Predictive Factors Associated with Access to Stroke Fast-Track Services: A Data Mining Approach in Community Hospital Settings

Abstract

This retrospective analytical study employed data mining techniques to identify key predictive factors influencing stroke fast-track service access within 4.5 hours. The study sample consisted of 424 stroke patients who received treatment at Ban Phue Hospital, Udon Thani Province, Thailand. Research instruments included electronic medical record databases, and hospital stroke reports. Data were analyzed using three machine learning algorithms:
k-Nearest Neighbors (KNN), Naïve Bayes, and Random Forest. Model performance was evaluated using 10-fold cross-validation and assessed by Classification Accuracy (CA), Recall, Specificity, and the Area Under the Receiver Operating Characteristic Curve (AUC). The majority of participants were male (60.14%), with 85.38% diagnosed with ischemic stroke; notably, 51.18% accessed stroke fast-track services within the 4.5-hour window. Comparison of model performance revealed that Naïve Bayes achieved the highest predictive performance, with a Classification Accuracy of 71.7%, Recall of 71.7%, Specificity of 71.9%, and an AUC of 0.787. The most important predictors of timely access included stroke type, patient’s subdistrict of residence, and travel time to the hospital. The findings demonstrate the potential of data mining techniques for analyzing and predicting access to stroke fast-track services. These findings may support healthcare service planning, improve access to care in remote areas, and inform community-level health policy decision-making.