215 Summary and general discussion 9 Another method to facilitate more efficient causal inference is by removing noise generated by including patients with divergent disease processes. In Chapter 8, latent class analysis was used to detect sub-phenotypes in patients with SAB. Five subphenotypes with different clinical characteristics, outcomes and treatment responses were identified. No (beneficial) treatment effect may be identified in heterogeneous diseases, because some subgroups of patients benefit, while other experience harm, leading to no average beneficial treatment effect. LCA is an example of a statistical method to identify subgroups that have different clinical characteristics. These subgroups may response differently to treatment. As was shown in the case of ARDS55, stratification of patients into subgroup may aid in the understanding of the disease, and subsequently allows for more targeted inclusion into clinical trials, which can increase the probability of detecting a beneficial treatment effect. The increasing availability of large data sets created opportunities to detect new patterns and associations16. Focusing on the field of infectious diseases, there are several areas in which LCA could increase our understanding of disease and response to therapy. Possible uses for LCA can be found in heterogeneously presenting diseases or syndromes, and in identifying subgroups who respond differently to treatment, such as severe influenza virus infection and response to corticosteroid treatment56. Every year, 290.000-645.000 people die from respiratory disease caused by influenza57. Most studies looking at the effect of corticosteroid use on mortality in severe influenza infection were observational, and suffer from significant bias as the indication for treatment with corticosteroid was unknown58. It is possible that, similar to ARDS55, some patient groups benefit, while others do not benefit and potentially experience harm. If this is the case, LCA could help define different, previously unrecognised subgroups. In the third part of this thesis, two different approaches were used to improve causal inference. S/F94 improved the assessment of the definitive outcome (mortality) compared to other commonly used outcome measures in COVID-19 clinical research. As discussed in Chapter 7, demonstrating trial level surrogacy would be an important future objective. Second, it could be that S/F94 could be used as an outcome measure in other viral respiratory infections. Similar to COVID-1959, if impairment of pulmonary oxygenation function indicates disease progression and is mechanistically linked to death, it may have similar benefits in improving causal inference. Moreover, using latent class analysis, the identification (and reproduction) of clinically relevant subgroups in SAB may aid future causal inference, by targeted inclusion in clinical trials. There are many ways in which efficient use of available resources and data can improve the efficiency of future research. Learning about and utilising these new methods, in collaboration with specialists from other fields, will create opportunities to advance our understanding of disease and ultimately provide better, personalised care for patients.
RkJQdWJsaXNoZXIy MTk4NDMw