This cross-sectional database study used the 2022 Medical Expenditure Panel Survey (MEPS) dataset. The MEPS involves a series of panel-style interviews over a 2-year time frame using a subsample of individuals from the previous year’s National Health Interview Survey. The MEPS includes 3 core components: the household component, the medical provider component, and the insurance component. MEPS staff can use data from the medical provider component or the insurance component to supplement data in the household component. The 2022 full-year consolidated data file used in this study included data for 22,431 individuals and contained a wide variety of data related to personal demographic characteristics and health status. The MEPS is approved annually by an institutional review board, and all individuals provide informed consent before participating [15-17].

Eligible individuals for this study included American adults (≥18 years of age) who were alive for the duration of the 2022 calendar year. Data for individuals under 18 years of age were not included in the MEPS data file to protect their anonymity. Individuals had to be alive for the full calendar year to ensure that they were within the scope for this analysis. Eligibility was determined using the person disposition status (MEPS variable: “PSTATS31/42/53”) and age (MEPS variable: “AGE22X”).

The dependent variable in this study was having MEPS-defined long COVID. As part of the priority conditions enumeration section of the MEPS, participants are asked if they have ever been diagnosed with COVID-19 (MEPS variable: “COVIDEVER53”). If an individual has ever been diagnosed with COVID-19, subsequent questions are asked to determine whether the individual had symptoms lasting 3 months or longer that they did not have prior to having COVID-19 (MEPS variable: “LCEVER53”). If this was the case, then the individual was deemed by the MEPS to have long COVID. This is described as “MEPS-defined long COVID” for the purposes of this study. Although some variables related to COVID-19 have been included in the MEPS datasets since 2020, the 2022 dataset is the first time that a variable related to long COVID has been included. Therefore, this represents the first opportunity to investigate MEPS-defined long COVID in the nationally representative MEPS dataset [16,17].

SAS (SAS Institute Inc) was used to perform the analysis of the data. As the MEPS is structured as complex survey data, the SAS survey procedures (eg, SAS PROC SURVEYFREQ and PROC SURVEYLOGISTIC), as well as cluster and stratum variables, were used. The cluster variable identifies the primary sampling unit, whereas the stratum variable identifies the stratification used in sampling. Furthermore, the appropriate weighting variable was used to obtain weighted population-based estimates [18]. The weighting variable provides the final annual person-level weight and helps calculate accurate SEs. Data that were coded as missing, inapplicable, refused, not ascertained, unknown, or incalculable (as indicated by negative codes in the codebook) were not included in this analysis. Differences between the characteristics of individuals in the 2 comparison groups (ie, those with MEPS-defined long COVID vs those without) were assessed using chi-square tests. A multivariable binomial logistic regression model was then developed to assess the association between each variable and having MEPS-defined long COVID. The event group was having MEPS-defined long COVID, and the reference group was not having MEPS-defined long COVID. An α value of .05 was selected a priori. Multicollinearity among variables was assessed and not detected.

The University of Arizona Institutional Review Board approved this study (00006677; July 16, 2025).

The key findings from this study were the prevalence of MEPS-defined long COVID and the identification of variables statistically associated with higher or lower odds of having MEPS-defined long COVID among American adults. On the basis of the weighted analysis, approximately 18,397,214 American adults had MEPS-defined long COVID in 2022. This figure is consistent with data from the 2022 National Health Interview Survey, which found that 6.9% of American adults had ever had long COVID, although both figures are likely underestimates due to inconsistent definitions, lack of definitive diagnostic tests, and mild or indistinguishable symptoms of long COVID [19]. Furthermore, variables identified in this study associated with higher or lower odds of having MEPS-defined long COVID included 3 predisposing variables (age, sex, and Asian race), 2 enabling variables (marital status and employment status), 3 need variables (number of chronic conditions, health status, and instrumental activity of daily living limitations), 1 personal health practices variable (having ever received the COVID-19 vaccine), and 1 external environmental variable (residence in the south region). These findings are interpreted and discussed below.

In this study, among the predisposing variables, age of 65 years or above was associated with lower odds of having MEPS-defined long COVID. Other research on this is mixed. For example, in the meta-analysis by Tsampasian et al [8], older age was associated with an increased likelihood of developing long COVID, whereas in the studies by Ford et al [20] and Kim [12], prevalence of long COVID was lower in older adults than in other groups. Nevertheless, the data demonstrate that there are perhaps additional factors associated with long COVID in older adults that require further investigation. Speculatively, it may be the case that older adults were able to better isolate themselves during the COVID-19 pandemic and, therefore, were less likely to develop long COVID. It may also be the case that older adults do have long COVID but attribute the symptoms to the aging process.

Regarding the other predisposing variables, the finding that male sex was associated with lower odds of having MEPS-defined long COVID was very consistent with other research: female sex is consistently identified as a risk factor for long COVID [6-8]. Among the racial and ethnicity factors, Asian race vs other or multiple races was the only factor significantly associated with lower odds of MEPS-defined long COVID. These findings were somewhat unexpected, but again, the research is not very clear in this area. Some studies have shown that White individuals are more likely to have a long COVID diagnosis. A study by Shi et al [21] identified that White individuals were more likely than people from other racial groups to develop long COVID, and research by Pfaff et al [22] examining patterns in International Classification of Diseases, 10th Revision, Clinical Modification, coding identified that White, non-Hispanic individuals were more likely to receive a long COVID–related diagnosis than other groups. In contrast, Glassman [23] examined the US Census Bureau Household Pulse Survey data, finding that Hispanic and Black respondents were more likely than White respondents to report long COVID symptoms. Notably, in the analysis by Glassman [23], Asian respondents were the least likely to report long COVID symptoms, which does align with the findings of our study. Similarly, lower rates of long COVID were identified among Asian individuals living in the United States compared with other racial groups in a study by Wu et al [24], consistent with the findings of our study.

