Predicting Cardiovascular Risk Using Social Media Data: Performance Evaluation of Machine-Learning Models

Area of Science:

• Cardiology
• Computational Linguistics
• Digital Health

Background:

• Current atherosclerotic cardiovascular disease (ASCVD) predictive models have limitations.
• There is a need to improve the discriminatory power of ASCVD risk prediction models.

Purpose of the Study:

• To evaluate the discriminatory power of social media posts in predicting 10-year ASCVD risk.
• To compare the predictive ability of social media language with traditional pooled cohort risk equations (PCEs).

Main Methods:

• 181 patients provided access to Facebook posts and electronic medical records (EMRs).
• A machine-learning model analyzed Facebook posts (up to 5 years prior) to predict 10-year ASCVD risk.
• The Linguistic Inquiry and Word Count (LIWC) tool assessed language patterns and their association with ASCVD risk.

Main Results:

• The machine-learning model achieved AUC values ranging from 0.57 to 0.78 across different ASCVD risk categories.
• The model distinguished between low (<10%) and high (>10%) ASCVD risk with an AUC of 0.69.
• Higher ASCVD risk scores correlated with increased use of words associated with sadness (r=0.32).

Conclusions:

• Social media language offers valuable insights into an individual's ASCVD risk.
• Analysis of social media content can inform novel approaches to ASCVD risk modification.