Background:Diabetes has been defi ned on the basis of diff erent biomarkers, including fasting plasma glucose (FPG),
2-h plasma glucose in an oral glucose tolerance test (2hOGTT), and HbA1c. We assessed the eff ect of diff erent
diagnostic defi nitions on both the population prevalence of diabetes and the classifi cation of previously undiagnosed
individuals as having diabetes versus not having diabetes in a pooled analysis of data from population-based health
examination surveys in diff erent regions.
Methods: We used data from 96 population-based health examination surveys that had measured at least two of the
biomarkers used for defi ning diabetes. Diabetes was defi ned using HbA1c (HbA1c ≥6·5% or history of diabetes
diagnosis or using insulin or oral hypoglycaemic drugs) compared with either FPG only or FPG-or-2hOGTT
defi nitions (FPG ≥7·0 mmol/L or 2hOGTT ≥11·1 mmol/L or history of diabetes or using insulin or oral
hypoglycaemic drugs). We calculated diabetes prevalence, taking into account complex survey design and survey
sample weights. We compared the prevalences of diabetes using diff erent defi nitions graphically and by regression
analyses. We calculated sensitivity and specifi city of diabetes diagnosis based on HbA1c compared with diagnosis
based on glucose among previously undiagnosed individuals (ie, excluding those with history of diabetes or using
insulin or oral hypoglycaemic drugs). We calculated sensitivity and specifi city in each survey, and then pooled
results using a random-eff ects model. We assessed the sources of heterogeneity of sensitivity by meta-regressions
for study characteristics selected a priori.
Findings: Population prevalence of diabetes based on FPG-or-2hOGTT was correlated with prevalence based on FPG
alone (r=0·98), but was higher by 2–6 percentage points at diff erent prevalence levels. Prevalence based on HbA1c was
lower than prevalence based on FPG in 42·8% of age–sex–survey groups and higher in another 41·6%; in the other
15·6%, the two defi nitions provided similar prevalence estimates. The variation across studies in the relation between
glucose-based and HbA1c-based prevalences was partly related to participants’ age, followed by natural logarithm of
per person gross domestic product, the year of survey, mean BMI, and whether the survey population was national,
subnational, or from specifi c communities. Diabetes defi ned as HbA1c 6·5% or more had a pooled sensitivity of
52·8% (95% CI 51·3–54·3%) and a pooled specifi city of 99·74% (99·71–99·78%) compared with FPG 7·0 mmol/L or
more for diagnosing previously undiagnosed participants; sensitivity compared with diabetes defi ned based on FPGor-
2hOGTT was 30·5% (28·7–32·3%). None of the preselected study-level characteristics explained the heterogeneity
in the sensitivity of HbA1c versus FPG.
Interpretation: Diff erent biomarkers and defi nitions for diabetes can provide diff erent estimates of population
prevalence of diabetes, and diff erentially identify people without previous diagnosis as having diabetes. Using an
HbA1c-based defi nition alone in health surveys will not identify a substantial proportion of previously undiagnosed
people who would be considered as having diabetes using a glucose-based test. |