Researchers from University College Dublin in Ireland have developed an AI model capable of identifying children with pneumonia who are at high risk for hospitalization, with very high accuracy. The findings were published in the journal PLOS Medicine in early June of this year.
Pneumonia remains one of the leading causes of death from infectious diseases among children under five worldwide, claiming the lives of nearly a million children annually. Therefore, identifying children who need special hospital care is a lifesaving measure for millions of children, especially in countries with limited health resources.
The researchers explained that the primary goal of this study was to test the AI model's ability to predict the likelihood of hospitalization for children - in primary healthcare centers - from the first visit of the child diagnosed with pneumonia, particularly children aged between two months and five years, by monitoring vital signs that indicate severe symptom progression, based on World Health Organization medical recommendations regarding pneumonia.
The algorithm uses a specific technique called random forest, as it can analyze a wide range of different factors simultaneously to reach a decisive medical decision regarding each child's condition based on symptom severity, including respiratory rate, body temperature, heart rate, blood oxygen levels, the child's ability to eat, and the home conditions in which the child exists.
The researchers examined data from 2,500 children from nine primary care centers in Malawi. The participants were children suffering from cough with breathing difficulties, all of whom had already been diagnosed with pneumonia, and they used the AI model to predict the progression of each child's condition over 7 days, determining the need for hospitalization or not.
The results showed that the AI model clearly outperformed traditional diagnostic methods - such as X-rays and tests - in predicting cases that need urgent hospitalization, successfully saving children at risk of respiratory failure and death by analyzing their vital signs.
Notably, this model was designed to work within Malawi’s health system, allowing its use without placing additional burdens on healthcare workers in a country like Malawi where there is one doctor for every 28,000 people, compared to one doctor for every 250 people in Ireland, emphasizing the immense importance of this model in countries facing healthcare staff shortages.
The researchers noted that current international guidelines for hospital admission of children with pneumonia could overlook some cases, particularly in poor countries. Previous studies have found that many children who died from severe pneumonia did not show the standard warning signs that are usually relied upon to refer patients for hospitalization, making the success of this model a significant medical achievement.
In conclusion, the study affirmed that what distinguishes this model is its ability to constantly update the algorithms through machine learning, enabling it to maintain its predictive capabilities regarding complications continuously as new medical recommendations arise, making it a significant support for physicians.