The validation of cardiovascular risk scores for patients with type diabetes mellitus

J. Van Der Leeuw*, S. Van Dieren, J. W.J. Beulens, H. Boeing, A. M.W. Spijkerman, Y. Van Der Graaf, D. L. Van Der A, U. Nöthlings, F. L.J. Visseren, G. E.H.M. Rutten, K. G.M. Moons, Y. T. Van Der Schouw, L. M. Peelen

*Corresponding author for this work

Research output: Contribution to journalArticleAcademicpeer-review


Objective Various cardiovascular prediction models have been developed for patients with type 2 diabetes. Their predictive performance in new patients is mostly not investigated. This study aims to quantify the predictive performance of all cardiovascular prediction models developed specifi cally for diabetes patients. Design and methods Follow-up data of 453, 1174 and 584 type 2 diabetes patients without pre-existing cardiovascular disease (CVD) in the EPIC-NL, EPICPotsdam and Secondary Manifestations of ARTerial disease cohorts, respectively, were used to validate 10 prediction models to estimate risk of CVD or coronary heart disease (CHD). Discrimination was assessed by the c-statistic for time-to-event data. Calibration was assessed by calibration plots, the Hosmer -Lemeshow goodness-of-fit statistic and expected to observed ratios. Results There was a large variation in performance of CVD and CHD scores between different cohorts. Discrimination was moderate for all 10 prediction models, with c-statistics ranging from 0.54 (95% CI 0.46 to 0.63) to 0.76 (95% CI 0.67 to 0.84). Calibration of the original models was poor. After simple recalibration to the disease incidence of the target populations, predicted and observed risks were close. Expected to observed ratios of the recalibrated models ranged from 1.06 (95% CI 0.81 to 1.40) to 1.55 (95% CI 0.95 to 2.54), mainly driven by an overestimation of risk in high-risk patients. Conclusions All 10 evaluated models had a comparable and moderate discriminative ability. The recalibrated, but not the original, prediction models provided accurate risk estimates. These models can assist clinicians in identifying type 2 diabetes patients who are at low or high risk of developing CVD.

Original languageEnglish
Pages (from-to)222-229
Number of pages8
Issue number3
Publication statusPublished - 1 Jan 2015

Cite this