Factors Associated with Long Covid Symptoms in an Online Cohort Study

Design, Setting, and Participants

The COVID-19 Citizen Science (CCS) study is an online cohort study that began enrolling participants on March 26, 2020.¹⁷ CCS is hosted on the Eureka Research Platform (University of California, San Francisco), a digital platform for clinical research studies including a mobile application (app) and web-based software. Participants are recruited through email invitations to participants in other Eureka Research Platform studies, press releases, word-of-mouth, and by partner organizations. Participants must be 18 years of age or older, register for a Eureka Research Platform account, have an iOS or Android smartphone with a cell phone number (or enroll in the web-based study launched January 21, 2021), agree to participate in English, and provide consent. There were no geographic restrictions, but 98% were United States residents. After providing electronic consent, participants complete baseline, daily, weekly, and monthly surveys. CCS methods have been previously described.¹⁷ For this analysis, we included data collected March 26, 2020 to May 18, 2022. The primary analyses included all individuals who reported a positive COVID-19 test result (PCR, Antigen, or Antibody) more than 30 days prior to May 4, 2022 and responded to a survey about Long COVID symptoms, but for longitudinal symptom comparison we also included those without a positive COVID-19 test and compared to those without SARS-CoV-2 infection. Sample size was not determined a priori. The study was reviewed and approved by the UCSF Institutional Review Board (#17-21879). Results are reported in accordance with STROBE guidelines.¹⁸

Long COVID Symptoms

Participants who reported a COVID-19 positive test more than 30 days prior were offered a cross-sectional survey about Long COVID Symptoms in January 2022 and May 2022 (surveys went out on different days depending on participant’s last answered survey). The survey asked about the presence, duration, and severity of Long COVID symptoms using a non-validated instrument. Severity was assessed using a Likert-scale asked for each reported symptom: “How bad do you think these symptoms were?” (1-5, very mild to very severe). Additionally, the survey asked about health care use and missed days of work or school due to Long COVID symptoms.

Longitudinal Symptoms

In addition to cross-sectional surveys, participants were surveyed on a daily or weekly basis regarding symptoms including: a scratchy throat, a painful sore throat, a cough, a runny nose, symptoms of fever or chills, a temperature >100.4°F or 38.0°C, muscle aches, nausea, vomiting, or diarrhea, shortness of breath, unable to taste or smell, and red or painful eyes. We did not use these surveys to classify people as having Long COVID. We categorized the surveys for each individual by time period relative to date of SARS-CoV-2 positive test into 30-60 days pre-infection, 0-30 days pre-infection, 0-30 days post-infection, 30-90 days post-infection, 90-180 days post infection, 180-365 days post infection, and >365 days post infection. For each period, we summed the proportion of respondents averaging _≥1 symptom and the average number of symptoms reported by Long COVID status among respondents to the cross-sectional survey and among SARS-CoV-2 infected non-respondents, and uninfected individuals (averaged over all time points since March 2020).

Other variables

Variant wave was classified by the date of first positive test: Initial (prior to March 11, 2021), Alpha (March 11, 2021-July 3 2021), Delta (July 4, 2021-December 25 2021), and Omicron (December 26, 2021-April 4, 2022).¹⁹ Participants self-reported demographics, medical history, and vaccine history via surveys. Most participants (969/1480, 65.5%) completed baseline surveys prior to SARS-CoV-2 infection, while the remainder enrolled after infection. Participants were queried about lifestyle factors before and after COVID including number of days per week they exercised, typical amount of sleep, and number of alcoholic drinks they consumed.

Standardized instruments including the PHQ-8 for depression²⁰ and GAD-7 for anxiety²¹ were used in a self-administered format and were assessed before and after COVID for those who completed surveys prior to infection. Socioeconomic status was assessed using the MacArthur Scale of Subjective Social Status.²²

Statistical Analysis

The presence of Long COVID symptoms was defined based _≥1 symptom reported more than 30 days after SARS-CoV-2 positive test on the cross-sectional survey. First, we compared demographics, pre-COVID medical history, socioeconomic variables, and lifestyle patterns among those with Long COVID symptoms compared to those without Long COVID symptoms using Chi-squared tests for categorical variables and T-tests for continuous variables. Then we described the prevalence of each reported symptom and patterns of symptom persistence. To assess risk factors associated with Long COVID and adjust for potential confounders, we constructed multivariable logistic regression models in a pre-specified staged approach using a complete case approach to missing data. In an initial (baseline) model (Stage 1) we included age, sex, variant wave, number of symptoms during acute infection, and past medical history. For the next model (Stage 2) we added vaccine status and timing, Hispanic ethnicity, and sociodemographic variables including socioeconomic status, education level, and employment in healthcare. In the final model (Stage 3), in which we prespecified including pre-COVID variables with p<0.10 in univariate analysis, we added pre-COVID anxiety, depression, and financial insecurity. We additionally conducted sensitivity analyses considering only those with pre-COVID baseline assessments, only those with persistent symptoms and only those with severe/very severe symptoms. Statistical significance was considered to be p<0.05 for all analyses other than the prespecified potential predictor selection process in Stage 3. All analyses were conducted with SAS version 9.4 (Cary, NC).

