State and Metropolitan Area Disparities in Long COVID-19 and Related Symptoms among US Adults, June-October 2022

2.1. Data

We used pooled data from five consecutive samples of the US Census Bureau's Household Pulse Survey (HPS) conducted during June to October 2022. The five independent HPS samples that included information on COVID-19 diagnosis, Long COVID, and severity of COVID symptoms among adults aged ≥18 years were: Week 46, June 1-13; Week 47, June 29-July 11; Week 48, July 27- August 8; Week 49, September 14-28; and Week 50, October 5-17, 2022.²⁷ The HPS is a national sample household survey in which data on socioeconomic, demographic, physical and mental health, food insecurity, childcare, and healthcare characteristics, including health insurance coverage, COVID-19 diagnosis and vaccination, and access to telehealth during the COVID-19 pandemic are collected in near real time via email and internet. The HPS was developed as a rapid response survey, in partnership with several federal statistical agencies, to track social and economic impacts of the COVID-19 pandemic on American households on a weekly or a bi-weekly basis.^28,29 Information collected in the survey is based on self-reports by respondents aged 18 years and older. The HPS employs a systematic sample design, ensuring its representativeness of the US civilian non-institutionalized population. Substantive and methodological details of the survey are available elsewhere. ^27-29

2.2. Measurement of the Outcome Variables: Long COVID and COVID Disease Severity

We used two COVID-related binary outcome variables. The first outcome variable, Long COVID, was based on the question, “Did you have any symptoms lasting 3 months or longer that you did not have prior to having coronavirus or COVID-19? Long term symptoms may include tiredness or fatigue, difficulty thinking, concentrating, forgetfulness, or memory problems (sometimes referred to as “brain fog,” difficulty breathing or shortness of breath, joint or muscle pain, fast-beating or pounding heart (also known as heart palpitations), chest pain, dizziness on standing, menstrual changes, changes to taste/smell, or inability to exercise.” COVID patients responding “yes” were defined as having Long COVID and those with “no” response were considered not to have Long COVID.

The second outcome variable, disease severity (i.e., severe COVID symptoms), was based on the question,“How would you describe your coronavirus symptoms when they were at their worst?” Respondents who reported severe symptoms were defined as having severe COVID symptoms, and those reporting no, mild or moderate symptoms as not having severe symptoms.

The pooled sample size from June-October 2022 HPS was 108,064 adults aged ≥18 years for whom information on the outcome variables and the covariates was available.²⁷

2.3. Measurement of Geographic Area and Covariates

The independent variable of interest was the respondent's state or Metropolitan Statistical Area (MSA) of residence at the time of the survey.²⁷ Data were available for all 50 states and DC and the 15 largest MSAs in the US, as shown in Tables 1-2 and Figures 1-2.

Close

Figure 1:

Prevalence¹ and Adjusted² Odds Ratios for Long COVID among Adults Aged ≥18 Years in the 15 Largest Metropolitan Statistical Areas (MSAs), United States, June-October 2022 (N=33,253)

¹Long COVID prevalence rates for all MSAs except Boston, Philadelphia, Seattle, and DC MSAs were significantly higher than the rate for the San Francisco MSA at p<0.05. ²Adjusted by logistic regression for age, race/ethnicity, gender, LGBT status, marital status, education, household income, housing tenure, and insurance status. The adjusted odds ratio for Atlanta, Chicago, Detroit, Phoenix, and Riverside MSAs were significantly higher than the San Francisco MSA at p<0.05.

Source: Data derived from June-October 2022 Household Pulse Survey

Export to PPT

Close

Figure 2:

Prevalence¹ and Adjusted² Odds Ratios for Severe COVID Symptoms among Adults Aged ≥18 Years in the 15 Largest Metropolitan Statistical Areas (MSAs), United States, June-October 2022 (N=33,202)

Prevalence rates for severe COVID symptoms for all MSAs except Boston, Philadelphia, Seattle, and DC MSAs were significantly higher than the rate for the San Francisco MSA at p<0.05. Adjusted by logistic regression for age, race/ethnicity, gender, LGBT status, marital status, education, household income, housing tenure, and insurance status. Adjusted odds ratios for Atlanta, Chicago, Detroit, Phoenix, and Riverside MSA were significantly higher than the San Francisco MSA at p<0.05.

