Post-acute sequelae of COVID-19 symptom phenotypes and therapeutic strategies: A prospective, observational study.

BACKGROUND: Post-acute sequelae of COVID-19 (PASC) includes a heterogeneous group of patients with variable symptomatology, who may respond to different therapeutic interventions. Identifying phenotypes of PASC and therapeutic strategies for different subgroups would be a major step forward in management.
METHODS: In a prospective cohort study of patients hospitalized with COVID-19, 12-month symptoms and quantitative outcome metrics were collected. Unsupervised hierarchical cluster analyses were performed to identify patients with: (1) similar symptoms lasting ≥4 weeks after acute SARS-CoV-2 infection, and (2) similar therapeutic interventions. Logistic regression analyses were used to evaluate the association of these symptom and therapy clusters with quantitative 12-month outcome metrics (modified Rankin Scale, Barthel Index, NIH NeuroQoL).
RESULTS: Among 242 patients, 122 (50%) reported ≥1 PASC symptom (median 3, IQR 1-5) lasting a median of 12-months (range 1-15) post-COVID diagnosis. Cluster analysis generated three symptom groups: Cluster1 had few symptoms (most commonly headache); Cluster2 had many symptoms including high levels of anxiety and depression; and Cluster3 primarily included shortness of breath, headache and cognitive symptoms. Cluster1 received few therapeutic interventions (OR 2.6, 95% CI 1.1-5.9), Cluster2 received several interventions, including antidepressants, anti-anxiety medications and psychological therapy (OR 15.7, 95% CI 4.1-59.7) and Cluster3 primarily received physical and occupational therapy (OR 3.1, 95%CI 1.3-7.1). The most severely affected patients (Symptom Cluster 2) had higher rates of disability (worse modified Rankin scores), worse NeuroQoL measures of anxiety, depression, fatigue and sleep disorder, and a higher number of stressors (all P<0.05). 100% of those who received a treatment strategy that included psychiatric therapies reported symptom improvement, compared to 97% who received primarily physical/occupational therapy, and 83% who received few interventions (P = 0.042).
CONCLUSIONS: We identified three clinically relevant PASC symptom-based phenotypes, which received different therapeutic interventions with varying response rates. These data may be helpful in tailoring individual treatment programs.

Introduction

Post-acute sequelae of COVID-19 (PASC) have been reported in 7–91% of patients following acute SARS-CoV-2 infection [1-10]. This wide prevalence range is related to heterogeneous definitions of PASC [11-13], differences in populations being assessed (e.g. severe versus mild-moderate versus asymptomatic COVID-19 patients), and types of symptoms or signs included in rate estimates. Because there is no current biological definition of PASC, many studies lump disparate symptoms and signs into the PASC diagnosis, without an assessment of clinical relevance. The resulting heterogeneity in PASC cohorts makes it difficult to assess treatment strategies, which are likely to vary depending on PASC symptomatology. Disaggregating PASC into meaningful clinical phenotypes may help development of future targeted therapies.

In this prospective study, we aimed to identify clinically important phenotypes of PASC in patients hospitalized 12-months prior with COVID-19 illness, examine associations of these phenotypes with quantitative measures of functional status and quality of life, and evaluate whether PASC clusters received different therapeutic interventions. We further aimed to determine the rates at which patients reported subjective improvement with a given treatment program. While there are cluster analyses evaluating symptoms in the acute phase of COVID-19 illness [14-17], and others focused on specific post-acute symptoms such as pulmonary complaints [18], brain fog [19], or neuropsychiatric symptoms [20], we chose to evaluate the entire post-acute COVID-19 symptom report form published by the World Health Organization (WHO) [21] and map symptom clusters to therapeutic interventions. These analyses may provide insight into the efficacy of various symptom-based treatment strategies and underscore the need for a multi-disciplinary holistic approach to post-COVID-19 care.

Methods

Study design and patient cohort

We conducted a prospective, observational outcome study of consecutive COVID-19 patients hospitalized at four New York City area hospitals within one academic system between March 10, 2020-May 20, 2020. Follow-up interviews were performed at 12-months (+/- 2 months) after initial SARS-CoV-2 diagnosis (window: January 10, 2021-July 20, 2021). Detailed enrollment, methodology and outcome measures have been previously reported [2, 3, 22, 23]. Inclusion criteria were: RT-PCR positive SARS-CoV-2 infection, age ≥18 years, hospital admission, and consent to participate in a follow-up interview. Exclusion criteria were: evaluation in an outpatient or emergency department setting only. PASC was defined according to Centers for Disease Control and Prevention (CDC) criteria as new or persistent symptoms occurring ≥4 weeks after SARS-CoV-2 infection [11].

Standard protocol approvals and patient consents

This study was approved by the NYU Grossman School of Medicine Institutional Review Board. All patients or their surrogates provided consent for participation. All patients or their surrogates provided verbal consent for participation. The IRB did not require written consent because follow-up visits were conducted via phone call due to social distancing guidelines in place during the pandemic. Additionally, this research was deemed low risk by the IRB and verbal consent was allowed due to the observational nature of the study. Consent was documented in REDCap.

12-month follow-up questionnaires

The post-COVID symptom interview was based on the clinical case report form for PASC developed by the World Health Organization (WHO)(21). Subjects were asked if they believed they had “post-COVID syndrome”, “long-hauler syndrome” or “long-COVID” characterized by symptoms occurring ≥4 weeks after initial COVID diagnosis. If the response was affirmative, subjects were asked to identify their symptoms from a list of 31 options [21] (S1 Table). Similarly, subjects were asked “What treatments have you used to deal with your prolonged COVID symptoms?” and were asked to indicate if they had used any of 14 possible therapeutic interventions, and whether the therapy improved their post-COVID symptoms (S2 Table). Subjects were also asked to indicate if they had experienced any of 15 stressors [1] within the month prior to the 12-month interview (S3 Table).

In addition to the above patient-reported outcomes, the following standardized batteries were administered at the 12-month interview: the modified Rankin Scale (mRS; 0 = no symptoms, 6 = dead) [24], the Barthel Index of activities of daily living (0 = completely dependent, 100 = independent for all activities) [25], the telephone-MoCA (22 = perfect score; ≤18 = abnormal cognition) [26], and Quality of Life in Neurological Disorders [27] (NeuroQoL) short form self-reported health measures of anxiety, depression, fatigue and sleep. NeuroQoL raw scores were converted into T-scores with a mean of 50 and standard deviation of 10 in a reference population (U.S. general population or clinical sample) [28]. Higher T-scores indicate worse self-reported health for the anxiety, depression, fatigue and sleep metrics.

Statistical analyses

Unsupervised hierarchical agglomerative cluster analyses were performed using Ward’s method to place patients into similar clusters suggested by the data, not into a priori pre-defined categories. Case similarity was measured using variance computed from a fourfold table as (b+c)/4n, where b and c represent the diagonal cells corresponding to cases present on one item but absent on the other, and n is the total number of observations. We selected unsupervised agglomerative hierarchical cluster analysis methods because they provided a flexible approach to assemble symptoms clusters (based on similarity of their taxonomic structures) without prior knowledge about any underlying relationships, and because they allowed us to generate easy-to-interpret views of the clustering structure, an advantage over some other clustering approaches. Agglomerative approaches are also efficient when datasets are smaller, such as ours [29]. Unlike k-means and some other clustering approaches, this method does not require advanced selection of number of clusters, and smaller clusters tend to be generated, which was useful in our exploration of PASC sub-phenotypes. We used Ward’s minimum variance methods to minimizes the total within-cluster variance.

