Tab7\" border=\"1\"\u003e\u003ccaption\u003e\n\u003cdiv class=\"CaptionNumber\"\u003eTable 7\u003c/div\u003e\n\u003cdiv class=\"CaptionContent\"\u003e\n\u003cp\u003eBest combination of RF hyperparameters in NP and OP implemented model.\u003c/p\u003e\n\u003c/div\u003e\n\u003c/caption\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth colspan=\"3\" align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eBest values\u003c/em\u003e\u003c/p\u003e\n\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eHPT\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eNP\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eOP\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003en estimators\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e200\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e100\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003emax depth\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e20\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e20\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003emin samples split\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e5\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003emin samples leaf\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003emax features\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003esqrt\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eauto\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\n\u003ch2\u003e3.5 Model evaluation\u003c/h2\u003e\n\u003cp\u003eTable\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e shows the values obtained from the model on the respective test sets of NP and OP, after training it with the remaining 70% of both data sets. A 10-fold cross-validation was employed to provide an approximate validation of models and to mitigate overfitting, and hyperparameters were optimized. The table evaluation metrics were used to analyze model performance in predicting the outcome of our study, both in terms of positive (patient home discharge) and negative instances (patient not home discharge). In both cases, values were calculated using SK-learn specific functions (accuracy_score, f1_score, precision_score, recall_score). They often get close to 1, showing that the model performs very well in predicting home discharge, both for NP and OP.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n\u003ctable id=\"Tab8\" border=\"1\"\u003e\u003ccaption\u003e\n\u003cdiv class=\"CaptionNumber\"\u003eTable 8\u003c/div\u003e\n\u003cdiv class=\"CaptionContent\"\u003e\n\u003cp\u003eModel Evaluation Metrics in NP and OP dataset.\u003c/p\u003e\n\u003c/div\u003e\n\u003c/caption\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth colspan=\"3\" align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eModel Evaluation\u003c/em\u003e\u003c/p\u003e\n\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eMetrics\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eNP\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eOP\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eAccuracy\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e93.8%\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e96.7%\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003ePrecision\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e94.4%\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e96.9%\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eRecall\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e93.9%\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e96.8%\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003ef1-score\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e93.7%\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e96.7%\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003c/div\u003e\n\u003c/div\u003e"},{"header":"4. DISCUSSION","content":"\u003cp\u003eThis study aimed at advancing the patient-centered approach to rehabilitation by integrating medical knowledge with supervised learning methods to predict probability of patients' home discharge considering modified Barthel Index and ADR-r\u0026rsquo;s data. Utilizing a substantial dataset for robust model training and employing machine learning techniques represent significant strengths of our analysis. However, a considerable challenge arose due to the unbalanced distribution of the target variable, which we addressed through extensive data preprocessing. This phase significantly contributed to the excellent performance evaluation metrics observed in the implemented classifier, detailed in Table\u0026nbsp;\u003cspan refid=\"Tab8\" class=\"InternalRef\"\u003e8\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eWithin the field of rehabilitation, selecting appropriate protocols for each patient remains a formidable challenge due to the lack of uniformity in their application among individuals showing similar demographic and clinical conditions. Despite this challenge, our work aims to simplify the selection process, narrowing down potentially applicable protocols to promote a more consistent approach to rehabilitation program choice and enhance overall personalization.