https://orcid.org/0009-0006-7823-5899
Figures
Abstract
Purpose
We aimed to develop a machine learning model to predict activities of daily living (ADL) at discharge in stroke patients and identify key predictors to guide rehabilitation decisions.
Materials and methods
Data of 589 stroke inpatients (2019–2024) were split into good (BI ≥ 60) and poor (BI < 60) ADL groups. Continuous variables were processed using Z-score normalization, followed by preliminary univariate regression screening (P < 0.05) and final feature selection via LASSO regression (lambda.1se = 0.0488). The screened features were used to train and validate ten machine learning algorithms; 30% of the dataset (n = 177) was allocated as an independent test set for model evaluation, and SHAP analysis was performed to interpret the optimal model.
Results
Six of 41 features were retained. Random forest achieved the best performance (AUC = 0.958; accuracy = 0.936; sensitivity = 0.934; specificity = 0.950). SHAP identified the top drivers: admission Barthel Index, standing balance, Brunnstrom stages (upper and lower limb), dressing, and grooming abilities.
Citation: Ye Q, Fang G, Li L, Li Q, Yang Y, Liu L (2026) Explainable machine learning for predicting activities of daily living at discharge in stroke patients: A retrospective study using SHAP interpretability. PLoS One 21(7): e0351468. https://doi.org/10.1371/journal.pone.0351468
Editor: Fan Zhang, Longhua Hospital Shanghai University of Traditional Chinese Medicine, CHINA
Received: November 28, 2025; Accepted: May 26, 2026; Published: July 2, 2026
Copyright: © 2026 Ye et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All relevant data for this study are publicly available from the Zenodo repository (https://doi.org/10.5281/zenodo.20629742).
Funding: The author(s) received no specific funding for this work.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Stroke remains the leading cause of adult disability worldwide, with 12.8 million incident cases annually [1]. In China, stroke is the leading cause of death and disability-adjusted life years (DALYs) among adults, with 3.95 million new cases annually and projected increases through 2040 [2–4]. Over 75% of surviving patients experience varying degrees of functional impairment, with approximately 42.3% unable to live independently due to disabilities. This reduces quality of life, increases utilization and readmission rates, and imposes a substantial socioeconomic burden [5–7].
ADL are used to quantify post-stroke functional deficits and rehabilitation outcomes. Accurate ADL prediction is clinically valuable for developing personalized rehabilitation strategies and improving long-term outcomes. Barthel Index (BI) is the most widely used standardized scale for ADL assessment in clinical practice [8–12]. While the BI demonstrates high inter-rater reliability (weighted κ = 0.93) and remains the gold standard for ADL assessment [13], functional recovery involves multifactorial determinants beyond single-scale metrics, including age, the type and location of the stroke, the stage of motor recovery, nutritional status, and complications [14–19]. Traditional statistical methods may inadequately capture these nonlinear interactions. Developing ML-based models that integrate multi-dimensional clinical data for ADL prediction is therefore a research priority.
Machine learning (ML), a subset of artificial intelligence, enables automatic pattern recognition from large datasets without explicit programming. ML has shown efficacy in predicting stroke rehabilitation outcomes, including motor recovery, walking ability, and swallowing function [20–22]. However, systematic ML-based predictive models for ADL at discharge in stroke patients remain limited. This study aims to develop an ML-based predictive model for ADL at discharge by integrating multi-source clinical features. The goal is to identify predictors of ADL outcomes to inform rehabilitation decisions and individualized patient management.
Materials and methods
Ethical Approval and Patient Selection
This study was conducted the TRIPOD statement and was approved by the Ethics Committee of the First Affiliated Hospital of Nanjing Medical University (Ethics No: 2025-SR-088). The completed TRIPOD checklist is included as Supplementary Material 1. As all the data were de-identified, informed consent was waived. Inclusion criteria: 1) age between 18 and 85 years; 2) complete clinical and laboratory examination records. Exclusion criteria: severe dysfunction in the heart, lungs, liver, kidneys, or other major systemic diseases. The study ensured the confidentiality of patient information, and all data were anonymized during the analysis process to protect patients’ privacy. An exemption for informed consent was obtained.
