Proceedings of the 2013 annual meeting of the Netherlands Epidemiology Society
Volume 1 Issue S1 Abstract 40
S. van Dieren, Julius Center, UMC Utrecht, the Netherlands
J.W.J. Beulens, Julius Center, UMC Utrecht, the Netherlands
H. Boeing, German Institute of Human Nutrition Potsdam- Rehbrücke, Nuthetal, Germany
A.M.W. Spijkerman, German Institute of Human Nutrition Potsdam- Rehbrücke, Nuthetal, Germany
D.L. van der A., RIVM, Bilthoven, the Netherlands
U. Nöthlings, University of Bonn, Germany
G.E.H.M. Rutten, Julius Center, UMC Utrecht, the Netherlands
K.G.M. Moons, Julius Center, UMC Utrecht, the Netherlands
Y.T. van der Schouw, Julius Center, UMC Utrecht, the Netherlands
L.M. Peelen, Julius Center, UMC Utrecht, the Netherlands
Several cardiovascular prediction models have been developed for application in patients with type 2 diabetes. Their predictive performance in a new set of patients is mostly lacking. We validated the cardiovascular prediction models, identified by a recent systematic review, in two cohorts of patients with type 2 diabetes.
Data from five years follow-up of 455 diabetes patients of the EPIC-NL cohort and 1175 diabetes patients of the EPIC-Potsdam cohort were used to validate 9 prediction models to predict cardiovascular disease (CVD) or coronary heart disease (CHD) among patients with type 2 diabetes. Discrimination and calibration were assessed by the c-statistic for survival data and calibration plots, respectively.
During 5-year follow-up, patients were followed-up for incidence of CVD and CHD. All nine prediction models showed a moderate discrimination, with c-statistics ranging from 0.55 (95% CI: 0.46 to 0.70) for the Fremantle risk score to 0.71 (95% CI: 0.61 to 0.80) for the UKPDS risk score. Most prediction models severely overestimated the risk between 23% to 189%, except for the ADVANCE risk score, which underestimated the risk. After recalibration of the models, the calibration of all models was good, with only a slight overestimation of the risk.
After recalibration, most of the models provided accurate CVD risk estimates. However, discrimination for almost all models between type 2 diabetes patients who did and those who did not develop CVD or CHD was only moderate. Before using these prediction models in clinical practice performance, especially discrimination, should be improved.
Published: 06 Jun, 2013