To the Editor: Davis and colleagues stated that the Framingham and United Kingdom Prospective Diabetes Study (UKPDS) cardiovascular risk equations are not suitable for predicting risk in an Australian population with type 2 diabetes.1 If confirmed, this would be extremely disappointing. However, before accepting this conclusion the following important considerations should be noted.

• Davis noted that the Fremantle Diabetes Study (FDS) patient group differed significantly from the UKPDS baseline group (eg, 38% of the FDS patients were aged outside the validated age range of the risk engine [25–65 years] and were assessed by non-validated extrapolation). Similarly, it cannot be assumed that the FDS group is representative of patients in general practice and hospital diabetes clinics around Australia. Moreover, it would be interesting to know how well the engine performs in FDS patients in the age group in which it was validated (ie, patients diagnosed with diabetes at age 25–65 years).

• It is likely that the low rate of cardiovascular events in the FDS (4.8% with at least one myocardial infarction, and 2.9% with at least one stroke)1 affects the accuracy of the results obtained with the UKPDS risk engine.

• The Framingham risk score has already been found to vary considerably in accuracy between populations, with predicted-to-observed ratios ranging from underprediction of 0.43 to overprediction of 2.87.2 Further, the UKPDS risk engine recently overestimated the risk of cardiovascular disease events in a UK general practice population.3

In purely pragmatic terms, most patients with type 2 diabetes aged over 50 years are at “high risk” for cardiovascular events (cardiovascular risk of more than 20% over 10 years),4 and the UKPDS risk engine is unlikely to influence prescribing practice significantly. However, we have found the engine to be a useful educational tool for explaining risk to patients. Even if the UKPDS risk engine is not optimally calibrated, the FDS analysis revealed that the coronary heart disease risk equation had modest discrimination (area under the receiver operating characteristic curve [AUC], 0.68), and the stroke risk equation had good discrimination (AUC ≥ 0.86),1 identifying those at highest risk.

We believe that, rather than being irrelevant in Australians, the UKPDS risk engine continues to identify those at highest risk for cardiovascular events, operates well within its validated age group, and provides a motivational tool for encouraging changes in patient behaviour. Until a large dataset is pooled from various Australian studies, we believe the UKPDS risk engine should not be discarded.

In reply: We thank McCallum and colleagues for their comments. In relation to their specific points:

• The Fremantle Diabetes Study (FDS) cohort is representative and drawn from a typical Australian urban centre.1 The 488 cardiovascular disease-free FDS participants with type 2 diabetes who were aged 25–65 years at both diagnosis and study entry had 22 coronary heart disease (CHD) events compared with 72 predicted, with a similar area under the receiver operating characteristic curve (AUC) to that for all 791 patients who were included in the analysis2 (0.66 v 0.68). Calibration indicated significant discrepancies between predicted and actual outcomes (P ≤ 0.02), and positive predictive values were low (≤ 3.5%). Therefore, restricting our patient sample to a “UKPDS” cohort did not alter our conclusions.

• We agree that the low observed CHD event rate in the FDS compared with that predicted by the UKPDS risk engine undermines its validity in Australians with type 2 diabetes. There was a similarly low CHD event rate in the FIELD study, which included many Australasians.3 Contemporary diabetes care clearly differs from that during the Framingham Study and UKPDS.

• Although the study cited by McCallum and colleagues, in a UK general practice population, is not strictly comparable to our study, it also found that the UKPDS cardiovascular disease risk engine performed only moderately (AUC, 0.72).4

Accurate risk prediction should be a basis for cost-effective care. We have developed an FDS risk calculator which should improve clinical management for Australians with diabetes.5