Risk-prediction models for incident type 2 diabetes in Chinese people with impaired glucose tolerance: a systematic literature review and external validation study

Abstract

BackgroundPreventing type 2 diabetes is a major challenge in China, which has the world's largest population with impaired glucose tolerance (IGT). Given that no validated diabetes risk-prediction tools exist for Chinese people with IGT to inform decision making when considering primary diabetes prevention measures, we evaluated the performance of published Asian-derived type 2 diabetes risk-prediction models in Chinese people with IGT. MethodsTo identify Asian-derived type 2 diabetes risk-prediction models we did a systematic MEDLINE and Embase literature search for publications from inception until July 13, 2019. The publications identified were divided into basic (clinical variables only) and extended (plus laboratory variables) models. Validation was done in two prospective Chinese IGT cohorts: the Acarbose Cardiovascular Evaluation (ACE) trial of participants with IGT and coronary heart disease treated with placebo (n=3241), and the Luzhou survey cohort with IGT in which 96% did not have coronary heart disease (n=1155). Model performance was assessed in terms of discrimination (C-statistic) and calibration (Hosmer-Lemeshow test and calibration plot). FindingsWe identified 36 Asian-derived models (13 basic and 23 extended). In the ACE trial type 2 diabetes developed over a median of 5 years (IQR 3·4–6·0) with a follow-up for 15·7% of participants and in the Luzhou cohorts the disease developed over 3 years (3·0–3·0) with a follow-up for 20·7%. Basic models showed poor discrimination in these populations (C-statistic 0·53–0·59 for the ACE trial participants and 0·52–0·58 for the Luzhou cohorts). Extended models showed poor-to-acceptable discrimination (C-statistic 0·56–0·73 for the ACE trial participants and 0·51–0·66 for the Luzhou cohorts). Most extended models overestimated diabetes risk in the ACE cohort and underestimated risk in the Luzhou cohort, whereas all basic models underestimated diabetes risk in both populations. Basic and extended models showed poor calibration, although performance was improved by recalibration. The navigator model showed the best discrimination in both populations but had poor calibration. InterpretationIn Chinese people with IGT (with or without coronary heart disease), Asian-derived basic models to predict type 2 diabetes did not exhibit good discrimination or calibration. Several extended models performed better, but a robust Chinese diabetes risk-prediction tool remains a major unmet need. FundingNone.