Reader Comments

Post a new comment on this article

Reducing the gap between noise and signal in a MR study

Posted by Sandeep_Grover on 21 Jul 2017 at 12:14 GMT

I read the recent article from Noyce et al. 2017 exploring the possibility of protective role of BMI on Parkinson’s disease using the modern approach of Mendelian Randomization (MR) with great interest. The article has been very well presented and provides support to some of the earlier published observational studies. Without undermining the impressive work done by the authors, I express some concern which may have been overlooked by the authors and that could have helped readers to better interpret the findings reported in the article.
Firstly, I note that authors have reported extraction of the genetic instrument from Locke et al. 2015. The current article claims that Locke et al. explored association of BMI in a GWAS of 339224 individuals of European descent. The GWAS study reported 97 significant signals from which the author of the current article could finally prioritize a set of 77 index SNPs explaining more than 2% variance in BMI for estimating the causal estimate using IVW and MR-Eggers approach. A careful inspection of the original study by Lock et al. 2015 on BMI revealed 332154 individual of European descent with only 77 loci surpassing significant threshold (p=5x 10-8 ). The remaining 20 variants were the result of either merging of the Europeans with Non-Europeans or as a result of signals originating from secondary analyses. Although such an approach to have more number of variants as a part of instrument variable is an attractive approach to increase the power of the study, it increases the risk of incorporating pleiotropic variants and relying on statistical tests to support absence of noise should always be the last resort. Another discrepancy was further observed in the Table 1 providing a descriptive summary of prioritized variants where P-values of some of these additional SNPs do not appear to match the corresponding effect estimates and standard errors as reported in the original GWAS study on BMI. It appears that authors have used effect estimates for all prioritized variants from the pooled population of Europeans and Non-Europeans. In such a scenario, it is recommend mentioning the minor allele frequency providing a comparison between the populations taken from different GWAS studies exploring the same variants with exposure and outcome in respective studies. And lastly but most importantly, a sensitivity analysis could have been performed by the authors to show the cumulative impact of these additional SNPs on the overall result.

A clear description of the method of prioritizing the SNPs including careful reporting of ethnicity and various SNP specific parameters utilized for the study is strongly recommended which could be extremely useful for someone not only attempting to replicate the findings but also in an ethically sound manner.

No competing interests declared.
report a concern respond to this posting