Abstract
Valid screening and diagnostic algorithms are needed to achieve 2030 targets proposed by the WHO's Global Diabetes Compact. We explored anthropometric thresholds to optimally screen and refer individuals for diabetes testing in rural South Africa. We evaluated screening thresholds for waist circumference (WC), body mass index (BMI), and waist-hip ratio (WHR) to detect dysglycemia based on a glycated hemoglobin (HbA1C) ≥6.5% among adults in a population-based study in South Africa using weighted, non-parametric ROC regression analyses. We then assessed the diagnostic validity of traditional obesity thresholds, explored optimal thresholds for this population, and fit models stratified by sex, age, and HIV status. The prevalence of dysglycemia in the total study population (n = 17,846) was 7.7%. WC had greater discriminatory capacity than WHR to detect dysglycemia in men (p-value<0.001) and women (p<0.001). WC had greater discriminatory capacity than BMI to detect dysglycemia in women (p<0.001). However, BMI and WC performed similarly for men (p = 0.589). Whereas traditional WC thresholds for women (>81cm) performed well (sensitivity 91%, positive predictive value [PPV] 14.9%), substantially lower thresholds were needed to achieve acceptable sensitivity and PPV among men (traditional >94cm, derived >79.5cm). WC outperforms BMI as an anthropometric screening measure for dysglycemia in rural South Africa. Whereas WC guideline thresholds are appropriate for women, male-derived WC cutoffs performed better at lower thresholds. In this rural South African population, thresholds that maximize specificity and PPV for efficient resource allocation may be preferred.