Long COVID symptoms change for 15 months after infection

By Vijay Kumar MalesuReviewed by Lauren HardakerNov 28 2025

A national RECOVER cohort reveals why Long COVID doesn’t follow a single recovery pattern. Instead, it shows who stays sick, who improves, and who unexpectedly worsens as symptoms evolve long after infection.

Study: Long COVID trajectories in the prospectively followed RECOVER-Adult US cohort. Image credit: Pixel-Shot/Shutterstock.com

In a recent study published in Nature Communications, a group of researchers defined and compared longitudinal symptom trajectories of Long coronavirus disease (COVID) after the first severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection using the Long COVID Research Index (LCRI).

Why Long COVID recovery paths remain unpredictable

One in three adults knows someone who is still unwell months after contracting COVID-19. Long COVID is recognized as an infection-associated chronic condition, including symptoms of fatigue, cognitive fog, and post-exertional malaise (PEM) that disrupt work, caregiving, and social life.

Early estimates suggest millions are affected, yet clinicians still lack clear guidance to counsel patients, plan follow-up, or design targeted trials. One-time assessments miss the ups and downs of relapsing symptoms, and clinic-based studies often skew toward people with more severe illness. And because most current cases occur in vaccinated, Omicron-era populations, it’s especially important to understand how Long COVID unfolds in this newer context.

National RECOVER cohort tracks symptoms across 15 months

Investigators analyzed adults in the National Institutes of Health (NIH) Researching COVID to Enhance Recovery (RECOVER) initiative, a prospective United States (US) cohort that follows individuals from first SARS-CoV-2 infection at 83 sites across 33 states, Washington, District of Columbia (DC), and Puerto Rico (PR). Participants completed standardized symptom surveys at approximately 3, 6, 9, 12, and 15 months after the index infection. The primary outcome was the 2024 LCRI, a weighted sum of 11 symptoms that incorporated severity; scores ranged from 0 to 30, and scores ≥11 denoted Long COVID.

Reinfections were identified, and surveys completed within 30 days of a reported reinfection were treated as active reinfection time points. Distinct longitudinal profiles were derived using finite mixture models fit with the expectation-maximization (EM) algorithm, modeling the LCRI as a Poisson-distributed continuous outcome.

Missing symptom data were handled using multiple imputation with random intercepts, assuming the missingness was random. The team used an averaged Bayesian Information Criterion (BIC) to choose the best model, then applied a consensus method to assign each participant to a trajectory across all imputed datasets. They also followed an uninfected comparison group on the same schedule to help interpret how common each profile was. All study procedures had IRB approval, and every participant provided informed consent.

Eight symptom patterns reveal diverging Long COVID courses

The analysis cohort included 3,659 adults followed prospectively from first infection; 69 % were female, 98 % were not hospitalized acutely, and 99.6 % were infected during the Omicron variant era.

At three months, 374 of 3,644 participants (10.3%) met the LCRI threshold; a related subgroup analysis considered 377 participants meeting this threshold under slightly different inclusion criteria. At 15 months, 324 of 2,970 (10.9 %) met the threshold after excluding active reinfections and missed surveys. Finite mixture modeling identified eight longitudinal profiles between months 3 and 15.

Profile A, persistent high symptom burden, comprised 5 % and consistently exceeded the threshold. Profile B, characterized by an intermittently high burden with fluctuations around the threshold, comprised 12 %. Profile C, improving moderate burden, comprised 10 % and trended downward over time. Profile D, improving low burden, comprised 9 % and typically approached zero by six months.

Profile E, characterized by a moderate worsening of burden, comprised 8 % of cases, with gradual increases. Profile F, delayed worsening, comprised 6 % and showed late increases at month 15, often featuring PEM. Profile G, consistent low burden, comprised 13 % with occasional symptoms usually below the threshold. Profile H, consistent minimal-to-none burden, comprised 36 % and never met the threshold.

Among the 377 participants who met the LCRI threshold at three months, 46 % followed Profile A, 35 % Profile B, 18 % Profile C, and 1 % Profile D, indicating that most early Long COVID cases remained persistent or intermittently symptomatic through 15 months, while a minority showed improvement. Participants with persistent high burden (Profile A) were more often female than those with minimal symptoms (77 % versus 64 %) and were more likely to have been hospitalized during acute infection (6 % versus 1 %). Overall, 36 % reported on-study reinfection by 15 months.

Reinfection rates were marginally higher in the worsening profiles (39-40 % in Profiles E and F) than in other profiles; however, the similar reinfection frequency elsewhere suggests that rising scores in these groups were not solely explained by reinfection.

In an uninfected comparator cohort followed on the same schedule, the profile distributions differed, with more individuals exhibiting minimal or low-burden patterns and a similar frequency of the moderate worsening pattern. A small proportion also met the LCRI ≥11 threshold at some visits. This raises the possibility that some worsening trajectories reflect background symptoms or other conditions.

Missing symptom data between months 6 and 15 ranged from 8 % to 16 %, but profile assignments were consistent across multiple imputed datasets and sensitivity analyses. Loss to follow-up was somewhat higher among participants with the highest symptom scores, which may modestly bias recovery estimates.

These findings reveal durable heterogeneity: a persistent-high group, a large intermittently high group, an improving minority, and a small delayed-worsening group, all of which require clinical vigilance. Trajectory labeling supports counseling and targeted trials.

Tailored Long COVID care needed

This national prospective analysis shows Long COVID is not a single course but a set of distinct, trackable pathways measurable with the LCRI. Roughly 5 % experienced a persistently high burden through 15 months, and more than 10 % cycled in and out of high burden without clear improvement, while others improved or worsened later.

These trajectories can help shape better clinical trials, guide clinic follow-up, and support more coordinated care, while also giving people clearer expectations for work, caregiving, and rest. Next steps include longer-term follow-up, careful interpretation beyond this mostly vaccinated, Omicron-era group, deeper analysis of biospecimens and digital measures to pinpoint predictive biomarkers, and ongoing testing of targeted interventions based on these symptom patterns.

Download your PDF copy now!