Predicting short and long-term mortality after acute ischemic stroke using EHR.

Area of Science:

• Medical Informatics
• Cardiovascular Research
• Public Health

Background:

• Stroke remains a significant cause of mortality and disability despite treatment advancements.
• Accurate prediction of post-stroke mortality is essential for patient management and resource allocation.

Purpose of the Study:

• To develop and validate predictive models for short- and long-term all-cause mortality after stroke using administrative data.
• To identify key factors associated with post-stroke mortality across different time windows.

Main Methods:

• Utilized patient-level electronic health record data for 7144 ischemic stroke survivors.
• Developed prediction models using three algorithms (e.g., Random Forest) across six prediction windows.
• Adhered to the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) guideline.

Main Results:

• Mortality rates ranged from 8% at 1 month to 25% at 2 years.
• The Random Forest model achieved an Area Under the ROC curve (AUROC) of 0.82 for 1-month mortality prediction.
• Age, hemoglobin levels, and body mass index were primary predictors; laboratory variables outweighed comorbidities. Hypercoagulation, smoking, and end-stage renal disease strongly predicted long-term mortality.

Conclusions:

• Trained algorithms effectively predict short- and long-term post-stroke mortality.
• Factors influencing mortality vary significantly based on the prediction window.
• The models emphasize the importance of managing risk factors, particularly those indicated by laboratory measures.