An asymptotically unbiased estimator of exposed versus non-exposed odds ratio from reported dose-response data

Abstract

Summary effect measures in meta-analysis of published epidemiological cohort or case control studies are often based on odds ratios reported for several exposure levels with varying arrangements and number of levels across primary studies. Usually only two-way contingency tables together with exposure specific adjusted odds ratios and corresponding standard errors are presented in articles. An asymptotically unbiased estimate of exposed versus non-exposed adjusted odds ratio from reported dose-response data is proposed. This estimate is based on the weighted sum of the exposure specific odds ratios, with the prevalences of the control group as weights. Large sample variance is derived accounting for the dependency between exposure specific adjusted odds ratios. The exposed versus non-exposed adjusted odds ratio could then be used in systematic reviewing.