We performed a meta-analysis of individual data from three studies: •

The first study included 81 consecutive nondiabetic patients aged <55 years who presented with acute myocardial infarction (AMI) in a tertiary hospital (16). This prospective study had a median follow-up duration of 4.1 years. Only 75 patients with first AMI were considered for the present analysis.

We analyzed eight genetic variants previously identified to be associated with CAD but not with the classical cardiovascular risk factors (total cholesterol, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, blood pressure, smoking, or diabetes mellitus), namely, single-nucleotide polymorphisms (SNPs) rs17464857 in MIA3, rs6725887 in WDR12, rs9818870 in MRAS, rs10455872 in SLC22A3-LPAL2-LPA, rs12526453 in PHACTR1, rs1333049 in CDKN2B-AS1, rs501120 in CXCL12, and rs9982601 in KCNE2-SCL5A3.

We also incorporated four variants of the ALOX5AP haplotype B, which has been reported to be associated with CAD in different populations (19–21). This haplotype consisted of rs10507391-A, rs9315050-A, rs17222842-G, and rs17216473-A. A GRS including these 12 genetic variants (eight SNPs and four haplotype variants) was defined as the sum of the number of risk alleles/haplotype across all genetic variants after weighting each one by its estimated effect size in the CARDIoGRAMplusC4D [Coronary Artery Disease Genome-Wide Replication and Meta-Analysis Plus the Coronary Artery Disease (C4D) Genetics] consortium (21). In addition, a weight of 0.131 [odds ratio (OR) = 1.14] was assigned to the ALOX5AP haplotype B.

DNA extraction was performed using standard techniques. The 12 genetic variants were determined, and the GRS was calculated using the CARDIO inCode Score commercial platform (GENinCode Plc, Oxford, UK). This GRS had been previously developed and validated in several populations in primary prevention (3, 6). Individual and genotype call rates and Hardy–Weinberg equilibria were also evaluated.

All the participants were followed up every 6 months at the outpatient clinic in the tertiary hospital where the patients for the first two studies were recruited. In the REGICOR Study, follow-up was conducted through data linkage with population-based myocardial infarction records, official mortality registries, and telephone interviews.

The primary outcome, defined as a recurrent cardiovascular event, was a composite of cardiovascular mortality and recurrent ACS, including both AMI and unstable angina. Nonculprit lesions that were revascularized were not considered recurrences. All outcomes were reviewed by two cardiologists in the first two studies and by an event committee in the REGICOR study.

The following variables were also collected: age; sex; maximum Killip class during index hospitalization; family history of CAD; personal history of hypertension, dyslipidemia, diabetes, and smoking; treatment with beta-blockers; angiotensin-converting enzyme inhibitors; and revascularization.

The quantitative variables were expressed as means and standard deviations, and the qualitative variables were expressed as counts and percentages. Student’s t-test and ANOVA were used to compare the quantitative variables, and the chi-squared test was used to compare the qualitative variables between the two groups.

In the multivariable analyses, Cox survival regression models were used in the first two studies and logistic regression was used in the case–control study. A clinical regression model, including age, sex, maximum Killip class during hospital stay, family history of CAD, and personal history of diabetes, hypertension, dyslipidemia, and smoking, was initially defined. The final model included those factors associated with the outcome in any of the studies. The GRS was added to this widely used clinical model. The added predictive value of the GRS was assessed by the improvement of the discriminative capacity [change in the C-statistic or area under the curve (AUC)] and reclassification (integrated discrimination improvement index and continuous net reclassification improvement).

We used the inverse variance method to weight the results of the three studies and the fixed effects model to estimate the summary effect size of the association and the predictive improvement metrics. We also assessed the heterogeneity of the results across studies.

A p-value of <0.05 was considered statistically significant. All the statistical analyses were performed using R version 3.3.2.

Participant and genotype call rates were >99%, and all the genetic variants analyzed followed the Hardy–Weinberg equilibrium. Table 1 shows the main characteristics of the participants in the three studies. Table 2 shows the differences between patients with and without an event during follow-up. The group of participants with events had a higher Killip class and standardized GRS. The association between a family history of CAD and a higher risk of events was inconsistent across studies, with two studies showing a lower proportion of family history among patients with events and the third showing the opposite. No other significant differences were observed.

In the multivariable analyses, the effect size of the association between the standardized GRS and the risk of events was hazard ratio (HR) = 1.43 [95% confidence interval (CI): 0.92–2.22], HR = 1.16 (95% CI: 0.92–1.46), and OR = 1.31 (95% CI: 0.94–1.81), respectively.

Table 3 shows the predictive capacity of the clinical model, which included age, sex, Killip class, family history of CAD, and personal history of hypertension, along with the added predictive value of the GRS across studies. The inclusion of the GRS did not improve the discriminative capacity of the clinical model. In two studies, we observed a correct reclassification of the patients after including the GRS in the model.

Table 3 and Figure 1 show the results of the meta-analysis performed. A 24% increase in the risk of the outcomes of interest per standard deviation of the GRS was observed (95% CI: 4%–47%), without heterogeneity between studies (heterogeneity index −I² = 0) (Figure 1A). The inclusion of the GRS did not increase the predictive capacity of the clinical model (change in C-statistic/AUC: 0.009; 95% CI: −0.007 to 0.025) (Figure 1B and Table 3) but improved its reclassification (continuous net reclassification index: 0.29; 95% CI: 0.08–0.51), without heterogeneity between studies (Figure 1C and Table 3).