Development and validation of multiple machine learning models for identifying factors associated with walking ability in ischemic stroke patients: a single-center retrospective study with SHAP approach.

OBJECTIVES: The aim of this study was to develop and validate machine learning (ML) models that integrate inflammatory biomarkers with clinical indicators to identify factors associated with walking ability in patients with ischemic stroke (IS). The major research question was which ML model achieves optimal discriminative performance for gait impairment in IS patients, and which factors are the key factors of walking ability in this population. METHODS: This retrospective cohort study enrolled 1,650 patients diagnosed with ischemic stroke. The participants were randomly allocated to a training set (70%) and a validation set (30%). Data on clinical, laboratory, and imaging variables were collected. Feature selection was performed using the Least Absolute Shrinkage and Selection Operator (LASSO) regression, the Boruta algorithm, and logistic regression. Six machine learning models were developed. SHapley Additive exPlanations (SHAP) analysis was applied to the best-performing model. RESULTS: Five significant factors were screened: the neutrophil-to-lymphocyte ratio (NLR), age, gender, occipital lobe and frontal lobe lesions. Random forest (RF) achieved optimal performance with an AUC of 0.868 in the training set and 0.681 in the test set. SHAP analysis prioritized NLR as the top contributor. CONCLUSIONS: The ML models show preliminary promise as screening tools for early risk stratification for gait impairment in IS patients.