Each year, community-acquired pneumonia claims more than 3 million lives in LMICs. Among patients receiving mechanical ventilation in intensive care units, around 15 per cent develop severe pneumonia, more than half of which is caused by highly drug-resistant pathogens.
Despite this burden, much of the evidence guiding everyday clinical care still comes from higher-income settings. Clinicians in LMICs are often required to adapt guidance developed for different healthcare systems, patient populations and diagnostic resources. In practice, that can mean making rapid decisions about respiratory support and antibiotic treatment with limited information.
To help address this gap, an international multidisciplinary team has launched PHENOM.AI, short for Discovering novel pneumonia phenotypes in Vietnamese patients using multimodal data and artificial intelligence. Supported by a Wellcome Discovery Award, the project brings together expertise from Vietnam and the UK in clinical research, pathogen diagnostics, artificial intelligence, data science, biostatistics and host-response biology.
From UK datasets to the bedside in Vietnam
Rather than importing off-the-shelf software, PHENOM.AI is building a foundation model designed specifically for infectious diseases in LMIC settings. The team will pre-train the model using large datasets such as MIMIC and UK Biobank, then fine-tune it using more than a decade of clinical data alongside newly recruited patients from Vietnamese hospitals.
At the heart of PHENOM.AI is a multimodal data platform that brings together three types of information: physiology and routine clinical care, pathogen diagnostics, and host response.
- The first includes continuous photoplethysmography (PPG), electrocardiograms (ECG), laboratory results and chest imaging.
- The second combines conventional microbiology with metagenomics, multiplex PCR and 16S rRNA sequencing.
- The third draws on multi-omic approaches, including whole-blood RNA sequencing, plasma proteomics, single-cell omics and sepsis response signature sub-phenotyping.
Together, these data streams allow researchers to view pneumonia not as a single diagnosis, but as a group of distinct diseases with different biological characteristics. Rather than simply predicting which patients are most at risk, the team aims to uncover the clinical and biological patterns that explain why some people deteriorate rapidly while others respond well to standard treatment.
The goal of PHENOM.AI is to create a lasting scientific and clinical resource for pneumonia research in Vietnam. This includes a foundation AI model tailored to Vietnamese clinical settings, a multimodal pneumonia dataset to support future critical care research, and biomarkers paired with decision-support algorithms to help ICU teams identify patients who may need earlier oxygen support or changes to antimicrobial treatment.
‘Pneumonia in critical care is rarely a one-size-fits-all condition. By combining deep host and pathogen profiling with continuous physiological data, we are building tools that reflect the reality of our ICUs. This collaboration between Vietnam and Oxford allows us to bring advanced data science directly into frontline clinical care, where the need is greatest.’
For clinicians, the ambition is simple: provide clearer evidence to support critical decisions at the bedside.
‘In the ICU, knowing precisely when to escalate mechanical ventilation or changing antibiotic regimens is a constant challenge. Having an AI system trained on our local patient profile will give us clearer, earlier signals, helping us tailor therapy to the individual right at the bedside.’
Recruiting patients in real-world ICUs
To reflect routine clinical practice, PHENOM.AI aims to recruit more than 1,100 patients with community-acquired or ventilator-associated pneumonia across four partner hospitals in Vietnam, representing secondary and tertiary care settings.
The study is operationally demanding. Recruiting patients and collecting samples within the first six hours after they arrive at hospital requires close coordination between research teams and partner hospitals.
Reflecting on these challenges, project manager Nguyễn Thị Lệ Thanh said:
‘The project is highly demanding, but it is also an invaluable opportunity for our team to strengthen our skills and capacity. Coordinating blood samples from multiple hospitals back to the OUCRU laboratory within two to four hours of collection, while maintaining strict temperature control to preserve sample quality, requires careful planning and seamless teamwork.
‘At the same time, our research nurses are learning to use up to ten digital applications simultaneously to capture different types of patient data. Each participant requires up to 80 data checkpoints, and every step must follow strict research protocols and ethical standards to ensure the quality, integrity and confidentiality of the study data. It is a huge step forward in building our capacity to deliver complex clinical studies in Vietnam.’
A decade in the making
PHENOM.AI builds on more than a decade of OUCRU research exploring how technology can improve critical care in Vietnam, led by Professor Louise Thwaites. Much of that work grew through the Vietnam ICU Translational Applications Laboratory (VITAL), which established the data platforms, clinical partnerships and technical expertise that now underpin PHENOM.AI.
The project reflects OUCRU’s long-term commitment to strengthening local research capacity and working in partnership with Vietnamese clinicians and scientists to address locally important health challenges.
Building on that foundation, PHENOM.AI continues to strengthen data systems, laboratory capacity and collaborative networks that will support future research on severe infections, while contributing to Ho Chi Minh City’s growing role as a regional health technology hub.
From 28 to 30 July 2026, researchers and clinical teams gathered in Ho Chi Minh City for the first PHENOM.AI annual project meeting.
The meeting brought together colleagues from OUCRU, the University of Oxford’s Institute of Biomedical Engineering and Computational Health Informatics Lab, the Hospital for Tropical Diseases, the National Hospital for Tropical Diseases, Trưng Vương Hospital and Nguyễn Thị Thập General Hospital. Alongside scientific discussions, the programme included visits to partner hospitals to review study implementation and strengthen collaboration across participating sites.
OUCRU thanks all investigators, collaborators and partner hospitals for their continued commitment to PHENOM.AI and to improving pneumonia care through collaborative research:
University of Oxford
OUCRU
Site Principal Investigators
- Dr Nguyễn Lê Như Tùng, Specialist II, Deputy Director, Hospital for Tropical Diseases
- Associate Professor Dr Trần Văn Giang, Director, Institute for Training and Research in Tropical Diseases, National Hospital for Tropical Diseases
- Dr Huỳnh Ngọc Hớn, Specialist II, Director, Trưng Vương Hospital
- Dr Lương Hoàng Liêm, Specialist II, Deputy Director, Nguyễn Thị Thập General Hospital
