Insight
- A novel score derived from a machine learning model using electronic health data demonstrated high accuracy in predicting coronary artery disease (CHD) outcomes across the disease spectrum.
Why this matters
- The results can help identify individuals with undiagnosed disease and support management in individuals with diagnosed disease.
Study design
- Cohort study for model development and validation using electronic medical records of 95.935 adults aged ≥ 40 years from the US-_ BioMe Biobank and the UK Biobank
- In both cohorts, 14 % of participants had a diagnosis of CHD.
- A machine learning model was applied to calculate the probabilities of CHD outcomes occurring in the form of an in-silico CAD [ISCAD] score.
- Main results: stenosis of the coronary arteries, obstructive-_ CHD, multi-vascular CHD, death of any cause, CHD sequelae (recurrent myocardial infarction, arrhythmia, heart failure after CHD diagnosis)
- Funding: NIH
Key results
- In the bioMe training/validation groups and holdout method groups, the model predicted the CHD with an area below the curve of 0,95 (94 % Sensitivity; 82 % specificity) and 0.93 (90 % Sensitivity; 88 % Specificity).
- In the external test group from the UK-_ Biobank predicted the model the CHC with an area below the curve of 0,91 (84 % Sensitivity; 83 % Specificity).
- The ISCAD score derived from the model included known CHD risk factors, pooled cohort equations and polygenic risk scores and ranged from 0 (lowest probability of outcome) to 1 (highest probability).
- The risks of occurrence of the following results increased with the ISCAD score quartile: coronary artery stenosis, obstructive CHD, multi-vessel concerning CHD and stenosis of the large coronary arteries.
- The risk of death from any cause increased with the ISCAD score decile.The aHR was 1.0 for participants in decile 1 (prevalence of 0.2 %), 11 for participants in decile 6 (prevalence of 4.2 %) and 56 for participants in decile 10 (prevalence of 11 %).
- The risk of recurrent myocardial infarction showed a similar increase.
- Almost half (46 %) of participants with undiagnosed CHD and an ISCAD score of ≥ 0.9 showed clinical signs of CHC according to Guidelines.
Expert comment
- In a Comment wrote Puneet Batra, Ph.D., Broad Institute of MIT &— Harvard and Verve Therapeutics, and Amit V.Khera, M.D., M.Sc., Brigham and Women’s Hospital: «Placing individuals on a spectrum of coronary artery disease that takes into account a variety of factors could allow for tailored interventions that better address the risks of coronary artery disease, unlike current scoring systems such as SYNTAX, which focus solely on coronary artery disease.»
Restrictions
- The CHD participants were identified using diagnostic codes, which may have led to misclassifications.
- Only a subset of participants underwent coronary angiography.
- Data for cause-specific mortality were not available.
- The generalizability to other populations is uncertain.
This full text is unfortunately reserved for medical professionals
You have reached the maximum number of articles for unregistered visitors
Comments are closed.