Activity Number:

88
 SPEED: Causal Inference and Related Methodology Part 2

Type:

Contributed

Date/Time:

Sunday, July 28, 2019 : 5:05 PM to 5:50 PM

Sponsor:

Section on Statistics in Epidemiology

Abstract #307509


Title:

Testing for Weak Instruments in Two Sample Summary Data Multivariable Mendelian Randomisation

Author(s):

Eleanor Sanderson* and Jack Bowden

Companies:

University of Bristol and University of Bristol

Keywords:

Mendelian Randomisation;
Ftest;
Weak Instruments;
Instrumental Variables

Abstract:

Multivariable Mendelian randomisation (MVMR) is a form of instrumental variable estimation which utilises genetic variants to estimate the direct causal effects of multiple exposures on an outcome. A key assumption required for consistent MVMR estimation is that the genetic variants used as instruments can strongly predict each exposures. For MVMR this means that the genetic variants must be able to predict each exposure conditional on having predicted all of the other exposures included in the model. In analysis using individual level data this assumption can be tested using a conditional Fstatistic. We consider how to define weak instruments in the twosample summary data MVMR setting and show that a summary data Fstatistic, calculated from an adjusted Q statistic could be used to test for weak instruments. We show that this summary data Fstatistic has the same distribution as the individual level conditional Fstatistic but that it depends on data that is generally not reported in standard summary statistics. We use simulations to explore how sensitive this summary data Fstatistic is to the missing data and explore strategies for estimation when this data is not available.