For symptom clustering, all 31 binary symptom variables were entered into the model in the order of their prevalence. Similarly, for therapy cluster analysis, all 14 binary therapy variables were entered into the model in the order of most to least frequently used. No other variables were entered into either cluster analysis. The number of clusters was determined based on inspection of the dendogram. To ensure cluster stability, cluster analyses were repeated in random subsamples of patients. In sensitivity analyses, we repeated hierarchical cluster analysis evaluating only incident symptoms (e.g. brain fog/confusion/memory loss, headache, anxiety, depression and shortness of breath could only be coded as present if the patient did not have a pre-COVID history of these symptoms). Because cluster analysis is performed on datasets without missing variables, patients with missing data were removed. To ensure the complete data sample was representative of the full sample, we contrasted the dataset used in the cluster analysis and the dataset of individuals with missing data using Chi-square tests for categorical measures and Mann-Whitney-U tests for continuous measures.

Demographics, stressors, comorbidities, hospital metrics and 12-month quantitative metrics were compared between symptom clusters, as well as between therapy groups, using Kruskal-Wallis tests for continuous, non-normally distributed data and Chi-squared or Fisher’s exact tests for dichotomous variables. The association of therapy groups with 12-month outcome metrics were assessed using binary logistic regression analyses. Multivariable logistic regression analyses were used to assess the relationship between each therapy cluster with the rate of subjective improvement with treatment, adjusting for severity of index COVID-19 illness (intubated vs. not-intubated). All analyses were conducted using IBM SPSS Statistics for Mac version 28 (IBM Corp., Armonk, NY).

Results

PASC symptoms and impact

Follow-up interviews 12-months after hospitalization for COVID-19 were attempted in 590 patients and completed in 242/590 (41%)(3). The median age was 65 (IQR 53–73), 64% were male, 81/242 (34%) required invasive mechanical ventilation during index hospitalization and 113/242 (47%) had a neurological complication during their hospitalization. Of these patients, 122/242 (50%) reported ≥1 prolonged COVID-19 symptom lasting a median of 12-months (range 1–15 months). Multiple symptoms were reported in 87/122 (71%), and the median number of symptoms was 3 (IQR 1–5). The most prevalent symptoms included shortness of breath in 73/122 (60%), headache in 54/122 (44%) and cognitive symptoms, including difficulty concentrating, “brain fog” and/or memory issues in 48/122 (39%, Fig 1A). PASC symptoms interfered with work activities in 55/122 (45%), household responsibilities in 61/122 (50%) and leisure activities in 66/122 (54%). There were no differences in age, sex, past medical history (hypertension, diabetes, lung disease, dementia, psychiatric illness, headache disorder), vaccination status or proportion of patients with neurological events when comparing patients with or without PASC symptoms (all P>0.05). However, the severity of index COVID-19 infection—as measured by requirement of invasive mechanical ventilation—was significantly associated with the development of PASC symptoms (73% of ventilated patients developed PASC compared to 43% of patients who were not mechanically ventilated, P<0.001). Of those who completed 12 month follow-up, data on post-acute symptoms was missing on 12 patients. When comparing data among included patients (N = 122) to those with missing data (N = 12), those with missing data were significantly older (median 74 versus 64 years, P = 0.002), had fewer life stressors (median 0 versus 1, P = 0.010) and had worse 12-month modified Rankin scores (median 4 versus 2, P<0.001, S5 Table).

Fig 1

Frequency of PASC symptoms, therapies received and rates of response to therapy.

A. Frequency of post-COVID symptoms lasting ≥ 4 weeks after initial diagnosis as reported by participants 12-months after hospitalization for COVID-19 (N = 242). B. Percent of patients who received, and responded to a given therapy for PASC symptoms within 12-months after hospitalization for COVID-19 (N = 242).

PASC therapies and response rates

Of 122 PASC patients, 85 (70%) received at least one therapy to manage PASC symptoms and 72/122 (59%) received multiple therapies. The most frequently prescribed interventions were physical or occupational therapy, exercise or pain medications. Psychiatric interventions included: talk therapy (9%), antidepressants (6%) and anti-anxiety medications (5%). Most patients reported a positive response to each therapy (range 75–100% reported symptom improvement), though a smaller proportion of patients (77%) felt that exercise alleviated their symptoms, and 7/85 (8%) reported no improvement with any intervention (Fig 1B).

Symptom cluster analysis

Hierarchical cluster analysis generated three symptom clusters (Fig 2A and 2B): Symptom Cluster 1 (N = 38, 31%) included patients with few symptoms, most often headache and anxiety though at lower rates than reported in other clusters; Symptom Cluster 2 (N = 16, 13%) consisted of patients with many symptoms, including high levels of anxiety and depression; and Symptom Cluster 3 (N = 68, 56%) included patients who predominantly reported shortness of breath, headache and cognitive abnormalities. Sensitivity analyses evaluating only index symptoms post-COVID yielded similar cluster results to the primary analysis (S1 Fig). Symptom clusters did not differ significantly by age, sex, race, pre-COVID-19 disability status (baseline mRS) or medical comorbidities (Table 1). However, patients in Symptom Cluster 2 had the highest number of symptoms and stressors, were least likely to report that their symptoms did not affect routine activities, and had worse 12-month quantitative outcomes as compared to the other groups, including higher mRS scores (indicative of worse disability), as well as increased levels of anxiety, depression, fatigue and sleep abnormalities, as measured by NeuroQoL metrics (all P<0.05). In contrast, Symptom Cluster 1 had the fewest number of symptoms, less severe index COVID-19 illness (significantly lower rates of mechanical ventilation and hypotension), was most likely to report that symptoms did not impact routine activities, and had the lowest measures of anxiety, depression, sleep abnormalities and fatigue.

Fig 2

Symptom clusters among patients with PASC (N = 122).

Cluster 1 has few symptoms; Cluster 2 has multiple symptoms with high rates of anxiety, depression and post-exertional malaise; Cluster 3 has high rates of shortness of breath, headache, and cognitive issues. Panel A. shows the frequency of each of the assessed symptoms in each cluster. Colors corresponding to each symptom are shown from left to right in the key. Panel B. shows a heat map for each patient, where red indicates presence of a symptom and blue indicates absence. SOB = shortness of breath; HA = headache.

Table 1

Demographics, stressors, co-morbidities, and 12-month metrics among symptoms clusters (N = 122 patients with PASC).

