Applications and reporting of causal inference modelling in infectious disease studies: A systematic review

Causal inference is increasingly employed in infectious disease (ID) epidemiology. Despite the increasing adoption of causal inference methods in infectious disease research, there has been no comprehensive review of their implementation trends, estimation approaches, and key specifications. A systematic examination of how these methods were being applied in practice could identify both successful strategies and common pitfalls. This systematic review aimed to describe the usage and reporting of causal methods in observational ID studies. The applications of causal methods in the analyses of ID observational data were identified from systematic searches of PubMed, Medline, Web of Science, and Scopus. Our analysis focused on detailing the adoption trends of causal inference methods and assessing the comprehensiveness of their reporting and publication between 2010 and 2023. Of the 172 studies, the majority utilised propensity score-based methods (n = 133, 77 %). We identified only 39 studies that explicitly described the use of causal frameworks and employed variations of causal analyses. The most common reason for using causal methods was to address time-varying variables that are prominent in ID research. Consequently, a common approach used was inverse probability treatment weighting with the marginal structural model; additionally, targeted maximum likelihood estimation has become popular in minimising bias.

There is substantial variation in reporting causal methods in ID research. Development of reporting guidelines is needed for clear reporting alongside training on how to use and appraise applications of causal inference in observational ID research. This is particularly important for ID modelling, where time-varying factors and complex transmissions and dynamics of treatment often necessitate complex modelling approaches.