Prevalence and Correlates of Long COVID Symptoms Among US Adults

Key Points Question How common are COVID-19 symptoms lasting longer than 2 months, also known as long COVID, among adults in the United States, and which adults are most likely to experience long COVID? Findings In this cross-sectional study of more than 16 000 individuals, 15% of US adults with a prior positive COVID-19 test reported current symptoms of long COVID. Those who completed a primary vaccination series prior to infection were less likely to report long COVID symptoms. Meaning This study suggests that long COVID is prevalent and that the risk varies among individual subgroups in the United States; vaccination may reduce this risk.


Introduction
For a subset of individuals with acute COVID-19 disease, symptoms may persist beyond 1 month, with some patients reporting symptoms at least 6 months later. 1 Initially referred to as postacute sequelae of COVID-19 or post-COVID-19 syndrome, 2 this phenomenon is now more commonly described as long COVID. 3 The World Health Organization 4 defined long COVID as generally occurring 3 months from the onset of COVID-19 with symptoms that last for at least 2 months.
Prevalence estimates for long COVID vary widely, in part because of variability in the definition and sampling frame. A self-report symptom tracking study among 4182 individuals found rates of symptomatic persistence of 13.3% at 1 month and 4.5% at 2 months. 5 In a United Kingdom COVID-19-focused survey, among 20 000 individuals with a positive SARS-CoV-2 test result, 13.7% reported symptom persistence at 12 weeks based on a single survey question. 6 Two studies have used administrative claims or electronic health records to examine long COVID symptoms among samples not limited to inpatients. One investigation using administrative data from the Veterans Affairs health system in the United States confirmed that a range of symptoms affecting multiple organ systems was common, among them respiratory, metabolic, cardiovascular, gastrointestinal, and neuropsychiatric diagnoses, 7 but did not report prevalence of the syndromes per se. More recently, a Centers for Disease Control and Prevention investigation using commercial electronic health record data found that 1 in 5 individuals aged 18 to 64 years and up to 1 in 4 individuals aged 65 years or older experienced new onset of a disease identified using a diagnostic code that could be associated with COVID-19 at or beyond 30 days from onset 8 ; however, that study did not otherwise account for age, gender, or a range of other confounding features. A key limitation in both of these studies is the reliance on coded diagnoses, which may miss individual symptoms that do not contribute to a medical encounter or are not coded as part of an encounter; a study using natural language processing to identify such symptoms of long COVID reported prevalences of 10% to 15%. 9 Numerous aspects of long COVID remain poorly understood, with reviews suggesting that this phenomenon may actually reflect multiple different syndromes. 3,10 In particular, it is not known which individuals will experience full recovery and which individuals will experience persistence of symptoms. One such concern is whether disadvantaged groups, such as individuals from racial and ethnic minority groups or socioeconomically disadvantaged groups, may have a disproportionately high prevalence of long COVID because they have experienced a disproportionately high burden of acute infection. 11 If high-risk individuals could be identified, it might be possible to develop strategies to mitigate or prevent symptom persistence, prompting calls for increased emphasis on investigation of postacute sequelae of COVID-19. 2,12 A prior self-report study identified older age and female gender as correlates of greater risk for persistent COVID-19 symptoms 13 ; associations with gender were further supported in a UK survey. 6 A particular correlate of interest has been the role of prior vaccination and long COVID risk (ie, the extent to which so-called breakthrough infections might be associated with differential risk). In a recent reanalysis of Veterans Health Administration data, 14 including more than 30 000 previously vaccinated individuals who experienced breakthrough infection, the risk of long COVID was modestly but statistically significantly diminished. A corresponding secondary analysis of a complementary study drew on self-report from more than 1 million users of a UK COVID-19 symptom tracking application, including approximately 8400 individuals who experienced COVID-19 infection after at least 1 vaccine dose 15 ; that study found marked protection after a second vaccine dose but no significant protection from an initial dose.
To better characterize symptom persistence with a broader and less select sampling frame, we used data from a multiwave US survey that included questions about COVID-19 encompassing 50 states and the District of Columbia. The survey did not focus only on COVID-19, yielding less likelihood of selection bias than more focused surveys of COVID-19 persistence in which participants opt in. We aimed to characterize the prevalence of long COVID in the general US population, to identify sociodemographic features associated with persistence of symptoms for at least 2 months after onset, and to estimate the protective association, if any, of vaccination.

Study Design
We included data collected from 8 waves of the COVID States Project, a large-scale internet survey conducted for an academic consortium approximately every 6 weeks between February 5, 2021, and

Measures
All respondents were asked if they had received a positive COVID-19 test result, which did not distinguish between polymerase chain reaction test or antigen test, and in which month they received this result. Those who reported any positive diagnosis were further asked whether their symptoms had resolved; for those who identified continued symptoms, they were asked to complete a checklist of commonly reported symptoms. Month of first and second vaccination, where applicable, was also identified via a checklist. All sociodemographic variables were collected by selfreport. Data on race and ethnicity were obtained from 5 US Census categories to confirm representativeness of the US population, with categories analyzed and reported in accordance with a recent medical publication guidance statement. 20

Statistical Analysis
We adapted the World Health Organization 4 definition of long COVID, including all individuals whose survey start date was more than 2 months after the month in which they initially identified a positive COVID-19 test result and defining casees as reporting continued symptoms at the time of the survey.
(A planned sensitivity analysis applied a stricter definition, excluding individuals who said that ongoing symptoms did not affect their life [answering "not at all" regarding effect], consistent with the World Health Organization reference to symptoms that "generally have an impact on everyday functioning," 4 and those for whom loss of smell was the only reported symptom). Vaccination prior to illness was defined by comparing the month of vaccination with the first identified month of illness. indicating 50% or more typed variants reflecting a given variant. For participants who responded to more than 1 survey wave, the most recent survey was included.
We applied multiple logistic regression in R, version 4.0 (R Project for Statistical Computing) 22 to examine the association of persistence of symptoms with sociodemographic features, and then we extended these models to include terms for vaccination status and predominant variant at month of infection. Post hoc analysis also examined the association of age by decade to detect possible nonlinear associations. In logistic regression models including sociodemographic features (Figure 1)    likelihood of long COVID (Figure 2).

