Are competing-risk models superior to standard Cox models for predicting cardiovascular risk in older adults? Analysis of a whole-of-country primary prevention cohort aged >= 65 years

Background Both cardiovascular disease (CVD) risk and deaths from non-CVD causes, which may preclude a CVD event, increase with age. We evaluated whether accounting for the competing risk of non-CVD death improves the performance of CVD risk-prediction equations in older adults. Methods All New Zealanders aged >= 65 years in 2012 without a prior CVD hospitalization were identified by anonymized linkage of eight routinely collected national health data sets. Sex-specific equations estimating the 5-year risk of a fatal or non-fatal CVD event were constructed using standard Cox and Fine-Gray (competing-risk) approaches. The pre-specified predictors were: age, ethnicity, deprivation level, diabetes, atrial fibrillation and baseline preventive pharmacotherapy. Model performance was evaluated by assessing calibration and discrimination in the overall cohort and in ethnic and age-specific subgroups. Results Among 360 443 people aged >= 65 years with 1 615 412 person-years of follow-up, 14.6% of men and 12.1% of women had a first CVD event, whereas 8.5% of men and 7.6% of women died from a non-CVD cause. Standard Cox models overestimated the mean predicted the 5-year CVD risk by similar to 1% overall and by 5-6% in the highest risk deciles. The mean predicted CVD risk from the Fine-Gray models approximated the observed risk overall, although slight underestimation occurred in some subgroups. Discrimination was similar for both models. Conclusions In a whole-of-country primary prevention cohort aged >= 65 years, standard Cox models overestimated the 5-year CVD risk whereas the Fine-Gray models were generally better calibrated. New CVD risk equations that take competing risks into account should be considered for older people.

Automated summary

This study investigated whether accounting for the competing risk of non-CVD death improves the performance of cardiovascular disease risk-prediction models in older adults. The results showed that the Fine-Gray models were generally better calibrated compared to the Cox models.