Development and calibration of the 2023 Australian cardiovascular disease risk prediction equations: a model updating study

Abstract

Objectives: To modify, recalibrate, and test the performance of the New Zealand cardiovascular disease (CVD) risk equations (PREDICT) for application in Australia.

Study design: Model updating study.

Setting, participants: New Zealand residents aged 30-79 years who presented to primary care practices without diagnosed CVD, congestive heart failure, or renal disease and whose CVD risk was assessed using PREDICT software during 1 October 2004 - 31 December 2016. For adapting the PREDICT equations to Australia, Māori, Pacific Islander, Middle Eastern, Latin American, and African people were excluded because of demographic differences between the two countries.

Intervention: The New Zealand PREDICT equations (general and diabetes-specific versions) were recalibrated for Australia, based on differences between Australia and New Zealand in CVD-specific mortality by age group and sex. Body mass index (BMI), ethnic background, and family history of CVD were omitted as variables in the general equation; BMI was retained in the diabetes-specific equation.

Main outcome measures: Risk prediction outcomes: first CVD-specific hospitalisation or death. Model performance measures: calibration of the modified equations, assessed by plotting mean 5-year predicted risk against observed 5-year risk; model discrimination, assessed with the Harrell C statistic.

Results: The modified New Zealand cohort for deriving the general AUS-PREDICT risk equation included 308 478 people (134 137 women, 43.5%); the modified cohort for deriving the diabetes-specific risk equation included 29 219 people with type 2 diabetes (13 246 women, 45.3%). For the general equation, predicted and observed CVD risks were closely aligned across risk deciles; discrimination was good for both women (C-statistic, 0.75; 95% confidence interval [CI], 0.74-0.76) and men (C-statistic, 0.74; 95% CI, 0.73-0.74). For the diabetes-specific equation, predicted and observed CVD risks were also closely aligned across risk deciles; discrimination was acceptable for both women (C-statistic, 0.73; 95% CI, 0.71-0.75) and men (C-statistic, 0.70; 95% CI, 0.68-0.71).

Conclusions: The internal validity of the new Australian CVD risk algorithm, recommended in the 2023 Australian CVD risk assessment and management guidelines, is good and has been recalibrated for use in Australia. The updated risk calculator is a landmark advance in the assessment of CVD risk in Australian primary care.

Keywords: Prevention and control; Risk assessment.