What is the clinical evidence from this study?

Study design: Cohort. Population: Atherosclerotic cardiovascular disease (n=262923). Intervention: Machine learning models (Gradient Boosting Machine) vs. Pooled cohort equations (PCE). Primary outcome: 5-year ASCVD risk prediction (Area Under the ROC Curve) (95% CI 0.825-0.846).

September 23, 2020npj Digital Medicine105 citationsOpen Access

Machine learning and atherosclerotic cardiovascular disease risk prediction in a multi-ethnic population

AWAndrew WardLehigh University ASAshish SarrajuPreventive Cardiology SCSukyung ChungStanford Medicine

Key Result

A gradient boosting machine model improved 5-year ASCVD risk prediction across a diverse cohort including PCE-ineligible patients (AUC 0.835) compared to the pooled cohort equations (AUC 0.775).

Study Design

Type

Cohort (n=262,923)

Structured PICO

Do machine learning models improve 5-year ASCVD risk prediction compared to the Pooled Cohort Equations in a multi-ethnic population?

Population

Multi-ethnic population including Asian and Hispanic subgroups, comprising both PCE-eligible and PCE-ineligible patients (4,309 total ASCVD events).

Intervention

Machine learning models (lasso penalty, random forest, gradient boosting machine, extreme gradient boosting) trained on electronic health record (EHR) variables

Comparator

Pooled Cohort Equations (PCE)

Outcome

5-year ASCVD risk prediction performance measured by area under the receiver-operating characteristic curve (AUC)

Machine learning models, particularly Gradient Boosting Machines, improve 5-year ASCVD risk prediction compared to standard Pooled Cohort Equations and can be applied to a broader group of patients, including those ineligible for PCE.

Main Result

Absolute Event Rate: 0.835% vs 0.775%

Limitations

Study population from a multipayer private health system in Northern California may not be generalizable across the United States
Used 5-year ASCVD risk rather than 10-year risk as typically reported by PCE
Did not perform nested cross-validation, which could potentially minimize bias in model comparison
Did not disaggregate Asian and Hispanic patients, which may mask heterogeneity within these subgroups
Race/ethnicity data were largely self-reported or inferred, which may contribute to misclassification

Abstract

lasso penalty, random forest, gradient boosting machine (GBM), extreme gradient boosting] and determined 5-year ASCVD risk prediction, including with and without incorporation of additional EHR variables, and in Asian and Hispanic subgroups. A total of 4309 patients had ASCVD events, with 2077 in PCE-ineligible patients. GBM performance in the full cohort, including PCE-ineligible patients (area under receiver-operating characteristic curve (AUC) 0.835, 95% confidence interval (CI): 0.825-0.846), was significantly better than that of the PCE in the PCE-eligible cohort (AUC 0.775, 95% CI: 0.755-0.794). Among patients aged 40-79, GBM performed similarly before (AUC 0.784, 95% CI: 0.759-0.808) and after (AUC 0.790, 95% CI: 0.765-0.814) incorporating additional EHR data. Overall, ML models achieved comparable or improved performance compared to the PCE while allowing risk discrimination in a larger group of patients including PCE-ineligible patients. EHR-trained ML models may help bridge important gaps in ASCVD risk prediction.

Ask AI

Helpful

Bookmark

View Full Paper