208 Chapter 9 infection, and the duration of mechanical ventilation, the P/F ratio and mortality were compared between patients who underwent surgery in a season with high ILI prevalence and who underwent surgery in a season with low ILI prevalence. Similar to the viral co-infection study, if this study was done by testing consecutive patients for the presence of a viral respiratory disease, a large sample size would be required. Moreover, as patient inclusion would be most effective in ILI season to increase the probability of finding patients who have a respiratory viral infection, this means that the study would be time-consuming. Utilising an existing source of observational data was an efficient method to further our understanding of the development of pulmonary complications and mortality after elective cardiac surgery in a season with high and low prevalence of ILI. In Chapter 6, several limitations of our approach were discussed. By extracting data from EHRs from multiple hospitals, as described in Chapter 2, the amount of information on possible confounders, that were not available in the NICE data set, could potentially be improved. However, as RT-PCR testing prior to elective surgery is unlikely to be done, this still means a proxy like ILI season would need to be used. Therefore, a prospective study, in which asymptomatic patients are routinely tested for viral respiratory infections at the start of the surgical procedure, may be needed to fully understand the relationship between viral respiratory infection and outcomes after cardiac surgery. Interestingly, a prospective multicentre study (VIRUS-ATTAC study, ClinicalTrials.gov identifier: NCT04562207) is currently being conducted in five hospitals in France. This study has completed recruitment in March 2023, but no results were available at the time of writing. Patients undergoing elective cardiac surgery were eligible for inclusion and were tested for an influenza virus infection. This prospective study is an important step to increase our understanding of the relationship between asymptomatic viral infection and pulmonary complications after cardiac surgery. Challenges that occur when trying to infer causality from observational data are generally well-known and have been described extensively26. In short, causal inference is possible when there are no differences in characteristics between the treated and untreated groups, conditional on confounders (conditional exchangeability), when the possibility of receiving treatment is non-zero in all groups (positivity) and when the treatment is clearly defined (consistency)40. These are called the identifiability criteria40. Below, considerations for the likeliness of a causal relation in the different studies in part II of this thesis are given. While in RCTs it is possible to achieve exchangeability between groups, observational studies mostly rely on conditioning for confounding variables to achieve conditional exchangeability42. In the viral co-infections chapter (Chapter 4), inclusion in the study (i.e., being tested for a viral co-infection) was dependent on disease severity. In other words, there was no exchangeability between the sampled and unsampled participants. In an aim to correct for this lack of exchangeability, inverse probability
RkJQdWJsaXNoZXIy MTk4NDMw