Causal modelling to support medical decision making in the treatment of multidrug resistant infections
Zusammenfassung
Inappropriate treatment of infections, even if only for a few days, may have severe consequences for the patients. In 2017, the WHO - for the first time ever - published a top priority list of pathogens for which novel treatments are urgently needed. Currently, antibiotics are prescribed according to the susceptibility of the pathogen causing the infection. However, medical decision making becomes increasingly complicated. The spread of multidrug resistant pathogens leads to loss of efficacy of the available standard therapies. Moreover, severely ill patients with infections due to (multi)resistant pathogens often receive a complex combination of antibiotics resulting in limited treatment options. For example, in a recent study, we found no evidence for the efficacy of early adequate treatment of ventilator associated pneumonia caused by Pseudomonas aeruginosa among invasively ventilated patients (sample size = 465). Other studies lead to similar dissatisfying conclusions. Evidence for optimal medical decision making is urgently needed. An attractive solution is the identification of personalized interventions that account not only for the type of pathogen, but also for the dynamically changing antibiotic treatment of the patient, the patient’s treatment response and the course of disease over time.
Dynamic treatment regimens (DTRs) can be regarded as a decision tree aiding the clinician to optimize the treatment for an individual patient over the course of time. In contrast to static treatments such as the prescription of a specific antibiotic for a specific pathogen, DTRs account for treatment response and the evolution of patient characteristics by allowing for an adaption of the treatment over the course of time. Causal inference techniques for evidence-based DTRs have become a major field of research in Statistics with main application to chronic diseases. The idea to adapt these methods to identify tailored antibiotic treatments of specific types of hospital acquired infections has been proposed only recently. However, applying these techniques to data on hospitalized patients bears major statistical challenges due to the time dynamics of the data situation and multiple possible outcomes (e.g. death, discharge alive, clinical cure) which result in a competing risks setting. Solid specialized statistical expertise and close collaboration with clinicians is indispensable to identify novel therapies and defeat multidrug resistance.
In this research project statistical inference techniques are adapted and advanced to identify optimal DTRs for infected hospitalized patients.