LAUSR

Background Cardiovascular disease (CVD) risk prediction equations are primarily used in clinical settings to inform individual risk management decisions. We sought to develop and validate alternative equations derived solely from linked routinely collected national health data that could be applied countrywide to inform population health planning. Methods Individual-level linkage of eight administrative health datasets identified all New Zealand residents aged 30-74 years in contact with publicly funded health services during 2006 with no previous hospitalizations for CVD or heart failure, and with complete data on eight pre-specified predictors. The linked health datasets encompassed demographic characteristics, hospitalizations, outpatient visits, primary care enrolment, primary care reimbursement, community laboratory requests, community pharmaceutical dispensing and mortality. Sex-specific Cox models were developed to estimate the risk of CVD death or hospitalization within 5 years and included sex, age, ethnicity, level of deprivation, diabetes, previous hospitalization for atrial fibrillation and baseline preventive pharmacotherapy (blood-pressure-lowering, lipid-lowering and antiplatelet/anticoagulant medications) as predictors. Calibration and discrimination were assessed in the whole cohort, in 15-year age bands, in different ethnic groups, in quintiles of deprivation, according to baseline dispensing of pharmacotherapy, and in regional sub-populations. Results First CVD events occurred in 62 031 of the 1 746 695 people during 8 526 024 person-years of follow-up (mean = 4.8 years). Median 5-year CVD risk was 1.1% in women and 2.6% in men. In both sexes, the risk equations were well calibrated throughout the risk range and had good risk discrimination in the national, regional and ethnic populations, within 15-year age bands, in deprivation quintiles and according to baseline medication dispensing. Conclusions Robust policy-focused CVD risk equations can be developed solely from administrative health data to inform population health planning, and will complement CVD primary prevention at the individual level using clinical risk tools. Similar policy-focused equations could be replicated in countries and regions with linked administrative health datasets.