Regarding enabling variables, only being married and being employed were associated with higher odds of MEPS-defined long COVID relative to those who were never married or not employed. It was notable that educational level, income or poverty status, and health insurance type were not associated with MEPS-defined long COVID in this study. These findings are interesting given that other research, such as the studies by Subramanian et al [10] and Feldman et al [11], has shown higher rates of long COVID among individuals with higher levels of economic hardships vs those in more economically advantaged groups. Being married and employed are generally more stable economic positions. This suggests that additional factors may be at play, perhaps underreporting or underrecognition of some of the nonspecific symptoms of long COVID, such as fatigue, myalgias, or memory difficulties, in this population. Further research could provide more insights into this finding.

Results for the need variables are mostly consistent with data published elsewhere. Having fewer chronic health conditions was associated with lower odds of having MEPS-defined long COVID compared with having 5 or more conditions. Similarly, having an excellent or very good health status was associated with lower odds of MEPS-defined long COVID compared with having a fair or poor health status. Furthermore, having instrumental activity of daily living limitations was associated with higher odds of MEPS-defined long COVID compared with no limits. All of these factors suggest a positive association between worsening health status and likelihood of MEPS-defined long COVID, which does align with previously published data. One notable area of difference was BMI, which was not associated with MEPS-defined long COVID in this study, although it has been reported as an associated factor elsewhere [8,10].

Regarding personal health practices, only a previous receipt of the COVID-19 vaccine was associated with lower odds of MEPS-defined long COVID. These findings mirror what is published in the literature on reduction of risk of long COVID with receipt of the COVID-19 vaccine. In future studies, it would be interesting to explore data on the COVID-19 vaccine in greater detail beyond the dichotomous data available in the MEPS, for instance, to determine whether there is an association between the number of doses of the vaccine, accounting for the initial 2 doses and any additional doses administered over time, as well as factors such as waning immunity and the emergence of additional SARS-CoV-2 variants. In this study, smoking status, reported exercise habits, and receipt of the influenza or shingles vaccine were not statistically significantly associated with MEPS-defined long COVID.

In this study, residing in the south vs the west was associated with lower odds of MEPS-defined long COVID. No significant associations were identified between the other regions. This finding is unexpected given that a study by Blanchflower and Bryson [25] determined that Alabama, Mississippi, and West Virginia had the highest rates of long COVID among all US states. Additional factors such as population racial or ethnic makeup, health care access, practice patterns of health care providers, and perceptions of SARS-CoV-2 infection may also be affecting this finding. Given the large number of states included in each region, variations in state-level policies and practices related to COVID-19, and a myriad of other considerations, the region variable may not be the most informative to advance a meaningful understanding of long COVID.

Several limitations may influence the broad applicability of this study. As a cross-sectional study, causality cannot be determined, only association. These associations may represent predictive or consequential factors for long COVID. In addition, MEPS data are self-reported, and as such, are susceptible to recall and self-report bias. This study used MEPS-defined long COVID. Alternative definitions, such as patient-reported long COVID or a clinical diagnosis, may result in different findings. These data are from 2022, a point when long COVID had only recently been recognized by the medical community; it is possible that some participants in the MEPS were not familiar with symptoms of long COVID at that time and, therefore, may not have recognized the symptoms to report. Furthermore, this may have been especially pertinent for individuals who had less contact with the health care system in general (noninsured individuals or those with low income). Only variables captured by the MEPS were available for analysis, so other factors may provide additional insights into risks of long COVID. Relatedly, some data were only available in a binary form, which prevented a more nuanced examination of the variables. For instance, the binary nature of the COVID-19 vaccination variable may miss some of the variability in this factor given that there were initially 2 doses of the vaccine followed by additional doses as immunity waned over time and additional variants of the virus became apparent. Data were missing for some variables (due to responses being not reported, ineligible, or not ascertained), which may have biased the results and should also be considered when interpreting the findings of this study. The comparison group in this study was any American adult without long COVID regardless of whether the individual had previously had COVID-19. It is possible that some of the associations detected in this analysis reflect differences in COVID-19 exposure, testing, and diagnosis or access to health services. Future work may contemplate a similar study with a comparison group of individuals who had previously had COVID-19 but did not develop long COVID.

This study analyzed self-reported MEPS data to identify the prevalence and factors associated with MEPS-defined long COVID among American adults. These findings offer additional insights into the limited literature available on long COVID in the United States. Some of the findings from this study reinforce what is already known about long COVID, whereas others indicate a contrary result that warrants further investigation of the complex environmental and social factors associated with MEPS-defined long COVID. Further research is necessary to monitor and address trends in the prevalence of this condition in the United States to further our understanding of the factors associated with it and to determine which factors cause long COVID and which are a consequence of it.