A method for combining results from several prospective epidemiologic studies

Abstract

An approach to the problem of a single study lacking adequate power to examine an important question is to combine the data from several studies to allow a more powerful test of the hypothesis in question. The present paper describes a method used by the International Collaborative Group to pool data on 54,492 men aged 40–69 from 10 population studies in 7 countries to examine the association between cholesterol level and risk of death from cancer. Furthermore, there was a suggestion that the inverse association observed in some studies might be the result of an effect of undetected cancer on cholesterol level, rather than an increase in cancer risk resulting from low cholesterol levels. This paper describes the process for selection of the method used to evaluate this possibility over other potential approaches and the results obtained in the analysis of the International Collaborative Group data that support this hypothesis. This report also indicates how one might expect some of the analyses described to compare with analyses based on a discrete version of the Cox proportional hazards model.