IDF Atlas: Did NCD-RisC Overestimate Worldwide Diabetes Prevalence? by Boyko EJ

The global diabetes estimates are compiled by three major groups:

1. International Diabetes Federation (IDF), which has released 11 estimates since 2000
2. Global Burden of Diseases (GBD) study, which published its latest estimate in 2021
3. Noncommunicable Diseases Risk Factor Collaboration (NCD-RisC), with its most recent data from 2022.

NCD-RisC estimate for 2022 differs by approximately 300 million persons compared to IDF Atlas 10^th edition and GBD 2021 and by 250 million compared to IDF Atlas 11^th edition 2024 estimates. IDF Atlas 11^th edition 20250 projection of 853 million comes closest to the NCD-RisC estimate, but 28 years later. Of the 3 groups attempting estimates of the diabetes world population, NCD-RisC is the outlier.

In the Pune Children’s study, HbA1c showed unexpected high prevalence of pre-diabetes and diabetes compared with OGTT (26% vs. 10.4%). India has a high prevalence of iron deficiency with an overall rate of 40% and 50% among women. A meta-analysis findings showed that Iron replacement therapy among person with iron deficiency anemia leads to substantial reduction in HbA1c.

The NCD-RisC estimated that 212 million people aged 18 and above in India had diabetes in 2022. In contrast, the IDF estimated only 74 million people aged 20–79 had diabetes in 2021. This results in a discrepancy of 138 million cases for India. Globally, the difference between these estimates is 291 million. Even after accounting for India’s difference, 153 million excess cases remain unexplained.

Conclusion:

The NCD-RisC overestimated diabetes prevalence due to factors such as falsely elevated HbA1c manly because of high prevalence of iron deficiency and prediction model extending HbA1c misclassification to other countries. Re-analysis of NCD-RisC data by removing HbA1c as a diabetes diagnostic criterion would help to resolve this discrepancy. On the other hand, while the IDF may underreport cases by not including HbA1c, its use is currently not recommended until the limitations of this method are better understood.