Data Collection for Predictor Variables
Data were extracted from the electronic medical record system for all acute stroke patients admitted to the Rehabilitation Medical Center between 8/2/2019 and 18/11/2024. The dataset comprised seven categories of indicators: 1. Demographic characteristics: gender, age; 2.Present illness history: location of stroke, type of stroke, duration of the disease; 3.Past medical history: hypertension, diabetes, previous strokes; 4.Personal history: smoking, alcohol use and stroke surgery; 5. Rehabilitation assessment indicators: Brunnstrom stages (upper limbs, hand, lower limbs), triceps muscle tone (modified Ashworth scale), admission BI score (total and for feeding, dressing, grooming, toileting, bathing, urination, defecation, transfer, walking, and climbing stairs), discharge BI total score, sitting balance, and standing balance function; 6. Clinical medication history: anticoagulant medication use and antiplatelet medication use; 7. Laboratory test indicators: total cholesterol (TC), triglycerides (TG), low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), total protein (TP), albumin (ALB), globulin (GLB), hemoglobin (HGB), creatine kinase (CK), D-dimer, and the presence of lower limb deep vein thrombosis (diagnosed by lower limb venous ultrasound).
Patients were stratified into functional independence (BI ≥ 60) and dependence (BI < 60) groups. This 60-point threshold was selected based on the statistically optimized cutoff identified by Zhou et al. (2021) in a large-scale Chinese multicenter study representing the clinical standard for ‘most self-care’ ability in Chinese stroke populations [23]. Notably, BI ≥ 60 is also commonly regarded as the minimum standard for ‘basic self-care ability in Chinese clinical practice [24, 25], and has been validated in Chinese stroke cohorts [23,26,27].
Data Preprocessing
Continuous variables including age, Barthel Index, creatine, TC, TG, HDL, LDL, TP, albumin, globulin, HGB, and D-dimer were normalized using Z-score transformation. Normalization parameters (mean and standard deviation) were calculated from the training set only and applied to the test set to prevent data leakage.
Model Construction and Evaluation
Feature dimension reduction was performed using a two-stage process. First, univariate logistic regression analysis was conducted (P < 0.05). Second, the Least Absolute Shrinkage and Selection Operator (LASSO) regression was applied using 10-fold cross-validation to determine the optimal regularization parameter. The dataset was divided into a training set (70%, n = 412) and an independent test set (30%, n = 177) using stratified random sampling. Based on the selected features, ten machine learning models were constructed to predict the risk of patients’ ADL at discharge. The models included Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), Decision Tree (DT), Neural Network (NNET), Extreme Gradient Boosting (XGBoost), Naive Bayes (NB), K-Nearest Neighbors (KNN), Linear Discriminant Analysis (LDA), and Generalized Linear Model (Glmnet). We evaluated model performance on the independent test set (not used during training or tuning) to obtain unbiased estimates. Ten-fold cross-validation was used for model training and hyperparameter optimization, while the overall model performance was assessed using metrics such as the area under the receiver operating characteristic curve (AUC), precision, accuracy, sensitivity, specificity and F1-score.
Statistical Methods
Variables with < 5% missing data had their values imputed using the random forest algorithm. Normally distributed continuous variables were expressed as mean±standard deviation (SD) and compared between groups using independent t-tests. Variables that did not follow a normal distribution were reported as medians (interquartile range) [M (P25, P75)], with group comparisons performed using rank-sum tests. Categorical variables were expressed as n(%) and compared using χ² test. All analyses were conducted using R software (version 4.2.2) for data preprocessing, feature selection, and modeling. SHAP (SHapley Additive exPlanations) interpretability analysis was performed using Python (version 3.11.3). P < 0.05 was considered statistically significant.
