Key result
Established diabetes risk models show poor discrimination for incident dysglycaemia, yielding c-statistics under 0.7.
Why the study?
Do established diabetes risk models accurately predict incident dysglycaemia in African and European populations?
Cohort (n=5,166)
Yes
Do established diabetes risk models accurately predict incident dysglycaemia in African and European populations?
Effect estimate: c-statistic < 0.7
Established diabetes risk models require recalibration to local risk levels to accurately predict incident dysglycaemia in Sub-Saharan African and European populations.
Aims We evaluated overall and sex‐specific performance of established diabetes risk models for predicting incident dysglycaemia in African and European populations. Materials and Methods We externally evaluated the African Diabetes Risk model, Finnish Diabetes Risk model, Epidemiological Study on the Insulin Resistance Syndrome model and the personal and clinical Framingham models. Evaluation was performed in three prospective cohorts: the Research on Obesity and Diabetes among African Migrants (RODAM‐Pros; migrant and non‐migrant Ghanaians; n = 1607), the Middle‐aged Soweto cohort (MASC; urban Black South Africans; n = 515) and the Västerbotten Intervention Programme (VIP; Sweden; n = 3044). Dysglycaemia was defined according to American Diabetes Association criteria. Discrimination (c‐statistic) and calibration were assessed before and after recalibration. Results Dysglycaemia developed in 17.7% of RODAM‐Pros participants, 35.0% of MASC participants and 27.5% of VIP participants, with median follow‐up of 6.7, 6.7 and 9.9 years, respectively. All models demonstrated poor‐to‐modest discrimination (c‐statistic < 0.7). Most models substantially underestimated risk. Recalibration eliminated systematic miscalibration and reduced calibration error to ≤ 2.3% in RODAM‐Pros and VIP and ≤ 3.4% in MASC. Performance varied by sex and cohort. Conclusions Existing diabetes risk prediction models can be applied to predict incident dysglycaemia in both Sub‐Saharan African and European populations, but only after recalibration to local risk levels.
No takes yet. Share an insight, caveat, or question.
Smidt et al. (2026) conducted a cohort in Incident dysglycaemia (n=5,166). Established diabetes risk models was evaluated on Incident dysglycaemia (c-statistic < 0.7). Established diabetes risk models demonstrated poor-to-modest discrimination (c-statistic < 0.7) for predicting incident dysglycaemia, requiring recalibration to local risk levels.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: