Framingham Risk Score Prediction at 12 Months in the STANDFIRM Randomized Control Trial.

Area of Science:

• Cardiovascular Medicine
• Neurology
• Data Science

Background:

• The STANDFIRM trial investigated chronic disease management for modifying Framingham Risk Score (FRS) in stroke/TIA patients.
• The primary outcome, change in FRS, was not achieved in the trial.
• This study aimed to identify baseline predictors of FRS reduction and assess if future FRS is predetermined.

Purpose of the Study:

• To determine baseline characteristics that predict a reduction in Framingham Risk Score (FRS) at 12 months post-stroke or transient ischemic attack.
• To investigate whether future FRS is predetermined at baseline.
• To evaluate the effectiveness of machine learning models in predicting FRS and its changes.

Main Methods:

• Utilized machine learning regression methods (random forest, XGBoost, CatBoost, SVR, MLP, KNN) to analyze 35 variables.
• Trained and tested models on data from 404 and 103 patients, respectively, matched for age, sex, and baseline/12-month FRS.
• Employed Shapley additive explanation (SHAP) to identify key predictive variables.

Main Results:

• Category boosting was the optimal model for predicting 12-month FRS (R²=0.712).
• Key predictors for FRS included baseline FRS, age, systolic blood pressure, male sex, and London Handicap score.
• Machine learning models demonstrated poor performance in predicting the change in FRS (R²<0.22).

Conclusions:

• Change in Framingham Risk Score (FRS) may have limited value as an endpoint in secondary stroke trials due to its strong baseline determination.
• Category boosting effectively predicted future FRS but not FRS change in this cohort.
• Baseline characteristics significantly influence future cardiovascular risk assessment in post-stroke patients.