Abstract / Summary
Existing studies of diabetes management have largely relied on dichotomous controlled/uncontrolled outcomes, which obscure the clinical heterogeneity of the population living with diabetes. We propose a composite classification combining HbA1c level and treatment stage, characterise its distribution and correlates among Korean adults, and quantify temporal changes from 2015 to 2024. The classification describes glycaemic–treatment phenotypes; it is not a prognostically validated severity staging. Using the Korea National Health and Nutrition Examination Survey (KNHANES), 2015–2024, we identified 8,702 adults aged ≥ 30 years with diabetes (physician diagnosis, glucose-lowering medication, fasting glucose ≥ 126 mg/dL, or HbA1c ≥ 6.5%) and a measured HbA1c value. Participants were classified into four glycaemic–treatment phenotypes: P1 (no or lifestyle-only treatment with HbA1c < 7%), P2 (oral medication with HbA1c < 7%), P3 (HbA1c 7 to < 9% and not receiving insulin), and P4 (insulin use or HbA1c ≥ 9%). Weighted proportions, multinomial logistic regression with P1 as reference, and trend models incorporating the complex survey design were applied. A formal change-point analysis over candidate years 2017–2023, a multiple-imputation sensitivity analysis for missing covariates, and an analysis restricted to previously diagnosed participants were also performed. The composition of each phenotype is reported cell by cell, and the temporal analysis is repeated under three alternative classification rules. The weighted distribution was 24.8% P1, 34.3% P2, 28.9% P3, and 12.0% P4. The composition changed abruptly in 2022. A formal change-point analysis selected 2022 for both P2 and P3 by design-based AIC and by the maximum Wald statistic, and 2022 was selected in 97.1% and 97.7% of 1,000 design-based bootstrap replicates respectively; a discrete level shift fitted better than a change in slope. No change-point was supported for P4. The P2 share rose from 28.2% (2015–2021) to 46.8% (2022–2024), a relative increase of 66.0% (Δ + 18.6% points; adjusted prevalence ratio 1.49, 95% CI 1.31–1.69), while P1 and P3 declined (adjusted prevalence ratios 0.81 and 0.68; both P ≤ 0.010) and P4 was unchanged. In the primary analysis of correlates, restricted to previously diagnosed participants, reduced kidney function (eGFR < 60) was the only full-sample association with the insulin-treated or markedly hyperglycaemic phenotype that retained its magnitude (OR 2.67, 95% CI 1.27–5.62); reported dyslipidaemia remained associated with P2 and P3 at roughly half its full-sample magnitude (P2 OR 1.79, 1.30–2.46), and the inverse associations of obesity and drinking seen in the full sample were absent. In the full sample, reported descriptively because its reference group is predominantly screen-detected, the corresponding estimates were 3.09 (2.11–4.53) and 3.11 (2.66–3.63). Multiple imputation reproduced every complete-case estimate closely (largest difference 0.17 on the odds-ratio scale). The temporal findings were unchanged under three alternative classification rules, and the estimated change-point remained 2022 under each of them. Of 42 interaction tests, none remained significant after false-discovery-rate correction. A composite HbA1c–treatment classification identifies a discrete redistribution of glycaemic–treatment phenotypes in 2022, with a 66% relative increase in the treated, at-target phenotype at the expense of the untreated and above-target phenotypes; because 2022 is also the first year of KNHANES cycle IX, the change cannot be attributed to diabetes care within this design. Reduced kidney function was the only correlate that persisted at its full-sample magnitude when the comparison was restricted to previously diagnosed participants, and reported dyslipidaemia persisted in attenuated form; the remaining associations appear to be generated largely by the composition of the reference phenotype, 88% of which consists of screen-detected, previously undiagnosed individuals. The scheme should therefore be read as a description of glycaemic–treatment phenotypes rather than as a validated severity staging, and its prognostic value remains to be tested against complications, cardiovascular events and mortality in longitudinal and independent populations. Not applicable.