	Symptom Cluster 1 (N = 38) Few symptoms	Symptom Cluster 2 (N = 16) Many symptoms with high levels of anxiety and depression	Symptom Cluster 3 (N = 68) Predominantly shortness of breath, headache and cognitive issues	P across all 3 groups
Demographics
Age, median (IQR)	62 (49–70)	63 (52–71)	64 (58–72)	0.351
Sex (male), N (%)	25/38 (66%)	7/16 (44%)	43/68 (63%)	0.285
Race (white), N (%)	19/32 (59%)	9/16 (56%)	33/51 (54%)	0.791
Education level >12 years, N (%)	27/33 (82%)	10/13 (77%)	53/64 (83%)	0.882
Stressors
At least one stressor, N (%)	22/38 (58%)	14/16 (88%)	38/68 (56%)	0.061
Number of stressors, median (IQR)	1 (0–2)	2 (1–5)	1 (0–2)	0.005
Social Isolation, N (%)	6/38 (16%)	5/16 (31%)	10/68 (15%)	0.277
Financial Insecurity, N (%)	6/38 (16%)	6/16 (38%)	12/68 (18%)	0.153
Unemployment, N (%)	5/38 (13%)	3/16 (19%)	9/68 (13%)	0.837
Food Insecurity, N (%)	1/33 (3%)	0/12 (0%)	0/39 (0%)	0.457
Homelessness, N (%)	0/38 (0%)	0/16 (0%)	1/68 (2%)	0.670
Domestic violence, N (%)	0/38 (0%)	0/16 (0%)	0/68 (0%)	—
Relationship problems in household, N (%)	2/38 (5%)	2/16 (13%)	5/68 (7%)	0.649
Education disruption, N (%)	1/38 (3%)	0/16 (0%)	3/68 (4%)	0.648
Increased caregiver responsibilities, N (%)	1/38 (3%)	2/16 (13%)	5/68 (7%)	0.378
Personal Illness, N (%)	8/38 (21%)	7/16 (44%)	22/68 (32%)	0.218
New Disability, N (%)	2/38 (5%)	5/16 (31%)	5/68 (7%)	0.008
Death of close contact, N (%)	0/38 (0%)	10/16 (63%)	6/68 (9%)	<0.001
Illness of close contact, N (%)	5/38 (13%)	3/16 (19%)	6/68 (9%)	0.494
Lack of access to child care, N (%)	0/38 (0%)	1/16 (6%)	0/68 (0%)	0.035
Political conflict with close contacts, N (%)	2/38 (5%)	0/16 (0%)	6/68 (9%)	0.407
Comorbidities
Pre-COVID disability (mRS), median (IQR)	0 (0–0)	0 (0–2)	0 (0–1)	0.488
Hypertension, N (%)	9/38 (24%)	8/16 (50%)	28/68 (41%)	0.102
Diabetes, N (%)	5/38 (13%)	5/16 (31%)	22/68 (32%)	0.087
COPD/Asthma, N (%)	3/38 (8%)	4/16 (25%)	7/68 (10%)	0.178
Headache Disorder, N (%)	4/38 (11%)	0/16 (0%)	2/68 (3%)	0.139
Dementia, N (%)	2/38 (5%)	0/16 (0%)	7/68 (10%)	0.306
Psychiatric history, N (%)	3/38 (8%)	2/16 (13%)	6/67 (9%)	0.864
Index COVID-19 Hospitalization
Neuro complication, N (%)	17/38 (45%)	6/16 (38%)	31/68 (46%)	0.840
Mechanically ventilated, N (%)	10/38 (26%)	8/16 (50%)	39/68 (57%)	0.009
Worst Sequential Organ Failure Assessment (SOFA) score, median (IQR)	3 (3–9)	6 (3–13)	6 (3–12)	0.062
Lowest % oxygen saturation, median (IQR)	86% (76–92%)	80% (68–87%)	78% (66–88%)	0.251
Lowest mean arterial blood pressure (mmHg), median (IQR)	70 (56–79)	59 (55–68)	60 (48–71)	0.020
Acute renal failure, N (%)	2/38 (5%)	4/16 (25%)	15/68 (22%)	0.061
Impact of Symptoms
Interfered with work, N (%)	13/38 (34%)	9/16 (56%)	33/68 (49%)	0.229
Interfered with household responsibilities, N (%)	16/38 (42%)	10/16 (63%)	35/68 (52%)	0.367
Impacted leisure activities, N (%)	15/38 (40%)	10/16 (63%)	41/68 (60%)	0.092
No impact of symptoms on routine activities, N (%)	16/38 (42%)	1/16 (6%)	8/68 (12%)	<0.001
Duration of symptoms in months, median (IQR)	12 (4–13)	12 (12–12)	12 (12–13)	0.496
12-month Quantitative Metrics
Number of symptoms, median (IQR)	1 (1–3)	16 (13–18)	3 (2–5)	<0.001
12-mo Barthel Index, median (IQR)	100 (100–100)	98 (68–100)	100 (90–100)	0.053
12-mo T-MoCA, median (IQR)	19 (16–20)	18 (14–19)	18 (15–20)	0.206
12-mo mRS, median (IQR)	2 (0–3)	3 (2–4)	2 (2–4)	0.047
12-mo NeuroQoL Anxiety, median T-score (IQR)	47 (36–54)	58 (52–62)	50 (44–55)	0.001
12-mo NeuroQoL Depression, median T-score (IQR)	40 (37–50)	50 (43–59)	47 (37–51)	0.047
12-mo NeuroQoL Fatigue, median T-score (IQR)	44 (39–51)	58 (48–62)	50 (42–55)	<0.001
12-mo NeuroQoL Sleep, median T-score (IQR)	46 (33–55)	57 (52–66)	50 (40–56)	0.003

Therapy cluster analysis

A separate hierarchical cluster analysis was performed to identify groups of patients who received similar therapeutic interventions. Three clusters were generated (Fig 3A and 3B): Therapy Group A (N = 72, 59%) received few therapeutic interventions, most often pain medications and exercise; Therapy Group B (N = 12, 10%) received several interventions, notably psychological talk therapy, antidepressants, and anti-anxiety medications; and Therapy Group C (N = 38, 31%) primarily received physical and occupational therapy. Patients in Therapy Group A were significantly younger than the other groups, and were less likely to have been mechanically ventilated, while Therapy Group C had significantly higher rates of mechanical ventilation (Table 2). Among those who received no therapeutic interventions (N = 37), the median age was 62 (IQR 51–72), 67% were male and 30% required mechanical ventilation during hospitalization. There were no significant differences in age or sex among those who did or did not receive a therapeutic intervention. However, those who did not receive any therapies were less likely to have required mechanical ventilation during index COVID-19 than those who received some intervention (30% versus 54%, P = 0.018).

Fig 3

Therapeutic groups among PASC patients (N = 122).

Group A received few therapeutic interventions; Group B received several interventions, most notably psychological talk therapy, anti-depressants, anti-anxiety medications and physical therapy; Group C primarily received physical and occupational therapy. Panel A. shows the percent of patients who received a given therapy stratified by therapy group. Colors corresponding to each therapy are shown from left to right in the key. Panel B. shows a heat map where red indicates a given therapy was received and blue indicates that a therapy was not utilized.

Table 2

Demographics, comorbidities, symptom clusters and 12-month outcome metrics among Therapy Groups (N = 122 PASC patients).

	Therapy group A: (N = 72) Received few therapeutic interventions	Therapy group B: (N = 12) Received several interventions, most notably psychological talk therapy, anti-depressants, anti-anxiety medications and physical therapy	Therapy group C: (N = 38) Primarily received physical and occupational therapy	P across all 3 groups
Demographics
Age, median (IQR)	62 (50–71)	68 (63–71)	66 (60–72)	0.035
Sex (male), N (%)	45/72 (63%)	5/12 (42%)	25/38 (66%)	0.314
Race (white), N (%)	33/55 (60%)	9/12 (75%)	19/32 (59%)	0.595
Education level >12 years, N (%)	9/51 (18%)	1/8 (13%)	3/29 (10%)	0.356
Comorbidities
Pre-COVID disability (mRS), N (%)	0 (0–1)	0 (0–0)	0 (0–2)	0.460
Hypertension, N (%)	23/72 (32%)	5/12 (42%)	17/38 (45%)	0.391
Diabetes, N (%)	16/72 (22%)	3/12 (25%)	13/38 (34%)	0.395
COPD/Asthma, N (%)	9/72 (13%)	2/12 (17%)	3/38 (8%)	0.647
Headache disorder, N (%)	5/72 (7%)	1/12 (8%)	0/38 (0%)	0.235
Dementia, N (%)	6/72 (8%	0/12 (0%)	3/38 (10%)	0.587
Psychiatric history, N (%)	4/71 (6%)	3/12 (25%)	4/38 (11%)	0.091
Neuro complication during index COVID-19 hospitalization, N (%)	32/72 (44%)	5/12 (42%)	17/38 (45%)	0.982
Mechanically ventilated during hospitalization for COVID-19, N (%)	24/72 (33%)	7/12 (58%)	26/38 (68%)	0.001
COVID-19 Vaccination, N (%)	46/70 (64%)	9/12 (75%)	27/38 (71%)	0.754
Symptom Clusters
Symptom Cluster 1, N (%)	24/34 (71%)	0/12 (0%)	9/38 (24%)	0.052
Symptom Cluster 2, N (%)	1/34 (3%)	10/12 (83%)	1/38 (3%)	<0.001
Symptom Cluster 3, N (%)	9/34 (27%)	2/12 (17%)	28/38 (74%)	0.015
Symptom Cluster 2 or 3 versus 1, N (%)	44/72 (61%)	11/12 (92%)	29/38 (76%)	0.052
Number of PASC symptoms, N (%)	2 (1–5)	12 (4–18)	3 (2–5)	<0.001
Number of therapies received, median (IQR)	0 (0–2)	7 (5–8)	3 (2–4)	<0.001
Duration of symptoms, median (IQR)	12 (10–12)	12 (12–13)	12 (12–13)	0.087
12-month Outcomes
Improved with therapy*, N (%)	29/35 (83%)	12/12 (100%)	37/38 (97%)	0.042
12-mo Barthel Index, median (IQR)	100 (100–100)	98 (61–100)	98 (68–100)	<0.001
12-mo T-MoCA, median (IQR)	18 (15–20)	18 (14–21)	19 (17–20)	0.480
12-mo mRS, median (IQR)	2 (1–3)	3 (1–4)	3 (2–4)	0.003
12-mo NeuroQoL Anxiety, median (IQR)	50 (44–56)	55 (39–63)	48 (36–53)	0.338
12-mo NeuroQoL Depression, median (IQR)	45 (37–51)	47 (38–57)	47 (37–51)	0.644
12-mo NeuroQoL Fatigue, median (IQR)	48 (42–54)	55 (45–60)	48 (42–54)	0.244
12-mo NeuroQoL Sleep, median (IQR)	52 (39–57)	51 (46–65)	46 (39–53)	0.195

