Triglyceride as a risk factor for ischaemic heart disease in British men: effect of adjusting for measurement error

Objective: To assess the influence of differential precision in the measurement of the correlated variables total cholesterol and high density lipoprotein (HDL) cholesterol on the estimates of the risk of ischaemic heart disease (IHD) associated with plasma triglyceride levels. Design, setting and participants: The Caerphilly Heart Disease Study (CHDS), a prospective cohort study of 2512 middle-aged men living in the town of Caerphilly, south Wales, UK. The results from two sub-studies were used to estimate the degree of measurement imprecision (laboratory error and within-person variation) in triglycerides, total cholesterol and HDL cholesterol. Main outcome measures: Multivariable risk estimates for major IHD calculated from logistic regression analysis, adjusted and not adjusted for measurement imprecision. Major IHD events were defined as death from IHD, clinical non-fatal myocardial infarction or electrocardiographic myocardial infarction. Results: There were 261 men with major IHD events during follow-up. In age-adjusted analyses, taking measurement imprecision into account strengthened associations with IHD for all lipid factors. The odds ratio (OR) for one S.D. increase in triglycerides, ignoring measurement imprecision, was 1.36 (95% confidence interval [95% CI] 1.20–1.55) but 1.57 (95% CI 1.30–1.89) when taking imprecision into account. The standardised odds ratio for triglycerides adjusted for measurement imprecision and the two other lipid factors was 1.35 (95% CI 1.09–1.69). In this model, the triglyceride level showed a stronger association than total cholesterol (OR 1.28; 95% CI 1.05–1.56) and HDL cholesterol (OR for one S.D. decrease 1.20; 95% CI 0.97–1.49). When adding fasting blood glucose and diastolic blood pressure, however, the effect of triglycerides was reduced and ceased to be statistically significant (OR 1.19; 95% CI 0.95–1.49). This was further attenuated upon inclusion of body mass index, smoking status and history of pre-existing IHD. Total cholesterol remained a statistically significant (P