Long COVID's Hidden Complexity: Machine Learning Reveals Why Personalized Care Remains Essential.

Area of Science:

• Medicine
• Infectious Diseases
• Public Health

Background:

• Long COVID, or post-acute sequelae of SARS-CoV-2 infection (PASC), affects individuals irrespective of initial infection severity.
• Existing research often focuses on mild cases and patient-centered multidisciplinary management.
• Current diagnostic algorithms do not account for potential symptom clusters in PASC patients.

Purpose of the Study:

• To investigate potential patient phenotypes and symptom clusters in individuals with severe COVID-19 using unsupervised machine learning.
• To assess the feasibility of identifying distinct patient groups for developing targeted diagnostic-therapeutic pathways.

Main Methods:

• Retrospective longitudinal study at two Italian hospitals (SMATTEO and CREMONA).
• Inclusion of patients discharged with severe COVID-19 diagnosis, followed up at 3 months.
• Application of unsupervised machine learning techniques, including Principal Component Analysis (PCA) and Partitioning Around Medoids (PAM), to identify patient phenotypes.

Main Results:

• 382 patients with severe COVID-19 were analyzed; over a third were >65 years old, and over 80% had comorbidities.
• Circulatory and endocrinopathies were the most frequent diagnoses (46% and 20%, respectively).
• Statistical analyses (PCA, hierarchical clustering, PAM) failed to identify distinct symptom clusters, with most patients falling into a single group.

Conclusions:

• The significant heterogeneity among patients with post-acute sequelae of SARS-CoV-2 infection complicates the identification of specific symptom clusters.
• The lack of identifiable clusters prevents the creation of common diagnostic-therapeutic pathways for Long COVID management.
• Further research is needed to understand and address the diverse manifestations of Long COVID.