*among those who had symptoms and received at least one intervention, logistic regression analyses adjusted for severity of index COVID-19 (as assessed by requirement for invasive mechanical ventilation)

Bold = significance with P<0.05. mRS = modified Rankin Scale; T-MoCA = telephone Montreal Cognitive Assessment, NeuroQoL = NIH Neurological Quality of Life patient reported outcomes.

Relationship of symptom and therapy clusters

Although generated independently, symptom clusters and therapeutic groups were closely aligned. Inclusion in Symptom Cluster 1 significantly predicted Therapy Group A (odds ratio [OR] 2.55, 95% confidence interval [CI] 1.10–5.89), Symptom Cluster 2 was associated with Therapy Group B (OR 15.70, 95% CI 4.14–59.70) and Symptom Cluster 3 was associated with Therapy Group C (OR 3.08, 95% CI 1.33–7.13, Fig 4, S4 Table).

Fig 4

Relationship of symptom clusters and therapy groups (N = 122).

Symptom Cluster 1 was significantly related to membership in Therapy Group A (Odds Ratio [OR] 2.55, 95% confidence interval [CI] 1.10–5.89), Symptom Cluster 2 was associated with Therapy Group B (OR 15.70, 95% CI 4.14–59.70) and Symptom Cluster 3 was associated with Therapy Group C (OR 3.08, 95% CI 1.33–7.13).

Subjective improvement among therapy groups

Among patients who received at least one therapeutic intervention (N = 85), those in Therapy Group B and C were significantly more likely to report improvement following therapy than Therapy Group A (100%, vs. 97% vs 83%, respectively, P = 0.042). This was observed despite the fact that Therapy Group A had significantly better measures of activities of daily living (Barthel Index) and less severe disability (mRS) than other groups. After adjusting for severity of index COVID-19 illness (intubation status) and age, Therapy Group A was associated with significantly lower rates of subjective improvement compared to other therapy groups (adjusted OR 0.07, 95% CI 0.01–0.68, P = 0.022). Conversely, belonging to Therapy Groups B or C (compared to A) was strongly associated with subjective improvement (aOR 14.5, 95% CI 1.5–142.9, P = 0.022), after adjusting for age and ventilator status. There was a significant association between the number of therapeutic interventions trialed and the subjective report of improvement following therapy (OR 3.00, 95% CI 1.22–7.35, P = 0.017).

Discussion

In this prospective study, hierarchical cluster analysis identified three distinct groups of PASC patients, who differed in type and frequency of post-COVID symptoms. These symptom clusters were clinically significant since they were associated with quantitative 12-month outcome measures of disability and quality of life. An independent cluster analysis also identified three therapeutic groups, and symptom clusters intuitively and statistically mapped to these therapy clusters. The most severely affected cluster (Symptom Cluster 2 representing 13% of PASC patients) had more symptoms, higher rates of disruption of routine activities due to symptoms, higher disability and higher quantitative measures of anxiety and depression, but also reported some of the highest rates of subjective symptom improvement with a therapeutic strategy that included psychiatric treatments (Therapy Group B). Given the high rates of incident anxiety disorders, depression, stress and adjustment disorders reported post-COVID-19 [30], and the stigma attached to mental health treatment [31], this excellent response rate suggests that psychiatric interventions for PASC are important in appropriately identified individuals. Since care for PASC patients may be siloed among different subspecialties, vigilance and aggressive management of psychiatric symptoms are warranted.

This study differs from other COVID-19 cluster analyses [15-20] in that, not only did we identify phenotypes of PASC symptoms, but we also mapped these clusters to corresponding therapeutic strategies and their subjective effectiveness. Disaggregating PASC, which is a highly heterogeneous condition, and identifying useful treatment strategies is critically needed, particularly since we found that 50% of hospitalized COVID-19 patients have PASC symptoms. The Centers for Disease Control and Prevention report 4,575,344 hospital admissions for COVID-19 since August 1, 2020 (which does not include the first wave in New York City) through February 2022 [32]. This would translate into approximately 2.3 million Americans who have had PASC symptoms within 12-months of hospitalization for COVID-19, in addition to the millions more who have PASC after mild-moderate COVID-19. Additionally, the economic and societal impact of PASC symptoms after hospitalization for COVID-19 is likely to be considerable since 80% of PASC patients in our study reported disruptions in their ability to work, manage household responsibilities or engage in leisure activities due to symptoms. Other strengths of this study include its prospective design, highly detailed characterization of the cohort from hospital stay through 12-month follow-up, and combination of patient reported outcomes and objective measures of functional status (mRS, Barthel Index, NeuroQoL).

We found that COVID-19 severity, even amongst a hospitalized cohort, was an overall predictor of PASC in the group of 242 patients that completed follow-up, as has been described in multiple other cohorts [33-35]. Although some studies have suggested that PASC may be more common in those with mild disease [36], we did not have a comparison group of non-hospitalized patients in our cohort to test this hypothesis. Among the subgroup of patients with PASC, the severity of index COVID-19 illness did not appear to predict which patients would have the most symptoms or disability. Indeed, the most severely affected group (Symptom Cluster 2) had not only the highest number of symptoms, but also worse measures of functional outcome, even though this group had less severe index COVID-19 illness (based on intubation rates), and similar rates of neurological complications compared to other clusters. Our findings imply that there are other factors that impact PASC severity. Notably, patients in Symptom Cluster 2 reported a significantly higher number of life stressors than other clusters. In a prior study of risk factors for PASC among U.S. community dwellers with and without mild COVID-19 conducted in February 2021, we identified multiple stressors (present within the month prior to interview) that were associated with the development of PASC, most notably financial insecurity and unemployment [1]. In that study, multivariable models predicting NeuroQoL measures of cognition, anxiety, depression, fatigue and sleep, demonstrated that several stressors were stronger predictors of abnormalities on quality of life testing than was COVID-19 infection itself. These data suggest an interplay of environmental and pandemic-related factors that may impact the development of PASC.

