Machine Learning Outperforms ACC/AHA CVD Risk Calculator in MESA
Problem
The standard ACC/AHA CVD risk calculator underestimates risk in high-risk individuals and overestimates it in low-risk ones, causing missed therapy or unnecessary statins.
Why it matters
Misjudged cardiovascular risk can mean preventable heart attacks or unneeded lifelong statin treatment.
What the AI did
An SVM-based machine-learning risk calculator was trained on 13-year MESA follow-up data (6459 participants), given identical inputs as the ACC/AHA calculator, and validated on the external FLEMENGHO study.
Results
Outperformed ACC/AHA calculator per title; performance metrics not stated in abstract
Impact
Title claims ML outperforms the guideline risk calculator, with external validation in FLEMENGHO.
Record
| Venue | Journal of the American Heart Association |
|---|---|
| Published | 2018 |
| Work type | article |
| Authors | Ioannis A. Kakadiaris, Michalis Vrigkas, Albert A. Yen, Tatiana Kuznetsova, Matthew Jay Budoff, Morteza Naghavi |
| Citations | 226 (OpenAlex, at import time) |
| References | 36 |
| Model or method | Support Vector Machines (SVM) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1161/jaha.118.009476 |
| Open access | gold |
| Keywords | machine learning, cardiovascular risk, SVM, MESA, risk calculator, statin therapy |
Importance rating
major — millions of lives, an industry or a field
Abstract quantifies guideline misclassification: 23.8% of hard CVD events occurred in those not recommended statins, while 46% were recommended statins, implying widespread missed prevention and unnecessary therapy.