Delineating COVID-19 subgroups using routine clinical data identifies distinct in-hospital outcomes

Area of Science:

• Infectious Disease Epidemiology
• Computational Biology
• Clinical Informatics

Background:

• The COVID-19 pandemic underscored the need for advanced predictive models to manage disease heterogeneity and guide clinical decisions.
• Existing models often lack the adaptability required for rapidly evolving infectious diseases.
• Identifying patient subgroups is crucial for risk stratification and resource allocation.

Purpose of the Study:

• To adapt and apply the unsupervised SuStaIn model for classifying COVID-19 patients based on clinical data.
• To uncover distinct patient subtypes and disease severity stages within hospitalized COVID-19 cases.
• To assess the predictive value of identified subtypes and stages for in-hospital mortality and treatment escalation.

Main Methods:

• Utilized the SuStaIn (Subtype and Stage) unsupervised learning model.
• Applied the model to 1344 hospitalized COVID-19 patients from the National COVID-19 Chest Imaging Database (NCCID).
• Data included 11 common clinical measures, with patients split into training and validation cohorts.

Main Results:

• Discovered three distinct COVID-19 patient subtypes: General Haemodynamic, Renal, and Immunological.
• Identified disease severity stages that, along with subtypes, predicted varying risks of in-hospital mortality or treatment escalation.
• A low-risk, 'Normal-appearing' subtype was also identified, aiding in comprehensive patient stratification.

Conclusions:

• The adapted SuStaIn model effectively identifies COVID-19 subtypes and severity stages with prognostic value.
• These data-driven classifications can improve decision-making and treatment prioritization for infectious diseases.
• The developed pipeline is adaptable for future infectious disease outbreaks, offering a valuable tool for public health preparedness.