Polygenic risk score modestly improves risk prediction for incident CAD
Predictive Accuracy of a Polygenic Risk Score–Enhanced Prediction Model vs a Clinical Risk Score for Coronary Artery Disease
Introduction and methods
Probability of developing CV outcomes in asymptomatic individuals can be estimated by risk prediction models . ACC/AHA risk assessment guidelines recommend lipid-lowering treatment for those with 10-year absolute risk of atherosclerotic CVD greater than 7.5% based on pooled cohort equations . In the last years, genetic variants/single-nucleotide polymorphisms (SNPs) have been identified that are associated with coronary artery disease (CAD) . Individual genetic variants only make a small contribution to disease risk, but genetic or polygenetic risk scores based on multiple genetic variants could potentially provide an added predictive value for CAD . However, the added value of polygenic risk score in addition to well-established and validated risk prediction models is currently unknown. This study evaluated the value of polygenic risk score in CAD risk prediction over pooled cohort equations.
This study used data from the UK biobank. The study population was divided into two cohorts. First, a case-control sample of 15 947 prevalent CAD cases and equal number of randomly selected age and sex frequency-matched controls was used for the optimization of parameters of the polygenic risk score calculation in order to acquire a polygenic risk score with maximal predictive ability. The polygenic risk score for CAD was derived as a weighted sum of risk alleles, using data from the largest genome-wide association study on CAD (CARDIoGRAMplusC4DREF ) that excluded participants from the present study. Second, an independent cohort of 352 660 individuals (mean age 55.9 years, 58.2% were women, median follow-up was 8 years, number of incident CAD events was 6272) was used as a cohort testing set to evaluate the predictive value of the polygenic risk score, pooled cohort equations, and a combination of both for incident CAD. Harrell’s C statistic and its 95%CI was used to assess the discrimination of each model [6-8]. Values range from 0.5 (no discrimination) to a theoretical maximum of 1.0. The primary endpoint of this study was CAD, defined as myocardial infarction and its related sequelae.
- Discrimination of CAD , calculated by C-statistic, was 0.61 (95%CI 0.60-0.62) for polygenic risk score, 0.76 (95%CI 0.75-0.77) for pooled cohort equations, and 0.78 (95%CI 0.77-0.79) for both polygenic risk score and pooled cohort equations combined.
- The associated change in C statistic between the pooled cohort equations and both polygenic risk score and pooled cohort equations combined was 0.02 (95%CI 0.01-0.03).
- When polygenic risk score was added to the pooled cohort equations model, predicted risk changed by <1% for 79.5% of participants, and changed by ≥5% for 1.1% of participants. When using a risk threshold of 7.5%, 526 of 6272 cases (8.4%) were correctly reclassified to the higher category and 250 of 6272 cases (4.0%) were incorrectly reclassified to the lower-risk category. For the noncases, 5284 of 346388 (1.5%) were correctly reclassified to the lower-risk category and 6723 of 346388 (1.9%) were incorrectly reclassified to the higher-risk category.
- The net reclassification improvement (NRI) was 4.4% (95%CI 3.5%-5.3%) for cases and -0.4% (95%CI -0.5% to -0.4%) for noncases. The overall NRI in the full population was 4.0% (95%CI 3.1%-4.9%).
- The increase in risk difference between cases and noncases was 0.0006 (95%CI 0.006-0.007), according to the integrated discrimination improvement (IDI) metric.
This study found a modest improvement in the predictive accuracy for incident CAD when a polygenic risk score for CAD was added to pooled cohort equations. Risk stratification improved in only a small portion of individuals.