Post‐COVID syndrome symptoms, functional disability, and clinical severity phenotypes in hospitalized and nonhospitalized individuals: A cross‐sectional evaluation from a community COVID rehabilitation service

Abstract There is currently limited information on clinical severity phenotypes of symptoms and functional disability in post‐coronavirus disease 2019 (COVID) Syndrome (PCS). A purposive sample of 370 PCS patients from a dedicated community COVID‐19 rehabilitation service was assessed using the COVID‐19 Yorkshire Rehabilitation Scale where each symptom or functional difficulty was scored on a 0–10 Likert scale and also compared with before infection. Phenotypes based on symptom severity were extracted to identify any noticeable patterns. The correlation between symptom severity, functional disability, and overall health was explored. The mean age was 47 years, with 237 (64%) females. The median duration of symptoms was 211 days (interquartile range 143–353). Symptoms and functional difficulties increased substantially when compared to before infection. Three distinct severity phenotypes of mild (n = 90), moderate (n = 186), and severe (n = 94) were identified where the severity of individual symptoms was of similar severity within each phenotype. Symptom scores were strongly positively correlated with functional difficulty scores (0.7, 0.6–0.7) and moderately negatively correlated with overall health (−0.4, −0.3, to −0.5). This is the first study reporting on severity phenotypes in a largely nonhospitalized PCS cohort. Severity phenotypes might help stratify patients for targeted interventions and planning of care pathways.


| Setting
This service evaluation study was conducted within a COVID Rehabilitation service in the North of England. This is a multidisciplinary rehabilitation service, which offers specialist assessment and treatment for PCS, delivered virtually and/or in person, delivered by a specialist multidisciplinary team. Patients are referred via General Practitioners (GPs) when symptoms have persisted beyond 12 weeks and cannot be explained through alternative medical assessment. Patients were not required to have had a positive polymerase chain reaction (PCR) test or antibody test as these were not widely available to the general population of the UK at the start of the pandemic. 16

| Participant identification
Consecutive patients referred to the COVID Rehabilitation service between February 2 and May 3, 2021 were considered for this service evaluation study regardless of whether they had been previously hospitalized with COVID-19. All patients had agreed for their data to be used anonymously for research purposes or service evaluation.
The eligible patients were prompted to complete the latest selfreport version of C19-YRS by one of the research team members. In total, responses were received from 370 patients.

| C19-YRS
The information gathered on C19-YRS includes demographic information, medical history, and 16 key symptoms of PCS (including breathlessness, persistent cough, fatigue, pain or discomfort, cognitive problems, anxiety, depression, symptoms of posttraumatic stress disorder [PTSD], palpitations, dizziness, weakness, and sleep problems) and their impact on five daily functions (including communication, mobility, personal care, wider activities of daily living, and social functioning). 8 Each symptom or functional ability is rated by the respondent on a scale from 0 to 10 (0 being no presence of symptom and 10 being most severe and life disturbing). Overall health status is also captured on a 0-10 numerical rating scale (NRS) scale. Unique to the C19YRS and this study, respondents are also asked to grade their pre-illness symptoms, functional abilities, and overall health to provide a general clinical baseline for comparison.

| Statistical analysis
Overall severity of the 12 most reported symptoms was defined as the mean symptom score, with a mean score of 6 or more considered "severe," 3 to 5.9 "moderate", and less than 3 "mild." Demographic details and brief clinical history were summarized overall and by symptom severity. Functional abilities were reported on the same 0-10 scale.
All pairwise Spearman's correlations across symptoms and functional abilities were calculated and presented graphically as a heat map. Symptoms were then categorized as severe (6 or more) or not, and the frequency of each combination of multiple severe symptoms was presented as an UpSet diagram.
Cluster analysis was used to identify any groupings or co-occurrence of symptoms that could indicate potential different phenotypes amongst the participants. Two approaches were used: k-means partition cluster analysis and a hierarchical agglomerative cluster analysis using average-linkage between clusters and the Euclidian dissimilarity measure. Robustness of results was assessed using 2, 3, and 4 clusters and with different starting values (k-means cluster analysis) and using different weights and dissimilarity measures (hierarchical cluster analysis). All statistical analyses were carried out using Stata version 16.1 and R version 4.0.5.  Patients were predominantly of white ethnicity (84%) with only 57 patients (16%) from black, Asian, other, or mixed ethnic groups, which is slightly lower than the proportion in the general population of the region served (https://observatory.leeds.gov.uk/ population/). Patients from black and particularly Asian ethnic groups reported experiencing more severe symptoms than patients from white ethnic groups.

