Project Details

Optimal targeting for community management of infants with possible serious bacterial infections

Yunwei Chen, PhD

Summary

BACKGROUND: The World Health Organization provides guidance for community health workers (CHWs) to identify sick young infants in community settings and recommends referral and hospital treatment for infants aged 0–59 days who present with clinical signs of possible serious bacterial infection (PSBI). However, in many low-resource settings, referral is often not feasible due to geographical, financial, or system barriers, potentially leaving many high-risk infants untreated while diverting scarce resources to lower-risk cases. GAP: Limited evidence exists on how to optimize scarce hospital resources and target inpatient treatment to infants most likely to benefit, while identifying low-risk cases that can be safely managed in the community. HYPOTHESIS: This study aims to assess the feasibility of developing targeted treatment policies for young infants who present with signs of PSBI using historical data and machine-learning-based policy optimization methods. The central hypothesis is that, given substantial heterogeneity in infant characteristics and episode severity, a one-size-fits-all approach is likely suboptimal. METHODS: We conduct a secondary analysis of 63,017 infants in Bangladesh, India, and Pakistan and 84,759 infants in DR Congo, Kenya, and Nigeria, followed over the first 60 days after birth by CHWs with rich data collected over up to ten visits. We apply policy learning algorithms to estimate optimal referral and treatment assignment rules that maximize infant survival while respecting real-world implementation constraints. RESULTS: Pending. IMPACT: Findings from this study can inform updates to guidelines on managing infants with PSBI in low-resource communities and support more efficient allocation of limited healthcare resources. The learned targeted treatment rules can also be integrated into clinical decision support tools to guide frontline health workers in making more accurate, data-driven decisions.