JMIR Publications

ABSTRACT

Background:

Neonatal mortality remains a pressing public health challenge in low- and middle-income countries (LMICs), with Africa being among the worst-hit continents. Existing triage systems cannot accurately identify and prioritize high-risk neonates to prioritize treatment to reduce neonatal deaths. Shukla et al. propose a machine learning (ML) model for neonatal risk prediction using datasets from a multi-country database in LMIC settings. Using eleven neonatal parameters/variables that spanned the period from delivery to day 1 and post-delivery to day 2, the model demonstrated high predictive accuracy, with an area under the curve (AUC) above 0.80 for neonatal mortality in the Indian context, whereby ‘birth weight’ was identified as the most important predictor.

Objective:

The study aims to evaluate the usability, effectiveness, and feasibility of the neonatal parameters/risk predictor variables using an ML model in Kenyan healthcare facilities as an LMIC setting, to inform its potential adoption, in alignment with national and global health goals.

Methods:

A mixed-methods approach will be employed to conduct a feasibility study through the real-world pre-implementation of the tool in three health facilities that provide neonatal services.

Results:

It is anticipated that the neonatal risk predictors from the ML model will demonstrate high feasibility for integration within triage systems in Kenyan health facilities, with minimal disruption to existing workflows. Usability testing is expected to show a positive user experience among healthcare professionals, highlighting ease of use, perceived usefulness, and acceptability. The model is projected to effectively identify high-risk deliveries and neonates within the first 48 hours of life, enabling timely clinical interventions. These findings are expected to support the model’s relevance and adequacy for neonatal risk assessment in low-resource settings, with potential to contribute meaningfully to reducing neonatal mortality, thereby aligning with national health goals and Sustainable Development Goal (SDG) Target 3.2.

Conclusions:

The study aligns with Kenya's national health priorities and contributes to global efforts to reduce neonatal mortality, thereby supporting the achievement of Target 3.2 of the SDGs. The study would generate actionable evidence on what it takes to move from ML innovation to real-world impact in newborn health, offering a model for evaluating and integrating AI tools into LMIC health systems effectively.