Results
Baseline Comparison
A total of 589 stroke patients met the inclusion criteria and were included in this study. Of these, 306 (51.95%) had BI < 60 and 283 (48.05%) had BI ≥ 60 at discharge. The specific baseline characteristics are detailed in Table 1.
Model Construction and Performance
Univariate screening identified 24 predictors from 41 candidate features.These features included age, surgical history, standing balance, dressing, transfer, walking, Brunnstrom lower limb, grooming, sitting balance, feeding, Brunnstrom upper limb, total admission BI score, Brunnstrom hand, bladder control, bowel control, going up and down stairs, bathing, serum albumin, antiplatelet drug use, serum globulin, D-dimer, lower extremity venous thrombosis status, hemoglobin, toilet use. Subsequently, LASSO regression with lambda.1se = 0.0488 selected six core features (Fig.1): admission Barthel Index, standing balance, Brunnstrom stage (upper limb), Brunnstrom stage (lower limb), dressing, and grooming. After training with ten machine learning models, the Random Forest (ntree = 1000, mtry = 2, nodesize = 3) exhibited the best performance on the test set (n = 177): AUC = 0.958; accuracy = 0.936; sensitivity = 0.934; specificity = 0.950 (Table 2). All models achieved AUC > 0.910, confirming the robustness of the selected features.
Upper panel: log (λ) values of the 41 features in the LASSO model. A coefficient profile plot was produced against the log (λ) sequence. Lower panel: parameter selection in the LASSO model used 10-fold cross validation via minimum criterion. Partial likelihood deviation (binomial deviation) curves and log (λ) curves were plotted.
Feature Importance and Interpretation
SHAP analysis revealed the contribution of each feature (Fig 2 and 3). Admission BI was the most important predictor, followed by standing balance, dressing, lower limb Brunnstrom stage, grooming, and upper limb Brunnstrom stage. Importantly, these SHAP values represent associations between baseline features and discharge ADL outcomes rather than causal relationships. For example, a high admission BI is associated with better discharge ADL, but this does not necessarily imply that improving the BI will cause better outcomes, as both may be influenced by unmeasured factors.
Each row represents a feature. A higher SHAP value indicates a greater contribution to model prediction.
Each dot corresponds to an individual sample. Horizontal dispersion of dots reflects the influence intensity of each feature: greater dispersion indicates a stronger impact. Colors denote feature values (red = high, blue = low).
The partial dependence plots (PDPs) (Fig.4) demonstrated that all six core predictors were positively associated with favorable discharge ADL outcomes. Admission BI was the strongest predictor, with a non-linear trend (sharp rise 20–60, plateau after 60) identifying BI < 20 as a high-risk threshold. Standing balance and lower limb Brunnstrom stage were the next key drivers, while dressing ability, upper limb Brunnstrom stage, and grooming ability also showed positive contributions to ADL prognosis.
The figure illustrates the marginal effects of six key predictive features on the model’s prediction of ADL independence. The y-axis represents the partial dependence value, indicating the average change in the predicted probability of ADL independence when varying the feature of interest while holding all other features constant. The x-axis shows the original range of each feature.
Discussion
Random forest outperformed nine other models, achieving the highest AUC (0.958), accuracy (0.936), sensitivity (0.934), and specificity (0.950). This performance reflects RF’s ability to model nonlinear relationships and interactions while remaining robust to outliers [28]. Bootstrap sampling and random feature subspaces reduce overfitting risk compared to single decision trees [29]. SHAP analysis enhances interpretability by quantifying feature contributions [30].
The core predictive factors were admission BI, standing balance ability, Brunnstrom stage(lower limb), dressing ability, Brunnstrom stage (upper limb), and grooming ability. These associations reflect cohort patterns, not causal relationships.
