Exploring inconsistencies between observational and experimental studies of selenium and diabetes risk.

Background: Observational and experimental epidemiologic studies that have addressed the relation between intake of the trace element selenium and cancer risk have yielded strongly conflicting results, as recently reported by a Cochrane review. Most observational studies suggest an inverse association, while randomized controlled trials (RCTs) have indicated a null or direct relation. Little is known about the replication of such inconsistencies when dealing with the risk of other chronic disease. Objectives: We investigated the results of observational and experimental studies linking selenium exposure to the occurrence of type 2 diabetes. Methods: After a literature search we identified 12 observational studies (8 cross-sectional and 4 cohort) and 5 RCTs. Using a random-effects model, we computed the summary relative risk (RR) of type-2 diabetes along with its 95% confidence interval (CI) in subjects with the highest versus the lowest selenium exposure category in observational studies, and in subjects allocated to selenium compared to placebo in the RCTs. Results: Summary RRs were 1.98 (95% CI 1.22-3.23) and 1.13 (0.15-8.45) for cross-sectional studies using serum and toenail selenium for exposure assessment, respectively. Cohort studies based on toenail selenium yielded a summary RR of 0.78 (0.62-0.98), while the only study assessing dietary selenium intake gave a RR of 2.39, (1.32-4.32). For RCTs, summary RR was 1.10 (1.00-1.21) among selenium-supplemented versus placebo. The distinctive feature of the two observational studies (one cross-sectional and one prospective) that failed to find an excess diabetes risk associated with higher selenium exposure was that the subjects were health professionals. Age, gender, study area and other demographic characteristics did not appear to have influenced the results. Conclusions: These results suggest that the ability of observational studies to predict results of RCTs when addressing the health effects of selenium may differ on the basis of the outcome studied (diabetes versus cancer) as well as the indicator used for exposure assessment and the type of population under study.