Using Mendelian randomization to investigate the causal effect of diet on prostate cancer


Year:

Session type:

Sarah Lewis1
1University of Bristol

Abstract

Background

Many dietary factors have been hypothesised to be risk factors for prostate cancer. However, despite a phenomenal amount of observational studies the evidence for many is still weak or limited. Even when there appears to be strong evidence for a particular risk factor, randomized controlled trials designed to test causality have given null results, such as the SELECT trial which was designed to test whether selenium and or vitamin E were risk factors for prostate cancer. The differences between observational studies and trials have been attributed to measurement error, reverse causation and confounding.

Method

We have performed two sample Mendelian randomization as an alternative to randomized trials to investigate whether selenium, lycopene, calcium, vitamin D and folate are risk factors for Prostate cancer. We used publicly available genome wide study (GWAS) data from a large cancer consortium (PRACTICAL) and from other GWAS studies of our exposures.

Results

Based on a very large sample size (up to 70,000 cases) we found no strong evidence that any of these dietary factors are causal risk factors for prostate cancer overall. Although, there was some weak evidence that calcium may reduce risk and that selenium may increase the risk of advanced disease. Some analyses were hindered by weak instruments (lycopene), for other analyses (selenium) the instruments were strong and the results mirrored those of large and expensive RCTs. I will present the findings from these analyses and discuss some of the strengths and limitations of using Mendelian randomization in this context.  

Conclusion

We suggest that resources for cancer research would be better diverted elsewhere rather than spent on yet more observational studies of these risk factors. In addition, it would be a good idea for future prostate cancer studies to focus on advanced prostate cancer or disease progression rather than incidence of disease.