Admission BI was the strongest predictor. Partial dependence analysis revealed consistently low predicted probabilities when admission BI fell below 40 points, particularly under 20 points, suggesting limited rehabilitation potential in this subset. Notably, a distinct inflection point was observed around 40 points, beyond which predicted values exhibited a gradual ascent within the 40–60 points, indicating emerging recovery potential that substantially plateaued at approximately 60 points. This threshold likely reflects differences in residual functional capacity. Patients scoring 20–40 points showed profound impairments in feeding, grooming, and sphincter control. Fewer than 50% achieved independence in these tasks [31]. Such severe deficits often indicate extensive neurological damage or poor physiological reserve, leaving limited substrate for rehabilitation-driven neuroplasticity. Conversely, patients entering the 40–60 points have generally established foundational self-care abilities and retain sufficient residual motor function that can be effectively harnessed through targeted interventions. This accelerated improvement aligns with established literature demonstrating [32–36] that early BI scores closely correlate with post-stroke rehabilitation prognosis and serve as robust indicators of functional independence [37–41]. An admission BI exceeding 40 points appears to serve as a critical prerequisite for meaningful functional gains, identifying patients with favorable recovery foundations who are likely to achieve superior ADL outcomes through systematic rehabilitation [31,42].
Standing balance ranked second in feature importance. The transition from static sitting balance to standing balance marked a critical threshold for functional recovery; only patients achieving standing balance capability demonstrated substantial functional improvement. This finding highlights the importance of trunk control in post-stroke rehabilitation [43]. Standing balance contributes to mobility-related activities; adequate trunk control is essential for transfers and ambulation [44], whereas impaired postural control restricts patients to dependent sitting positions [45]. Poor balance increases fall risk during rehabilitation, with consequent injury and recovery setbacks [46]. These findings support the inclusion of early, targeted trunk control interventions in stroke rehabilitation protocols.
Brunnstrom lower limb staging demonstrated a nonlinear association with ADL independence, characterized by a steep ascent from Stage II to IV, followed by a decelerated rise toward Stage V and a subsequent plateau [47]. The rapid improvement during Stage II–IV reflects the critical window of neuroplasticity, wherein patients sequentially acquire sitting balance, standing ability via extensor synergy, and dissociated movements essential for functional ambulation, resulting in stepwise gains in high-weighted ADL items (transfers and locomotion) [48]. Beyond Stage IV, recovery shifts toward fine motor control and gait quality optimization, yielding diminishing returns on standard scales [49], while Stage V exhibits a marked ceiling effect of the Barthel Index and Functional Independence Measure, which lack sensitivity to residual deficits. These findings suggest that rehabilitation should intensify task-specific gait and transfer training before Stage IV to maximize functional independence, whereas post-Stage V care should focus on community integration and advanced ADL goals rather than staging progression alone.
Dressing and grooming abilities exhibited rapid ascent. Independence in these activities indicates substantial functional recovery [19]. De Wit et al. [50] reported that independence in dressing and bathing at discharge predicted long-term functional independence: independent dressers achieved 74% self-care rates at 5 years versus 6% for dependent patients.
Brunnstrom upper limb stage revealed patients at stage II demonstrated worse discharge ADL independence than those at stage I. This may reflect learned non-use [51,52]. During stage I (flaccidity), rehabilitation focuses on passive movement [53,54]. Upon entering stage II (spasticity emergence), patients may attempt active use but experience repeated failures due to unstable motor control, reinforcing limb abandonment [55,56]. Spasticity and synergistic movements during stage II may create functional interference, with the paretic limb becoming a burden [57–59]. The steep stage II-IV ascent aligns with motor control literature: recovery potential concentrates in the transition from spasticity-dominated to dissociated movement [60].The subsequent stage IV-VI plateau decline suggests decreasing marginal improvement as upper limb function approaches ceiling effects.
Limitations
Our study has several limitations. First, as a single-center retrospective study, selection bias may limit the sample representativeness and generalizability. Second, reliance on retrospective electronic medical record data extraction may introduce information bias due to incomplete or non-standardized records. Third, despite statistical adjustments, residual confounding from unmeasured variables cannot be excluded. Future studies should consider multi-center collaboration and prospective design to enhance reliability and generalization.
