Prediction of cardiovascular disease risk in a cohort of older Chinese adults in Hong Kong

Abstract

Introduction: Risk prediction models for cardiovascular disease (CVD) derived from Western cohorts allow stratification of CVD risk based on an individual's risk factors and the population's CVD risk. Chinese have relatively low rates of ischaemic heart disease. The predictive performance of such tools in Chinese populations should be evaluated before application. Project Objectives: We compared the predictive performance of the widely used Framingham Risk Score (FRS) and Globorisk, a new risk prediction model, for adults aged ≥65 years in the Hong Kong Department of Health's Elderly Health Service (EHS) cohort. Methods: The cohort consisted of 66,820 adults aged ≥65 years enrolled from 1998 to 2001. We tracked CVD deaths prospectively through linkage with death registration until 2012, and assumed a ratio of 8:1 for nonfatal to fatal events. We assessed the discrimination and calibration of the FRS and Globorisk among participants without CVD at baseline. Discrimination was measured using the C-Statistic, with 0.5 equating to chance and 1 indicating perfect discrimination. Calibration was measured by the Hosmer-Lemeshow test, with Χ2<20 indicating good calibration. We used a random sample of 20,671 participants to generate a new Globorisk function based on mean risk factor levels and CVD death rates in our cohort, and validated this function in the remaining participants. Results: The FRS overestimated fatal and nonfatal CVD events at 10 years by 7.5% points in men (predicted risk 38.5% vs. observed risk 31.0%; Χ2 367.6) and 3.0% points in women (predicted risk 21.6% vs. observed risk 18.6%; Χ2 258.6). Recalibration using EHS cohort data improved its performance slightly in men (predicted risk 36.1%; Χ2 218.6), but not women (predicted risk 22.2%; Χ2 303.0). The calibration of Globorisk using EHS data overestimated CVD deaths at 10 years by 0.4% point in men (predicted risk 4.5% vs. observed risk 4.1%; Χ2 34.2) and 1.1% point in women (predicted risk 4.0% vs. observed risk 2.9%; Χ2 187.7). The new Globorisk function improved the calibration (Χ2 21.1 in men, 61.1 in women) with moderate discrimination (C-statistic 0.668 in men, 0.700 in women). Conclusions: The FRS and Globorisk overestimated CVD risk by 16% to 19% and 4% to 38% relatively in this cohort of older Hong Kong Chinese despite recalibration. Because our cohort included survivors and case-fatality ratios are decreasing, these models should be used with precaution. New models for CVD with incidence data are needed for CVD prevention and disease burden estimation. Project Number: 12133051