An independent validation of QRISK on the THIN database.
Collins GS., Altman DG.
Scope and purpose: QRISK is a new cardiovascular disease risk prediction algorithm published in 2007. This document has two main objectives: To report the findings of an independent validation of the QRISK algorithm on the THIN database; To report the findings of an independent review of the unpublished Robins et al paper and identify the reasons for discrepancies between that paper and the QRISK investigators' own validation study described in a paper published in "Heart". Background: In July 2007, Hippisley-Cox et al published a paper in the BMJ describing the derivation and validation of QRISK, a new cardiovascular disease risk prediction algorithm. This was a novel risk prediction algorithm which includes traditional risk factors (age, sex, systolic blood pressure, smoking status, serum cholesterol/HDL ratio, presence of diabetes, ECG-LVH) included in the Framingham equation but also includes body mass index, family history of cardiovascular disease, social deprivation and the use of blood pressure treatment. The QRISK algorithm then underwent a revision to account for statin use and changes to the multiple imputation method to account for missing data. This revised algorithm was then used in a second validation study designed to test the performance of QRISK in practices contributing to a second independent THIN dataset. This study was published in "Heart" and the paper includes a comparison of the model performance statistics from the original QRESEARCH cohort and the THIN cohort. Subsequently, an unpublished paper by Robins et al from the University of East Anglia (UEA) was submitted to NICE in response to its consultation on the Lipid Modification guidelines, which included an independent evaluation of QRISK in the THIN database. The authors concluded that they could not replicate the results published in the "Heart" paper. That disagreement has highlighted the need for an independent evaluation of the QRISK algorithm in the THIN dataset to address the apparent discrepancies between these papers and to consider whether any of the issues raised in the Robins et al paper are substantial with regard to QRISK. We report here the findings of independent analyses conducted at the Centre for Statistics in Medicine (CSM), in Oxford.