Markers of imminent myocardial infarction

Abstract

Myocardial infarction is a leading cause of death globally but is notoriously difficult to predict. We aimed to identify biomarkers of an imminent first myocardial infarction and design relevant prediction models. Here, we constructed a new case-cohort consortium of 2,018 persons without prior cardiovascular disease from six European cohorts, among whom 420 developed a first myocardial infarction within 6 months after the baseline blood draw. We analyzed 817 proteins and 1,025 metabolites in biobanked blood and 16 clinical variables. Forty-eight proteins, 43 metabolites, age, sex and systolic blood pressure were associated with the risk of an imminent first myocardial infarction. Brain natriuretic peptide was most consistently associated with the risk of imminent myocardial infarction. Using clinically readily available variables, we devised a prediction model for an imminent first myocardial infarction for clinical use in the general population, with good discriminatory performance and potential for motivating primary prevention efforts.

Fig. 1. Derivation of the sample representing 169,053 individuals without previous cardiovascular disease from six European population-based cohorts.

The distribution of MIMI participants across Europe is shown, with the participating countries and cohort centers indicated. Cases (n = 420) were initially sampled, and center-specific strata based on sex and median age were constructed. From each cohort center, up to four subcohort representatives were drawn for each case from the same stratum. A subcohort (n = 1,598) weighted to represent the total cohort (N = 169,053) based on the number of individuals in the age and sex strata in the total cohort was thus assembled. NA, not applicable.

Fig. 2. Associations of proteins, metabolites and clinical variables with IMI risk.

The associations of 817 proteins, 1,025 metabolites and 16 clinical variables with the risk of a first myocardial infarction within 6 months in the full MIMI study, adjusted for technical covariates, are shown by biomarker category (clinical, metabolite or protein). HR relates to a doubling of the concentration of proteins and metabolites and a one-unit higher level of clinical biomarkers on their original scale (for example, years, mmol l⁻¹). The top 25 biomarkers that passed external validation and ranked on how many internal validation splits the biomarker passed the replication criteria in the model adjusted for technical covariates in addition to the external validation are highlighted. ^aIL-6 and ^bKIM1 were measured on multiple Olink panels and tested in separate statistical tests. n = 420 cases and 1,598 noncases.

Fig. 3. Top variables associated with IMI risk.

The top 25 biomarkers that passed external validation and ranked on how many internal validation splits the biomarker passed the replication criteria in the model adjusted for technical covariates in addition to the external validation are shown. Each predictor is represented by two rows, with the discovery result (blue) presented first and the validation result presented second (red). The results are sorted by predictor type (clinical, metabolite or protein) and effect size from the combined analysis of the discovery and validation samples. P value was calculated based on a 2 d.f. Wald test for metabolites analyzed using the missing indicator method (biomarker and missing indicator) and a 1 d.f. Wald test otherwise (biomarker only), two-sided in both cases. The 95% CI of the point estimate (log(HR)) was calculated for the biomarker only and might include 1 even if P < 0.05 from the 2 d.f. (biomarker + indicator) Wald test. ^aIL-6 and ^bKIM1 values were determined from multiple Olink panels and tested in separate statistical tests. n = 296 cases and 1,121 noncases in the discovery sample; n = 124 cases and 477 noncases in the validation sample.

Fig. 4. Nomogram of the model for the clinical prediction of an IMI.

A nomogram for predicting IMI risk based on the final clinical model is shown. Each variable value contributes points (ruler at the top) that are summed up and translated to the predicted risk of a myocardial infarction within 6 months (bottom two rulers). Equation, β coefficients, 6-month survival and mean variable values are provided in Supplementary Table 4. A worked example is shown in Extended Data Fig. 8. The model is also presented as an interactive web application at miscore.org.

Extended Data Fig. 1. Causal assumptions for developing bias- minimized models.

Directional acyclic graph (DAG) showing the best-guess relationship between the exposure (BNP) and the outcome (IMI) and other factors influencing this relationship together with the expected direction.

Extended Data Fig. 2. Variables associated with risk of an imminent myocardial infarction, further adjusted for age and sex.

Scatter plot comparing the point estimate (log[HR]) and corresponding 95% confidence interval from models adjusting for technical covariates (x-axis) and models additionally adjusting for age and sex (y-axis) in the full MIMI study. The 95% C.I. is calculated for the biomarker only whereas the p-value in the tables is based on the 2 d.f. Wald test (two-sided) of biomarker and missing indicator (when used), hence the 95% C.I. might overlap with null for a p-value < 0.05. N=420 cases and 1598 non-cases.

Extended Data Fig. 3. Association of BNP with Risk of an Imminent Myocardial Infarction.

Kaplan–Meier graph of unadjusted cumulative hazard of an imminent myocardial infarction (IMI) by fourths (Q) of brain natriuretic peptide (BNP), weighted by sampling weights.

Extended Data Fig. 4. Leave-one-out analyses of the association of BNP with imminent myocardial infarction.

Model with BNP and technical covariates in leave-one-out analyses where one cohort was omitted at time.

Extended Data Fig. 5. Associations of BNP and NT-proBNP with imminent myocardial infarction in the individual cohorts.

Forest plot of the regression results in the model adjusting for technical covariates performed per cohort for all available BNP measurements (BNP and N-terminal pro form). Hazard ratio and corresponding 95% confidence interval is presented. NT-proBNP is measured on both the CVD2 (*) and Metabolism panel (**). Individual cohort sample sizes are given in Supplementary Table 1.

Extended Data Fig. 6. Comparison of associations of biomarkers with risk of IMI within 3 vs 6 months.

Regression estimates from models with time to IMI within 3 vs 6 months as outcomes. MI, myocardial infarction.

Extended Data Fig. 7. Calibration of the prediction model.

a Internal calibration. Cross-validated calibration curves for predicted probabilities of an imminent myocardial infarction from the MIMI model (solid black line) and the SCORE2 model (dashed black line). b External calibration. Calibration curves for predicted probabilities of an imminent myocardial infarction from the original MIMI model (solid black line), the MIMI model recalibrated to the UKBB data (dotted black line), and the original SCORE2 model (dashed black line). The diagonal gray line in both panels is the line of ideal calibration where predicted probabilities match the observed fraction experiencing the event.

Extended Data Fig. 8

Worked example of nomogram use. A 78-year-old (73 points) smoking (13 points) low-educated (10 points) man (23 points) with diabetes (8 points), height 1.71 m (28 points), waist circumference 110 cm (22 points), LDL cholesterol 4.5 mmol/L (21 points) and HDL cholesterol 1.2 mmol/L (39 points) will have a total score of 73 + 13 + 10 + 23 + 8 + 28 + 22 + 21 + 39 = 237 points, corresponding to a 6- month risk of a first myocardial infarction of circa 1.58%. The model is also presented as an interactive web application on miscore.org.