\u003c/p\u003e \u003cp\u003eA common limitation of ML work is the absence of a model validation set. While both training and testing steps are crucial for verifying model learning and performance in supervised ML analysis, a complete model evaluation requires the use of an unseen input dataset. In the context of our work, this limitation is planned to be addressed by collecting a substantial number of new ADR-r forms related to clinic-admitted patients over the next few years. These will be utilized as unseen data for model validation, thereby enhancing the actual functionality of the implemented tool. Despite that, an approximate model validation is obtained performing a 10-fold Cross Validation.\u003c/p\u003e \u003cp\u003eMoving to the specific findings of our analysis, we aim to provide a predictive model for patient home discharge at the end of the rehabilitation period. The dataset was divided into two subsets based on the nature of the basic pathology encountered: Neurological (NP) and Orthopedic (OP). Our analysis identified the top 20 characteristics with the most significant impact on outcomes in both subsets, detailed in Table\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eThe results align with previous studies, confirming Age as the leading predictive feature for both subsets, consistent with recent pioneering work targeting mBI at the end of the rehabilitation program [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan additionalcitationids=\"CR48 CR49\" citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e]. Notably, patient independence (mBI) has a more substantial impact on NP predictions than on OP, highlighting the lower autonomy levels in performing ADLs among individuals undergoing rehabilitation for basic neurological pathologies. Despite its importance decreases among orthopedic ones, it still ranks among the top 20 most influential predictors, thus confirming the findings in Masaru Uragami's study on readmission of patients with hip fracture[\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn the case of OP, the most influential variable ranking reveals that the level of urinary system impairment is the second most important feature in predicting home discharge. This suggests that elderly patients facing urinary system issues are likely to have a lower probability of home discharge after rehabilitation, as specific mBI sub-items relate to intestinal and urinary continence, as well as toilet use.\u003c/p\u003e \u003cp\u003eThe consistent identification of age as the leading predictive feature for both NP and OP subsets reinforces the importance of age in assessing patient outcomes and discharge disposition [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e]. Moreover, the observation that patient independence exerts a more substantial impact on predictions for NP compared to OP highlights the complexity of rehabilitation needs among individuals with neurological conditions.\u003c/p\u003e \u003cp\u003eThe significant influence of urinary system condition as the second most important feature in predicting discharge destination among OP underscores the importance of comprehensive assessment and targeted interventions to address patient-specific needs [\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e, \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e]. Elderly patients with urinary system issues may require multidisciplinary care and support to facilitate their transition to home settings post-rehabilitation.\u003c/p\u003e \u003cp\u003eThese implications emphasize the necessity for tailored interventions aimed at improving patient outcomes and reducing healthcare utilization post-rehabilitation. Future research may focus on exploring additional factors contributing to discharge destination after rehabilitation, refining predictive models, and implementing targeted interventions to optimize patient care and reduce healthcare costs.