Abstract
Pooling (or combining) and analysing observational, longitudinal data at the individual level facilitates inference through increased sample sizes, allowing for joint estimation of study- and individual-level exposure variables, and better enabling the assessment of rare exposures and diseases. Empirical studies leveraging such methods when randomization is unethical or impractical have grown in the health sciences in recent years. The adoption of so-called "causal" methods to account for both/either measured and/or unmeasured confounders is an important addition to the methodological toolkit for understanding the distribution, progression, and consequences of infectious diseases (IDs) and interventions on IDs. In the face of the Covid-19 pandemic and in the absence of systematic randomization of exposures or interventions, the value of these methods is even more apparent. Yet to our knowledge, no studies have assessed how causal methods involving pooling individual-level, observational, longitudinal data are being applied in ID-related research. In this systematic review, we assess how these methods are used and reported in ID-related research over the last 10 years. Findings will facilitate evaluation of trends of causal methods for ID research and lead to concrete recommendations for how to apply these methods where gaps in methodological rigor are identified. We will apply MeSH and text terms to identify relevant studies from EBSCO (Academic Search Complete, Business Source Premier, CINAHL, EconLit with Full Text, PsychINFO), EMBASE, PubMed, and Web of Science. Eligible studies are those that apply causal methods to account for confounding when assessing the effects of an intervention or exposure on an ID-related outcome using pooled, individual-level data from 2 or more longitudinal, observational studies. Titles, abstracts, and full-text articles, will be independently screened by two reviewers using Covidence software. Discrepancies will be resolved by a third reviewer. This systematic review protocol has been registered with PROSPERO (CRD42020204104).
Highlights
Pooling and analysing observational, longitudinal data at the individual level facilitates inference through increased sample sizes, allowing for joint estimation of studyand individual-level exposure variables, and better enabling the assessment of rare exposures and diseases
We will apply MeSH and text terms to identify relevant studies from EBSCO (Academic Search Complete, Business Source Premier, CINAHL, EconLit with Full Text, PsychINFO), EMBASE, PubMed, and Web of Science. Eligible studies are those that apply causal methods to account for confounding when assessing the effects of an intervention or exposure on an infectious diseases (IDs)-related outcome using pooled, individual-level data from 2 or more longitudinal, observational studies
Growing use of these methods in epidemiology can enhance the internal validity while maintaining the value of an observational cohort’s external validity. They do so by improving our ability to control for observed and/or unobserved confounders. These causal methods include but are not limited to instrumental variables (IV), which, in simple terms, ‘looks for a randomized experiment embedded in the observational study’ [8]; propensity scores (PS), which can be implemented in several ways, including weighting, matching, or subclassification, e.g., to adjust for covariates, allowing the exposed and causal inference methods in pooled longitudinal observational infectious disease studies—A protocol unexposed to be more comparable; difference-in-differences (DiD) models are well-suited for pre-/post-interventions or data with shocks in between; and regression models are widely used in medicine to control for observed confounding
Summary
Pooling (or combining) and analysing observational, longitudinal data at the individual level facilitates inference through increased sample sizes, allowing for joint estimation of studyand individual-level exposure variables, and better enabling the assessment of rare exposures and diseases. Empirical studies leveraging such methods when randomization is unethical or impractical have grown in the health sciences in recent years. No studies have assessed how causal methods involving pooling individuallevel, observational, longitudinal data are being applied in ID-related research. Findings will facilitate evaluation of trends of causal methods for ID research and lead to concrete recommendations for how to apply these methods where gaps in methodological rigor are identified.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
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 call
