# Automated Abdominal Aortic Calcification Scores and Atherosclerotic Cardiovascular Disease in the UK Biobank Imaging Study

## TL;DR

This study shows that a machine learning algorithm can predict heart disease risk using routine bone density scans.

## Contribution

Validated a machine learning AAC score as a novel screening tool for atherosclerotic cardiovascular disease.

## Key findings

- High ML-AAC24 scores correlated with a 2.87 times higher risk of incident ASCVD compared to low scores.
- The ML-AAC24 remained predictive even after adjusting for traditional cardiovascular risk factors.
- Consistent risk patterns were observed for coronary artery disease, myocardial infarction, and stroke.

## Abstract

Abdominal aortic calcification (AAC) is a subclinical measure of atherosclerotic cardiovascular disease (ASCVD). AAC can be captured on lateral spine images obtained from bone density machines during routine osteoporosis screening. Identifying individuals with AAC provides a new opportunity to prevent disease progression.

The aim of the study was to externally validate a machine learning-derived AAC 24-point algorithm (ML-AAC24) with incident ASCVD.

Middle-aged individuals from the UK Biobank Imaging Study with lateral spine images, obtained via dual-energy x-ray absorptiometry, were included. ML-AAC24 scores were grouped as low (<2), moderate (2 to <6), and high (≥6). Linked health records were used to identify ASCVD-associated events, including hospitalizations and death.

Among 53,611 participants (52% female; mean age 65 years), 78.2% had low, 16.4% had moderate, and 5.4% had high ML-AAC24. After excluding people with prevalent ASCVD or missing data, 1,163 (2.3%) of 50,923 people had an incident ASCVD event over a median follow-up of 4.1 [3.0-5.5] years. In age- and sex-adjusted analysis, compared to those with low ML-AAC24, those with moderate (HR: 1.80 [95% CI: 1.57-2.08]) and high ML-AAC24 (HR: 2.87 [95% CI: 2.39-3.44]) had a higher HR for incident ASCVD. Results remained comparable after adjustment for established ASCVD risk factors. Consistent patterns were observed when considering incident coronary artery disease, myocardial infarction, and stroke.

Assessing ML-AAC24 on lateral spine images offers a new and promising screening method to identify people with higher risk of incident ASVD events.