\u003c/p\u003e \u003cp\u003eThe extensive data pre-processing phase, including the use of the ROS technique and hyperparameter tuning of the RF classifier, enhanced our predictive model performance. Its evaluation metrics, such as accuracy, precision, recall, and F1 score, yielded promising results for both OP and NP sets (Table\u0026nbsp;\u003cspan refid=\"Tab8\" class=\"InternalRef\"\u003e8\u003c/span\u003e). The differences in performance are rooted in clinical principles, where neurological conditions impact a wider range of human systems and functions than orthopedic pathologies. The greater diversity of instances among NP patients results in a reduced model accuracy for the NP classifier (93.8%) compared to OP (96.7%), highlighting the intricate and multifaceted impact of neurological conditions on patient health. This underscores the challenge faced by AI engineers in predicting rehabilitation outcomes for individual neurologic patients.\u003c/p\u003e"},{"header":"5. CONCLUSIONS","content":"\u003cp\u003eThis study introduces a medical tool leveraging ADR-r data to predict patient clinic home discharge post-rehabilitation, facilitating the personalization of treatment plans. The RF output could be used to align rehabilitation programs with patient needs, enabling the full recovery of ADLs. Additionally, it provides timely alerts to family members and caregivers, optimizing home care management. Successful home discharge predictions could be linked to in-home health assistance, while the other case predictions support preparing patients, physically and mentally, for a not home discharge. The model could also be used for clinical resource management, improving bed allocation, medical equipment, and staffing. Future hopes spin around gathering new ADR-r forms to validate more robustly the implemented tool, which will persist in being refined and enhanced along with the significant advancements in AI and ML. This ongoing process will simplify the exploration of new clusters, pathways, and targets, thereby improving the personalization of rehabilitation.\u003c/p\u003e \u003cp\u003eThe successful integration of machine learning techniques into clinical practice offers promising avenues for enhancing patient-centered care and improving rehabilitation outcomes. By accurately predicting patient\u0026rsquo;s home discharge, healthcare providers can tailor treatment plans to individual patient needs, optimize resource allocation, and provide timely support to patients and their caregivers. This study represents a significant step forward in employing artificial intelligence and machine learning to optimize patient care and improve healthcare delivery.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003e\u003cstrong\u003eOP \u003c/strong\u003eOrthopedic Patients;\u003cstrong\u003e NP \u003c/strong\u003eNeurologic Patients;\u003cstrong\u003e ML \u003c/strong\u003eMachine Learning;\u003cstrong\u003e AI \u003c/strong\u003eArtificial Intelligence;\u003cstrong\u003e ADL \u003c/strong\u003eActivities of Daily Living\u003cstrong\u003e; mBI \u003c/strong\u003eModified Barthel Index;\u003cstrong\u003e RF \u003c/strong\u003eRandom Forest;\u003cstrong\u003e XGBoost \u003c/strong\u003eeXtreme Gradient Boosting;\u003cstrong\u003e GB \u003c/strong\u003eGradient Boosting;\u003cstrong\u003e CatBoost \u003c/strong\u003eCategorical Boosting;\u003cstrong\u003e LightGBM \u003c/strong\u003eLight Gradient Boosting Model;\u003cstrong\u003e ADR-r \u003c/strong\u003eAcceptance and Discharge Report of rehabilitation;\u003cstrong\u003e ICD-9-CM \u003c/strong\u003eInternational Classification of Diseases, revision number 9 with Clinical Modifications;\u003cstrong\u003e HPT \u003c/strong\u003eHyper-Parameter Optimization;\u003cstrong\u003e k-CV \u003c/strong\u003ek-fold Cross Validation;\u003cstrong\u003e ROS \u003c/strong\u003eRandom Over Sampling;\u003cstrong\u003e SMOTE \u003c/strong\u003eSynthetic Minority Oversampling TEchnique;\u003cstrong\u003e ADASYN \u003c/strong\u003eADAptive SYNthetic;\u003cstrong\u003e RUS \u003c/strong\u003eRandom Under Sampling;\u003cstrong\u003e CC \u003c/strong\u003eCluster Centroids;\u003cstrong\u003e HPT\u003c/strong\u003e Hyper-Parameter Tuning.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAUTHORSHIP\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eMF has made substantial contributions to conception and design of the study. SP and CD carried out the data collection. FI, PR and LB designed the algorithm for data analysis. SP and FI participated in the study design and coordination. LB, PR, ESC performed the statistical analysis. LB has elaborated the original draft of the manuscript. MF, FI, SP, and PR and ESC participated in the manuscript revisions. MF gave the final approval of the version. All authors contributed to the article and approved the submitted version.