High proportion of post-acute sequelae of SARS-CoV-2 infection in individuals 1-6 months after illness and association with disease severity in an outpatient telemedicine population

Study population

Our study population consisted of patients enrolled in Emory Healthcare’s Virtual Outpatient Management Clinic (VOMC) in Atlanta, Georgia between March 24 and September 20, 2020. Patients were offered enrollment in VOMC during results notification calls following a positive SARS-CoV-2 nasopharyngeal reverse transcription polymerase chain reaction (RT-PCR) test at an Emory site (two large throughput outpatient testing centers and the emergency departments of four acute care hospitals). Patients who received a positive test at an outside site (rapid test or RT-PCR) could be referred to VOMC by their Emory provider or by calling the institutional COVID Hotline and hospitalized patients could be referred at discharge if within their isolation period. Patients in VOMC underwent standardized telemedicine assessment by Emory Healthcare providers as previously described (31). Care through VOMC continued for up to 21 days depending on symptom severity and risk factors for hospitalization. Our inclusion criteria were (a) adults at least 18 years of age, (b) previous positive test for SARS-CoV-2, (c) with a valid email address in the Emory electronic health record, and (d) discharge from the Emory VOMC.

Data collection

A quantitative, cross-sectional survey was created using Qualtrics® Online Survey Software (Qualtrics, Provo, UT) to collect participant-reported information on persistent COVID-19 symptoms and recall of acute symptoms. From August to November of 2020, an email was sent to eligible participants inviting them to complete the survey using an identified link. Non-respondent patients were sent reminder emails and called once by study staff. The survey included research consent, general information (race, sex, education, income level), aspects of the participant’s experience with SARS-CoV-2 infection such as severity of their acute disease, which symptoms were ongoing at the time of the survey (33 individual symptoms), questions related to physical and emotional quality of life, and access to healthcare. Specific symptoms of interest are listed in the results tables. Data on specific comorbidities (see results tables) and provider-assessed severity of acute illness at VOMC intake visit were extracted from the medical record for each participant. Provider-assessed severity was defined as “mild”: cough without shortness of breath or wheeze and / or fever, chills and other mild systemic symptoms; “moderate”: any mild symptoms plus dyspnea on exertion, wheezing and / or mid-chest tightness; “severe”: shortness of breath at rest, pulse oximetry < 92%, pleuritic pain and concerning systemic symptoms like syncope, severe weakness, confusion or acute decline in functional status (29,31). The survey was closed after three months of gathering responses.

Data analysis

Data from the survey were exported from Qualtrics into Microsoft Excel and analyzed using SAS v9.4 (Carey, NC). A p-value of 0.05 was considered statistically significant for all analyses. Summary statistics were reported using frequencies and means / medians where appropriate. Bivariate analyses using chi-sq, t-test or crude logistic regression (single variable) were performed between the main outcome (persistent symptoms), the main exposure (provider-assessed severity of acute illness) and all covariates. Covariates included age, sex, race, comorbid conditions, hospitalization, and income. Multivariate logistic regression was then performed using any persistent symptom as an outcome with provider-assessed severity (mild versus moderate or severe) as the main exposure and age, sex, hospitalization and comorbidities as confounders. Tests of collinearity and interactions were performed and variables dropped if necessary. Backward stepwise elimination was then undertaken to determine the final model. A second model was conducted limited to participants who answered the survey at 90 days or greater.

In order to determine if certain categories of symptoms had unique associations with potential risk factors for persistence of these symptoms, we divided them into respiratory (dry or wet cough, dyspnea on exertion, shortness of breath at rest, chest tightness or pain), systemic (fatigue, chills, weakness, joint pains, muscle pains, dizziness), neuropsychiatric (headache, mental fog, difficulty sleeping, irritability, feeling depressed, feeling anxious), and loss of taste and/or smell. We then performed bivariate analyses with these outcomes and all main predictors as per the initial analyses above. Also compared to these categories of symptoms were individual rating of severity of acute symptoms related to that specific persistent symptom (like lower respiratory symptoms for persistent respiratory symptoms).