Conclusion
This study developed an ML-based predictive model for ADL at discharge in stroke patients, demonstrating strong discriminative performance(AUC 0.958) in internal validation. SHAP interpretability analysis identified key features influencing functional outcomes and their contribution patterns. By combining predictive power with clinical interpretability, the model is expected to assist in identifying high-risk individuals and elucidating the primary causes of functional limitations.
References
- 1. GBD 2021 Nervous System Disorders Collaborators. Global, regional, and national burden of disorders affecting the nervous system, 1990-2021: a systematic analysis for the Global Burden of Disease Study 2021. Lancet Neurol. 2024;23(4):344–81.
- 2. Ma Q, Li R, Wang L. Temporal trend and attributable risk factors of stroke burden in China, 1990-2019: an analysis for the Global Burden of Disease Study 2019. Lancet Public Health. 2021;6(12):e897–906.
- 3. National Clinical Research Center for Neurological Diseases, China Stroke Center Alliance. Summary of China Stroke Center Report 2022. Chin J Cerebrovasc Dis. 2024;21(8):565–76.
- 4. Wang SH, Lai FX, Zhang ZY. Prediction of the epidemic trend of ischemic stroke in China from 2020 to 2040. Chinese J Neuroimmunol Neurol. 2024;31(6):459–67.
- 5. Tsalta-Mladenov M, Andonova S. Health-related quality of life after ischemic stroke: impact of sociodemographic and clinical factors. Neurol Res. 2021;43(7):553–61. pmid:33637026
- 6. Ramos-Lima MJM, Brasileiro I de C, Lima TL de, Braga-Neto P. Quality of life after stroke: impact of clinical and sociodemographic factors. Clinics (Sao Paulo). 2018;73:e418. pmid:30304300
- 7. Lackland DT, Roccella EJ, Deutsch AF, et al. Factors influencing the decline in stroke mortality: a statement from the American Heart Association/American Stroke Association. Stroke. 2014;45(1):315–53.
- 8. Mahoney FI, Barthel DW. Functional evaluation: the Barthel index. Md State Med J. 1965;14:61–5. pmid:14258950
- 9. Granger CV, Hamilton BB, Gresham GE, Kramer AA. The stroke rehabilitation outcome study: Part II. Relative merits of the total Barthel index score and a four-item subscore in predicting patient outcomes. Arch Phys Med Rehabil. 1989;70(2):100–3. pmid:2916925
- 10. Bahrmann A, Benner L, Christ M, Bertsch T, Sieber CC, Katus H, et al. The Charlson Comorbidity and Barthel Index predict length of hospital stay, mortality, cardiovascular mortality and rehospitalization in unselected older patients admitted to the emergency department. Aging Clin Exp Res. 2019;31(9):1233–42. pmid:30406920
- 11. Martín Moreno V, Martínez Sanz MI, Martín Fernández A, Guerra Maroto S, Sevillano Fuentes E, Pérez Rico E, et al. Proposal for a revised Barthel index classification based on mortality risk assessment in functional dependence for basic activities of daily living. Front Public Health. 2025;12:1478897. pmid:39877909
- 12. Lee EY, Sohn MK, Lee JM. Changes in long-term functional independence in patients with moderate and severe ischemic stroke: comparison of the responsiveness of the modified Barthel index and the functional independence measure. Int J Environ Res Public Health. 2022;19(15):9612.
