AI Tool Identifies Children With Pneumonia Requiring Hospital Care

Early recognition of severe cases in primary care is difficult, and current international referral guidelines can miss dangerously ill patients. A new study shows an artificial intelligence approach can better identify young children with pneumonia who require hospital care.

University College Dublin (UCD) researchers developed BIOTOPE (BIOmarkers TO diagnose PnEumonia), a machine-learning algorithm designed to identify children with pneumonia who are at high risk of needing hospitalization. The tool significantly outperformed existing risk assessment methods used to determine which children require urgent referral. It was also designed to operate within Malawi’s Integrated Community Health Information System (iCHIS) to avoid adding administrative burden for health workers.

BIOTOPE applies a random forest model to evaluate multiple factors concurrently, including breathing rate, temperature, heart rate, oxygen levels, nutritional status, and household conditions. By integrating physiologic and contextual variables, the model is intended to assist frontline clinicians with consistent, data-driven referral decisions. Researchers noted that many children who died from severe pneumonia in prior studies did not display the standard warning signs that typically trigger referral, underscoring the need for multiparametric assessment.

Details of the work were published in PLOS Medicine on June 9, 2026. The international team trained and validated the algorithm using data from more than 2,500 children attending primary care clinics in Malawi. The BIOTOPE project, led by UCD, included collaborators from Mzuzu University (Malawi), the University of Galway, Queen’s University Belfast, the World Health Organization, the Malawi Ministry of Health, and Luke International Norway. The initiative also incorporated extensive public participation, with parents and caregivers helping to shape study priorities. According to the project team, embedding the tool in existing health information systems offers a practical, scalable route to frontline use in low-resource settings.

“What this research shows is that we can do much better for children who are severely ill with pneumonia. A child's life can depend on whether a health worker in a clinic correctly identifies how sick they are. We now have a tool that could help make that decision easier in the real world,” said Dr. Joe Gallagher, UCD School of Medicine, who led the study.

“Machine learning gives us the ability to create something that improves over time rather than becoming obsolete. This algorithm can be continuously retrained as new data accumulates, keeping it relevant across changing disease patterns and health system contexts,” said Professor Cathal Seoighe, University of Galway.

Related Links
UCD School of Medicine