Multivariate logistic regression was performed for each category of symptom using the main exposure of provider rated illness severity and other covariates as listed above. Model diagnostics were also undertaken including testing of collinearity, interactions and confounding. Individual ratings of symptom severity were not included in the multivariate model.

Lastly, all symptoms were compared to measures of physical and emotional health to determine associations with persistent symptoms and three different self-reported markers of quality of physical and emotional health – whether they report their physical health as worse or much worse than before their COVID-19 (versus same or better), whether their physical health limits their daily activities, and whether their emotional health limits their daily activities (yes or no). Bivariate chi square analyses were performed with each symptom except for those persistent symptoms that were reported in less than 5 individuals.

Ethical Approvals

The study was approved by the Emory University Institutional Review Board. All participants provided electronic consent at the outset of the emailed survey.

Basic descriptive factors

Two hundred ninety individuals consented and completed the questionnaire, of whom 216 (75%) were women and 194 (67.4%) reported a comorbid condition (Table 1). The median age was 44 (range 18, 84) and 125 (43.1%) were Black. The median length of time after diagnosis of their acute COVID-19 was 119 days (range 26, 220), with 70% responding at more than 90 days after their acute illness. Of the sample, 35 (12.6%) were hospitalized, and at the initial intake VOMC visit, 182 (68.7%) were determined to have mild or resolved symptoms, 81 (30.1%) moderate, and 2 (0.8%) severe. The most common comorbid condition was obesity (BMI > 30 kg / m²) followed by hypertension and asthma / COPD (Table 1). Half of the study sample experienced self-reported weight loss during their illness with a mean loss of 10.7 lbs (SD 7.4) (Table 2). About 25% of individuals said their thinking was worse or much worse than before their acute COVID-19 illness and 25.6% said their physical health was somewhat or much worse than before their acute illness (Table 2).

Symptoms

Overall, 115 (39.7%) had persistent symptoms at the time of the survey (Supplemental Table 1). The most common persistent symptom was fatigue (n= 59, 20.3%) followed by dyspnea on exertion (n=41, 14.1%), and mental fog (n=39, 13.5%) (Supplemental Table 1). Table 3 shows the study sample split into two time periods: 26-90 days after the acute COVID-19 phase (“early,” about 1-3 months) and 90-220 days after the acute COVID-19 phase (“late,” over 3 months). There were similar proportions of patients with continued symptoms in the early group (n=34, 38.6%) and the late group (n=79, 39.9%), (p = 0.84). Twenty-nine symptoms are tabulated in table 3 and all of them were very similarly ranked in frequency whether the participant was in the early or late group. Loss of smell was common at 11.9% (n=33) even in the late group (n=21, 10.6%), twice as frequent as loss of taste (n=18, 6.3%). To account for the wide range of days post-COVID-19 that participants answered the survey, t-test analyses were performed for all symptoms based on days post-illness and did not show statistically significant differences based on when they took the survey.

Predictors of Persistent Symptoms

Since the two time periods (early and late) did not differ in the frequency of persistent symptoms or frequency of specific symptoms, we analyzed the study sample first in aggregate. In addition, we repeated the logistic regression model with only those “late” participants who answered the survey at 90 days or later from acute illness. Our main predicted exposure of moderate to severe disease severity at time of diagnosis (intake visit) was highly associated with persistent symptoms on both unadjusted and adjusted models (Table 4), with an aOR of 3.24 (95% CI 1.75, 6.02). This was similar in the 90 days and greater group. Being hospitalized was also more common in those with persistent symptoms but did not reach statistical significance on the adjusted logistic regression models. In the overall group, female sex was not statistically significantly associated with persistent symptoms, but when the model was limited to the late reporters, female sex (aOR 2.73, 95% CI 1.10, 6.79) was associated. Middle-age (aOR 2.08, 95% CI 1.07, 4.03) was associated with having persistent symptoms in both adjusted models. The only comorbid condition to have a significant result in the model was hypertension which had a negative association with persistent symptoms (aOR 0.45, 95% CI 0.22, 0.94), that remained when limited to the longer duration of symptoms (Table 4).

