Statistical Analysis

We performed a pooled analysis of the SKS and the C-PROBE cohorts. We tabulated baseline participant characteristics for each cohort and overall. We compared baseline characteristics of the study population across quartiles of each biomarker. A multivariable linear regression analysis was performed to test the associations of these baseline characteristics with GDF-15, gal-3, and sST2 concentrations. We used unadjusted Spearman correlations and scatterplots to determine correlations among the cardiac biomarkers, eGFR, and urine albumin-creatinine ratio (UACR). Kaplan-Meier curves were generated to examine the probabilities of the primary and secondary outcomes in participants above and below the median level of GDF-15, gal-3, and sST2. Unadjusted incidence rates for each outcome across quartiles of each biomarker were calculated as the number of events divided by person-years at risk. We assessed the functional forms of associations of GDF-15, gal-3, and sST-2 levels with each outcome (mortality, heart failure, and atherosclerotic CVD) using cubic spline models. There was no evidence of departure from linearity; therefore, we assessed these markers as continuous linear exposures. Cox proportional hazards regression were used to test the association of each biomarker level (modeled continuously per standard deviation increase) at baseline and time until first occurrence of mortality, heart failure, or atherosclerotic CVD. Covariates for adjustment were selected a priori based on previous literature regarding potential confounders as well as examining the covariates that differed greatly across quartiles of each biomarker. Nested models were used to progressively explore the confounding effects of demographics, known cardiovascular risk factors, and kidney function. The first model adjusted for baseline age, gender, race, and study cohort. The second model additionally adjusted for prevalent cardiovascular disease, current smoking status, diabetes, eGFR and urine albumin-creatinine ratio. Participants were censored at first event, death, or end of study period (July 31, 2013 for SKS and July 29, 2016 for C-PROBE). Two-sided probability values ≤ 0.05 were considered statistically significant. All analyses were performed using Stata version 13.1 (College Station, TX) and IBM SPSS Statistics for Windows version 23.0 (Armonk, NY).