A positive finding of this paper is that most patients who received a given therapy or group of therapies reported improvement, and only 8% reported no improvement at all with any trialed intervention. These data are encouraging, particularly since no patient received a therapy specifically engineered to target PASC symptoms. Practitioners appeared to prescribe therapies directed towards symptoms, and were largely successful with this approach. However, not all patients reported the same rate of symptom improvement with a given therapeutic strategy. Notably, the patients in Therapy Group A, who received the fewest therapeutic interventions, and were the youngest and least disabled (based on mRS and Barthel Index scores), were also the least likely to report improvement following therapy. These data suggest that these patients may have had relatively mild PASC and small improvements were difficult to detect, or that too few or inadequate interventions were trialed. We did identify a significant association between the number of interventions used and the reported rates of symptomatic improvement. Whether this relationship is merely related to trial and error resulting in selection of effective treatments, increased interaction with the healthcare system in those that received more therapies, or other factors such as PASC severity, remains to be determined. Of the 14 interventions tested, exercise appeared to be the least likely to lead to symptomatic relief. This is noteworthy because of the similarities PASC shares with chronic fatigue syndrome, in which symptoms are characteristically worsened with exertion [37]. However, only 16% of PASC patients in our study reported post-exertional malaise, suggesting that phenotypic similarities with chronic fatigue syndrome may only occur in a subset of PASC patients.

There are several limitations to this study. First, the study population had COVID-19 requiring hospitalization and results of our study may not generalize to non-hospitalized COVID-19 patients. However, the types and prevalence of PASC symptoms we reported are similar to those described following mild-moderate COVID-19 [1, 35, 38–40]. Second, data was missing in 12 patients who were excluded from cluster analyses. Those excluded from analyses were largely similar to those included, however excluded patients were significantly older, had worse 12-month mRS scores, and had fewer life stressors than those included. Additionally, we were only able to contact 242/590 (41%) patients eligible for 12-month follow-up due to defunct contact information, language barriers, lack of patient/surrogate consent or unanswered phone calls. Though we made three phone attempts at contact, our data may represent a biased sample. Third, because this was not a controlled, clinical trial, we are unable to determine if one therapeutic intervention was superior to another. Indeed, since the majority of patients received multiple treatments and these treatments may augment or interact with one another, the only way to deal with this heterogeneity was with cluster analysis. Additionally, there were too few patients in certain treatment categories to make any statements regarding efficacy. While those in Therapy Group A reported lower rates of subjective improvement compared to Groups B or C, confidence intervals were wide suggesting imprecision in point estimates. Fourth, though patients reported subjective improvement with therapy, we do not have objective measures of improvement, and we cannot exclude the possibility that responses to interventions may represent a “placebo effect”. Fifth, we did not have detailed information on the severity of symptoms or the degree of improvement with therapy, since these data were collected in a dichotomous fashion. It is possible that those in Therapy Group A who reported lower rates of symptomatic improvement with therapy had less severe symptoms, and hence, less room for dramatic improvements with intervention. Conversely, those with several symptoms of varying severity might require only modest interventions to notice results. We attempted to mitigate against this bias by adjusting for initial severity of COVID-19 illness, and we still demonstrated a lower likelihood of subjective improvement in Therapy Group A. Finally, the spectrum of therapies we assessed was wide, but not comprehensive. For example, we did not evaluate cognitive therapies, and we did not have details regarding which psychiatric medications were most efficacious. Future studies focusing on specific therapeutic interventions are warranted.

Conclusions

PASC following severe COVID-19 is a heterogeneous condition that can be phenotyped into three clinically relevant symptom clusters. These symptom clusters differed in number and impact of symptoms, disability status and quantitative measures of anxiety, depression, sleep and fatigue, and were significantly associated with independently generated therapy clusters. However, response rates to therapeutic strategies differed significantly among groups. Matching PASC symptom phenotypes with corresponding therapeutic strategies may be helpful in tailoring individual treatment programs. Importantly, identifying the most severely affected PASC patients (i.e. Symptom Cluster 2) and implementing psychiatric treatments should be considered in clinical practice.

Symptom questionnaire.

Subjects were asked 12-months after hospitalization for COVID, “Do you believe you have "post-COVID syndrome" also called "long hauler" syndrome or did you continue to have symptoms for at least 4 weeks after your initial diagnosis of COVID-19?” If the response was “yes”, they were presented with the options shown in this table.

(DOCX)

Click here for additional data file.

Therapy questionnaire.

Subjects were asked, “What treatments have you used to deal with your prolonged COVID symptoms? Please check left column if ever used, and also right column if therapy helped post-COVID symptoms”.

(DOCX)

Click here for additional data file.

Stressor questionnaire.

Subjects were asked “Have you experienced any of the following stressors within the last month? (check all that apply in the past month)”.

(DOCX)

Click here for additional data file.

Univariate odds ratios and 95% confidence intervals representing the association of demographics, comorbidities, symptom clusters and 12-month outcome metrics with therapy groups (N = 122 PASC patients).

(DOCX)

Click here for additional data file.

Comparison of patients included in analyses to patients with missing data.

(DOCX)

Click here for additional data file.

Sensitivity analysis of incident symptom clusters using hierarchical cluster analysis (Ward’s method, variance measurement).

The three symptom clusters generated are similar to the primary analysis.

(TIFF)

Click here for additional data file.

7 Jul 2022

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Bart Ferket

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. Please amend your current ethics statement to address the following concerns:

a) Did participants provide their written or verbal informed consent to participate in this study?

b) If consent was verbal, please explain i) why written consent was not obtained, ii) how you documented participant consent, and iii) whether the ethics committees/IRB approved this consent procedure.

3. Thank you for stating the following in the Competing Interests section:

“Potential competing interest: JAF receives funding for the following COVID-19-related grants: NIH/NINDS 3U24NS11384401S1, NIH/NHLBI 1OT2HL161847-01, NIH/NIA 3P30AG066512-01; LET, NMS and LJB receive funding for the following COVID-19-related grant: NIH/NHLBI 1OT2HL161847-01; and TW receives funding for the following COVID-19-related grant:  NIH/NIA 3P30AG066512-01.  AdH, SY, SS, DY, AL, KM, and SLG do not report any relevant conflicts of interest.”

Please confirm that this does not alter your adherence to all PLOS ONE policies on sharing data and materials, by including the following statement: "This does not alter our adherence to  PLOS ONE policies on sharing data and materials.” (as detailed online in our guide for authors http://journals.plos.org/plosone/s/competing-interests).  If there are restrictions on sharing of data and/or materials, please state these. Please note that we cannot proceed with consideration of your article until this information has been declared.

Please include your updated Competing Interests statement in your cover letter; we will change the online submission form on your behalf.

4. Please know it is PLOS ONE policy for corresponding authors to declare, on behalf of all authors, all potential competing interests for the purposes of transparency. PLOS defines a competing interest as anything that interferes with, or could reasonably be perceived as interfering with, the full and objective presentation, peer review, editorial decision-making, or publication of research or non-research articles submitted to one of the journals. Competing interests can be financial or non-financial, professional, or personal. Competing interests can arise in relationship to an organization or another person. Please follow this link to our website for more details on competing interests: http://journals.plos.org/plosone/s/competing-interests

5. In your Data Availability statement, you have not specified where the minimal data set underlying the results described in your manuscript can be found. PLOS defines a study's minimal data set as the underlying data used to reach the conclusions drawn in the manuscript and any additional data required to replicate the reported study findings in their entirety. All PLOS journals require that the minimal data set be made fully available. For more information about our data policy, please see http://journals.plos.org/plosone/s/data-availability.

Upon re-submitting your revised manuscript, please upload your study’s minimal underlying data set as either Supporting Information files or to a stable, public repository and include the relevant URLs, DOIs, or accession numbers within your revised cover letter. For a list of acceptable repositories, please see http://journals.plos.org/plosone/s/data-availability#loc-recommended-repositories. Any potentially identifying patient information must be fully anonymized.

Important: If there are ethical or legal restrictions to sharing your data publicly, please explain these restrictions in detail. Please see our guidelines for more information on what we consider unacceptable restrictions to publicly sharing data: http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data-access-restrictions. Note that it is not acceptable for the authors to be the sole named individuals responsible for ensuring data access.

