OP1 The validation and comparison of prevalent diabetes risk prediction models in a mixed-ancestry population of South Africa

Abstract

Introduction: The use of multivariable risk prediction models have been advocated as practical and potentially affordable approaches for improving the detection of undiagnosed diabetes or screening for future risk of diabetes. However, no one single model can perform well in all settings and available models must be tested before implementation in new populations. The objective is to assess and compare the performance of six prevalent diabetes risk models updated with intercept adjustment in mixed-ancestry South Africans Methods: The data from the Cape Town Bellville-South cohort served as the basis for this study. Models were identified via recent systematic reviews. Models’ discrimination was assessed and compared using C-statistic and non-parametric methods. Calibration was assessed via the Hosmer and Lemeshow statistics, before and after recalibration through simple intercept adjustment. Results: 737 participants (27% male), aged 52.2 years, were included, among who 130 (17.6%) had prevalent undiagnosed diabetes. The highest C-statistic was recorded with the San Antonio Heart Study (SAHS) model [C-statistic 0.925 (95% confidence: 0.895–0.954)] and the lowest with the Framingham model [0.557 (0.528–0.626)]; with always significant statistical differences when SAHS was compared with the other models (best fit-MESA Framingham, best-fit MESA San Antonio, Cambridge and KORA risk predictive models). The intercept adjustment improved calibration across all models, most significantly in the SAHS. Conclusion: The wide range of performances of different models in our sample highlights the challenges of selecting an appropriate model for diabetes risk prediction in different settings.