View article

Background: This study prospectively characterized subjective and objective attributes in people with post-acute sequelae COVID-19 (PASC), leveraging machine learning techniques. Subjective attributes included symptoms of PASC and were analyzed with topic modeling, while objective attributes included laboratory findings, frailty and physical function and were analyzed with clustering. PASC attributes were used to explore predictors of a pre-defined PASC recovery definition at follow-up.

Methods: This was a prospective, observational study of patients consecutively attending a multidisciplinary PASC clinic from Jun 2020 to Jun 2023. Each patient was longitudinally described by the combination of domains and phenotypes. The former were built from fine-grained subjective attributes depicting symptoms' intensity changes; whilst the latter were synthesized from objective attributes comprising clinical variables collected during visits.

Findings: A total of 1,012 people were evaluated, mean age was 60.3 years, 42.1% were females. 70.8% met the recovery definition 15.9% lost at follow-up and 13.2% were still in charge after a mean of 1.76±0.62 years after the acute infection. 368 people were visited at least twice, while 2 people died before the follow-up. Patients with at least two visits generated phenotype trajectories, enabling the identification of clinical characteristics associated with PASC recovery. Metabolic variables and lifestyle discriminated at baseline PASC recovery.

Interpretation: This study underlined the multifaceted nature of PASC using a patient-centered clinical approach which combined symptoms domains and patient phenotypes …