Refitting of the UKPDS 68 risk equations to contemporary routine clinical practice data in the UK

Economic evaluations of new diabetes therapies rely heavily upon the UK Prospective Diabetes Study (UKPDS) equations for prediction of cardiovascular events; however, concerns persist regarding their relevance to current clinical practice and appropriate use in populations other than newly diagnosed patients. This study refits the UKPDS 68 event equations, using contemporary data describing low- and intermediate-risk patients.

Anonymized patient data describing demographics, risk factors and incidence of cardiovascular and microvascular events were extracted from The Health Improvement Network (THIN) database over the 10-year period from 1 January 2000 to 31 December 2009. Following multiple imputation of missing values, accelerated failure-time Weibull regression equations were refitted to produce new coefficients for each risk group. Discriminatory performance was assessed and compared with both UKPDS 68 and UKPDS 82 risk equations, and the implication of coefficient choice within an economic evaluation was assessed using the Cardiff type 2 diabetes model.

When applied to patient-level data, the three sets of coefficients (UKPDS, THIN low-risk and intermediate-risk) lead to fairly consistent predictions of the 5-year risk of events. Exceptions include lower predicted rates of myocardial infarction and higher rates of ischaemic heart disease, congestive heart failure and end-stage renal disease with both sets of revised THIN coefficients compared with UKPDS. Over a modelled lifetime, the coefficients derived from the low-risk data predict fewer total cardiovascular events compared with UKPDS, while those from the intermediate-risk data predict a greater number. The areas under the receiver-operating characteristic curves demonstrated a marginal improvement in the discriminatory performance of the refitted equations. The incremental cost-effectiveness ratio associated with dapagliflozin versus sulphonylurea in addition to metformin changed from £7,708 to £7,519 and £6,906 per QALY gained, using the THIN intermediate- and low-risk coefficients, respectively.

The results suggest that while the UKPDS equations perform best in newly diagnosed patients, they may overpredict the lifetime risk in this group and underpredict it in patients with more advanced diabetes. Implementation of the revised coefficients will result in different absolute numbers of predicted diabetes-related events; however, they are not expected to significantly affect the conclusions of economic modelling.