When persistent symptoms were divided into categories of symptoms (respiratory, neuropsychiatric, systemic, and loss of taste / smell), the provider-assessed severity at VOMC intake visit remained associated with these persistent symptoms (Table 5) in unadjusted bivariate analyses. Also included were individual patient rating of severity of acute symptoms (recall at time of survey) related to that specific persistent symptom (e.g. patient-rated severity of acute lower respiratory symptoms for persistent respiratory symptoms). Severe ratings for these were also consistently associated with the persistent chronic system-based symptoms.

Interestingly, comorbid lung disease was associated with persistent systemic symptoms (OR 2.18, 95% CI 1.13, 4.22) and taste / smell symptoms on bivariate analyses (OR 2.23, 95% CI 1.01, 4.94), but not with persistent respiratory symptoms (0.88, 95% CI 0.41, 1.90). These did not hold in the final logistic models, however. For those that reported moderate to severe taste and smell disturbance during their acute illness (n=190, 65.5%), 81% of those had resolution of their symptoms at the time of their completion of this survey (Table 5).

Symptoms associated with physical and mental health

Table 6 shows associations with individual symptoms and three different markers of quality of physical and emotional health. For physical health being worse than before acute COVID-19, the commonly reported persistent symptoms most associated were fatigue (OR 10.48, 95% CI 5.47, 20.07), muscle aches (OR 10.47, 95% CI 3.95, 27.81), weakness (OR 10.76, 95% CI 3.75, 30.84), feeling depressed (OR 10.35, 95% CI 4.14, 25.9) and mental fog (OR 9.82, 95% CI 4.63, 20.85. Although less commonly reported and with wide confidence intervals, joint pains (OR 17.1, 95% 5.56, 52.59) and shortness of breath at rest (OR 11.19, 95% CI 2.27, 55.18) had even stronger associations. For physical health limiting daily activities, the common symptoms most associated were dyspnea on exertion (OR 13.68, 95% CI 5.98,31.28), mental fog (OR 12.41, 95% CI 5.41, 28.48), headache (OR 12.24, 95% CI 4.81, 31.17), and joint pains (OR 12.06, 95% CI3.95, 36.85). While not a commonly reported symptom, heart palpitations (OR 16.17, 95% 4.62, 56.53) was strongly associated with limitations of daily activities. For emotional health limiting daily activities, the following symptoms were most associated: irritability (OR 6.02, 95% CI 2.02, 17.9), feeling depressed (OR 5.53, 95% CI 2.35, 12.99), back pain (OR 5.40, 95% CI 1.78, 16.32), feeling anxious (OR 4.71, 95% CI 1.78. 12.43), and ear fullness (OR 3.86, 95% 1.06, 14.04).

Principal Findings

We find that chronic symptoms are common in a follow-up survey of patients previously enrolled in a COVID-19 outpatient telemedicine monitoring program, with 40% of patients reporting persistence of at least one symptom. We hypothesized based on prior work (29) that initial symptom severity (at time of telemedicine intake visit) would predict the likelihood of long-term symptoms and we find this association is significant. Patient recall of symptom severity (reported at time of survey) finds similar relation strengthening the internal validity of this finding. We do not find association between other factors such as demographics or comorbidities and the presence of symptoms at the time of follow-up survey except for female sex (increased likelihood) and hypertension (decreased likelihood), nor do we find a significant difference between patients responding to the survey closer to their acute illness versus those farther out from their acute illness.