- 13. Duffy L, Gajree S, Langhorne P, Stott DJ, Quinn TJ. Reliability (inter-rater agreement) of the Barthel Index for assessment of stroke survivors: systematic review and meta-analysis. Stroke. 2013;44(2):462–8. pmid:23299497
- 14. Sturm JW, Donnan GA, Dewey HM, Macdonell RAL, Gilligan AK, Srikanth V, et al. Quality of life after stroke: the North East Melbourne Stroke Incidence Study (NEMESIS). Stroke. 2004;35(10):2340–5. pmid:15331799
- 15. Mahesh PKB, Gunathunga MW, Jayasinghe S, Arnold SM, Liyanage SN. Factors influencing pre-stroke and post-stroke quality of life among stroke survivors in a lower middle-income country. Neurol Sci. 2018;39(2):287–95. pmid:29103178
- 16. Safaz İ, Kesikburun S, Adigüzel E, Yilmaz B. Determinants of disease-specific health-related quality of life in Turkish stroke survivors. Int J Rehabil Res. 2016;39(2):130–3. pmid:26795717
- 17. Kelly PJ, Furie KL, Shafqat S, Rallis N, Chang Y, Stein J. Functional recovery following rehabilitation after hemorrhagic and ischemic stroke. Arch Phys Med Rehabil. 2003;84(7):968–72. pmid:12881818
- 18. Katrak PH, Black D, Peeva V. Do stroke patients with intracerebral hemorrhage have a better functional outcome than patients with cerebral infarction?. PM R. 2009;1(5):427–33. pmid:19627929
- 19. Paolucci S, Antonucci G, Grasso MG, Bragoni M, Coiro P, De Angelis D, et al. Functional outcome of ischemic and hemorrhagic stroke patients after inpatient rehabilitation: a matched comparison. Stroke. 2003;34(12):2861–5. pmid:14615613
- 20. Zu W, Huang X, Xu T, Du L, Wang Y, Wang L, et al. Machine learning in predicting outcomes for stroke patients following rehabilitation treatment: A systematic review. PLoS One. 2023;18(6):e0287308. pmid:37379289
- 21. Ma J, Xie Y. Machine learning techniques for independent gait recovery prediction in acute anterior circulation ischemic stroke. J Neuroeng Rehabil. 2025;22(1):19.
- 22. Jeong SY, Kim JM, Park JE, Baek SJ, Yang SN. Application of deep learning technology for temporal analysis of videofluoroscopic swallowing studies. Sci Rep. 2023;13(1):17522. pmid:37845272
- 23. Zhou M, Liu X, Zha F, Liu F, Zhou J, Huang M, et al. Stroke outcome assessment: Optimizing cutoff scores for the Longshi Scale, modified Rankin Scale and Barthel Index. PLoS One. 2021;16(5):e0251103. pmid:33984006
- 24.
Wang YL. Rehabilitation Function Evaluation. Beijing: People’s Medical Publishing House. 2008.
- 25.
Zhuo DH. Chinese rehabilitation medicine. 2nd ed. Beijing: Huaxia Publishing House. 2003.
- 26. Ouyang YT, Zhan J, Wang QW. Construction and validation of a nomogram model based on influencing factors of activities of daily living in patients with ischemic stroke. Chin J Rehabil Med. 2025;40(12):1808–15.
- 27. Wang SH, Shi JJ, Sun Y. Reliability and validity of the simplified Chinese version of the modified Barthel Index in stroke patients during rehabilitation period. Chin J Rehabil. 2020;35(4):179–82.
