Random Survival Forest Machine Learning for the Prediction of Cardiovascular Events Among Patients With a Measured...

Area of Science:

• Cardiology
• Biostatistics
• Genomics

Background:

• Established cardiovascular risk models may not accurately assess patients with elevated lipoprotein(a) [Lp(a)] levels.
• Lipoprotein(a) is a causal risk factor for atherosclerotic cardiovascular disease.

Purpose of the Study:

• To develop and evaluate a novel risk prediction model for cardiovascular events.
• To compare the performance of a random survival forest model against traditional risk factor models in patients with measured Lp(a) levels.

Main Methods:

• A model development study using data from Nashville Biosciences Lp(a) dataset.
• Included patients with measured Lp(a) between 1989-2022 and at least one year of prior EHR data.
• Compared a random survival forest model with Cox proportional hazards models using traditional risk factors, evaluating discrimination with Harrell's C-index.

Main Results:

• The study included 4369 patients (49.5% female, mean age 51 years, mean Lp(a) 33.6 mg/dL).
• The random survival forest model achieved a higher C-index (0.82) compared to primary prevention (0.69) and secondary prevention (0.80) models.
• These findings were consistent across primary prevention populations and competing risk analyses.

Conclusions:

• A random survival forest model demonstrates superior performance in predicting cardiovascular events for individuals with measured Lp(a) levels.
• This suggests that incorporating Lp(a) into advanced models can enhance cardiovascular risk stratification.