Sex-Dependent and Asymmetric Associations of Bodyweight History in the Twenties with Later HbA1c Trajectories in a Japanese Occupational Cohort

Figures

• 

  Figure 1. Participant flow for the trajectory and time-to-event analyses. The full cohort in this study included adults aged 20–59 years with valid sex information. Participants were then restricted to those with at least five BMI records, at least five HbA1c records, and a follow-up duration of at least 5 years. Among these, individuals with at least one BMI measurement in their 20s were identified. After exclusion of participants with both underweight and obesity history in their 20s, the final three-group trajectory cohort comprised 2923 participants (2126 women and 797 men), classified as NW20s, UW20s_ever, and OW20s_ever. For the time-to-event analysis, participants with HbA1c ≥ 5.6% at study entry were further excluded, yielding a final event-analysis cohort of 2753 participants (1996 women and 757 men).

• 

  Figure 2. HbA1c trajectories and local slopes at ages 25, 35, and 45 years by BMI-history cohort in the 20s, stratified by sex. (A) Estimated mean HbA1c (%) at ages 25, 35, and 45 years with 95% confidence intervals, obtained from a sex-stratified linear mixed-effects model: HbA1c ~ cohort3 × ns (Age) + (1|id). Estimated marginal means (EMMs) of HbA1c with 95% confidence intervals at ages 25, 35, and 45 years were obtained from sex-stratified linear mixed-effects models. Participants with follow-up ≥5 years were included. Cohorts were defined by body-weight history in the 20s: NW20s (no underweight/overweight), UW20s_ever (ever underweight), and OW20s_ever (ever overweight). (B) Local slopes (dHbA1c/dAge; % per 1-year increase) at ages 25, 35, and 45 years (95% confidence intervals), derived from marginal trends (emtrends) of the same model. Cohorts were defined using BMI measurements in the 20s: NW20s (no underweight or overweight), UW20s_ever (ever underweight without overweight), and OW20s_ever (ever overweight without underweight). Estimated marginal means and local slopes were computed using the R package (version 4.5.2) emmeans, based on linear mixed-effects models fitted with lme4.…

• 

  Figure 3. Kaplan–Meier cumulative incidence of incident HbA1c ≥ 5.6% by BMI-history cohort in the 20s, stratified by sex. (A) Kaplan–Meier cumulative incidence of reaching HbA1c ≥ 5.6 (supporting analysis). Cumulative incidence (1 − S(t)) is shown using age as the time scale with left truncation (delayed entry at each participant’s first observed age ≥30 years). The event was defined as the first observed HbA1c ≥ 5.6%. Shaded areas indicate 95% confidence intervals. Numbers at risk are displayed below each panel. The inset text reports the log-rank–equivalent p-value based on the Cox score test and the sex-specific Cox hazard ratios (HRs) with 95% confidence intervals for UW20s_ever and OW20s_ever compared with NW20s. The numbers at risk indicate the number of participants who remained under observation and had not yet reached HbA1c ≥ 5.6% at each age point. Because the risk set became small at older ages, especially in the obesity history group, estimates in the later age range should be interpreted with caution. (B) Forest plot of sex-stratified Cox hazard ratios. Cox proportional hazards regression was performed with age as the time scale using a counting-process formulation to account…