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

Area of Science:

• Computational biology and bioinformatics
• Pediatric infectious diseases
• Clinical trial data analysis

Background:

• Azithromycin is used to treat watery diarrhea in children, but optimal targeting strategies are needed.
• Personalized medicine approaches can improve treatment efficacy by identifying subgroups most likely to benefit.
• Machine learning offers powerful tools for analyzing complex clinical data to develop predictive treatment rules.

Purpose of the Study:

• To develop and validate personalized treatment rules for azithromycin in children with watery diarrhea using machine learning.
• To identify child-specific characteristics that predict benefit from azithromycin treatment.
• To assess the impact of targeted azithromycin treatment on diarrhea duration, re-hospitalization, and mortality.

Main Methods:

• Utilized data from a randomized trial (NCT03130114) involving 6692 children with watery diarrhea.
• Employed an ensemble machine learning procedure to estimate child-level expected treatment benefit based on diagnostic, child, and clinical characteristics.
• Developed personalized treatment rules and estimated the proportion of children treated and average treatment benefits.

Main Results:

• Personalized treatment rules recommended azithromycin for approximately one-third of children.
• Targeted treatment reduced the risk of diarrhea on day 3 by 10.1% (NNT: 10) and day 90 re-hospitalization/death by 2.4% (NNT: 42).
• Host characteristics, rather than pathogen diagnostics, were stronger predictors for benefits related to re-hospitalization and death.

Conclusions:

• Machine learning can effectively identify children who will benefit most from azithromycin for watery diarrhea.
• Personalized targeting of azithromycin can significantly improve clinical outcomes and reduce healthcare utilization.
• Effective targeting for severe outcomes may be achievable without relying on pathogen diagnostics, simplifying treatment strategies.