Chest radiograph deep-learning model for cardiac event risk estimation

1. In this retrospective study, the chest X-ray cardiovascular disease (CXR CVD)-risk model performed similarly to the standard atherosclerotic (AS) CVD risk score at predicting 10-year risk for major adverse cardiovascular events (MACE).

2. Among patients with an unknown ASCVD risk score, those deemed statin-eligible by the model were nearly twice as likely to have incident MACE.

Evidence Rating Level: 2 (Good)

Study Rundown: Current guidelines recommend using a regression model based on nine variables to estimate a person’s 10-year risk for developing ASCVD. While this score is useful for prognostication and shared decision-making at the individual level, there is limited utility for population-based screening due to the high number of required inputs. This study tested the ability of CSR CVD-Risk, a deep learning model that only analyzes routine CXRs, to estimate patients’ risk of MACE within the next decade. It was found that CXR CVD-Risk and the traditional ASCVD risk score were concordant for statin eligibility in nearly three-quarters of patients who had complete data.  Approximately half of the discordant cases were up-classified by the new model and half by the ASCVD risk score. The ASCVD risk score methodology superficially appeared to have slightly better discrimination than CXR CVD-Risk at predicting incident MACE, although this difference was not statistically significant. Notably, nearly half of patients whose ASCVD risk scores could not be calculated because of missing inputs were identified by CXR CVD-Risk as statin-eligible. These statin-eligible patients were found to be nearly twice as likely as their statin-ineligible counterparts to suffer a MACE within 10 years. Some limitations of this study include a retrospective design, a racially homogenous patient cohort, and a narrow age window.

Click here to read the study in AIM

In-Depth [retrospective cohort study]: In this retrospective study, a deep-learning model based exclusively on chest radiographs was first trained on data from the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial and then tested on a separate cohort of outpatients aged 50 to 75 years who were imaged within the Mass General Brigham health care system. The major exclusion criteria were prevalent diabetes, current use of a statin, and history of MACE, or coronary/peripheral artery disease. The study design consisted of two cohorts: one with complete data for calculating ASCVD score (2,132 patients) and one with unknown ASCVD risk due to missing data (8,869 patients). Among patients with known ASCVD scores, the overall MACE incidence was 4.8% (102 of 2132) over a median follow-up of 10.4 years. CXR CVD-Risk identified 37.0% (787 of 2132) of this cohort as statin-eligible, whereas ASCVD score identified 36.8% (782 of 2132). According to CXR CVD-Risk, 7.6% of those who were statin-eligible developed incident MACE, versus 3.1% of those who were ineligible (Hazard Ratio [HR], 2.77; 95% Confidence Interval [CI], 1.87 to 4.11). According to ASCVD scoring, 8.7% of statin-eligible patients developed incident MACE compared to 2.5% (HR, 3.88; 95% CI, 2.57 to 5.85). Among those with unknown ASCVD risk, 47.9% were statin-eligible according to CXR CVD-Risk. These findings suggest that CXR CVD-Risk may represent a cost-effective supplement to existing strategies for cardiovascular disease prevention.

Image: PD

©2024 2 Minute Medicine, Inc. All rights reserved. No works may be reproduced without expressed written consent from 2 Minute Medicine, Inc. Inquire about licensing here. No article should be construed as medical advice and is not intended as such by the authors or by 2 Minute Medicine, Inc.