We will update your Data Availability statement to reflect the information you provide in your cover letter.

6. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: http://journals.plos.org/plosone/s/supporting-information.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: I read this manuscript with a lot of interest as the subject is one that is currently not well understood with regards to the COVID-19 pandemic . The paper presents very interesting results regarding outcomes of SARS-CoV-2 infection and the response to treatment. This study is among few studies that have attempted to explore the clusters of post-Covid condition especially the treatment outcomes.

I have few suggestions to make the paper more valuable;

1. In the race to develop a case definition for long Covid, there is a growing interest in its symptom clusters. Accordingly, provide the symptoms that characterises each cluster you identified in your study (you did this for cluster 2 and 3 but not for cluster 1). Reporting that cluster 1 had few symptoms is not enough.

2. Report vaccinationation status in Table 1, preferably by number of doses. COVID-19 vaccination has been demonstrated to produce favorable outcomes. No such findings was demonstrated when adjusting for vaccination status in your study. You need to check this for this especially by dosage of COVID-19 vaccine and not vaccination status.

3. The quality of images are poor and needs a lot of improvement. Increase image quality to more than 300 dpi.

Reviewer #2: This interesting paper is about different subgroups/phenotypes of post-acute sequele of Covid-19. The authors performed unsupervised cluster analysis in order to identify subgroups that may help to management/treatment of long term symptom following the infection.

The study was based on the cohort of patients collected in the prospective manner. There were 242 patients whereas 50% reported at least one long term symptom after Covid-19 at 12 months follow-up. The study identified three cluster with few remining symptoms, with several symptoms and with specific symptoms, such as shortness of breath, headache and cognitive symptoms. The cluster with several symptoms were the most affected as they had the highest score of disability. All patients in this groups got psychiatric therapy and the major part received also physical/occupational therapy.

The title is appropriate and well inform on the study.

The abstract is very well written and contain the most important information.

The introduction is clear and consistent; however, I would suggest it can be completed with a paragraph on the previous study on subgroups/clusters as there are several studies on it already in the literature.

The method a well written. The detailed description of the cohort was published before.

-What kind of psychiatric intervations did the patients received? Please add information.

-What kind of occupational/physiotherapy did the person reciecived?

-Did you ask about in the survey and did you revive the patients records for these informations?

The results.

- I would suggest you divide the results part in the section, then it will be easier to follow-up the results text.

- I would suggest to add the simple figure with for example circles where you first have three circles of clusters (divided by symptoms) and then on them you have three circles regarding therapy.

- In the table 2 caputure, you should describe what all the rows of numbers means in each cell. As the groups are small you might consider write only OR with 95% CI and p and the numbers in the supplementary materials.

-there are several types of cluster analysis, discuss why did you use the one you selected.

The discussion:

-Is this correct that the study is the first on post covid phenotypes corresponging to the theraputic strategies? The majority of references in the paper is from 2020-2021 and you should add some more new reference and refered also on some recent paper on the phenotypes.

- You write: ‘’ Although some studies have suggested that PASC may be more common in those with mild

disease(28), we did not have a comparison group of non-hospitalized patients in our cohort to test this hypothesis.’’ Is this write with the current stage of knowladge? The patients after more severe Covid-19 used to have more somatic remaining symptoms whereas the patients after mild Covid-19 used to have cognitive impairment and fatigue.

I suggest the major revision of this paper and in particularly suggest to review the recent literature on post Covid phenotypes and complete with the references from the current year 2022. You should also mention some more other studies on cluster analysis in patients with long term symptom after Covid-19.

Thank you to letting me review this paper and good lack!

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

**********

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

9 Aug 2022

Response to Editor and Reviewers

Editor Comments:

1. Add some description of cluster 1 symptom rates in the results section based on Fig 2A (and suppl Fig 1, also see Reviewer 1 comment)

We have modified the abstract as follows (page 2, results):

“Cluster analysis generated three symptom groups: Cluster1 had few symptoms (most commonly headache)”

We have modified the results section as follows (page 6, paragraph 3, and page 7, paragraph 2)

“Hierarchical cluster analysis generated three symptom clusters (Figure 2 A and B): Symptom Cluster 1 (N=38, 31%) included patients with few symptoms, most often headache and anxiety though at lower rates than reported in other clusters; Symptom Cluster 2 (N=16, 13%) consisted of patients with many symptoms, including high levels of anxiety and depression; and Symptom Cluster 3 (N=68, 56%) included patients who predominantly reported shortness of breath, headache and cognitive abnormalities.”

“A separate hierarchical cluster analysis was performed to identify groups of patients who received similar therapeutic interventions. Three clusters were generated (Figure 3 A and B): Therapy Group A (N=72, 59%) received few therapeutic interventions, most often pain medications and exercise; Therapy Group B (N=12, 10%) received several interventions, notably psychological talk therapy, antidepressants, and anti-anxiety medications; and Therapy Group C (N=38, 31%) primarily received physical and occupational therapy.”

2. Please redo Table 2 completely, because there are a lot of redundant data included (also see Reviewer 2 comment):

o The N for the denominator can be given when different from given in the header or provide N (%) with missing data for applicable subsets.

o Would it make sense to assess differences in variable distributions among the Therapy groups using univariable analysis (chi squared tests and nonparametric/anova test for continuous variables)?

We appreciate these suggestions. Table 2 has been restructured to be more similar to Table 1, where we compare distributions across clusters using chi-squared and non-parametric Kruskal Wallis tests. We moved odds ratios for each demographic, comorbidity, symptom cluster and 12-month outcome variable and its relationship with each therapeutic cluster to supplemental Table 4 as suggested by Reviewer 2.

o For the symptom clusters: it may be interesting to compare odds of falling within cluster 2 and 3 vs cluster 1 (reference group) among Therapy groups.

We appreciate this thought. We have analyzed membership in Symptom Cluster 2 or 3 versus Cluster 1 (reference) and added these data to Table 2. Patients in symptom Clusters 2/3 were more likely to be represented in Therapy Groups B and C, though this did not reach statistical significance (P=0.052).

o For the 12-month outcomes, e.g., "improved with therapy", it may be interesting to compare Therapy group B and C vs A, while adjusting for cluster 2 and 3 dummy variables as well as severity of disease + testing for interaction between the therapy dummy variables (Therapy group B and C) and cluster dummy variables. Perhaps these analyses based on above 2 suggestions could be added separately.

The editor makes an interesting suggestion. We performed a logistic regression analysis predicting “improved with therapy” testing therapy groups B/C vs A, adjusting for covariates symptom cluster 2/3 vs 1, ventilator status (severity of index disease), age and the interaction of these symptom cluster and therapy dummy variables. However, the model was unstable, likely due to the few numbers of patients analyzed and the relatively high number of covariates. When running the model including only covariates therapy groups B/C vs A, age and ventilator status, belonging to therapy groups B/C strongly predicted subjective improvement (aOR 14.5, 95% CI 1.5-142.9, P=0.022).

We added the following to the results section (page 8, paragraph 1):

“Conversely, belonging to Therapy Groups B or C (compared to A) was strongly associated with subjective improvement (aOR 14.5, 95% CI 1.5-142.9, P=0.022) after adjusting for age and ventilator status.”

