Statistical Analysis

Educational status was modeled as the maximum (at any visit) reported years completed. Smoking was classified as current versus former/never. To account for typical dietary consumption, we averaged consumption of TMAO precursor food groups, including eggs and red meat, from years 7 (1992–1993) and 20 (2005–2006) of CARDIA, as well as a measure of diet quality.17

We natural log‐transformed CRP and triglycerides to normalize data. Estimated glomerular filtration rate (eGFR) (mL/min per 1.73 m²) was calculated from serum creatinine using the 2009 CKD‐EPI (CKD Epidemiology Collaboration) equation.18 Urine albumin‐creatinine ratios (UACRs) were standardized to sex and race and expressed in milligrams per gram of creatinine. We estimated insulin resistance using homeostasis model assessment for insulin resistance (HOMA‐IR).19

Multivariable‐adjusted regression models controlled for 2000–2001 covariate values. Variables were included in the model as continuous, unless noted. A minimally adjusted model (model 1) controlled for participant sex (male, female), study center (Birmingham, Chicago, Minneapolis, or Oakland), race (white, black), and age. Additional adjustment (model 2) included years of education, current cigarette smoking (current, never/former), intensity units of physical activity, systolic blood pressure, LDL‐C, HDL‐C, natural log‐transformed triglycerides, HOMA‐IR, BMI, natural log‐transformed high‐sensitivity CRP, eGFR, and UACR. We tested the sensitivity of models additionally adjusted for the use of lipid‐lowering or blood pressure mediation, sICAM‐1 and F₂‐isoprostanes, and dietary variables.

Incident CAC was defined as having an Agatston score of 0 in 2000–2001 (Year 15) and a score >0 at follow‐up. We used Poisson regression with an offset of time to incident CAC to estimate the effect of TMAO on 10‐year CAC incidence. To account for CT error in detection of CAC, we reran models based on a ≥10 score increase. We also defined 10‐year CAC progression as any CAC increase over follow‐up (including prevalent CAC in 2000–2001) and estimated relative rates of 10‐year CAC progression with respect to TMAO using Poisson regression.

We assessed the sensitivity of results to the specification of TMAO in regression models. In addition to our primary analysis of sample‐based quartiles, we included linear and quadratic specifications of continuous TMAO, and TMAO categories with cutpoints from a previous study of TMAO and CVD (<2.43, 2.43–3.66, 3.67–6.18, and >6.18 μmol/L).2

We additionally considered prospective associations between TMAO and 10‐year incidence of nonfatal CVD events using Cox proportional hazards regression, association between TMAO in 2000–2001 and level of common cIMT in 2005–2006 using linear regression, and the prospective association between TMAO and 10‐year changes in eGFR and UACR using linear regression. We assessed potential selectivity of our study sample by comparing those eligible for our analysis (required to attend all 3 exams in the study period) and all participants who attended the 2000–2001 exam. Statistical analyses were completed with SAS (v9.4; SAS Institute Inc., Cary, NC) and Stata/MP (v14.0; StataCorp LP, College Station, TX) software.