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study has been funded by the Italian Ministry of Health (Ricerca Corrente).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflict of interest disclosure\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAVAILABILTY OF DATA AND MATERIALS\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe dataset supporting the conclusion of this article is available in Zenodo repository with this DOI: 10.5281/zenodo.10991206 and this link: https://zenodo.org/records/10991206\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eETHICS APPROVAL STATEMENT\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study protocol was approved by the ethical committee of the IRCCS San Raffaele Pisana of Rome on 18/07/2018 (code number 07/18).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePATIENT CONSENT STATEMENT\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eEach patient has received and signed informed consent.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePERMISSION TO REPRODUCE MATERIAL FROM OTHER SOURCES\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNo data have been reproduced.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eOrganization WH. WHO global disability action plan 2014-2021: Better health for all people with disability [Internet]. World Health Organization; 2015 [cited 2024 Mar 26]. Available from: https://apps.who.int/iris/bitstream/handle/10665/199544/9789241509619_eng.pdf\u003c/li\u003e\n\u003cli\u003eGimigliano F, Negrini S. The World Health Organization\u0026quot; rehabilitation 2030: a call for action\u0026quot;. European Journal of Physical and Rehabilitation Medicine. 2017;53:155\u0026ndash;68. \u003c/li\u003e\n\u003cli\u003eNegrini S, Meyer T, Arienti C, Kiekens C, Pollock A, Selb M, et al. The 3rd Cochrane Rehabilitation Methodology Meeting:\u0026quot; Rehabilitation definition for scientific research purposes\u0026quot;. European journal of physical and rehabilitation medicine. 2020;56:658\u0026ndash;60. \u003c/li\u003e\n\u003cli\u003eRahgozar P. AIM in Rehabilitation. Artificial Intelligence in Medicine [Internet]. Springer; 2022 [cited 2024 Mar 27]. p. 1809\u0026ndash;17. Available from: https://link.springer.com/content/pdf/10.1007/978-3-030-64573-1_177.pdf\u003c/li\u003e\n\u003cli\u003eMennella C, Maniscalco U, De Pietro G, Esposito M. The Role of Artificial Intelligence in Future Rehabilitation Services: A Systematic Literature Review. IEEE Access. 2023;11:11024\u0026ndash;43. \u003c/li\u003e\n\u003cli\u003eSantilli V, Mangone M, Diko A, Alviti F, Bernetti A, Agostini F, et al. The Use of Machine Learning for Inferencing the Effectiveness of a Rehabilitation Program for Orthopedic and Neurological Patients. International Journal of Environmental Research and Public Health. 2023;20:5575. \u003c/li\u003e\n\u003cli\u003eKoh GC-H, Chen C, Cheong A, Choo TB, Pui CK, Phoon FN, et al. Trade-offs between effectiveness and efficiency in stroke rehabilitation. International Journal of Stroke. 2012;7:606\u0026ndash;14. \u003c/li\u003e\n\u003cli\u003eChow P, Chen C, Cheong A, Fong NP, Chan KM, Tan BY, et al. Factors and trade-offs with rehabilitation effectiveness and efficiency in newly disabled older persons. Archives of physical medicine and rehabilitation. 2014;95:1510-1520. e4. \u003c/li\u003e\n\u003cli\u003eDamiani C, Mangone M, Paoloni M, Goffredo M, Franceschini M, Servidio M, et al. Trade-Offs with rehabilitation Effectiveness (REs) and Efficiency (REy) in a sample of Italian disabled persons in a in post-acuity rehabilitation unit. Ann Ig. 2020;32:327\u0026ndash;35. \u003c/li\u003e\n\u003cli\u003eCollin C, Wade DT, Davies S, Horne V. The Barthel ADL Index: a reliability study. International disability studies. 1988;10:61\u0026ndash;3. \u003c/li\u003e\n\u003cli\u003eHsueh I-P, Lin J-H, Jeng J-S, Hsieh C-L. Comparison of the psychometric characteristics of the functional independence measure, 5 item Barthel index, and 10 item Barthel index in patients with stroke. Journal of Neurology, Neurosurgery \u0026amp; Psychiatry. 2002;73:188\u0026ndash;90. \u003c/li\u003e\n\u003cli\u003eKrittanawong C, Zhang H, Wang Z, Aydar M, Kitai T. Artificial intelligence in precision cardiovascular medicine. Journal of the American College of Cardiology. 2017;69:2657\u0026ndash;64. \u003c/li\u003e\n\u003cli\u003eRech J, Althoff K-D. Artificial intelligence and software engineering: Status and future trends. KI. 2004;18:5\u0026ndash;11. \u003c/li\u003e\n\u003cli\u003eRajkomar A, Dean J, Kohane I. Machine learning in medicine. New England Journal of Medicine. 