Investigating the transparency of reporting in two-sample summary data Mendelian randomization studies using the MR-Base platform

Abstract

Background Two-sample Mendelian randomization (2SMR) is an increasingly popular epidemiological method that uses genetic variants as instruments for making causal inferences. Clear reporting of methods employed in such studies is important for evaluating their underlying quality. However, the quality of methodological reporting of 2SMR studies is currently unclear. We aimed to assess the reporting quality of studies that used MR-Base, one of the most popular platforms for implementing 2SMR analysis. Methods We created a bespoke reporting checklist to evaluate reporting quality of 2SMR studies. We then searched Web of Science Core Collection, PsycInfo, MEDLINE, EMBASE and Google Scholar citations of the MR-Base descriptor paper to identify published MR studies that used MR-Base for any component of the MR analysis. Study screening and data extraction were performed by at least two independent reviewers. Results In the primary analysis, 87 studies were included. Reporting quality was generally poor across studies, with a mean of 53% (SD = 14%) of items reported in each study. Many items required for evaluating the validity of key assumptions made in MR were poorly reported: only 44% of studies provided sufficient details for assessing if the genetic variant associates with the exposure (‘relevance’ assumption), 31% for assessing if there are any variant-outcome confounders (‘independence’ assumption), 89% for the assessing if the variant causes the outcome independently of the exposure (‘exclusion restriction’ assumption) and 32% for assumptions of falsification tests. We did not find evidence of a change in reporting quality over time or a difference in reporting quality between studies that used MR-Base and a random sample of MR studies that did not use this platform. Conclusions The quality of reporting of two-sample Mendelian randomization studies in our sample was generally poor. Journals and researchers should consider using the STROBE-MR guidelines to improve reporting quality.

Exploring the lifetime effect of children on wellbeing using two-sample Mendelian randomisation

“…2) That there is no sample overlap. The effect of the second assumption is to make weak instrument bias deflationary (41)(42)(43).…”

The UK BiLEVE and Mendelian randomisation: Using multivariable instrumental variables to address “damned if you, dammed if you don’t” adjustment problems

“…For readers less familiar with the MR literature, it is worth noting that the intercept-free weighted regression is equivalent to an inverse variance weighted meta-analysis of the IV Wald ratios for each SNP. In addition, (1) and (2) only used the first 50 simulated SNPs (i.e. those which associated with the exposure), while (3) used all 100 simulated SNPs (i.e.…”

“…One of the most popular resources for conducting MR analyses is the UK Biobank (UKB) (2). The UKB is a large (approximately half a million participants) population cohort study of Britons.…”

Reporting and methodological quality of studies that use Mendelian randomisation in UK Biobank: a meta-epidemiological study

“…Of the 116 articles included in the study, 65% (75) completely reported the genetic variants and weights used to construct the IV, while 9% (10) partially reported this and 27% (31) did not. 40% (46) reported that variants were identified in a different sample as that used in the analysis, or externally weighted, 10% (12) reported only that variants were identified in the same sample and unweighted or identified and weighted in a larger sample which includes the sample used in the MR analysis, 7% (8) reported that variants were identified and weighted from the same sample used in the MR analysis and 43% (50) did not report enough detail to assess this. Whether the variants used had been independently replicated had lower reporting quality across articles; 9% (10) of articles reported that variants were independently replicated, while 8% (9) reported they were partially replicated (eg, replicated in a partially overlapping sample), 13% (15) used unreplicated variants and 71% (82) did not report this.…”