Selection of 51 predictors from 13,782 candidate multimodal features using machine learning improves coronary artery disease prediction

Saaket Agrawal; Marcus D R Klarqvist; Connor Emdin; Aniruddh P Patel; Manish D Paranjpe; Patrick T Ellinor; Anthony Philippakis; Kenney Ng; Puneet Batra; Amit V Khera

doi:10.1016/j.patter.2021.100364

Selection of 51 predictors from 13,782 candidate multimodal features using machine learning improves coronary artery disease prediction

Patterns (N Y). 2021 Oct 4;2(12):100364. doi: 10.1016/j.patter.2021.100364. eCollection 2021 Dec 10.

Authors

Saaket Agrawal^{1

2

3}, Marcus D R Klarqvist⁴, Connor Emdin^{1

2

3}, Aniruddh P Patel^{1

2

3}, Manish D Paranjpe^{1

2

3}, Patrick T Ellinor^{1

2

3}, Anthony Philippakis⁴, Kenney Ng⁵, Puneet Batra⁴, Amit V Khera^{1

2

3}

Affiliations

¹ Cardiovascular Disease Initiative, Broad Institute of MIT and Harvard, Cambridge, MA, USA.
² Center for Genomic Medicine, Department of Medicine, Massachusetts General Hospital, 185 Cambridge Street, Simches Research Building | CPZN 6.256, Boston, MA 02114, USA.
³ Department of Medicine, Harvard Medical School, Boston, MA, USA.
⁴ Data Sciences Platform, Broad Institute of MIT and Harvard, Cambridge, MA, USA.
⁵ Center for Computational Health, IBM Research, Cambridge, MA, USA.

Abstract

Current cardiovascular risk assessment tools use a small number of predictors. Here, we study how machine learning might: (1) enable principled selection from a large multimodal set of candidate variables and (2) improve prediction of incident coronary artery disease (CAD) events. An elastic net-based Cox model (ML4H_EN-COX) trained and evaluated in 173,274 UK Biobank participants selected 51 predictors from 13,782 candidates. Beyond most traditional risk factors, ML4H_EN-COX selected a polygenic score, waist circumference, socioeconomic deprivation, and several hematologic indices. A more than 30-fold gradient in 10-year risk estimates was noted across ML4H_EN-COX quintiles, ranging from 0.25% to 7.8%. ML4H_EN-COX improved discrimination of incident CAD (C-statistic = 0.796) compared with the Framingham risk score, pooled cohort equations, and QRISK3 (range 0.754-0.761). This approach to variable selection and model assessment is readily generalizable to a broad range of complex datasets and disease endpoints.

Keywords: Framingham risk score; QRISK3; coronary artery disease; machine learning; pooled cohort equations; risk prediction.

Abstract

Grants and funding