2019;380:1347\u0026ndash;58. \u003c/li\u003e\n\u003cli\u003eSaria S, Butte A, Sheikh A. Better medicine through machine learning: What\u0026rsquo;s real, and what\u0026rsquo;s artificial? PLoS medicine. Public Library of Science San Francisco, CA USA; 2018. p. e1002721. \u003c/li\u003e\n\u003cli\u003eMitchell TM. Machine Learning. New York: McGraw-Hill; 1997. \u003c/li\u003e\n\u003cli\u003eHandelman GS, Kok HK, Chandra RV, Razavi AH, Lee MJ, Asadi H. eD octor: machine learning and the future of medicine. Journal of internal medicine. 2018;284:603\u0026ndash;19. \u003c/li\u003e\n\u003cli\u003eSafaei N, Safaei B, Seyedekrami S, Talafidaryani M, Masoud A, Wang S, et al. E-CatBoost: An efficient machine learning framework for predicting ICU mortality using the eICU Collaborative Research Database. Plos one. 2022;17:e0262895. \u003c/li\u003e\n\u003cli\u003eRufo DD, Debelee TG, Ibenthal A, Negera WG. Diagnosis of diabetes mellitus using gradient boosting machine (LightGBM). Diagnostics. 2021;11:1714. \u003c/li\u003e\n\u003cli\u003eYan L, Zhang H-T, Goncalves J, Xiao Y, Wang M, Guo Y, et al. An interpretable mortality prediction model for COVID-19 patients. Nature machine intelligence. 2020;2:283\u0026ndash;8. \u003c/li\u003e\n\u003cli\u003eSeccia R, Boresta M, Fusco F, Tronci E, Di Gemma E, Palagi L, et al. Data of patients undergoing rehabilitation programs. Data in brief. 2020;30:105419. \u003c/li\u003e\n\u003cli\u003eAl-Shehari T, Alsowail RA. An insider data leakage detection using one-hot encoding, synthetic minority oversampling and machine learning techniques. Entropy. 2021;23:1258. \u003c/li\u003e\n\u003cli\u003eEl Affendi MA, Al Rajhi KHS. Text encoding for deep learning neural networks: A reversible base 64 (Tetrasexagesimal) Integer Transformation (RIT64) alternative to one hot encoding with applications to Arabic morphology. 2018 sixth international conference on digital information, networking, and wireless communications (DINWC). IEEE; 2018. p. 70\u0026ndash;4. \u003c/li\u003e\n\u003cli\u003ePournajaf S, Pellicciari L, Proietti S, Agostini F, Gabbani D, Goffredo M, et al. Which items of the modified Barthel Index can predict functional independence at discharge from inpatient rehabilitation? A secondary analysis retrospective cohort study. International Journal of Rehabilitation Research. 2023;46:230\u0026ndash;7. \u003c/li\u003e\n\u003cli\u003eLiu F, Tsang RC, Zhou J, Zhou M, Zha F, Long J, et al. Relationship of Barthel Index and its Short Form with the Modified Rankin Scale in acute stroke patients. Journal of stroke and cerebrovascular diseases. 2020;29:105033. \u003c/li\u003e\n\u003cli\u003eOcagli H, Cella N, Stivanello L, Degan M, Canova C. The Barthel index as an indicator of hospital outcomes: A retrospective cross‐sectional study with healthcare data from older people. Journal of Advanced Nursing. 2021;77:1751\u0026ndash;61. \u003c/li\u003e\n\u003cli\u003eShah S. Modified Barthel index or Barthel index (expanded). Compendium of Quality of Life Instruments, Part II Chichester, UK: Wiley and Sons. 1998; \u003c/li\u003e\n\u003cli\u003eSulter G, Steen C, De Keyser J. Use of the Barthel index and modified Rankin scale in acute stroke trials. Stroke. 1999;30:1538\u0026ndash;41. \u003c/li\u003e\n\u003cli\u003eWitten IH, Frank E. Data mining: practical machine learning tools and techniques with Java implementations. Acm Sigmod Record. 2002;31:76\u0026ndash;7. \u003c/li\u003e\n\u003cli\u003eKotsiantis SB, Zaharakis ID, Pintelas PE. Machine learning: a review of classification and combining techniques. Artif Intell Rev. 2006;26:159\u0026ndash;90. \u003c/li\u003e\n\u003cli\u003ePedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O, et al. Scikit-learn: Machine learning in Python. the Journal of machine Learning research. 2011;12:2825\u0026ndash;30. \u003c/li\u003e\n\u003cli\u003eVon Elm E, Altman DG, Egger M, Pocock SJ, G\u0026oslash;tzsche PC, Vandenbroucke JP. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. The Lancet. 2007;370:1453\u0026ndash;7. \u003c/li\u003e\n\u003cli\u003eJalaeian Zaferani E, Teshnehlab M, Khodadadian A, Heitzinger C, Vali M, Noii N, et al. Hyper-parameter optimization of stacked asymmetric auto-encoders for automatic personality traits perception. Sensors. 2022;22:6206. \u003c/li\u003e\n\u003cli\u003eYounis EM, Zaki SM, Kanjo E, Houssein EH. Evaluating ensemble learning methods for multi-modal emotion recognition using sensor data fusion. Sensors. 2022;22:5611. \u003c/li\u003e\n\u003cli\u003eTangadpalliwar SR, Vishwakarma S, Nimbalkar R, Garg P. ChemSuite: A package for chemoinformatics calculations and machine learning. Chemical Biology \u0026amp; Drug Design. 2019;93:960\u0026ndash;4. \u003c/li\u003e\n\u003cli\u003eCao XH, Stojkovic I, Obradovic Z. A robust data scaling algorithm to improve classification accuracies in biomedical data. BMC Bioinformatics. 