The PRESCO study is a multi-center, prospective, 3-month cohort study designed to identify clinical and molecular signatures associated with progression to severe COVID-19. There were up to five study visits: (1) enrollment during initial presentation to a hospital or ambulatory clinic, and if occurred, (2) 2 days after hospitalization; (3) the day of intensive care unit (ICU) admission; (4) the day of hospital discharge; and (5) 3 months after enrollment (Figure 1A). Adults with laboratory-test confirmed SARS-CoV-2 infection (RT-PCR or antigen testing) who received care at eight sites (Table 1) between May 2020 and June 2021 were invited to participate. Eligible participants were adults that (1) were 18 years old or older in age, (2) were U.S. residents, (3) confirmed positive for COVID-19, (4) received care at a participating site, (5) were willing and able to provide informed consent, and (6) were willing and able to complete all study procedures. Participants were excluded if they were pregnant. Enrollment was completed before the delta variant became predominant in the United States in the summer of 2021 and before the availability of nirmatrelvir and ritonavir treatments. Later on in the study as the pandemic evolved, collection of PASC information and symptom information at the 3-month follow-up visit (Visit 5) were added. See the Supplementary material for more details.

The PRESCO study was approved by a central Western Institutional Review Board (Protocol# 20201016) and at each of the eight sites. Written informed consent was obtained from all participants or their legally authorized representatives before study-related procedures were performed.

Participants were asked to select symptoms present from a list of 22 symptoms at enrollment and a list of 30 at the 3-month follow-up (Supplementary Tables 1, 2). Symptoms were grouped and analyzed by System Organ Class (SOC) according to the Medical Dictionary for Regulatory Activities (MedDRA), which groups symptoms by etiology, manifestation site, and/or purpose. At the 3-month follow-up visit, participants were also asked how many weeks had passed since their last study visit until they felt at their usual state of health. We defined PASC as those individuals who did not recover to their usual state of health for four or more weeks after the start of COVID-19, which was determined by the earliest of several non-self-reported dates, including enrollment, first positive SARS-CoV-2 test, hospital presentation, and hospitalization. WHO clinical severity scale was used to measure COVID-19 severity (8). Additionally, participants’ demographics and longitudinal clinical characteristics were collected. See the Supplementary material for more details.

The nomenclatures for the populations used in the analysis are provided in Figure 1B. The enrolled population included participants who signed the informed consent and were enrolled in the study. The PASC analysis population included those who had sufficient data to be categorized as having PASC or without PASC (non-PASC), and the cluster analysis population included those who provided symptom information at the 3-month follow-up visit. The PASC analysis population included 354 participants out of the 494 (71.7%) participants enrolled. Descriptive statistics included mean (standard deviation, SD) or median (range), and frequencies (percentages), as appropriate. Continuous data were compared using Wilcoxon rank sum tests, and categorical data were compared using Chi-square or Fisher’s exact tests, as appropriate. Unadjusted univariate tests were conducted for all demographic information, clinical characteristics, and clinical labs to search for statistically significant differences between PASC and non-PASC groups. Association analysis of PASC with comorbidities, concomitant medication, and clinical labs were further adjusted for potential confounders. Multiplicity was corrected in the association analysis of PASC with comorbidities, concomitant medications, and clinical labs, controlling the false discovery rate (FDR) at 0.05 with the Benjamini-Hochberg procedure. The cluster analysis population included 152 participants out of the 494 (30.8%) enrolled. Clustering analysis was based on symptoms collected in the questionnaire at the 3-month follow-up visit. Hierarchical clustering of participants was performed with Ward’s method using hamming distance. Three clusters were determined from visual evaluation of the heatmap, and the dendrogram was then cut at an appropriate height to generate resulting clusters. See the Supplementary material for more details.

Due to delayed implementation of the amendment to outcome survey, PASC information and 3-month follow-up symptoms were not collected from participants who exited the study before May 2021. Missing data was not imputed given the observational nature of the study.

Based on the PASC analysis population, we analyzed the clinical characteristics that were associated with the development of PASC. Participants who developed PASC were significantly older than participants without PASC (p < 0.001; Table 2), but there were no sex differences. The PASC group had a greater proportion of Non-Hispanic White people and a lower proportion of Asians, Black people, and Hispanic people when compared to the non-PASC group (p < 0.01). Among comorbidities, there was a greater proportion of obesity and tobacco use in the PASC group compared to the non-PASC group (p = 0.007 and p = 0.04, respectively; Table 2). Hypertension and gastroesophageal reflux disease (GERD) were associated with PASC (Table 2 and Supplementary Table 3). Individuals who developed PASC had more severe COVID-19 than participants without PASC (p < 0.001; Table 2). The PASC group also had a higher proportion of patients that were hospitalized (p = 0.02; Table 2) and required a longer duration of hospitalization (p < 0.001; Table 2).

