Development and Validation of Body Mass Index-Specific Waist Circumference Thresholds in Postmenopausal Women : A Prospective Cohort Study

Abstract

Background: A 2020 consensus statement proposed body mass index (BMI)-specific waist circumference (WC) thresholds to improve patient care.

Objective: To determine whether stratifying BMI categories by BMI-specific WC thresholds improves mortality risk prediction.

Design: Prospective cohort study.

Setting: Women's Health Initiative multicenter, population-based U.S. study, with enrollment from 1993 to 1998 and follow-up through 2021.

Participants: 139 213 postmenopausal women aged 50 to 79 years were included in a development cohort (n = 67 774) and 2 external validation cohorts. Validation Cohort 1 had high prevalence of overweight or obesity (n = 48 335), and Validation Cohort 2 included diverse, geographically separate centers (n = 23 104).

Measurements: Height, weight, and WC measured at enrollment. BMI categories were normal weight (18.5 to <25 kg/m²), overweight (25 to <30 kg/m²), obesity-1 (30 to <35 kg/m²), obesity-2 (35 to <40 kg/m²), and obesity-3 (≥40 kg/m²), with further stratification by prespecified WC thresholds (≥80, ≥90, ≥105, ≥115, and ≥115 cm, respectively). Mortality was ascertained annually and was supplemented with serial National Death Index queries. Ten- and 20-year mortality prediction models that included BMI categories were compared to models with BMI categories stratified by WC thresholds using c-statistics and continuous net reclassification improvement (NRI).

Results: Over a median of 24 years of follow-up, 69 297 participants died. Multivariable-adjusted mortality risk was consistently greater for BMI categories with large WC than those with normal WC. Compared with women with normal weight and normal WC, women with normal or overweight BMI but large WC (hazard ratios [HRs], 1.17 [95% CI, 1.12 to 1.21] and 1.19 [CI, 1.15 to 1.24], respectively) had risk similar to those with obesity-1 but normal WC (HR, 1.12 [CI, 1.08 to 1.16]). Mortality associated with obesity-1 and large WC (HR, 1.45 [CI, 1.35 to 1.55]) was similar to that with obesity-3 and normal WC (HR, 1.40 [CI, 1.28 to 1.54]). Models with BMI-specific WC thresholds improved discrimination and risk stratification at 10 years for Validation Cohort 1; c-statistics improved by 0.7% (CI, 0.3% to 1.0%) to 61.3% (CI, 60.2% to 62.5%), and continuous NRI was 20.4% (CI, 17.3% to 23.6%). Results were mixed for Validation Cohort 2; risk stratification improved (continuous NRI, 12.3% [CI, 8.5% to 16.0%]), but not discrimination. Results were similar at 20 years.

Limitation: The study did not include men or younger women.

Conclusion: Further stratifying BMI categories by WC thresholds modestly improved mortality risk stratification, with larger WC predicting greater mortality, although the degree of improvement varied by cohort. Discrimination did not improve consistently.

Primary funding source: National Heart, Lung, and Blood Institute of the National Institutes of Health.

Figure 1.. Baseline* risk for all-cause mortality at 10 and 20 years, estimated from the BMI-WC model on the development cohort (n = 67 774).

The BMI-WC model comprised the mortality model plus BMI categories further stratified by WC thresholds. Estimates were similar but not identical for the BMI and mortality models (Supplement Table 4, available at Annals.org). BMI = body mass index; WC = waist circumference. BMI-WC model, mortality model plus BMI-categories further stratified by waist circumference thresholds. * Baseline risk corresponds to participants whose predictor values were 0 (i.e., the reference group for predictors). † Prior disease included history of cardiovascular disease (myocardial infarction, coronary artery bypass graft, percutaneous coronary intervention, or stroke) or cancer. ‡ Preexisting conditions included history of diabetes, hypertension, or hyperlipidemia.

Figure 2.. HRs from multivariable regression models predicting all-cause mortality in the development cohort (n = 67 774).

Estimates are from the mortality model (left), the BMI model (middle), and the BMI-WC model (right). There were no missing data across models. All 3 models were developed on 67 774 participants and 34 127 all-cause deaths through 31 December 2021. Ten- and 20-year survival ($Sk(10)$ and $Sk(20)$) for age and disease stratum $k$ can be obtained by subtracting the corresponding 10- and 20-year risks (reported in Figure 1) from 1. Risk for all-cause mortality within 10 or 20 years for age and disease stratum $k$ can be computed as $1-Sk(t)\times exp(PI)$, where $t=10$ or 20 years and $PI=\text{the}\text{sum}\text{of}\text{estimated}\text{β}\mathrm{s}$ (i.e., log-HRs) multiplied by their respective predictor values (coded 0 for referent categories and 1 otherwise). Internal validation (apparent performance) of the development cohort at 10 and 20 years showed c-statistics of 60.5% (95% CI, 59.7% to 61.4%) and 58.6% (CI, 58.1% to 59.0%), respectively, for the mortality model; 61.3% (CI, 60.4% to 62.1%) and 59.1% (CI, 58.6% to 59.5%), respectively, for the BMI model; and 61.7% (CI, 60.8% to 62.5%) and 59.4% (CI, 59.0% to 59.9%), respectively, for the BMI-WC model. The BMI model comprised the mortality model plus BMI categories. The BMI-WC model comprised the mortality model plus BMI categories further stratified by WC thresholds. BMI = body mass index; HR = hazard ratio; ref = reference; WC = waist circumference.

Figure 3.. Stratified c-statistics and corresponding differences and associated 95% CIs for all-cause mortality, computed biennially.

The upper panels correspond to c-statistics and 95% CIs evaluated with increasingly longer periods of cumulative follow-up; prediction models were exclusively developed using follow-up through 2021. The lower panels correspond to their differences and 95% CIs. Confidence intervals were based on the infinitesimal jackknife estimator for variances and covariances. The BMI model comprised the mortality model plus BMI categories. The BMI-WC model comprised the mortality model plus BMI categories further stratified by WC thresholds. BMI = body mass index; WC = waist circumference.

Figure 4.. Continuous NRIs and 95% CIs for all-cause mortality, computed biennially.

Continuous NRIs were evaluated with increasingly longer periods of cumulative follow-up; prediction models were exclusively developed using follow-up through 2021. Confidence intervals were estimated using the percentile method with 10 000 stratified bootstrap samples. The BMI model comprised the mortality model plus BMI categories. The BMI-WC model comprised the mortality model plus BMI categories further stratified by WC thresholds. BMI = body mass index; NRI = net reclassification improvement; WC = waist circumference.