Design: Cross-sectional analysis of nationally representative surveys
Population: 598,883 adults aged 25 and over, BMI 18.5-29.9, in 82 low- and middle-income countries
Primary outcome: Classification of diabetes status by waist circumference, relative fat mass and BMI
Effect: AUC in men: waist 0.69, relative fat mass 0.69, BMI 0.64; in women 0.69, 0.69, 0.63

A cross-sectional analysis in Diabetes Care (September 2026) used nationally representative surveys from 82 low- and middle-income countries, covering 598,883 adults aged 25 and over with a BMI between 18.5 and 29.9 — the intermediate range where BMI tells a clinician least.

Pooled diabetes prevalence was 9.1%. Waist circumference and relative fat mass (a sex-specific index calculated from height and waist) classified diabetes status better than BMI in both men (AUC 0.69, 0.69 and 0.64) and women (0.69, 0.69 and 0.63). Relative fat mass matched or beat waist across most regions.

The gain is real but moderate: an AUC of 0.69 is still a weak classifier on its own. The value is in who gets tested. A patient with a 'normal' BMI and a large waist is the one BMI-based screening rules skip.

This lands squarely in Indian practice, where diabetes at lower BMI and central adiposity are the norm. A tape measure costs nothing and needs no laboratory.

Measure waist circumference in adults with a BMI of 18.5-29.9, not just weight and height
Offer glucose or HbA1c testing to a patient with a normal BMI and a raised waist
Relative fat mass needs only height and waist, and performed at least as well as waist alone
Treat an AUC of about 0.69 as a reason to test, never as a diagnosis
In South Asian patients, do not let a normal BMI close the question of diabetes risk

Why it matters

BMI-based screening thresholds miss exactly the lean-but-centrally-obese adults who make up much of the diabetes burden in India.

The statistics, in plain English

AUC is the chance that a measure ranks a randomly chosen person with diabetes above one without it: 0.5 is a coin toss, 1.0 is perfect. Moving from 0.63-0.64 to 0.69 is a useful gain for a free measurement, but 0.69 alone would still misclassify many people, which is why it points to testing rather than replacing it.