In outbreaks, the most valuable information is often the simplest: Who walks into the emergency department, what their first test results show and what happens next.

Admission-day hospital data captures this critical moment in determining treatment pathways and interventions. If this early snapshot can reliably flag who is at higher risk, it can guide triage, monitoring and staffing. For a country like Lebanon, where hospitals have had to manage repeated shocks while keeping services running, turning routine data into actionable insight is one way to protect patients when resources are stretched. 

In a study published in the DIGITAL HEALTH Journal, Dr. Christopher El Hadi, resident at the LAU Gilbert and Rose-Marie Chagoury School of Medicine, alongside collaborators from the Saint Joseph University of Beirut, set out to see whether hospitalized COVID-19 patients can be grouped into clear profiles using information available within the first 24 hours of admission.

Titled “Unsupervised clustering reveals a tri-phenotype model of hospitalized COVID-19 patients: Beirut cohort study and literature synthesis,” the study examined 556 confirmed cases admitted to Hôtel Dieu de France Hospital in Beirut from March 2020 through October 2021.

The team employed an unsupervised machine learning approach that searches for natural similarities across variables at once, effectively grouping patients with similar clinical traits. They coupled their approach with standard quality checks to determine whether the final grouping was coherent rather than arbitrary.

“This model takes into account the diverse clinical parameters we encounter in medical practice and research,” noted Dr. El Hadi. “This helped the results gain rigor and be as generalizable as possible to every infected human being, thanks to its ability to find unfiltered core patterns.” 

The result was a three-group model that aligns with what frontline teams often observe. One group reflected patients with more intense inflammation and signs of lung infections, many of whom needed to be transferred to the Intensive Care Unit (ICU) and stayed in the hospital longer. A second group captured the oldest patients with the highest rates of chronic conditions such as high blood pressure, diabetes, heart and kidney disease and lung disease. This group also showed stronger signals linked to superinfections and other complications. The third group represented younger, generally healthier patients with fewer chronic conditions and shorter hospital stays.

Statistically, imaging findings and key blood-test patterns differed significantly across groups, as did outcomes and complications. In the ICU, some outcomes, including intensive care length of stay and intubation risk, did not differ by cluster. However, the older group with multiple chronic illnesses showed significantly greater odds of death than the remaining two groups, reinforcing the profound impact that age and chronic disease have on COVID-19 risk in hospitalized settings. Interestingly, the survival rates of both the hyperinflammatory and middle-aged groups were comparable.

The study offers a practical path to data-driven decision-making. “Another remarkable discovery the research found is that the pattern of stratification we obtained is highly identical to multiple analyses in COVID-19 literature,” said Dr. El Hadi. “This translates into additional guidance and support to our intuition when we first encounter a COVID-19 patient.” 

By relying on patient information collected during the first 24 hours of admission, and when choices on monitoring intensity and escalation are most consequential, the approach offers a timely tool for supporting care planning.

To browse more scholarly output by the LAU community, visit our open-access digital archive, the Lebanese American University Repository (LAUR).