We also examined dichotomized 12-month outcomes evaluating the variable therapy group B/C vs A adjusting for adjusting for covariates cluster 2/3 vs 1, ventilator status (severity of index disease), age and the interaction of these cluster and therapy dummy variables with the following results (all non-significant):

For outcome mRS 4-6: therapy group B/C vs A, aOR 2.14 (0.34-13.30) P=0.415

For outcome Barthel<100: therapy group B/C vs A, aOR 4.3 (0.62-30.3) P=0.141

For tMOCA≤18: therapy group B/C vs A, aOR 0.64 (0.091-4.56) P=0.658

For anxiety NeuroQoL≥60: therapy group B/C vs A, aOR 3.06 (0.22-42.84) P=0.406

For depression NeuroQoL≥60: therapy group B/C vs A, model unstable

For sleep NeuroQoL≥60: therapy group B/C vs A, model unstable

For fatigue NeuroQoL≥60: therapy group B/C vs A, aOR 4.33 (0.22-84.9) P=0.334

Because none of the above models reached statistical significance, we did not include these in the results section.

o Were continuous variables (incl age, continuous 12-mo outcomes) analyzed as binary outcomes with the median as cut-off value? No wonder the OR are close to 1 when there are sufficient data to estimate the median, and when Ns above vs below median are compared in each cell (i.e., 0.5/(1-0.5)=1).

For Table 2, continuous variables were entered into logistic regression to predict the dichotomous outcoming of belonging to a specific therapy group or not. We did not dichotomize continuous variables at the median.

3. Please include results as supplemental material for: "To ensure the complete data sample was representative of the full sample, we contrasted the dataset used in the cluster analysis and the dataset of individuals with missing data using Chi-square tests for categorical measures and Mann-Whitney-U tests for continuous measures."

Of those who completed 12 month follow-up, data on post-acute symptoms was missing on 12 patients. We have added a supplemental table, as well as revised the results and discussion section to include these data.

We have added the following to the results section (page 6, paragraph 1):

“Of those who completed 12 month follow-up, data on post-acute symptoms was missing on 12 patients. When comparing data among included patients (N=122) to those with missing data (N=12), those with missing data were significantly older (median 74 versus 64 years, P=0.002), had fewer life stressors (median 0 versus 1, P=0.010) and had worse 12-month modified Rankin scores (median 4 versus 2, P<0.001, Supplemental Table 5).”

4. Limitations: Mention the use of a complete case analysis, excluding those with missing risk factors, and the incomplete 12-month survey data in 59% of eligible patients. For the latter, the response rate was quite low (41%) and this may have led to a selected sample.

We appreciate the suggestion and have added the following to the discussion limitations section (page 10, paragraph 2):

“Second, data was missing in 12 patients who were excluded from cluster analyses. Those excluded from analyses were largely similar to those included, however excluded patients were significantly older, had worse 12-month mRS scores, and had fewer life stressors than those included. Additionally, we were only able to contact 242/590 (41%) patients eligible for 12-month follow-up due to defunct contact information, language barriers, lack of patient/surrogate consent or unanswered phone calls. Though we made three phone attempts at contact, our data may represent a biased sample.”

Reviewer #1: I read this manuscript with a lot of interest as the subject is one that is currently not well understood with regards to the COVID-19 pandemic . The paper presents very interesting results regarding outcomes of SARS-CoV-2 infection and the response to treatment. This study is among few studies that have attempted to explore the clusters of post-Covid condition especially the treatment outcomes.

I have few suggestions to make the paper more valuable;

1. In the race to develop a case definition for long Covid, there is a growing interest in its symptom clusters. Accordingly, provide the symptoms that characterises each cluster you identified in your study (you did this for cluster 2 and 3 but not for cluster 1). Reporting that cluster 1 had few symptoms is not enough.

Thank you for this comment.

We have modified the abstract as follows (page 2, results):

“Cluster analysis generated three symptom groups: Cluster1 had few symptoms (most commonly headache)”

We have modified the results section as follows (page 6, paragraph 3, and page 7, paragraph 2):

“Hierarchical cluster analysis generated three symptom clusters (Figure 2 A and B): Symptom Cluster 1 (N=38, 31%) included patients with few symptoms, most often headache and anxiety though at lower rates than reported in other clusters; Symptom Cluster 2 (N=16, 13%) consisted of patients with many symptoms, including high levels of anxiety and depression; and Symptom Cluster 3 (N=68, 56%) included patients who predominantly reported shortness of breath, headache and cognitive abnormalities.”

“A separate hierarchical cluster analysis was performed to identify groups of patients who received similar therapeutic interventions. Three clusters were generated (Figure 3 A and B): Therapy Group A (N=72, 59%) received few therapeutic interventions, most often pain medications and exercise; Therapy Group B (N=12, 10%) received several interventions, notably psychological talk therapy, antidepressants, and anti-anxiety medications; and Therapy Group C (N=38, 31%) primarily received physical and occupational therapy.”

3. The quality of images are poor and needs a lot of improvement. Increase image quality to more than 300 dpi.

We appreciate this suggestion and have modified the figures to be 400 dpi.

Reviewer #2: This interesting paper is about different subgroups/phenotypes of post-acute sequele of Covid-19. The authors performed unsupervised cluster analysis in order to identify subgroups that may help to management/treatment of long term symptom following the infection.

The study was based on the cohort of patients collected in the prospective manner. There were 242 patients whereas 50% reported at least one long term symptom after Covid-19 at 12 months follow-up. The study identified three cluster with few remining symptoms, with several symptoms and with specific symptoms, such as shortness of breath, headache and cognitive symptoms. The cluster with several symptoms were the most affected as they had the highest score of disability. All patients in this groups got psychiatric therapy and the major part received also physical/occupational therapy.

The title is appropriate and well inform on the study.

The abstract is very well written and contain the most important information.

The introduction is clear and consistent; however, I would suggest it can be completed with a paragraph on the previous study on subgroups/clusters as there are several studies on it already in the literature.

The reviewer makes an excellent suggestion. When the paper was originally submitted in early 2022 there were not many cluster analyses evaluating post-acute long COVID symptoms. We have incorporated several references in our introduction (as below), and noted some of the differences in our study.

We have modified the introduction (page 3, paragraph 2):

“While there are cluster analyses evaluating symptoms in the acute phase of COVID-19 illness(14-17), and others focused on specific post-acute symptoms such as pulmonary complaints(18), brain fog(19), or neuropsychiatric symptoms(20), we chose to evaluate the entire post-acute COVID-19 symptom report form published by the World Health Organization (WHO)(21) and map symptom clusters to therapeutic interventions. These analyses may provide insight into the efficacy of various symptom-based treatment strategies and underscore the need for a multi-disciplinary holistic approach to post-COVID-19 care.”

The method a well written. The detailed description of the cohort was published before.

-What kind of psychiatric intervations did the patients received? Please add information.

We have added the following to the Results section (page 7, paragraph 2):

Of 122 PASC patients, 85 (70%) received at least one therapy to manage PASC symptoms and 72/122 (59%) received multiple therapies. The most frequently prescribed interventions were physical or occupational therapy, exercise or pain medications. Psychiatric interventions included: talk therapy (9%), antidepressants (6%) and anti-anxiety medications (5%). Most patients reported a positive response to each therapy (range 75-100% reported symptom improvement), though a smaller proportion of patients (77%) felt that exercise alleviated their symptoms, and 7/85 (8%) reported no improvement with any intervention (Figure 1B.).

-What kind of occupational/physiotherapy did the person reciecived?

The reviewer bring up an important query, however, we do not have details on the specific types of occupational and physical therapy utilized.

-Did you ask about in the survey and did you revive the patients records for these informations?

We obtained information about therapies used directly from the patient or surrogate because many patients in the NYC area seek care at a variety of institutions and some of the therapies we were studying are not well captured in the EMR (e.g. exercise, pain meds, meditation, acupuncture, chiropracter etc).

The results.

- I would suggest you divide the results part in the section, then it will be easier to follow-up the results text.

We appreciate this suggestion and have divided the results into sections (page 6, paragraph 1 to page 8 paragraph 3)

- I would suggest to add the simple figure with for example circles where you first have three circles of clusters (divided by symptoms) and then on them you have three circles regarding therapy.

We agree with this recommendation. A Venn diagram was difficult to construct, but we did create a figure which summarizes the relationship between symptom clusters and therapy groups (Figure 4).

