Predictive performance of cardiovascular disease risk prediction models in older adults: a validation and updating study.

Abstract

Background: Current cardiovascular disease (CVD) risk prediction models tailored for older adults are inadequate. This study aimed to validate, update and assess the utility of widely used CVD risk prediction models including American College of Cardiology/American Heart Association, 2008 Framingham, GloboRisk, National Vascular Disease Prevention Alliance and Predict1 originally developed for middle-aged population, as well as an age-specific Systematic COronary Risk Evaluation 2-Older Person model, in Australian and the US community-dwelling older adults.

Methods: Participants, without history of CVD events, dementia or physical disability, enrolled in the ASPREE (ASPirin in Reducing Events in the Elderly) clinical trial and ASPREE-eXTention observational post-trial follow-up, were considered for CVD risk prediction. The main outcome was predicted CVD risk from adjudicated CVD events. The performance of the original, recalibrated (adjusting models' intercept and slope) and updated (adjusting models' coefficients) models was evaluated by discrimination (C statistic), calibration (calibration plots) and clinical utility (decision curves). Models were extended by incorporating predictors including serum creatinine, depression and socioeconomic status index (Index of Relative Socio-economic Advantage and Disadvantage, IRSAD) into models' equation, and the changes in discrimination were evaluated.

Results: Among 15 618 adults (mean age 75 (4.4) years), 520 men and 498 women experienced CVD events over a median follow-up of 6.3 (IQR: 5.2-7.7) years. Following updating, the discrimination power of models increased for both sexes (C statistics ranged 0.62-0.64 for men and 0.68-0.69 for women). Updated models indicated good calibration, with an added net benefit at the risk thresholds ranging from 4%-10% for women to 5%-12% for men. Incorporating IRSAD, depression and serum creatinine did not improve CVD risk discrimination of updated models.

Conclusions: Updating models, by adjusting model coefficients to better reflect the characteristics and risk factors of older adults, improves CVD risk prediction in a large cohort of relatively healthy Caucasian population aged 70+. Further external validation in diverse older populations including those with frailty and multimorbidity is recommended before clinical implementation.