Mycoplasma pneumonia affects thousands of children in Singapore and across Asia every year. Most cases are mild, but some children develop severe disease that progresses quickly and becomes dangerous. The problem: early warning signs of mild and severe cases look almost identical, making it hard for doctors to know which children need closer monitoring or stronger treatment.
Researchers at Frontiers in Pediatrics tackled this challenge using artificial intelligence—computer systems trained to recognize patterns in medical data. They studied over 1,000 children admitted with Mycoplasma pneumonia, including 205 with severe disease.
An AI model that spots 95% of at-risk children
The researchers fed routine medical information into their AI system: blood test results, how long the child had been coughing, physical exam findings like wheezing or crackling lung sounds, and chest X-ray observations. These are tests doctors in Singapore already order regularly for children with pneumonia.
The best-performing AI model correctly identified 95% of children who would develop severe disease. Just as importantly, it was highly accurate at identifying children who would not develop severe disease—this prevents unnecessary intensive treatment for low-risk children and reduces overuse of antibiotics.
Certain blood markers proved especially useful for predicting severity: liver enzymes, kidney function markers, and blood counts. All of these are already part of standard care in Singapore's healthcare system.
AI supports doctors, but doesn't replace clinical judgment
The researchers emphasize this tool is meant to help doctors make faster decisions, not replace their expertise. A doctor still applies their clinical knowledge, discusses risks with families, and makes the final call based on the whole picture of the child's health.
Better risk prediction matters because some children currently receive more aggressive treatment than necessary (leading to unnecessary antibiotics and side effects), while others might not receive prompt care when they need it. This AI system could help strike the right balance.
Before hospitals use it, testing must continue across multiple centres
While results are promising, the researchers note that this AI system needs testing across multiple medical centres and different patient populations before it's ready for Singapore's hospitals and clinics. The current study, though large with over 1,000 children, was conducted in a single setting.