Our results indicate that the estimated prevalence of long COVID in a population representative sample differs depending on whether pre-infection symptoms are accounted for. In the U.S. population, most people with COVID return to their pre-infection symptom level after the acute phase of the disease. However, more than one-fifth (23%) experience long COVID, with at least one symptom originating around the time of SARS-CoV-2 infection lasting for more than 12 weeks. Without adjusting for pre-infection symptoms, the prevalence is estimated to be 40%, which suggests the potential for a significant over-estimation of long COVID in previous studies.
The most frequently experienced persistent new-onset symptoms among those with long COVID include headache (22%), runny or stuffy nose (19%), abdominal discomfort (18%), fatigue (17%), and diarrhea (13%). The fully adjusted logistic regression model indicates that the likelihood of experiencing long COVID is not significantly associated with sociodemographic or behavioral factors including age, gender, race/ethnicity, education, current smoking status or the presence of chronic conditions. COVID long haulers are more likely to experience hair loss, headache, and sore throat at the time of infection compared to their counterparts whose symptoms reduce more quickly. Also, those who are obese are at higher risk of experiencing persistent new-onset symptoms.
To our knowledge, this is the first study that defined long COVID accounting for pre-infection baseline symptoms. Even before SARS-CoV-2 infection, more than two-fifths (44%) of our sample experienced at least one symptom that can be potentially linked to COVID. Also, among the infected, the prevalence of most symptoms returned to the pre-infection level at the post-infection stage. It means that people report many symptoms both before and after COVID that may be due to other conditions. So, while around 40% of the COVID-infected persons have at least one symptom 12 weeks after COVID diagnosis, this may overestimate the prevalence of long COVID if these persons are all classified as long COVID. The longitudinal nature of data, from pre-infection to post-infection stage, made it possible to distinguish new onset symptoms from the symptoms that might be experienced by someone without SARS-CoV-2 infection. Admittedly, the current approach only picks up on new-onset symptoms and not the changing severity of symptoms. But still, due to the relatively high prevalence of symptoms in our sample even before COVID, our longitudinal and conservative approach can help avoid possible overestimation.
Compared to the estimates of long COVID prevalence based on other nationally representative studies, our final estimate (23%) based on UAS data is between the U.K. ONS estimate (10%) , and the Whitaker et al. estimate (38%) . The three studies are similar in study design and population representativeness, so the difference in estimates may reflect to the different number of symptoms used in each study. Specifically, the ONS estimate is based on 12 symptoms, while the Whitaker et al. estimate on 29 symptoms. The current study included 18 symptoms, which is roughly between the other two studies. The symptoms included in REACT-2 but not in UAS include sudden swelling to face or lips, sore eyes, purple scores/blisters on feet, numbness/tingling, hoarse voice, heavy arms/legs, dizziness, difficulty sleeping, chills, and appetite loss. However, these symptoms generally have low prevalence among the SARS-CoV-2 infected, and/or diminish quickly after initial infection . Hence, the lack of these symptoms in the questionnaire is not likely to cause a significant difference in the estimated prevalence of long COVID.
Our estimated prevalence is also similar to the estimate of 27% based on never-hospitalized COVID symptomatic Californians , and the estimate of 30% based on a sample combining hospitalized patients and outpatients in Seattle, Washington . While it is notably lower than the estimated prevalence of at least 50% using hospitalized patient sample in Michigan . These differences may reflect the fact that we adjust for pre-infection symptoms, and we do not represent the hospitalized population.
The significant association between long COVID and obesity is consistent with previous studies [22, 25, 26]. Both Whitaker et al.’s and the ONS studies found that existing health conditions are associated with elevated long COVID risk. Our results do not show any link between the presence of health conditions and long COVID.
We differ from some existing studies, in that we did not find a significant association between long COVID and any sociodemographic factors included in this study. It is probably because the analytic approaches used by the ONS , Whitaker et al. , and Sudre et al.  to assess risk factors for long COVID either are based on bivariate comparisons, or do not include the effects of existing health conditions as we do. Hence, the age differences and gender differences they found may be explained by health differences across gender and age groups or other uncontrolled factors. Also, Sudre et al. collected data from an international sample including respondents from the U.K., the U.S., and Sweden, while Whitaker et al. focus on England and the ONS focused on the U.K. population. The discrepancy in results may also reflect differences in socioeconomic and demographic context across countries. We found some symptoms reported at the time of infection to be associated with experiencing long COVID, but the symptoms we found (hair loss, headache, and sore throat) are different from the ones identified by Augustin et al.  (anosmia and diarrhea). It is probably because we used new-onset symptoms as the predictors in our regression model, but the previous study was not able to distinguish new-onset symptoms from those started even before SARS-CoV-2 infection.
By limiting our sample to those who had complete data at the pre-infection stage, the time of infection, and post-infection stage, we excluded those who were diagnosed in the last 5 waves of the survey. It means that for the respondents in our sample, the latest possible date of diagnosis was between November 25 to December 23, 2020 (the 19th survey wave). Since COVID vaccines were available for only a small number of health care workers at that time, we do not think the vaccination would change our estimate in any meaningful way, so we did not include information on vaccination status.
Our study has limitations. We are likely to have missed some severe COVID cases since they likely would not have answered the survey while suffering from severe illness. Information on hospitalization is not available in the UAS. Since hospitalized COVID patients generally experience moderate or severe disease outcomes, it is reasonable to assume that they are more likely to have missing data in the UAS and to be excluded from our final sample. This would lead to an underestimate of the prevalence of long COVID. Given that around 5% of the SARS-CoV-2 infected population are hospitalized , and since long COVID is highly prevalent (50%-90%) among hospitalized patients [6–11], we believe that the real prevalence at population level may be higher than our estimate of 23%, ranging from 24–26%. Some limitations of our study are due to the nature of the secondary data we use. The UAS COVID National Survey does not have information on brain fog, which is considered to be a long COVID symptom. So, we failed to include the COVID long haulers who suffered from only persistent brain fog. Finally, Our assessment of long COVID is based on self-reports, instead of clinical diagnoses, and we do not have a clear set of clinical indicators of long COVID. However, self-reported symptoms are still valuable for gaining insights into what is happening in the population. but it provides little information on the pathology or mechanism.
With the availability of vaccines and the onset of new variants, the nation has moved into new stages of the pandemic. The vaccinated population has tripled since the last wave of data used in the current study, and by March 2022, more than 65% of the total US population have been fully vaccinated . The Omicron variant, a variant which spreads more easily than the original virus and the Delta variant , emerged in the US in December 2021 and by February 2022, almost all the new cases were driven by Omicron lineages . It remains unclear how vaccination affects long COVID , and there is limited evidence on whether the Omicron wave has changed what we know about long COVID [37, 38]. Nevertheless, long COVID is still a public health concern. More knowledge on its prevalence, persistent symptoms, and risk factors may help healthcare professionals allocate resources and services to help long haulers get back to normal lives.