- 28. Chen X, Ishwaran H. Random forests for genomic data analysis. Genomics. 2012;99(6):323–9. pmid:22546560
- 29. Couronné R, Probst P, Boulesteix A-L. Random forest versus logistic regression: a large-scale benchmark experiment. BMC Bioinformatics. 2018;19(1):270. pmid:30016950
- 30. Chen H, Lundberg SM, Lee S-I. Explaining a series of models by propagating Shapley values. Nat Commun. 2022;13(1):4512. pmid:35922410
- 31. Granger CV, Dewis LS, Peters NC, Sherwood CC, Barrett JE. Stroke rehabilitation: analysis of repeated Barthel index measures. Arch Phys Med Rehabil. 1979;60(1):14–7. pmid:420565
- 32. Duncan PW, Lai SM, van Culin V, Huang L, Clausen D, Wallace D. Development of a comprehensive assessment toolbox for stroke. Clin Geriatr Med. 1999;15(4):885–915. pmid:10499941
- 33. Shah S, Vanclay F, Cooper B. Improving the sensitivity of the Barthel Index for stroke rehabilitation. J Clin Epidemiol. 1989;42(8):703–9. pmid:2760661
- 34. Li Q-X, Zhao X-J, Wang Y, Wang D-L, Zhang J, Liu T-J, et al. Value of the Barthel scale in prognostic prediction for patients with cerebral infarction. BMC Cardiovasc Disord. 2020;20(1):14. pmid:31931720
- 35. Kwakkel G, Veerbeek JM, Harmeling-van der Wel BC, van Wegen E, Kollen BJ, Early Prediction of functional Outcome after Stroke (EPOS) Investigators. Diagnostic accuracy of the Barthel Index for measuring activities of daily living outcome after ischemic hemispheric stroke: does early poststroke timing of assessment matter?. Stroke. 2011;42(2):342–6. pmid:21183748
- 36. Hung M-C, Hsieh C-L, Hwang J-S, Jeng J-S, Wang J-D. Estimation of the long-term care needs of stroke patients by integrating functional disability and survival. PLoS One. 2013;8(10):e75605. pmid:24124500
- 37. Rahayu UB, Wibowo S, Setyopranoto I, Hibatullah Romli M. Effectiveness of physiotherapy interventions in brain plasticity, balance and functional ability in stroke survivors: a randomized controlled trial. NeuroRehabilitation. 2020;47(4):463–70. pmid:33164953
- 38. Koch G, Bonnì S, Casula EP, Iosa M, Paolucci S, Pellicciari MC, et al. Effect of Cerebellar Stimulation on Gait and Balance Recovery in Patients With Hemiparetic Stroke: A Randomized Clinical Trial. JAMA Neurol. 2019;76(2):170–8. pmid:30476999
- 39. Wu W-X, Zhou C-Y, Wang Z-W, Chen G-Q, Chen X-L, Jin H-M, et al. Effect of early and intensive rehabilitation after ischemic stroke on functional recovery of the lower limbs: a pilot, randomized trial. J Stroke Cerebrovasc Dis. 2020;29(5):104649. pmid:32115341
- 40. Wang X, Qiu J, Zhou Y, Liu W, Zhang S, Gong Y, et al. Effects of virtual reality-assisted and overground gait adaptation training on balance and walking ability in stroke patients: a randomized controlled trial. Am J Phys Med Rehabil. 2024;103(6):480–7. pmid:38063309
- 41. Nasb M, Li Z, S A Youssef A, Dayoub L, Chen H. Comparison of the effects of modified constraint-induced movement therapy and intensive conventional therapy with a botulinum-a toxin injection on upper limb motor function recovery in patients with stroke. Libyan J Med. 2019;14(1):1609304. pmid:31032717
- 42. So C, Lage DE, Slocum CS, Zafonte RD, Schneider JC. Utility of Functional Metrics Assessed During Acute Care on Hospital Outcomes: A Systematic Review. PM R. 2019;11(5):522–32. pmid:30758920
- 43. Franchignoni FP, Tesio L, Ricupero C, Martino MT. Trunk control test as an early predictor of stroke rehabilitation outcome. Stroke. 1997;28(7):1382–5. pmid:9227687
- 44. Song JW, Kim JM, Cheong YS, Lee Y-S, Chun SM, Min Y-S, et al. Balance Assessment in Subacute Stroke Patients Using the Balance Control Trainer (BalPro). Ann Rehabil Med. 2017;41(2):188–96. pmid:28503450
- 45. Cabanas-Valdés R, Cuchi GU, Bagur-Calafat C. Trunk training exercises approaches for improving trunk performance and functional sitting balance in patients with stroke: a systematic review. NeuroRehabilitation. 2013;33(4):575–92. pmid:24018373
- 46. Verheyden G, Vereeck L, Truijen S, Troch M, Herregodts I, Lafosse C, et al. Trunk performance after stroke and the relationship with balance, gait and functional ability. Clin Rehabil. 2006;20(5):451–8. pmid:16774097
- 47. Chen X, Hu D, Zhang R, Pan Z, Chen Y, Xie L, et al. Interpretable evaluation for the Brunnstrom recovery stage of the lower limb based on wearable sensors. Front Neuroinform. 2022;16:1006494. pmid:36156985
- 48. Nishida T, Miyata Y, Tanaka T, et al. The Impact of Trunk Function and Lower Limb Paralysis on Independence in Activities of Daily Living Among Stroke Patients. Cureus. 2025;17(3):e40135023.
