Abstract
The causal effect of an exposure on an outcome of interest in an observational study cannot be estimated directly if the confounding variables are not controlled. Many approaches are available for estimating the causal effect of an exposure. In this manuscript, we demonstrate the advantages associated with using inverse probability weighting (IPW) and doubly robust estimation of the odds ratio in terms of reduced bias. IPW approach can be used to adjust for confounding variables and provide unbiased estimates of the exposure’s causal effect. For cluster-structured data, as is common in animal populations, inverse conditional probability weighting (ICPW) approach can provide a robust estimation of the causal effect. Doubly robust estimation can provide a robust method even when the specification of the model form is uncertain. In this paper, the usage of IPW, ICPW, and doubly robust approaches are illustrated with a subset of data with complete covariates from the Australian-based National Bovine Respiratory Disease Initiative as well as simulated data. We evaluate the causal effect of prior bovine viral diarrhea exposure on bovine respiratory disease in feedlot cattle. The results show that the IPW, ICPW and doubly robust approaches would provide a more accurate estimation of the exposure effect than the traditional outcome regression model, and doubly robust approaches are the most preferable overall.
Highlights
In veterinary science, the goal of many observational studies is to estimate the causal effect of exposures on disease outcomes
We only focus on three methods to estimate the average causal effect: inverse probability weighting (IPW), inverse conditional probability weighting (ICPW) and the doubly robust approach, and we refer them as the causal inference estimation approaches throughout the paper
The rationale for this paper is to introduce causal inference estimation approaches to veterinary researchers using realistic example data and to illustrate the advantages when compared to traditional outcome regression model-based approaches to estimation
Summary
The goal of many observational studies is to estimate the causal effect of exposures on disease outcomes. Comparing the estimates of effect obtained from statistical causal inference methods used to adjust for confounding in observational studies can be classified into two categories: G-methods and stratification-based methods. G-methods include IPW, standardization and G-estimation, where the conditional exchangeability has been used in subsets defined by covariates to estimate the causal effect of exposures on outcomes in the entire population (marginal). Stratification-based methods include stratification, restriction and matching, but the conditional exchangeability is used in subsets defined by covariates to estimate the association between exposures and outcomes in those subsets only (conditional) [1]. The rationale for this paper is to introduce causal inference estimation approaches to veterinary researchers using realistic example data and to illustrate the advantages (reduced bias in the estimation of the average causal effect) when compared to traditional outcome regression model-based approaches to estimation
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
