An interpretable machine learning scoring tool for estimating time to recurrence readmissions in stroke patients.

Area of Science:

• Medical Informatics
• Cardiovascular Medicine
• Public Health

Background:

• Stroke recurrence readmission incurs significant patient and healthcare burdens.
• Accurate risk stratification is crucial for targeted interventions to reduce readmissions.
• Predicting short- and intermediate-term readmission risks aids temporal risk stratification.

Purpose of the Study:

• To develop and validate an interpretable machine learning risk scoring tool for predicting stroke recurrence readmission.
• To provide accurate time-to-readmission information for enhanced risk stratification.

Main Methods:

• Retrospective study using Shanghai Health and Health Development Research Centre database (2015-2019).
• Developed the Score for Stroke Recurrence Readmission Prediction (SSRRP) using interpretable machine learning for time-to-event outcomes.
• SSRRP incorporates six variables: sequelae, length of stay, stroke type, random plasma glucose, medical expense payment, and number of hospitalizations.

Main Results:

• 339,212 stroke admission episodes analyzed; 9.97% experienced 90-day readmission.
• The SSRRP demonstrated an integrated area under the curve of 0.730 in the temporal validation dataset.
• The tool showed good calibration and clinical benefit rate.

Conclusions:

• The SSRRP is a parsimonious, point-based scoring tool for predicting recurrent stroke readmission risk.
• The tool accurately provides time-to-readmission data, facilitating temporal risk stratification.
• SSRRP supports informed clinical decision-making for stroke patient management.