The mechanism(s) underlying the persistence of symptoms beyond acute COVID-19 remain unclear; the finding of a relationship between the severity of acute symptoms (recorded at the end of the first week of illness) and the prevalence of symptoms months later suggests a possible role for acute inflammation and/or viral burden. Subsequent hospitalization may predict risk in the unadjusted comparison but is not significant in the adjusted model. This may be due to the low number of hospitalizations in our cohort, but also that risk factors for hospitalization (age, male sex, comorbidities) are not directly associated with risk of long-term symptoms.

These findings are significant in public health messaging and individual risk communication for patients with COVID-19. The factors predicting persistent symptoms remain unclear and it appears that patients and the public cannot be reassured that individuals are unlikely to develop persistent symptoms until their initial course (i.e. first week of illness) is past. Furthermore, we do not find a difference in the frequency of persistent symptoms in our “early” (day 26-89) and “late” (day 90-220) groups. While it is possible that there are differences between these groups based on the “wave” of the pandemic, it suggests that there is not a clear time for expected resolution of symptoms that persist, and further study will be needed to determine long-term outcomes.

Many respondents to the survey reported lower quality of life on questions regarding limitations due to physical and emotional health, similar to a prior report indicating disrupted home, work, and social life for patients after acute COVID-19(6). These responses were associated with certain common symptoms (e.g. fatigue, dyspnea on exertion, mental fog) more than others (e.g. cough, change in smell or taste). These findings may suggest that specific symptom profiles are more significant in quality of life and may warrant particular emphasis in studies of PASC. More than half sought care for their symptoms indicating a level of concern leading to medical attention. Prior work indicates that patients encounter challenges due to limited recognition of PASC (5,32) and providers should be aware of the likelihood of PASC and evolving treatment guidelines.

Comparison to prior studies

Similar to previous studies (23,25–27,33,34), this study represents a single institution report of persistent symptoms in patients after acute COVID-19 and finds a high rate of persistent symptoms. A novel aspect of this work is the prospective enrollment in the telemedicine program with standardized data collection by providers in the acute phase of illness, such that risk factors for later outcomes could be systematically analyzed. We found that the severity of symptoms at the time of the acute telemedicine visit was a significant predictor, which corroborates the prior reports relating acute illness profile to prolonged symptoms (24–26). We do not find a strong association with hospitalization, suggesting that the term “severity” may need further clarification as a symptom descriptor distinct from “disease severity” (often referring to highest level of care, which is more strongly associated with age, sex, and medical comorbidity). The use of the physical health questions also gives more description and nuance to our findings, which is an added strength over large cohort studies of reported frequency of persistent symptoms.

We find that female gender is associated with PASC in our cohort, especially in the later group, similar to prior studies (23–25). It is well described that males are at higher risk for severe acute COVID-19 (35), and it has been proposed that genetic differences (including expression of Toll-like receptor 7 on the X chromosome) may account for this(36), but a mechanism relating to increased risk for PASC is not clear. There may be a common mechanism with other post-infectious syndromes: female gender has been reported as associated with lower six minute walk distance in survivors of SARS (9) as well higher rate of post-dengue fatigue (13), post-chikungunya arthralgia (12) but not consistently for post-Ebola symptoms(11).

Middle age (40-59) was significant in the adjusted model, similar to prior report (26). Given the strong association of illness severity (i.e. risk of hospitalization and death) with age(35), hypothesized to relate to immune system changes across age groups(36), it is plausible that immune-system mediated factors during acute illness increase the risk for PASC in individuals in middle age more than other age categories.

The role of hypertension in acute COVID-19 is not clear - initially reported as a possible risk factor for severe COVID-19 (in excess in severe versus nonsevere cases)(37), has not been shown to be a robust predictor of severity with age-hypertension interactions even decreasing risk(35) and improved blood pressure control correlating to worse outcomes (38). Antihypertensive medications may impact disease course, with preliminary results indicating effects that may vary by race(39). In this context, it is possible that there is a blood pressure (or antihypertensive) correlation with PASC and this relationship merits further study.