Discussion
In this cross-sectional study of a cohort of 16 091 adults surveyed between February 2021 and July 2022 in all 50 states in the US and the District of Columbia, we estimated that 14.7% of those who reported a positive COVID-19 test result more than 2 months previously continued to describe symptoms that they associated with acute infection, or 13.9% after reweighting to reflect the US adult population. These point prevalence estimates were similar when the cohort was restricted to those whose acute illness was 6 and 12 months in the past.
Our results are broadly similar to those previously reported among nonhospitalized cohorts. In a study using app-based symptom recording for 4182 patients with COVID-19, only 4.5% reported symptoms for more than 8 weeks, broadly similar to our results-as in the inpatient cohorts, fatigue was among the most common symptoms, along with dyspnea and headache. 13 Our results are consistent with that app-based study in identifying age and female gender as factors associated with risk, even though our sampling frame is markedly different. Our design more closely resembles that of a very large-scale UK survey, which found the greatest risk for persistence among female respondents and younger respondents. 6 More broadly, the differences among these 3 studies, which all used self-report, may reflect differences in ascertainment and question design and may indicate the importance of multiple convergent methods to characterize long COVID. The associations with income, educational level, and race and ethnicity that we identified in our sample highlight the importance of considering these features in understanding the differential longer-term, as well as shorter-term, outcomes of infection.
Two recent studies directly examined the question of protection afforded by prior vaccination, using different designs. One study examined more than 33 000 previously vaccinated individuals with breakthrough COVID-19 infection from Veterans Affairs electronic health records. 14 The Veterans Affairs population may not fully reflect the general adult population in the US, and coded clinical data may be less sensitive to symptoms than narrative notes or patient-reported symptoms. 9 Still, despite these differences, the approximately 24% reduction in odds of long COVID that we observed after a single vaccination (approximately 33% when applying a stricter definition of long COVID) does approximate the 15% reduction in hazard of long COVID in that study. 14 A complementary study drew on self-report from more than 1 million users of a UK COVID-19 symptom app, including approximately 8400 users who experienced COVID-19 infection after at least 1 vaccine dose. 15 In that study, no apparent protective association of an initial vaccine dose with symptoms beyond 28 days was detected (OR, 1.03; 95% CI, 0.85-1.24), consistent with our findings after a single dose, although marked protection was observed after a second dose (OR, 0.51; 95% CI, 0.32-0.82).

Limitations and Strengths
This study has some limitations. First, because this study used preempaneled respondents in a nonprobability design, we cannot reliably calculate the response rate; as such, nonresponse bias cannot be estimated. However, in other domains, these nonprobability surveys have closely mirrored results from more traditional designs, 24 and prior work with this survey found results that closely approximate estimates obtained using other methods, including probability polls and administrative data. 25 apparent long COVID symptoms would be identified as a consequence of other illnesses. Lacking detailed assessment of respondents' medical history, we also cannot examine the associations of comorbid medical illness or acuity of acute illness with risk for long COVID, which could explain some of the observed associations. Finally, we relied on self-report of symptoms rather than objective physiological or cognitive measures. As such, our results must be seen as complementing, rather than replacing, analyses using administrative claims 14 or electronic health records. Prospective studies will be necessary to confirm our results; the National Institutes of Health RECOVER (Researching COVID to Enhance Recovery) study, for example, will be valuable in providing systematic and objective measurement of sequelae. 27 Despite these limitations, we also emphasize the strengths of this systematic assessment, namely, that by design it should be more representative than other single-cohort studies because it captures individuals drawn from every state. Moreover, because the survey is not specifically aimed at individuals with COVID-19 or symptom persistence, it may be less biased toward those with a greater interest in long-term symptoms than (for example) symptom tracking applications. 13 That is, because recruitment materials did not specify COVID-19 or persistence, our results are less likely to reflect individuals with greater interest in COVID-19 persistence. At the other extreme, our approach is less likely to overestimate prevalence than investigations based solely on artifacts of clinical care.
A key question for further investigation will be the differences by race and ethnicity in the prevalence of long COVID that we observed, even after accounting for a range of sociodemographic correlates. These differences cannot be explained by a lack of access to COVID-19 testing because our outcome definition was contigent on obtaining such a test. The finding that greater educational levels, greater income, and urban vs rural setting are associated with diminished long COVID risk highlights the importance of accounting for nonbiological associations in understanding this phenomenon, a limitation of prior investigations. Finally, the suggestion that rates of long COVID may vary by predominant variant at time of infection also merits further investigation because it may help to inform efforts to understand the mechanisms underlying the development of this syndrome.

Conclusions
In aggregate, the results of this cross-sectional study provide an estimate of the mean point prevalence of long COVID in a large, representative population sample of individuals in the United States, complementing studies using administrative claims, electronic health records, or COVID-19focused self-report apps and surveys. They support the potential protective association of vaccination in reducing but not eliminating long COVID risk. If confirmed in prospective studies, these results may facilitate risk stratification, with a goal of early intervention to minimize the effect of long COVID, and could contribute to efforts to prevent this syndrome altogether.