Macrolide combination therapy for hospitalised CAP patients? An individualised approach supported by machine learning

Abstract

@ERSpublicationsThe article by König and co-workers paves the way for the selection of a subset of patients with CAP in whom combination initial therapy including macrolides could improve outcomes using a new and interesting mathematical approach http://bit.ly/2p8RdCDCite this article as: Garcia-Vidal C, Cillóniz C, Torres A. Macrolide combination therapy for hospitalised CAP patients? An individualised approach supported by machine learning. Eur Respir J 2019; 54: 1902111 [https://doi.org/10.1183/13993003.02111-2019.The use of macrolides in combination with betalactams to improve outcomes in patients with community-acquired pneumonia (CAP) remains a topic of controversy, mainly because most of the evidence comes from observational studies rather than from randomised clinical trials (RCTs). Some recent studies have suggested that macrolides are effective in patients with Streptococcus pneumoniae infection plus a high systemic inflammatory response [1]. However, macrolides can be harmful [2] and there is a clear need to identify CAP phenotypes that would benefit from macrolides without suffering negative effects. We are unlikely to solve this question using conventional study designs; rather, we need to explore new technologies, among them the use of models obtained with machine learning methodology.In an interesting paper, KÖNIG et al.[3] used new statistical tools, combining propensity matching score with tree decision models (not specified) using non-supervised machine learning (also not specified) to identify patients in whom macrolide use would be beneficial. They found that a simple decision tree of pneumonia patients' characteristics, comprising chronic cardiovascular and chronic respiratory co-morbidities and leukocyte counts in the respiratory secretion at admission, was able to identify the ones who would achieve better outcomes with the addition of macrolide treatment. The results are important and open up new perspectives for conducting optimised, randomised trials to demonstrate the beneficial role of macrolides in patients with CAP.But probably one of the most interesting features of this paper is the incorporation of new statistical tools in medical research. The machine learning approach involves the study of algorithms used by computer systems to perform a specific task, most of them associated with predictions, without using explicit instructions that rely on patterns. Machine learning builds a mathematical model based on training data in order to make predictions in a test dataset. It includes a heterogeneous range of approaches, such as