Aims: The aim was to carry out a systematic screening of interactions between the traditional risk factors and to evaluate which interactions are truly relevant for estimation of cardiovascular disease (CVD) risk.
Methods: Cox regression was used in a meta-analysis of five independent, population-based health examination surveys (the National FINRISK Study). End-points were 10-year incidence of coronary heart disease (CHD), ischemic stroke (IS), and CVD in a population free of cardiovascular disease (n = 35,460).
Results: In addition to expected age interactions, systolic blood pressure was found to be a markedly stronger risk factor for CVD (and for CHD) among subjects with normal BMI (BMI < 25: HR 1.42 [1.30-1.55] for one SD increase in systolic blood pressure) when compared to obese subjects (BMI > 30: HR 1.10 [1.01-1.19]) (P < 0.001 for interaction) and among subjects with highest high-density lipoprotein (HDL) (33% tertile: HR 1.43 [1.29-1.58]) when compared to subjects with low HDL (lowest 33% tertile: HR 1.20 [1.13-1.28]) (P < 0.001 for interaction). Interactions improved risk prediction of CVD (cross-validated continuous net reclassification improvement [NRI] 49.4% with 95% CI 44.7%-54.1%, P < 0.0001 and clinical NRI 4.7%, with 95% CI 2.8%-6.5%, P < 0.0001). The C-statistic improved from 0.8438 to 0.8455 (P = 0.010). No significant interaction was associated with the risk of IS.
Conclusions: There are significant effect modifications between major risk factors, and accounting for them leads to significantly more accurate estimation of cardiovascular risk.
Keywords: Cardiovascular disease; coronary heart disease; interactions; risk prediction; stroke.