Article Text
Abstract
Background Cardiovascular disease (CVD) risk prediction models are primarily used in clinical settings, but may also have potential applications in population health. For example, these models can be used to estimate 10-year CVD mortality within a region. In order to expand the applications of such models, we developed a CVD risk prediction model for population health planning that can account for temporal changes in mortality.
Methods The Evidence for Cardiovascular Prevention from Observational Cohorts in Japan (EPOCH-JAPAN) study is an individual participant data meta-analysis of cardiovascular epidemiology in the Japanese population. This project comprises 15 cohort studies involving 147,465 Japanese people, with a total of 5,543 CVD deaths. Before constructing the CVD risk prediction model for the population, we grouped these cohort studies into three categories according to their year of cohort initiation (1990–1994, 1995–1999, and 2000 or later), and used two groups for model construction (Group 1: 1990–1994; Group 2: 1995–1999). First, we constructed a group-specific CVD risk prediction model based on a Cox model that included age (year), systolic blood pressure (SBP; mmHg), total cholesterol (TC; mg/dl), diabetes (DM), smoking status, and study cohort as independent variables. Next, we checked the homogeneity of the model parameters using hazard ratios, and developed a common parameter using weighted mean values. Finally, to eliminate the temporal discrepancies between the model results and contemporary mortality (as of 2015), we calculated calibrating factors using government vital statistics of Japan.
Results Among the 15 cohort studies included in EPOCH-JAPAN, seven were categorized into Group 1 and three were categorized into Group 2. For CVD in men, the results showed similar hazard ratios for age (1.12), SBP (1.01), and TC (1.00); furthermore, the hazard ratios were similar in both groups for DM (Group 1: 1.43, Group 2: 1.64) and current smokers (Group 1: 1.52, Group 2: 1.55). This homogeneity in model parameters was also observed in women (age: 1.14, SBP: 1.01, TC: 1.00, DM [Group 1: 1.56, Group 2: 2.27], and current smokers [Group 1: 1.63, Group 2: 1.37]), and a common parameter was developed and included in the final CVD risk prediction model. The calibrating factors to adjust for contemporary mortality in 2015 were calculated to be 0.74 in men and 0.55 in women.
Conclusion We constructed a CVD risk prediction model for population health planning that can be used to estimate current mortality in the Japanese population.