Longitudinal Patient-Reported Outcome Trajectories in Long COVID: Findings From the STOP-PASC Clinical Trial

Long COVID is a heterogeneous post-infectious condition. Although patient-reported outcome (PRO) measures for diagnosis or therapeutic monitoring have been adapted from related complex chronic illnesses, no PRO has been validated specifically in Long COVID. The STOP-PASC randomized, placebo-controlled trial of nirmatrelvir/ritonavir (NMV/r) in adults with Long COVID showed no overall treatment effect. This exploratory analysis aimed to identify distinct symptom trajectories and clinical characteristics associated with improvement or worsening over time.

We performed latent class trajectory modeling (LCTM) on PRO measures-including the Patient Global Impression of Severity (PGIS), Patient Global Impression of Change (PGIC), PROMIS domains, and core symptoms-among 155 randomized participants. Participants were followed for 15 weeks with serial symptom assessments. Trajectory groups were identified using Bayesian Information Criteria and characterized using descriptive statistics and absolute standardized differences.

LCTM revealed heterogeneity in symptom trajectories. Two groups emerged for PGIS (improving n = 17, persistent/severe n = 136) and PGIC (improving n = 130; worsening n = 22). PROMIS-Physical Function modeling identified four groups (improving, normal/mild, moderate, and severe), fatigue core symptom modeling identified three (improving; moderate; severe). Worsening groups had higher proportions of NMV/r-treated participants and greater prevalence of cardiovascular symptoms and low-dose naltrexone use. Improving groups had shorter time since infection and higher baseline physical function. No subgroup showed a clear benefit from NMV/r.

Distinct PRO trajectories reflect the clinical heterogeneity of Long COVID. NMV/r showed no clear benefit across subgroups. These findings emphasize the need for validated, Long COVID-specific PRO instruments and targeted therapeutic trials tailored to Long COVID subtypes.