medwireNews: Researchers assessing 22 predictive models of cardiovascular disease (CVD) risk find that none accurately predict its occurrence in a population of people with type 2 diabetes.
The findings underline “the difficulties of accurately predicting CVD in a relatively high-risk population,” write Katarzyna Dziopa (University College London, UK) and study co-authors in Diabetologia.
The models’ discrimination – the ability to distinguish between individuals who did and did not develop CVD – ranged from 61.9% to 67.0%. This poor discrimination was despite most of the models achieving good calibration (ie, correctly predicting the number of outcomes in the target population).
The “almost uniform” discriminative accuracy was despite the derivation dates of the models encompassing more than 3 decades, leading the researchers to conclude that healthcare changes over time do not explain the poor performance.
The team assessed the models’ performance in 168,871 people (average age 59.3 years, 46% women) with type 2 diabetes and no CVD at baseline identified in English healthcare databases. During a median follow-up of 10 years, 22.7% of the study population developed CVD, heart failure, or atrial fibrillation.
Discrimination was equally poor in the nine models developed in people with type 2 diabetes, and was no better in models with a larger number of predictors.
medwireNews is an independent medical news service provided by Springer Healthcare Ltd. © 2022 Springer Healthcare Ltd, part of the Springer Nature Group