A machine learning model for predicting cognitive improvement after scalp cluster needling therapy in patients with post-stroke mild cognitive impairment.

A retrospective cohort of 750 hospitalized PSCI-MCI patients was split into training and validation sets to develop a model predicting cognitive improvement after SCNT-PR.

• Patients were hospitalized with PSCI-MCI between January 1, 2010, and January 1, 2023.
• Random assignment gave a training set of n = 526 and a validation set of n = 224.
• The outcome was an increase of at least 2 points in corrected MoCA score after 1 month of treatment compared with baseline.

Seven key predictors were identified from the intersection of LASSO regression and the Boruta algorithm.

• Baseline data covered demographics, vascular risk factors, stroke severity, imaging features, cognitive and functional assessments, emotional and sleep indicators, treatment-related variables, and laboratory findings.
• Seven supervised machine learning algorithms were applied to develop pretreatment prediction models using these predictors.

LightGBM showed the best, acceptable performance among the evaluated models, with an AUC of 0.759.

• Seven supervised machine learning algorithms were compared.
• Performance was assessed in terms of discrimination, calibration, and clinical utility.
• The authors describe the discrimination as 'acceptable', not high.

Calibration curves and decision curve analysis indicated acceptable calibration and favorable clinical utility for the LightGBM model.

• Calibration and clinical utility were evaluated alongside discrimination.
• The authors suggest the model may provide preliminary support for patient stratification and individualized clinical assessment.

SHAP analysis identified seven major contributors to model predictions: baseline corrected MoCA score, years of education, NIHSS total score, Barthel Index, age, frontal lobe lesion, and diabetes.

• These seven variables correspond to the predictors selected by LASSO and Boruta.
• SHAP was used for model interpretation, ranking the contribution of each feature to predictions.

Sensitivity analyses showed consistent predictor patterns under alternative outcome definitions and predictor configurations.

• Alternative definitions of cognitive improvement were tested.
• Alternative predictor configurations were also tested, and the predictor patterns remained consistent.

The model was only internally validated, and the authors state external validation in independent cohorts is required before clinical implementation.

• The study was retrospective and used data from hospitalized patients.
• Validation used a random split of the same cohort.