In 2024, 589 million adults aged 20–79 years were living with diabetes mellitus. That is 11.11% (95% uncertainty interval 11.05–11.17) of the adult population in this age band. The estimate belongs to the 11th edition of the IDF Diabetes Atlas (2025) and is reproduced in Lancet Diabetes Endocrinol. 2026;14(2):149–156. By 2050 the Atlas projects 853 million (12.96%). One in nine adults today; almost one in eight in the projection.
This number is not "all diabetes in one pile without type". The IDF estimates type 1 diabetes separately: more than 9.5 million people in 2024, of whom 1.9 million were under 20 years of age (Diabetes Atlas, 11th ed., Summary). T2DM accounts for more than 90% of all diabetes cases — this is how both the Atlas and the WHO fact sheet of 14 November 2024 put it ("More than 95% of people with diabetes have type 2 diabetes"). The difference of 90% vs 95% is not a reason to choose the "nicer" number: these are different sources with different methods. A study answer names the source.
In the same fact sheet the WHO gives a different global estimate: the number of people with diabetes rose from 200 million in 1990 to 830 million in 2022; in 2022, 14% of adults aged 18 and older were living with diabetes. This is neither a refutation of the IDF nor "a more recent truth". The WHO age band starts at 18 years, the estimate year is 2022, and the method is NCD-RisC, not the Atlas. Mixing 589 and 830 in one sentence without a caveat is a reading error, not a refinement.
Three consequences for the clinician are already visible from epidemiology, before pathogenesis.
First. Many cases are undiagnosed. The IDF Atlas 11th edition estimates that 43% of adults with diabetes (252 million) are unaware of their disease; almost 90% of them live in low- and middle-income countries. This is an observational estimate of prevalence, not proof that "screening everyone reduces mortality". But it explains why ADA 2026 requires screening of asymptomatic adults and why in T2DM the fundus examination is done at the time of diagnosis rather than "in five years": by the time it is detected, hyperglycaemia may already have lasted for years (ADA 2026, §12.4).
Second. City and age shift the curve. In the 2024 Atlas, prevalence is higher in men than in women (11.55% vs 10.68%), higher in urban than rural areas (12.26% vs 9.23%), and reaches 23.72% in adults over 65. The peak is 24.79% in the 75–79 age group. This is not "diabetes of old age": it is accumulated duration of exposure plus demography. T2DM in a 35-year-old with obesity is the same nosology, with a different risk of missing an autoimmune variant.
Third. The Kazakhstan figure that can be quoted is a registry figure, not a population figure. In material based on the MoH RK electronic dispensary registry and published on 13 November 2025, 535,033 patients with diabetes mellitus were registered for 2025, of whom 504,456 (95%) had T2DM and 30,577 had type 1 diabetes, including 5,625 children under 18. These are recorded cases under dispensary follow-up, not standardised prevalence. A population survey of "how many there really are" in Kazakhstan was not found in open sources as of the passport date. A registry is always smaller than hidden morbidity — this is a property of registries, not a proven coefficient.
Mortality and economics are also given in the Atlas, and they too cannot be restated as "diabetes kills one in nine". For 2024 the IDF estimates 3.4 million diabetes-related deaths among adults aged 20–79 and direct health-system expenditure of more than 1 trillion US dollars. For 2021 the WHO reports 1.6 million deaths directly attributed to diabetes, plus 530,000 deaths from kidney disease caused by diabetes. Again the difference is methodological: "death from diabetes" and "diabetes-related death" are different counters.
For study practice one rule is enough. When you are asked to "state the burden of T2DM", you name the source, the year, the age band and the type of estimate (Atlas modelling, WHO estimate, national registry). In this course a number without its frame counts as no answer.