External validation of cardiovascular risk models in 23,685 adults with diabetes

Abstract

<p><b>Background</b><br>There is an abundance of risk models for predicting cardiovascular outcomes in people with type 2<br>diabetes. Independent external validations of cardiovascular disease risk prediction models are crucial<br>for assessing their applicability. We externally validated risk prediction models developed for people<br>with diabetes in a large population-based prospective cohort (UK Biobank).<br><b>Methods</b><br>We externally validated 17 cardiovascular risk scores (from seven risk prediction models: Chang et al,<br>Framingham, HKU-SG, Li et al, RECODe, SCORE, UKPDS OM2) identified from systematic reviews<br>using UK Biobank (n= 23,685; 40-71 years old; 63.5% male) data from 2006 to 2021. We evaluated<br>model performance by assessing discrimination and calibration for mortality, cardiovascular mortality,<br>congestive heart failure, myocardial infarction, stroke, and ischemic heart disease.<br><b>Results</b><br>Over a total of 269,430 person-years of follow-up (median: 11.89 years), the models showed low-tomoderate discrimination performance on external validation (c-indices ranging from 0.58-0.71). Most<br>models had low calibration with overprediction of the observed risk. RECODe outperformed other<br>models across five comparable endpoints for discrimination: all-cause mortality (c-index=0.67, [95% CI:<br>0.65-0.69]), cardiovascular mortality (0.70, [0.67-0.72]), congestive heart failure (0.71, [0.69-0.72]),<br>myocardial infarction (0.67, [0.65-0.68]), and stroke (0.65, [0.62-0.68]; and for calibration (except for all cause mortality). Models performed better for younger participants and somewhat better for non-white<br>ethnicities. Models developed from non-Western datasets had lower performance in our UK-based<br>validation set.<br><b>Conclusions</b><br>The RECODe equations had better risk estimations in this predominantly white European population.<br>Further validation is needed on non-Western populations to assess generalizability for other<br>populations.</p>