Yale School of Public Health researchers have developed an AI-powered patient triage system capable of predicting disease severity and hospitalization duration during viral outbreaks, with a focus on COVID-19. Utilizing metabolomics and clinical data, this tool aids in resource allocation and patient management. While promising, the study’s limitations include data predating widespread COVID-19 treatments and vaccine availability and potential racial disparities among patient cohorts. This research reflects the broader trend of employing advanced technologies, such as ML, to enhance outbreak response, as seen in the University of Pittsburgh’s EDS-HAT system for infectious disease outbreak detection.
Researchers at the Yale School of Public Health (YSPH) have developed an AI-driven patient triage system that can accurately predict the severity of diseases and the length of hospitalization during viral outbreaks. This innovative platform, discussed in a recent publication in Human Genomics, harnesses machine learning (ML) and metabolomics data to enhance patient management and resource allocation.
Dr. Vasilis Vasiliou, a professor of epidemiology at YSPH and the senior author of the study, emphasized the importance of predicting which patients can be safely sent home and which may require intensive care during outbreaks. This capability is crucial for optimizing patient outcomes and efficiently utilizing hospital resources.
The platform primarily focuses on COVID-19 as a model disease, utilizing a combination of untargeted plasma metabolomics data, patient comorbidity information, and routine clinical data to make predictions. To develop this tool, the research team analyzed data from 111 COVID-19 patients admitted to Yale New Haven Hospital and 342 healthy, COVID-negative individuals between March and May 2020. Patients were categorized based on their treatment needs, ranging from those not requiring external oxygen to those needing intubation.
This analysis allowed the researchers to identify a panel of metabolites strongly correlated with COVID-19 severity and disease progression. Notably, they found substances such as kynurenine, hydroxytryptophan, picolinic acid, glucuronic acid, and allantoin as well as elevated blood eosinophil levels to be potential biomarkers for disease severity. Patients requiring positive airway pressure or intubation were also observed to have reduced plasma serotonin levels.
The study’s findings highlight the promising potential of predictive AI in public health, not only for COVID-19 management but also for future viral outbreaks. Dr. Vasiliou underscored that this AI-powered platform represents a significant step toward a more effective and data-driven public health response.
However, it’s important to note several limitations of the study. Data were collected before the widespread availability of COVID-19 treatments and vaccines, which could affect the observed changes in metabolites. Additionally, the findings may be influenced by the racial and ethnic composition of the patient cohorts, with a higher proportion of Black patients in the COVID group and predominantly white individuals in the healthy cohort.
This research aligns with the broader trend of leveraging advanced technologies to bolster outbreak response efforts. In 2021, researchers from the University of Pittsburgh School of Medicine and Carnegie Mellon University demonstrated the effectiveness of ML and whole genome sequencing in improving infectious disease outbreak detection. Their Enhanced Detection System for Healthcare-Associated Transmission (EDS-HAT) clusters patients with similar infections and identifies potential transmission routes. Over two years, this system was estimated to have prevented numerous infectious disease transmissions and saved UPMC Presbyterian Hospital significant financial resources.