2016;17:359. \u003c/li\u003e\n\u003cli\u003eFern\u0026aacute;ndez A, Garc\u0026iacute;a S, Luengo J, Bernad\u0026oacute;-Mansilla E, Herrera F. Genetics-based machine learning for rule induction: state of the art, taxonomy, and comparative study. IEEE Transactions on Evolutionary Computation. 2010;14:913\u0026ndash;41. \u003c/li\u003e\n\u003cli\u003eHaixiang G, Yijing L, Shang J, Mingyun G, Yuanyue H, Bing G. Learning from class-imbalanced data: Review of methods and applications. Expert systems with applications. 2017;73:220\u0026ndash;39. \u003c/li\u003e\n\u003cli\u003eLaurikkala J. Improving identification of difficult small classes by balancing class distribution. Artificial Intelligence in Medicine: 8th Conference on Artificial Intelligence in Medicine in Europe, AIME 2001 Cascais, Portugal, July 1\u0026ndash;4, 2001, Proceedings 8. Springer; 2001. p. 63\u0026ndash;6. \u003c/li\u003e\n\u003cli\u003eDuan F, Zhang S, Yan Y, Cai Z. An oversampling method of unbalanced data for mechanical fault diagnosis based on MeanRadius-SMOTE. Sensors. 2022;22:5166. \u003c/li\u003e\n\u003cli\u003eHaibo He, Yang Bai, Garcia EA, Shutao Li. ADASYN: Adaptive synthetic sampling approach for imbalanced learning. 2008 IEEE International Joint Conference on Neural Networks (IEEE World Congress on Computational Intelligence) [Internet]. Hong Kong, China: IEEE; 2008 [cited 2024 Mar 27]. p. 1322\u0026ndash;8. Available from: http://ieeexplore.ieee.org/document/4633969/\u003c/li\u003e\n\u003cli\u003eWu Y, Fang Y. Stroke prediction with machine learning methods among older Chinese. International journal of environmental research and public health. 2020;17:1828. \u003c/li\u003e\n\u003cli\u003eAli Z, Hayat MF, Shaukat K, Alam TM, Hameed IA, Luo S, et al. A proposed framework for early prediction of Schistosomiasis. Diagnostics. 2022;12:3138. \u003c/li\u003e\n\u003cli\u003eSwana EF, Doorsamy W, Bokoro P. Tomek link and SMOTE approaches for machine fault classification with an imbalanced dataset. Sensors. 2022;22:3246. \u003c/li\u003e\n\u003cli\u003eKirk D, Catal C, Tekinerdogan B. Precision nutrition: A systematic literature review. Computers in Biology and Medicine. 2021;133:104365. \u003c/li\u003e\n\u003cli\u003eShah AA, Devana SK, Lee C, Bugarin A, Lord EL, Shamie AN, et al. Machine learning driven identification of novel patient factors for prediction of major complications after posterior cervical spinal fusion. Eur Spine J. 2022;31:1952\u0026ndash;9. \u003c/li\u003e\n\u003cli\u003eAdhiya J, Barghi B, Azadeh-Fard N. Predicting the risk of hospital readmissions using a machine learning approach: a case study on patients undergoing skin procedures. Front Artif Intell. 2024;6:1213378. \u003c/li\u003e\n\u003cli\u003eLin Y, Wu JY, Lin K, Hu YH, Kong GL. [Prediction of intensive care unit readmission for critically ill patients based on ensemble learning]. Beijing Da Xue Xue Bao Yi Xue Ban. 2021;53:566\u0026ndash;72. \u003c/li\u003e\n\u003cli\u003eZhu Y, Venugopalan J, Zhang Z, Chanani NK, Maher KO, Wang MD. Domain Adaptation Using Convolutional Autoencoder and Gradient Boosting for Adverse Events Prediction in the Intensive Care Unit. Front Artif Intell. 2022;5:640926. \u003c/li\u003e\n\u003cli\u003eUragami M, Matsushita K, Shibata Y, Takata S, Karasugi T, Sueyoshi T, et al. A machine learning-based scoring system and ten factors associated with hip fracture occurrence in the elderly. Bone. 2023;176:116865. \u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"journal-of-neuroengineering-and-rehabilitation","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"jner","sideBox":"Learn more about [Journal of NeuroEngineering and Rehabilitation](http://jneuroengrehab.biomedcentral.com/)","snPcode":"12984","submissionUrl":"https://submission.nature.com/new-submission/12984/3","title":"Journal of NeuroEngineering and Rehabilitation","twitterHandle":"@BioMedCentral","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Personalized rehabilitation, machine learning, classification algorithm, random forest, unbalance, hyperparameters optimization, accuracy","lastPublishedDoi":"10.21203/rs.3.rs-4383785/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4383785/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eIn recent years, the fusion of the medical and computer science domains has gained significant traction in the scientific research landscape. Progress in both fields has enabled the generation of a vast amount of data used for making predictions and identifying interesting clusters and pathways. The Machine Learning model's application in the