Sensitivity Analyses

First, we considered whether selection bias bias from those who joined the study and completed baseline surveys after SARS-CoV-2 infection impacted the findings. Among 969 individuals with pre-infection baseline surveys, 239 (25%) reported Long COVID symptoms compared to 237/511 who completed baseline surveys after SARS-CoV-2 infection (46%; p<0.0001), which suggests that individuals with Long COVID symptoms after SARS-CoV-2 infection were more likely to participate than those without Long COVID symptoms, consistent with selection bias. Importantly, mean levels of reported physical activity, sleep duration, alcohol intake, financial insecurity, depression, and anxiety were similar between those with baseline surveys obtained prior to SARS-CoV-2 infection and those who joined the cohort after infection, suggestive that differential recall is unlikely to bias these variables (eTable 2). The general pattern of results for the multivariable models were unchanged when only individuals included with baseline surveys prior to infection (n=239 with and = 730 without symptoms) were included (eTable 3).

Second, we considered only persistent symptoms (n=227) compared to those no symptoms (n=1004), excluding those whose symptoms resolved (eTable 3). Overall results were similar with one exception: female sex was associated with persistent symptoms (OR 1.34, 95%CI 1.21-1.49; eTable 3). We also compared those with severe or very severe symptoms (n=62) to those without symptoms (n=1004), with no substantive differences in our findings (eTable 3).

Comparison of Symptom Patterns and Persistence

Our findings of common symptoms (fatigue, shortness of breath, confusion, and headache) are consistent with prior reports.^1-3 Similarly, we found that symptoms were persistent for more than 12 months among approximately half of those infected with SARS-CoV-2 who reported symptoms lasting at least one month. This is consistent with the prior literature that symptoms present for more than 3 months tend not to self-resolve,¹ but higher than estimates that only 15% of individuals with symptoms at 3 months continue to have symptoms beyond 1 year.²

Risk of Long COVID by Symptoms and Variant Wave

The number of symptoms during acute infection was associated with Long COVID consistent with prior reports that acute illness severity is associated with Long COVID,⁸ and raises the question of whether reducing acute symptoms through acute treatment might modify the risk of developing Long COVID. A second interesting finding is that the variant wave is associated with Long COVID symptoms even with adjustment for timing of vaccination (pre-infection, post-infection, or not vaccinated) and number of symptoms during acute infection; more recent variants were associated with lower odds of Long COVID. One prior study suggested that there may be some subtle differences in Long COVID symptoms by variant wave (more dyspnea with ancestral strain, more neuropsychiatric and myalgic symptoms with Alpha, and hair loss with Delta for example).^23,24 Our findings are consistent with three prior studies that suggested that there may be a lower prevalence of Long COVID with the more recent variants (Epsilon, Omicron).^15,25,26

Demographics, Social Determinants of Health and Long COVID

Even within a relatively homogenous cohort, we found that Hispanic ethnicity, lower socioeconomic status and financial insecurity were associated with Long COVID symptoms. In contrast to prior reports, we did not identify female sex as associated with Long COVID symptoms after adjustment, except in sensitivity analysis of persistent symptoms.^10,28 Despite extensive research documenting the role of adverse social determinants of health increasing risk of acute SARS-CoV-2 infection, there are limited prior data regarding associations between social determinants and Long COVID. Consistent with our study, one prior study found that financial concerns were associated with worse health utility and quality of life among those recovering from COVID-19.²⁷ Similarly, an online survey-based study found that graduate education and urban residence were associated with lower odds of Long COVID.¹⁵ The implications are that clinical trials of potential therapeutics should make intentional efforts to include those at highest risk, including those of lower socioeconomic status, and social determinants of health should be considered in public health approaches to address Long COVID.

Depression, Anxiety and Long COVID

In our adjusted analysis, pre-COVID depression was associated with Long COVID symptoms. Importantly, we found that anxiety and depression scores did not increase after COVID among those with pre-infection baseline surveys, suggesting that individuals with depression may be at elevated risk of Long COVID. One prior study from three large cohorts identified that depression, anxiety, perceived stress, and loneliness measured before the pandemic were associated with post-COVID conditions, although we found that anxiety decreased among those with Long COVID.²⁹ Our findings are consistent with studies that lack pre-infection assessments that have found concurrent depression or anxiety to be associated with Long COVID symptoms.^{10,11,28,30,31} Further research into mechanisms of how depression may be an antecedent factor to Long COVID are needed. Fluvoxamine, a selective serotonin reuptake inhibitor (SSRI), may have protective effects in acute SARS-CoV-2 infection,^32,33 but understanding whether treatment with antidepressants or naltrexone³⁴ may prevent or treat Long COVID requires clinical trials.³⁵

Strengths and Limitations

Strengths of this online cohort are a large sample size, data collection prior to infection in many participants, and data from participants infected during different variant waves; most prior studies included predominantly individuals infected with earlier variants. The primary limitation arises from responder bias: those with Long COVID are more likely to respond to surveys although over 1000 individuals without symptoms also responded. This may induce bias in our estimate of Long COVID symptom prevalence but is less likely to bias ascertainment of factors associated with Long COVID. The second limitation is that a subset of individuals joined the study after SARS-CoV-2 infection, limiting the ability for prospective comparisons to those who had already joined the study. Symptoms may be misclassified, and some may be attributable to specific medical conditions rather than Long COVID. We did not assess the effect of repeat SARS-CoV-2 infection. Finally, external generalizability may be limited as the study sample overrepresented those who identify as white, female, and of higher socioeconomic status.