Dexamethasone and remdesivir usage were significantly greater in the PASC group compared to the non-PASC group (p < 0.001 and p = 0.002, respectively; Table 2) (9, 10). After adjusting for propensity score and final COVID-19 severity, the odds ratio (95% confidence interval [CI]) of developing PASC with treatment when compared to without treatment was 2.23 (95% CI; 1.14, 4.36) and 1.78 (95% CI; 0.97, 3.27) for dexamethasone and remdesivir, respectively.

Participants who developed PASC had significantly lower absolute lymphocyte counts at Visit 1, during the acute illness period (p = 0.005; Table 2). Although this was no longer statistically significant after adjusting for age and baseline COVID-19 severity (Supplementary Table 4), the effect size is large (effect size = −0.932). We also found significantly higher serum bicarbonate at the time of hospital discharge in the PASC group than the non-PASC group (p = 0.05; Supplementary Table 4). However, the difference was no longer significant after adjustment for confounders.

Multivariable modeling was conducted using clinical risk factors that were found significant in the univariate analysis (Supplementary Table 5). Dexamethasone administration, hospital duration, WHO score, lymphocyte count at hospital presentation, serum bicarbonate levels at hospital discharge, and body mass index were associated with PASC (Supplementary Table 5). Sensitivity analysis was performed excluding serum bicarbonate levels (available in only 122 patients) and revealed that in addition to the above PASC associations, GERD, tobacco use, and race-ethnicity were found to be associated with PASC, whereas BMI was excluded from the final model (Supplementary Table 5).

A subset of 152 (42.9%) participants of the 354 PASC analysis population provided the type of symptoms they experienced during the 3-month follow-up period, which was used to perform cluster analyses (termed the “cluster analysis population”; Figure 1B). The cluster analysis population was slightly younger and had a greater proportion of women, a lower median WHO score, and a shorter time-period between COVID-19 onset and presentation to a health system compared to the enrolled population (Table 1).

Hierarchical clustering of the presence or absence of long-term symptoms revealed three distinct clusters of individuals (cluster 1: remitting, cluster 2: persistent, cluster 3: incident) (Figure 2A). All three clusters had similar symptom burden during the acute illness, but differed with regards to symptom burden at the 3-month follow-up visit, underscoring three different disease trajectories of COVID-19 when assessing temporal trends (Figures 2B, C). Participants in cluster 2 were on average older than those in cluster 3 (p = 0.015). There was a greater proportion of Asians in cluster 1 than in the other two clusters (Supplementary Table 6). There were no differences in sex, obesity, other comorbidities, hospitalization rate, or concomitant medication across the clusters. Interestingly, there was no significant difference in COVID severity (WHO ordinal scores) across clusters. Yet, cluster 1 had significantly lower rates of PASC compared to the other two clusters and a significant reduction in symptom number from the acute illness period, suggesting a “remitting” temporal phenotype (Figure 2C and Table 3). In contrast, cluster 2 demonstrated persistent symptomatology at 3-months compared to the acute illness period, suggesting a “persistent” temporal phenotype (Figure 2C and Table 3), and cluster 3 showed an increase in symptoms at 3 months that would suggest an “incident” temporal phenotype despite lower hospitalization rates (Figure 2C and Table 3). A preponderance of symptoms involving the neurological, respiratory, and general symptoms distinguished cluster 3 from cluster 1 (Supplementary Figure 2). While the number of days from the start of COVID-19 to Visit 5 was significantly greater in cluster 1 than cluster 3 (145.6 days in cluster 1 vs. 112.3 days in cluster 3; Table 3), cluster 1 still had a significantly larger symptom reduction compared to cluster 3 following regression analysis adjusting for that time influence (Supplementary Table 7). Multiple differences in clinical laboratory results at enrollment and at 3 months across clusters were also observed (Supplementary Table 8).

To further characterize these phenotypic clusters, we compared the type of symptoms present at 3 months classified by SOC. Constitutional symptoms included under “General disorders and administration site conditions” were the most frequent symptoms reported by patients in all three clusters at the 3-month follow-up visit, although their prevalence was significantly higher in clusters 2 and 3 compared to cluster 1 (Figure 2D and Supplementary Tables 2, 9). The majority of individuals in cluster 3 experienced symptoms relating to the Nervous system, and Respiratory, thoracic and mediastinal disorders. Cluster 3 participants also had a significantly higher prevalence of symptoms relating to Gastrointestinal disorders compared to the other two clusters (Figure 2D and Supplementary Tables 2, 9).

To better characterize the evolution of symptoms, we analyzed the longitudinal changes in symptomatology by SOC for each cluster. We found that cluster 3 was characterized by more persistent symptoms in multiple SOCs at 3 months after hospital presentation, while individuals in cluster 1 recovered from their acute symptoms (Supplementary Figure 2 and Supplementary Table 10).