Predicting cardiovascular death in overweight/obese people with prediabetes using machine learning-a proof-of-concept...

Area of Science:

• Cardiology
• Endocrinology
• Public Health

Background:

• Individuals with prediabetes have a higher risk of cardiovascular (CV) complications and premature mortality.
• Current risk stratification tools are not specifically designed for the prediabetes population.
• There is a need for targeted risk assessment in prediabetes patients with established CV disease.

Purpose of the Study:

• To develop a simple, explainable model to predict fatal cardiovascular outcomes in individuals with prediabetes.
• To assess the performance of this prediabetes-specific model against existing general cardiovascular risk calculators.

Main Methods:

• A binary logistic regression model was trained using data from 5636 participants aged ≥45 years with prediabetes, established CV disease, and overweight/obesity.
• The model utilized seven demographic and clinical variables.
• Model performance was evaluated using receiver operating characteristic (ROC) area under the curve (AUC), precision, and recall, with threefold cross-validation.

Main Results:

• The study included 5636 participants, with 182 (3.2%) experiencing CV death over a mean follow-up of 3.3 years.
• The developed prediabetes-specific model achieved an acceptable discrimination (ROC AUC) of 0.730.
• Existing benchmark models (SCORE2, SCORE2-Diabetes) showed significantly lower performance for predicting CV death (ROC AUC: 0.630-0.643) and major adverse cardiovascular events (MACE) in this cohort (ROC AUC: 0.596-0.603) compared to their original populations.

Conclusions:

• A proof-of-concept, prediabetes-specific model for predicting cardiovascular death has been developed.
• This model demonstrates potential for improved early prevention in high-risk prediabetes patients.
• External validation is essential to confirm the model's generalizability before real-world implementation.