Random forest achieved optimal discriminative performance for gait impairment in ischemic stroke patients with an AUC of 0.868 in training and 0.681 in testing, with SHAP analysis identifying neutrophil-to-lymphocyte ratio as the top contributing factor among five key variables.
Random forest was the best-performing machine learning model for identifying gait impairment in ischemic stroke patients.
Five significant factors associated with walking ability in ischemic stroke patients were identified through feature selection.
SHAP analysis identified the neutrophil-to-lymphocyte ratio (NLR) as the top contributing factor to the model's predictions of gait impairment.
Lesion location in the occipital and frontal lobes was associated with walking ability in ischemic stroke patients.
Age and gender were identified as significant clinical factors associated with walking ability after ischemic stroke.
The machine learning models were described as showing preliminary promise as screening tools for early risk stratification for gait impairment in ischemic stroke patients.
This research suggests that machine learning models combining inflammation markers from blood tests with clinical information can help identify which ischemic stroke patients are at risk for difficulty walking. The study analyzed data from 1,650 stroke patients and found that five factors were most important for predicting walking problems: a blood test ratio called the neutrophil-to-lymphocyte ratio (NLR, which reflects inflammation), age, sex, and whether the stroke affected the frontal or occipital lobes of the brain. Among six different machine learning approaches tested, the 'random forest' method performed best, correctly distinguishing patients with and without gait impairment about 68% of the time in new patients. The neutrophil-to-lymphocyte ratio stood out as the single most influential factor, highlighting the potential role of inflammation in determining stroke recovery outcomes related to walking. The researchers used an interpretability technique called SHAP analysis to make the model's reasoning transparent, showing exactly how much each factor contributed to individual predictions. This kind of explainability is important for clinical tools, as it allows clinicians to understand why the model flags a particular patient as high risk. This research suggests that combining routine blood inflammation markers with brain imaging and basic patient information could assist clinicians in earlier identification of stroke patients who may need more intensive rehabilitation support for walking recovery. However, since this was a single-center, retrospective study and the model's performance was notably lower when tested on new patients than during training, the findings would need to be confirmed in larger, multi-center studies before being applied in clinical practice.
Qiu J, Yang X, Guo M, Guo W, Guo X, Jia X, et al.. (2026). 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.. Frontiers in neurology. https://doi.org/10.3389/fneur.2026.1854073