- In the table 2 caputure, you should describe what all the rows of numbers means in each cell. As the groups are small you might consider write only OR with 95% CI and p and the numbers in the supplementary materials.

We appreciate these suggestions. Table 2 has been restructured to be more similar to Table 1, where we compare distributions across clusters using chi-squared and non-parametric Kruskal Wallis tests. We moved odds ratios for each demographic, comorbidity, symptom cluster and 12-month outcome variable and its relationship with each therapeutic cluster to supplemental Table 4 as suggested.

-there are several types of cluster analysis, discuss why did you use the one you selected.

We have added the following to the Methods section (page 5, paragraph 2):

“We selected unsupervised agglomerative hierarchical cluster analysis methods because they provided a flexible approach to assemble symptoms clusters (based on similarity of their taxonomic structures) without prior knowledge about any underlying relationships, and because they allowed us to generate easy-to-interpret views of the clustering structure, an advantage over some other clustering approaches. Agglomerative approaches are also efficient when datasets are smaller, such as ours. Unlike k-means and some other clustering approaches, this method does not require advanced selection of number of clusters, and smaller clusters tend to be generated, which was useful in our exploration of PASC sub-phenotypes. We used Ward’s minimum variance methods to minimizes the total within-cluster variance.”

The discussion:

-Is this correct that the study is the first on post covid phenotypes corresponging to the theraputic strategies? The majority of references in the paper is from 2020-2021 and you should add some more new reference and refered also on some recent paper on the phenotypes.

We appreciate this suggestion and have added more recent references to the introduction (as mentioned above) as well as to the discussion as below (page 9, paragraph 1):

“This study differs from other COVID-19 cluster analyses(15-20) in that, not only did we identify phenotypes of PASC symptoms, but we also mapped these clusters to corresponding therapeutic strategies and their subjective effectiveness.”

We did not identify any papers evaluating therapeutic groups, nor correlating symptom clusters with therapeutic strategies, so we feel our paper is unique in this regard.

- You write: ‘’ Although some studies have suggested that PASC may be more common in those with mild

disease(28), we did not have a comparison group of non-hospitalized patients in our cohort to test this hypothesis.’’ Is this write with the current stage of knowladge? The patients after more severe Covid-19 used to have more somatic remaining symptoms whereas the patients after mild Covid-19 used to have cognitive impairment and fatigue.

We agree that most of the literature supports a strong association of COVID severity with PASC symptoms (including our own). However, there is some conflicting data showing more PASC in patients with milder COVID. This could simply be because there are many more patients with mild COVID than those requiring hospitalization, or that patients with more severe COVID are too disabled to report their symptoms. The discussion section on this topic is as follows (page 10, paragraph 2):

“We found that COVID-19 severity, even amongst a hospitalized cohort, was an overall predictor of PASC in the group of 242 patients that completed follow-up, as has been described in multiple other cohorts(32-34). Although some studies have suggested that PASC may be more common in those with mild disease(35), we did not have a comparison group of non-hospitalized patients in our cohort to test this hypothesis.”

I suggest the major revision of this paper and in particularly suggest to review the recent literature on post Covid phenotypes and complete with the references from the current year 2022. You should also mention some more other studies on cluster analysis in patients with long term symptom after Covid-19.

We agree with this suggestion and have updated the introduction and discussion section with more recent references as mentioned above.

Submitted filename: Response to Editor and Reviewer PLOSOne (1).docx

Click here for additional data file.

30 Aug 2022

Submitted filename: Reviewer 1 comments.docx

Click here for additional data file.

1 Sep 2022

Reviewer 1:

I appreciate the improvements the authors have made on the manuscript. I have one concern on the following paragraph (sections highlighted in yellow).

1. Can the authors explain why the estimated aOR as reported in this paragraph lacks precision? Could this be a chance finding?

We appreciate this query. When contrasting cluster A (n=72) to B-C combined (n=50), we detected large differences, with low probability that these were due to chance, despite relatively small sample sizes. While the exact magnitude of difference may be imprecise, the likelihood of differences between these two groups is strong. Nonetheless, we agree that there is a wide confidence interval and the point estimate is imprecise. We have added the following to the limitations section of the discussion:

“Additionally, there were too few patients in certain treatment categories to make any statements regarding efficacy. While those in Therapy Group A reported lower rates of subjective improvement compared to Groups B or C, confidence intervals were wide suggesting imprecision in point estimates.” (page 12, paragraph 1)”

2. The last sentence that starts with “When ……. is incomplete!!!!

We thank the reviewer for catching this typo and have deleted it.

Subjective Improvement among Therapy Groups

Among patients who received at least one therapeutic intervention (N=85), those in Therapy Group B and C were significantly more likely to report improvement following therapy than Therapy Group A (100%, vs. 97% vs 83%, respectively, P=0.042). This was observed despite the fact that Therapy Group A had significantly better measures of activities of daily living (Barthel Index) and less severe disability (mRS) than other groups. After adjusting for severity of index COVID-19 illness (intubation status) and age, Therapy Group A was associated with significantly lower rates of subjective improvement compared to other therapy groups (adjusted OR 0.07, 95% CI 0.01-0.68, P=0.022). Conversely, belonging to Therapy Groups B or C (compared to A) was strongly associated with subjective improvement (aOR 14.5, 95% CI 1.5-142.9, P=0.022),

after adjusting for age and ventilator status. There was a significant association between the number of therapeutic interventions trialed and the subjective report of improvement following therapy (OR 3.00, 95% CI 1.22-7.35, P=0.017). When comparing

3. Can the authors justify the adequacy of the sample size to perform a cluster analysis of therapy groups? It seems the sample size used to perform the clustering was not adequate.

We appreciate this query. For cluster analyses, the goal is to determine whether subgroups exist within a sample vs a null hypothesis of a single group, and power is generally defined as the probability of correctly detecting that subgroups are present, which increases as a function of both ‘effect size’ (magnitude of difference between groups) and sample size. When separation between groups is large, small sample sizes are not a major concern. Studies have shown that most algorithms can perform optimally even with lower separations with a minimum sample size of 20–30 observations per subgroup (Dalmaijer, E.S., Nord, C.L. & Astle, D.E. Statistical power for cluster analysis. BMC Bioinformatics 23, 205 (2022). https://doi.org/10.1186/s12859-022-04675-1). In our case, we had sample sizes of 12-72 per cluster. When contrasting cluster A (n=72) to B-C combined (n=50), we detected large differences, with low probability that these were due to chance, despite relatively small sample sizes. While the exact magnitude of difference may be imprecise, the likelihood of differences between these two groups is strong.

We have added the above reference to the methods section (page 5, paragraph 2, reference 29)

4. The authors should report briefly on those who received no therapy.

The reviewer makes a good suggestion. We have added the following to the results section (page 8, paragraph 2):

“Among those who received no therapeutic interventions (N=37), the median age was 62 (IQR 51-72), 67% were male and 30% required mechanical ventilation during hospitalization. There were no significant differences in age or sex among those who did or did not receive a therapeutic intervention. However, those who did not receive any therapies were less likely to have required mechanical ventilation during index COVID-19 than those who received some intervention (30% versus 54%, P =0.018).”

Submitted filename: Response to Editor and Reviewers R2.docx

Click here for additional data file.

13 Sep 2022

Post-Acute Sequelae of COVID-19 Symptom Phenotypes and Therapeutic Strategies:  A Prospective, Observational Study

PONE-D-22-12045R2

Dear Dr. Frontera,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Chong Chen

Academic Editor

PLOS ONE

19 Sep 2022

PONE-D-22-12045R2

Post-Acute Sequelae of COVID-19 Symptom Phenotypes and Therapeutic Strategies:  A Prospective, Observational Study

Dear Dr. Frontera:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Chong Chen

Academic Editor

PLOS ONE