52 Does testosterone shorten life? triangulation of a much-debated association

Objectives Conflicting data about the association between testosterone (T) and mortality risk in men has fuelled intense public controversy, including regulatory actions. Given the lack of definite evidence, the integration of results from several different methodological approaches might strengthen the conclusions about the potential causal influence of T on male mortality risk. Method The present triangulation approach was constructed in accordance with suggested criteria for triangulation in aetiological epidemiology, including: (1) the comparison of results from two or more different approaches, (2) addressing the same underlying causal question, (3) with key sources of bias and their expected direction explicitly acknowledged. Individual knowledge of literature was combined with a PubMed search to identify relevant studies. Risk estimates from four different methodological approaches addressing the association between T and mortality were integrated: (1) the largest prospective cohort study, (2) meta-analyses of prospective cohort studies, (3) genetic Mendelian randomization (MR), and (4) meta-analyses of RCTs. Based on the expected direction of potential bias sources, stronger effects in the prospective cohort study and their meta-analyses (residual confounding likely), compared to the MR approach and meta-analyses of RCTs (major bias unlikely) were anticipated. Results Observational evidence from the largest cohort study (n=2,639), as well as the meta-analysis of 11 cohort studies (n=16,184) suggests an inverse association between exogenous T and mortality (hazard ratio: 1.65, 95% confidence interval (95% CI): 1.29–2.12) and relative risk (RR): 1.35, 95% CI: 1.13–1.62, respectively). Mendelian randomization analysis reported a non-significant association between T and mortality (RR: 1.23, 95% CI: 0.40–3.82). Finally, the meta-analysis of 20 RCTs showed no association between endogenous T and mortality in men (odds ratio: 0.88, 95% CI: 0.55–1.42). Plotting the different risk estimates against their potential risk of bias yielded a graded decrease in magnitude of effect and significance level (Figure 1). Varying durations of exposure across the different approaches, ranging from weeks (RCT), to years (cohort studies) and lifetime exposure (MR), are suspected to explain the differences in magnitude of effect only to a small degree. Conclusions Integrating evidence from four different methodological approaches, the present triangulation provides little support for a causal association between T and mortality risk in men. As the triangulation approach provides a qualitative assessment of the strength of evidence, further quantitative research would strengthen this conclusion and establish stronger evidence.