This study combined individual data from four cohort studies: the Atherosclerosis Risk in Communities (ARIC) Study [14], the Cardiovascular Health Study (CHS) [15], the Multi-Ethnic Study of Atherosclerosis (MESA) Study [16], and the Health, Aging and Body Composition (HABC) Study [17]. Subjects were excluded from the study if they had abnormal ECG results, clinical MI, coronary artery bypass graft, or coronary angioplasty history at baseline, missing data in ECG or FG, or did not attend any follow-up visits.

ARIC is a community-based prospective cohort study that was started in 1987 to examine the risk factors and evolution of CVD. At the baseline visit (1987–1989), a total of 15,792 individuals aged between 45 and 64 from four US communities were enrolled. Follow-up visits were conducted every three years: 1990–1992 (visit 2), 1993–1995 (visit 3), 1996–1998 (visit 4), and visit 5 was conducted from 2011–2013. Data from the baseline period (1987–1989) through 2013 (visit 5), were used in the present investigation. After exclusion, data from a total number of 13,137 individuals were enrolled for analysis.

CHS is a cohort study designed to assess the risk factors of CVD in the elderly. The study recruited 5888 individuals aged 65 or older from four U.S. communities (Maryland, North Carolina, Pennsylvania, and California), and performed annual clinical examinations between 1989 and 1999, followed by semi-annual phone calls. The present study included data obtained from baseline (1989–1990 for the original cohort and 1992–1993 for the additional recruitment) through 1999, while in-person ECG examinations were conducted.

MESA is a prospective, population-based observational cohort study that included 6814 subjects from four different racial/ethnic groups (White, Black, Hispanic, and Asian) and 6 medical centers in the U.S., aged 45 to 84, and free of CVD at the time of enrollment (2000–2002). Follow-up visits were conducted every two years, and ECG examinations were conducted during baseline visits and visit 5 (2010–2011). Data obtained from the first 5 exams (2000–2011) was used for the present study.

HABC is a longitudinal cohort study initiated in 1997–1998. The study enrolled 3075 men and women aged 70–79 at baseline, with 45% of the women and 33% of the men identified as African-American. All participants were randomly recruited as a sample of Medicare beneficiaries from two clinical centers in Pittsburgh, PA, and Memphis, TN. We included data from the first four annual visits (1997–2001) and ECG examinations were conducted in baseline and visit four.

We selected these cohorts because they are all prospective cohort studies designed to investigate the incidence and risk factors of CVD with repeated FG measurement and available ECG exams to determine SMI status.

We excluded the following participants: (1) subjects with missing or incomplete ECGs; (2) subjects with missing data on FG at all visits; (3) subjects with CVD history at enrollment which were defined as the presence of ECG evidence of MI, a self‐reported history of physician‐diagnosed MI, coronary artery bypass surgery, and coronary angioplasty. After all exclusions, the final analysis cohorts consisted of 13,137 ARIC study participants, 4850 CHS participants, 4271 MESA participants, and 2259 HABC participants.

All participants provided written informed consent at enrollment, and all four studies were approved by the institutional review boards at corresponding participating institutions. The institutional review board at USF deemed this study non-human subjects research.

Each participant had a standard 12-lead resting ECG taken at the study examination, at least 1 h after consuming any coffee or smoking, using MAC PC ECG equipment (Marquette Electronics, Milwaukee, WI). Two medical professionals who had undergone training in interpreting ECGs following the Minnesota code (MC) and were blinded to the outcome status read all ECGs. A senior practitioner, not involved in the initial review, made the final assessments regarding any discrepancies between the two professionals.

SMI was defined as ECG evidence of new MI during follow‐up visits that were not present at enrollment and in the absence of recorded clinical MI (including diagnosis and/or hospitalization for an MI). The newly introduced MC ECG classifications were used to characterize ECG evidence of MI as a major Q/QS wave abnormality (MC 1.1 or 1.2) or minor Q/QS wave abnormality (MC 1.3) alongside a major STT abnormality (MC 4.1, MC 4.2, MC 5.1, or MC 5.2) [18].

Our primary exposure was longitudinal measures of FG. Covariates included age, gender, race, smoking status (current, former, never), body mass index (BMI), blood pressure, low-density lipoprotein (LDL), serum creatinine, the use of lipid-lowering medication, antihypertensive medication, antidiabetic medication, and insulin treatment. All independent variables, except for age, gender, race, and smoking were measured at each follow-up visit over the length of the studies. Additional file 1: Fig. S1 illustrates the time points at which FG and ECG measurements were taken. For the majority of the FG measurements, they align with the timing of ECG exams, with one exception noted in the MESA study. In the MESA study, ECG assessments were only conducted at the baseline and Visit 5. Missing covariate data were imputed using the value from the nearest visit for the same individual.

The primary outcome of our study was the incidence of SMI which was interval censored between two sequential visits. The exposure and covariates were time-dependent variables, therefore, we used Cox proportional-hazards models with time-dependent variables to estimate the association between time-varying FG and SMI, with adjustment for potential confounders, including age, gender, race, study, smoking status, diastolic blood pressure (DBP), LDL, blood creatinine, and BMI (Model 1). In Model 2, we further adjusted for the use of lipid-lowering medication, antihypertensive medication, antidiabetic medication, and insulin treatment as covariates to see if adjustment for treatments had any further impact on the association between FG and SMI. We applied the penalized splines for FG in the Cox regression model to test for a potential non-linear association between FG and the incidence of SMI. Subgroup analyses were conducted by gender and by age stratified at 65 years at baseline. We also performed sensitivity analyses by examining the association taking the average FG over time and the intra-individual variability of FG over time of each individual as the exposure. The variability of FG was calculated as the coefficient of variation (CV, sd/mean across visits). All the analysis work was conducted in R with Survival package version 3.4–0, and rms package version 6.3–0.