Predicting post-stroke cognitive impairment using electronic health record data.

Area of Science:

• Neurology
• Data Science
• Public Health

Background:

• Post-stroke cognitive impairment (PSCI) poses a significant challenge, necessitating early identification of at-risk individuals for targeted secondary prevention.
• Effective risk factor modification strategies are crucial for reducing the incidence and severity of PSCI.

Purpose of the Study:

• To develop and validate a predictive model for identifying patients at elevated risk of developing PSCI within five years post-stroke.
• To utilize data readily available in electronic health records (EHRs) for model development and application.

Main Methods:

• A cohort study involving primary care patients aged 45+ from two academic medical centers, with incident ischemic stroke between 2003-2016 (development/internal validation) and 2010-2022 (external validation).
• Predictors of PSCI were extracted from EHRs; the outcome was defined as incident dementia/cognitive impairment within five years post-stroke using ICD-9/10 codes.
• A least absolute shrinkage and selection operator (LASSO) penalized Cox proportional hazards model with 10-fold cross-validation was employed for variable selection.

Main Results:

• The predictive model demonstrated good performance with a C-statistic of 0.731 in the internal validation cohort and 0.724 in the external validation cohort.
• A risk score stratified patients into low, intermediate, and high-risk groups for PSCI.
• High-risk individuals exhibited a significantly increased hazard ratio (HR) for incident PSCI (HR: 6.2 internal, 6.1 external) compared to low-risk groups.

Conclusions:

• Routinely collected electronic health record data can accurately predict the five-year risk of post-stroke cognitive impairment.
• The developed model provides a valuable tool for risk stratification, enabling targeted preventive interventions for individuals at increased risk of PSCI.