BACKGROUND

Rapidly phenotyping patients can enhance healthcare management during new pandemic outbreaks. This can be accomplished through data-driven unsupervised methods that do not require clinical outcomes to be available. This study aimed to identify and compare phenotypes of COVID-19 patients and the subset of those patients who visited emergency departments using clustering techniques based on a limited set of easily accessible variables across different stages of the pandemic.

METHODS

We conducted a population-based retrospective study that included all reported adult COVID-19 patients in the Basque Country from March 1, 2020, to January 9, 2022. Phenotypes were identified separately for the pre-Omicron and Omicron periods in an unsupervised manner using clustering techniques based on easily obtainable clinical and sociodemographic variables. The clinical characteristics of the phenotypes were compared, and subsequently their association with the clinical outcomes was assessed.

RESULTS

Four phenotypes were identified in both the general population and the emergency department sub-group in the pre-Omicron period, whereas three phenotypes were extracted in Omicron. Within each scenario, these phenotypes varied significantly in age and comorbidity rates, leading to varying associations with COVID-19 outcomes. Despite their similarities, the emergency department phenotypes consistently experienced worse outcomes than their general population counterparts. Moreover, the population and emergency department phenotypes identified during the Omicron period resembled those from the pre-Omicron stage, suggesting stable phenotypic structures throughout the pandemic.

CONCLUSIONS

This study highlights the potential of phenotype identification based on a few accessible variables for a meaningful segregation of patients. This approach could be extended to future pandemics as a preventive public health strategy, especially considering the growing likelihood of facing new ones.