Estimation of time-varying causal effects with multivariable Mendelian randomization: some cautionary notes
International Journal of Epidemiology · 2023
DOI: 10.1093/ije/dyac240 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Epidemiology, 2023. Verfasst von Haodong Tian und 1 weiteren Autor:innen. Beteiligte Einrichtung: MRC Biostatistics Unit, School of Clinical Medicine, University of Cambridge, Cambridge, UK. Hauptthemen laut MEDLINE: Mendelian Randomization Analysis. Die Arbeit wurde bisher 21-mal zitiert (Crossref) und nennt 30 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Wellcome Trust, Department of Health, Medical Research Council.
Abstract (Originalsprache)
Introduction
For many exposures present across the life course, the effect of the exposure may vary over time. Multivariable Mendelian randomization (MVMR) is an approach that can assess the effects of related risk factors using genetic variants as instrumental variables. Recently, MVMR has been used to estimate the effects of an exposure during distinct time periods.
Methods
We investigated the behaviour of estimates from MVMR in a simulation study for different time-varying causal scenarios. We also performed an applied analysis to consider how MVMR estimates of body mass index on systolic blood pressure vary depending on the time periods considered.
Results
Estimates from MVMR in the simulation study were close to the true values when the outcome model was correctly specified: i.e. when the outcome was a discrete function of the exposure at the precise time points at which the exposure was measured. However, in more realistic cases, MVMR estimates were misleading. For example, in one scenario, MVMR estimates for early life were clearly negative despite the true causal effect being constant and positive. In the applied example, estimates were highly variable depending on the time period in which genetic associations with the exposure were estimated.
Conclusions
The poor performance of MVMR to study time-varying causal effects can be attributed to model misspecification and violation of the exclusion restriction assumption. We would urge caution about quantitative conclusions from such analyses and even qualitative interpretations about the direction, or presence or absence, of a causal effect during a given time period.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Autor:innen und Institutionen
- MRC Biostatistics Unit, School of Clinical Medicine, University of Cambridge, Cambridge, UK.
- British Heart Foundation Cardiovascular Epidemiology Unit, Department of Public Health and Primary Care, University of Cambridge, Cambridge, UK.
ORCID-Profile: Haodong Tian, Stephen Burgess
Förderung
- Wellcome Trust
- Department of Health
- Medical Research Council
- United Kingdom Research and Innovation Medical Research Council
- National Institute for Health Research Cambridge Biomedical Research Centre
- Royal Society
- National Institute for Health Research
- Department of Health and Social Care
Zitieren
APA
Tian, H., & Burgess, S. (2023). Estimation of time-varying causal effects with multivariable Mendelian randomization: some cautionary notes. International Journal of Epidemiology. https://doi.org/10.1093/ije/dyac240
Vancouver
Tian H, Burgess S. Estimation of time-varying causal effects with multivariable Mendelian randomization: some cautionary notes. International Journal of Epidemiology. 2023. doi:10.1093/ije/dyac240
BibTeX
@article{Tian2023,
author = {Tian, Haodong and Burgess, Stephen},
title = {Estimation of time-varying causal effects with multivariable Mendelian randomization: some cautionary notes},
journal = {International Journal of Epidemiology},
year = {2023},
doi = {10.1093/ije/dyac240}
}
EUVIMED