- 49. Yun SM, Lee SY, Sohn MK, Lee J, Kim DY, Lee S-G, et al. Factors Associated with Changes in Functional Independence after Six Months of Ischemic Stroke. Brain Neurorehabil. 2020;13(3):e19. pmid:36741795
- 50. De Wit L, Putman K, Devos H, Brinkmann N, Dejaeger E, De Weerdt W, et al. Long-term prediction of functional outcome after stroke using single items of the Barthel Index at discharge from rehabilitation centre. Disabil Rehabil. 2014;36(5):353–8. pmid:23692390
- 51. Hirsch T, Barthel M, Aarts P, Chen Y-A, Freivogel S, Johnson MJ, et al. A first step toward the operationalization of the learned non-use phenomenon: a delphi study. Neurorehabil Neural Repair. 2021;35(5):383–92. pmid:33703971
- 52. Ballester BR, Winstein C, Schweighofer N. Virtuous and vicious cycles of arm use and function post-stroke. Front Neurol. 2022;13:804211. pmid:35422752
- 53. Shahid J, Kashif A, Shahid MK. A Comprehensive review of physical therapy interventions for stroke rehabilitation: impairment-based approaches and functional goals. Brain Sci. 2023;13(5):717. pmid:37239189
- 54. Mariana de Aquino Miranda J, Mendes Borges V, Bazan R, José Luvizutto G, Sabrysna Morais Shinosaki J. Early mobilization in acute stroke phase: a systematic review. Top Stroke Rehabil. 2023;30(2):157–68. pmid:34927568
- 55. Li S, Francisco GE, Rymer WZ. A new definition of poststroke spasticity and the interference of spasticity with motor recovery from acute to chronic stages. Neurorehabil Neural Repair. 2021;35(7):601–10. pmid:33978513
- 56. McPherson LM, Dewald JPA. Abnormal synergies and associated reactions post-hemiparetic stroke reflect muscle activation patterns of brainstem motor pathways. Front Neurol. 2022;13:934670. pmid:36299276
- 57. Lackritz H, Parmet Y, Frenkel-Toledo S, Baniña MC, Soroker N, Solomon JM, et al. Effect of post-stroke spasticity on voluntary movement of the upper limb. J Neuroeng Rehabil. 2021;18(1):81. pmid:33985543
- 58. Hadjiosif AM, Branscheidt M, Anaya MA, Runnalls KD, Keller J, Bastian AJ, et al. Dissociation between abnormal motor synergies and impaired reaching dexterity after stroke. J Neurophysiol. 2022;127(4):856–68. pmid:35108107
- 59. Dromerick AW, Geed S, Barth J, Brady K, Giannetti ML, Mitchell A, et al. Critical Period After Stroke Study (CPASS): A phase II clinical trial testing an optimal time for motor recovery after stroke in humans. Proc Natl Acad Sci U S A. 2021;118(39):e2026676118. pmid:34544853
- 60. Li S. Stroke recovery is a journey: prediction and potentials of motor recovery after a stroke from a practical perspective. Life (Basel). 2023;13(10):2061. pmid:37895442