Personalized azithromycin treatment rules for children with watery diarrhea using machine learning.

Introduction: We used machine learning to identify novel strategies to target azithromycin to the children with watery diarrhea who are most likely to benefit. Methods: Using data from a randomized trial of azithromycin for watery diarrhea, we developed personalized treatment rules given sets of diagnostic, child, and clinical characteristics, employing a robust ensemble machine learning-based procedure. For each rule, we estimated the proportion treated under the rule and the average benefits of treatment. Results: Among 6,692 children, treatment was recommended on average for approximately one third of children. The risk of diarrhea on day 3 was 10.1% lower (95% CI: 5.4, 14.9) with azithromycin compared to placebo among children recommended for treatment. For day 90 re-hospitalization and death, risk was 2.4% lower (95% CI: 0.6, 4.1) with azithromycin compared to placebo among those recommended for treatment. While pathogen diagnostics were strong determinants of azithromycin effects on diarrhea duration, host characteristics were more relevant for predicting benefits for re-hospitalization or death. Conclusion: The ability of host characteristics to predict which children benefit from azithromycin with respect to the most severe outcomes suggests appropriate targeting of antibiotic treatment among children with watery diarrhea may be possible without access to pathogen diagnostics.