209 Summary and general discussion 9 weighting was used. In inverse probability weighting, more weight is given to certain individuals, to make the groups more balanced. Additionally, there were some differences between the exposed (co-infected) and unexposed (mono-infected) group for several confounders, such as age. We conditioned on these variables to achieve conditional exchangeability. However, there were also unmeasured confounders, such as vaccination status. As with all observational studies, there is an (inherent) possibility of residual confounding, so it is possible that (conditional) exchangeability was not achieved completely. In the IL-6 treatment chapter (Chapter 5), preference for a specific treatment from a physician was prevented, as treatment groups were based on the time of hospitalisation. However, there are some confounding factors that influence the timing of SARS-CoV-2 infection, such as age and sex. While we were able to correct for some of these factors, we were unable to correct for all of them, which makes residual confounding likely (e.g. vaccination status, SARS-CoV-2 variant). In Chapter 6, the exposure was the season in which surgery took place, which was dependent on the ILI prevalence in the community. While there were probably no variables (measured or unmeasured) that influence season, this study is slightly more complex, because season was used as a proxy for respiratory viral infection. The assumptions in this study can be visualised in a Directed Acyclic Graph (DAG), see Figure 1. If this DAG were true, season could be used as an instrumental variable to study the relation between respiratory infection and duration of mechanical ventilation. Instrumental variables can be used to estimate causal effects in observational studies43. However, there are several issues, that were also shortly discussed in Chapter 6. First of all, an instrumental variable needs to predict ‘treatment’ status, in our case the respiratory infection43. While there is a clear association between ILI season and respiratory infections, this association is weak. Not everyone who underwent surgery in ILI season will have had a respiratory infection, and while rare, Duration of mechanical ventilation Respiratory infection Confounders Season Figure 1: Simplified Directed Acyclic Graph for the study in Chapter 6
RkJQdWJsaXNoZXIy MTk4NDMw