A meta-regression analysis of insulin and cardiovascular disease

Article type
Ruige JB, Assendelft WJJ, Dekker JM, Kostense PJ, Heine RJ, Bouter LM
Objective: To summarize the strength of the longitudinal relationship between hyperinsulinaemia and cardiovascular disease from the available literature, and to identify study characteristics which modify this relationship.

Methods: Articles were identified using Index Medicus and Embase and by citation tracking. Eligible studies were prospective population-based (cohort) studies and nested case-control studies on fasting and/or non-fasting insulin levels and coronary hearth death, myocardial infarction or electrocardiographic abnormalities. Data were extracted with regard to fasting or non-fasting insulin measurements, 'type of outcome studied', correction for confounding, insulin assay used, follow-up years, age, gender and genetic background. Associations of insulin and cardiovascular disease were re-expressed in a uniform manner (an estimate of relative risk and 95% confidence interval) to be used in meta-regression analysis.

Results: Of fourteen potentially eligible studies, eight provided enough information. Overall, a weak positive association was found. The meta-regression analysis revealed that a difference of 50 pmol/1, the difference between the 25th and the 75th percentile of the fasting insulin distribution, was accompanied by an overall estimate of the relative risk (95% confidence interval) of 1.22 (1.09 - 1.36) for fasting insulin. Likewise, a difference of 250 pmol/l was accompanied by an estimated relative risk of 1.21 (1.11 - 1.32) for non-fasting insulin. Race, 'the type of outcome studied', and the mean age of the study population were identified as study-characteristics, which borderline significantly modified the relationship of non-fasting insulin level and cardiovascular disease.

Discussion: A weak relationship was shown between insulin and the occurrence of cardiovascular disease. Study characteristics which probably affected this relationship were race, "type of outcome studied' and mean age of the study population.