| Ethical approval
Half (49%) of patients were still employed or studying on the same hours as before infection with COVID-19, and 41 (11%) were in the same role as homemakers, still on maternity leave, retired or unemployed. However, 144 (40%) had reduced their work hours, were still on sick leave, or had stopped work altogether because of ill health. Table 2 shows the three severity phenotypes identified based on their mean symptom score (6 or more considered "severe," 3-5.9 "moderate", and less than 3 "mild" as recommended by the C19-YRS scale). There was a tendency for patients with the most severe symptoms to be more likely female, older, with higher weight or BMI.
Those with the most severe symptoms were half as likely to remain employed on the same hours as those with the least severe symptoms.
Radar plot mapping of the average individual symptom score for each of the three categories ( Figure 1) showed there was a gradient in all 12 mean symptom scores, with patients who had greater overall symptoms having each separate symptom higher on average with no single symptom driving this association. For patients with milder overall severity scores, fatigue was, on average, the dominant symptom but all mean symptom scores were lower than those deemed to have moderate severity overall. For patients with greater overall severity scores, fatigue was joined by high mean symptom scores across all 12 symptoms recorded, with all mean scores higher than for the moderate category. There was a similar gradient in functional difficulties, depending on the severity of the symptoms  of all symptoms (bad days) and symptom-free days (good days). We did not find any one symptom determining this association or driving other symptoms which are again supportive of underlying common mechanisms in the condition. We have some pointers in the literature towards these common mechanisms in PCS such as vascular damage (hypercoagulability), 17 immune dysregulation, 18 and dysautonomia. 19 The reason for such a heterogeneous presentation of symptoms in individuals needs to be explored in future studies. There is a concern in PCS that many symptoms such as fatigue or anxiety or mood disorders may already be present in much of the general population before infection. We were however able to show in this study that, for a large number of symptoms, these were not pre-existing before infection, albeit scored retrospectively. This supports the de novo (new-onset) nature of the symptoms attributable to the condition of PCS. The data collected in this study also suggests that the C19-YRS scale can be used to capture pre-illness symptoms even though there is likely to be a certain degree of recall bias.
Given the growing number of people with PCS (already more than one million in the UK alone), the findings of severity phenotypes in this study could have widespread implications for the provision and resourcing of services to support people living with the condition.
The stratification based on the severity of cases could help national and local providers to plan services and interventions that might be directed towards these categories. Mild cases can be investigated in investigations and interventions. 16,22 One weakness of many previous studies has been the reliance on cohorts entirely comprised of individuals previously hospitalized with COVID-19. Our inclusion of a large proportion of nonhospitalized patients, and the consistency of their symptoms with those of hospitalized patients, implies that our reported symptoms are unlikely to be a result of the hospital or intensive care experience but are due to the unique underlying pathophysiological mechanism of PCS. This is in keeping with other studies which have shown a similar burden of symptoms in nonhospitalized patients. 23 It is important to remember that PCS is a fluctuating or episodic condition and that symptom severity can vary over time within the same individual. 24 Capturing this fluctuation of severity and personal triggers (physical, cognitive, emotional) may help individuals stay within the limits of these triggers and pace their activities accordingly, to avoid worsening of symptom severity and its functional impact. We recommend complex multifaceted rehabilitation interventions to manage symptom severity fluctuation seen in PCS. 16 There are some limitations to this study. First, a relatively small sample size precludes determining distinct symptom-based phenotype patterns to be estimated with sufficient precision to be identified. It is possible that rarer phenotypes exist that did not present with sufficient numbers in our study sample. Instead, all groupings of individual symptoms based on correlations between them were dominated by the overall severity across all symptoms.
Second, it is worth noting that symptoms and their severity were selfreported by patients, so there could be a degree of subjectivity in their recording, that they tend to grade severity similar across the symptoms. It is possible that milder cases not presenting to PCS centers for management might present with more distinct symptom clusters that would indicate different underlying pathophysiological mechanisms. Finally, there is an element of recall bias in reporting pre-illness scores, but this had no bearing on the findings of this study.

| CONCLUSIONS
This is the first study in the current literature reporting on severity phenotypes in a largely nonhospitalized PCS cohort. Severity phenotypes might help stratify patients for targeted interventions and planning of care pathways. Further research is needed to understand the common mechanisms and pathophysiological basis of PCS.