Discovering novel pneumonia phenotypes in Vietnamese patients using multimodal data and artificial intelligence

A Vietnamese patient with pneumonia is more likely to receive the wrong treatment because the risk scores and disease categories used to guide care were developed in Europe or North America, not in Ho Chi Minh City or Hanoi. This matters because community-acquired pneumonia and ventilator-associated pneumonia are the top reasons for hospital and ICU admission in low- and middle-income countries, yet almost all research on them comes from high-income settings. Doctors in Vietnam lack validated tools to predict which patients will need intensive respiratory support, or to decide who with suspected ventilator-associated pneumonia truly needs antibiotics. Without these tools, they either overtreat with antibiotics—fueling resistance—or undertreat, risking death. Over the past eight years, this team has built AI systems in Vietnam that analyse clinical data, medical images, and continuous physiological signals from custom monitoring platforms. They have already uncovered clinical and inflammatory patterns linked to outcomes in life-threatening infections. Now they will combine multimodal data with advanced pathogen detection and host immune profiling to identify new pneumonia phenotypes tied to the need for ICU respiratory support or antibiotic therapy. If successful, the research could produce risk scores and diagnostic strategies tailored to LMIC settings, reducing unnecessary antibiotic use and improving survival. It will also advance fundamental understanding of why some pneumonia patients deteriorate while others recover.