medical domain is one of the most compelling and challenging topics to explore, bridging the gap between Artificial Intelligence (AI). and healthcare. The combination of AI and medical information offers the possibility to create tools that can benefit both healthcare providers and physicians. This enables the enhancement of rehabilitation therapy and patient care. In the rehabilitation context, this work provides an alternative perspective: prediction of patients\u0026rsquo; home discharge upon completing the rehabilitation protocol. Demographic and clinical data were collected on 7282 inpatients from electronic Medical Record, each record was categorized into Neurological Patients (NP, N\u0026thinsp;=\u0026thinsp;3222) or Orthopedic Patients (OP, N\u0026thinsp;=\u0026thinsp;4060).\u003c/p\u003e \u003cp\u003eTo comprehend the most suitable machine learning model, an extensive data preprocessing phase was essential. This included steps such as variable recoding, scaling, and comparing various dataset balancing methods to enhance the model\u0026rsquo;s performance. Random Forest model was selected after a careful review and comparison of algorithms commonly utilized in the clinical-rehabilitative domain. Following a significant hyperparameter optimization phase through a grid search technique, we achieved an accuracy rate of 97% for OP and slightly lower (94%) for NP. This work points out the increasing importance of AI in medicine, especially in the realm of personalized rehabilitation. The use of such approaches could signify a transformative shift in healthcare. The integration of machine learning not only enhances the precision of treatment but also opens new possibilities for patient-centered care, improving outcomes and quality of care for individuals undergoing rehabilitation.\u003c/p\u003e","manuscriptTitle":"Enhancing Patient Rehabilitation Outcomes: Artificial Intelligence-Driven Predictive Modeling for Home Discharge in Neurological and Orthopedic Conditions","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-05-17 18:24:41","doi":"10.21203/rs.3.rs-4383785/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-03-17T09:08:47+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-03-16T14:50:01+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-03-03T04:13:21+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"25808211633619750224307028713409079751","date":"2025-03-03T04:11:57+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"201877200175110036229811409134543718121","date":"2025-02-26T13:12:11+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"178323377212864715885331369722541009458","date":"2024-12-16T13:28:39+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-09-09T13:11:17+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"316865920889802187291977981789667878373","date":"2024-08-20T06:43:41+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"11519169296694631246492346483106432406","date":"2024-05-24T10:44:03+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-05-08T18:00:53+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-05-08T06:44:24+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-05-08T06:44:24+00:00","index":"","fulltext":""},{"type":"submitted","content":"Journal of NeuroEngineering and Rehabilitation","date":"2024-05-07T14:20:55+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"journal-of-neuroengineering-and-rehabilitation","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"jner","sideBox":"Learn more about [Journal of NeuroEngineering and Rehabilitation](http://jneuroengrehab.biomedcentral.com/)","snPcode":"12984","submissionUrl":"https://submission.nature.com/new-submission/12984/3","title":"Journal of NeuroEngineering and Rehabilitation","twitterHandle":"@BioMedCentral","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"af192201-23c5-4c41-b7b7-acb029636d43","owner":[],"postedDate":"May 17th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-06-02T16:10:08+00:00","versionOfRecord":{"articleIdentity":"rs-4383785","link":"https://doi.org/10.1186/s12984-025-01654-4","journal":{"identity":"journal-of-neuroengineering-and-rehabilitation","isVorOnly":false,"title":"Journal of NeuroEngineering and Rehabilitation"},"publishedOn":"2025-05-26 15:57:27","publishedOnDateReadable":"May 26th, 2025"},"versionCreatedAt":"2024-05-17 18:24:41","video":"","vorDoi":"10.1186/s12984-025-01654-4","vorDoiUrl":"https://doi.org/10.1186/s12984-025-01654-4","workflowStages":[]},"version":"v1","identity":"rs-4383785","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4383785","identity":"rs-4383785","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}