Participant’s descriptive characteristics were summarized as means and standard deviation, medians, and interquartile ranges, or frequencies and percentages. We explored sex differences using independent t tests, Mann-Whitney U tests, or χ2 tests for normally distributed, skewed or dichotomous variables, respectively. We assessed the normality of variables and logarithmically or reciprocally transformed skewed variables before further analyses.
We investigated the separate longitudinal associations of cfPWV and cIMT (predictors) at 17.7 years with each of fasting LDL, HDL, triglyceride, insulin, and glucose categories (outcomes) at 24.5 years using binary logistic regression models (Supplemental Material ). We also examined the separate associations of the 7-year progression in cfPWV and cIMT with the longitudinal progression in each of the metabolic outcomes from ages 17.7 to 24.5 years using linear mixed-effect models for repeated measures. Analyses were adjusted for sex, age at 17.7 years, and covariates repeatedly measured at ages 17.7 and 24.5 years, viz, resting heart rate, systolic blood pressure, fat mass, lean mass, high-sensitivity C-reactive protein, smoking status, family history of hypertension, diabetes, high cholesterol or vascular disease and moderate to vigorous physical activity at 15.5 and 24.5 years as well as fasting plasma samples; LDL, HDL, insulin, triglyceride, or glucose, depending on the outcome.
Lastly, we used structural equation modeling with autoregressive cross-lagged path analysis (detailed in theSupplemental Material and published earlier13 (link)) to examine the separate temporal causal associations of cfPWV and cIMT with metabolic outcomes, adjusting for covariates listed above. All covariates were selected based on previous studies.3 (link),11 (link)–13 (link),18 (link),19 (link) We examined sex interactions and presented sex-stratified results. We also presented body mass index–weight stratified results, cross-sectional results, and age- and sex-adjusted partial correlation analyses in Tables S4 through S7. Differences and associations with a 2-sided P<0.05 were considered statistically significant with conclusions based on effect estimates and their CI or SE. Analyses involving 40% of a sample of 10 000 ALSPAC children at 0.8 statistical power, 0.05 alpha, and 2-sided P value would show a minimum detectable effect size of 0.049 standard deviations if they had relevant exposure for a normally distributed quantitative variable.20 (link) All statistical analyses were performed using SPSS statistics software, Version 27.0 (IBM Corp, Armonk, NY), and structural equation modeling were conducted using IBM AMOS version 27.0.
We investigated the separate longitudinal associations of cfPWV and cIMT (predictors) at 17.7 years with each of fasting LDL, HDL, triglyceride, insulin, and glucose categories (outcomes) at 24.5 years using binary logistic regression models (
Lastly, we used structural equation modeling with autoregressive cross-lagged path analysis (detailed in the