Source: Data derived from June-October 2022 Household Pulse Survey

Export to PPT

Based on previous research on COVID-related outcomes and the social determinants of health framework, we selected the following covariates for the two outcome variables: age (18-24 years, 25­34, 35-44, 45-54, 55-64, 65-74, ≥75), race/ethnicity (non-Hispanic White, non-Hispanic Black/African American, Asian, Hispanic, other and multiple races), gender (male, female), lesbian, gay, bisexual, and transgender (LGBT) status (lesbian or gay, bisexual, transgender, straight/heterosexual, other), marital status (married, widowed, divorce/separated, single), region of residence (Northeast, South, Midwest, West), educational attainment (less than high school, high school, some college, college degree, graduate degree or higher), annual household income (<$25,000; 25,000-34,999; 35,000-49,999; 50,000-74,999; 75,000-99,999; 100,000-149,999; 150,000-199,999; ≥200,000; unknown), housing tenure (homeowner, renter), and health insurance status (insured, not insured).^{6,9,12,22,27,30}

2.4. Statistical Methods

Given the binary outcome variables, multivariable logistic regression was used to model disparities in COVID-related outcomes by geographic area after controlling for socioeconomic, demographic, and health care characteristics. The Chi-square statistic tested the overall association between each covariate and Long COVID or severe COVID symptoms prevalence, while the two-sample t-test assessed differences in prevalence between any two geographic areas. To account for the complex sample design of the HPS, SUDAAN software was used to conduct all statistical analyses, including the logistic modeling procedure RLOGIST.³¹ Adjusted prevalence estimates for the two outcome variables were derived by the logistic model at the mean values of the covariates.

We used two health disparity indices,the coefficient of variation and the root-mean-square-deviation (RMSD) to summarize geographic disparities in Long COVID and severe COVID symptoms among 50 states and DC or among 15 metropolitan areas.^32-34 We used root-mean-square-deviation (RMSD) as a summary measure of geographic disparities among 50 states and DC or among 15 metropolitan areas.^32,33 The RMSD, developed by Singh and colleagues, is similar to the square root of the variance, except that the average-squared-deviations are calculated using a “standard” estimate other than the sample mean. The RMSD is given by the formula:

where C_ri is the prevalence of Long COVID or severe COVID symptoms for the ith state (i=1,2 51) or MSA (i=1,2 15), C_rl is the corresponding statistic for the “standard” state or metropolitan area with the lowest prevalence (i.e., DC or San-Francisco for Long COVID), and 1 is the number of states (51) or metropolitan areas (15) being compared. While RMSD is a measure of absolute health disparity, the coefficient of variation (CV) of the where C_ri is the prevalence of Long COVID or severe COVID symptoms for the ith state (i=1,2 51) or MSA (i=1,2 15), C_rl is the corresponding statistic for the “standard” state or metropolitan area with the lowest prevalence (i.e., DC or San-Francisco for Long COVID), and 1 is the number of states (51) or metropolitan areas (15) being compared.

While RMSD is a measure of absolute health disparity, the coefficient of variation (CV) of the RMSD provides an estimate of relative disparity and is given by

3.1. State and Metropolitan Area Disparities in Long COVID Prevalence

During June-October 2022, an estimated 35.4 million or 32.2% of US adults aged ≥18 years with COVID-19 infection reported experiencing Long COVID symptoms (Table 1). This amounts to 14.4% (or 1 in 7) of all US adults developing Long COVID at some point (data not shown). Among those who reported experiencing Long COVID, 18.2 million or 51.6% reported having Long COVID symptoms currently (data not shown).

The prevalence of Long COVID ranged from a low of 24.0% in DC, 25.4% in Hawaii to a high of 39.2% in Alabama, 39.9% in Wyoming, and 43.6% in West Virginia. Adjusted for individual-level covariates, adults with COVID-19 diagnosis in Alabama, Wyoming, and West Virginia had 48-178% higher adjusted odds of developing Long COVID than their counterparts in DC (Table 1). State maps on unadjusted and adjusted prevalence show higher rates of Long COVID in the Southeast, Southwest, and Northern Plains states and lower rates of Long COVID in states along the East and West coasts of the United States (Figure 3).

Close

Figure 3:

(a-d) State Differences in the Prevalence of Long COVID and Severe COVID Symptoms (%) among Adults Aged ≥18 Years During the COVID-19 Pandemic, United States, June - October 2022

Adjusted prevalence estimates for Long COVID and severe COVID symptoms were derived by logistic regression models that con­trolled for age, race/ethnicity, gender, LGBT status, marital status, education, household income, housing tenure, and health insurance status

Export to PPT

Large disparities in Long COVID prevalence among the 15 largest metropolitan areas ranged from 21.9% in San Francisco to 38.0% in Riverside, California (Figure 1).Adults with COVID-19 diagnosis in Atlanta, Phoenix, Chicago, Riverside, and Detroit metropolitan areas had 43-69% higher adjusted odds of developing Long COVID than their counterparts in the San Francisco Metropolitan Area (Figure 1).

Health disparity indices in Table 1 summarize geographic disparities in Long COVID and provide an assessment of the extent to which the observed geographic disparities can be attributed to state or metropolitan area differences in individual-level socioeconomic, demographic, and health insurance characteristics. The relative summary index of state disparity, CV(RMSD), decreased from 41.0% for the crude prevalence to 27.2% for the adjusted prevalence, suggesting that the adjustment for the individual-level sociodemographic and healthcare factors accounted for 34% of the state disparity in Long COVID. Adjustment for sociodemographic factors accounted for 45% of the metropolitanarea disparity in Long COVID (data not shown). The simple index of CV indicates a 21% reduction in state disparity (13.1% vs.10.4%) and a 29% metropolitan-area disparity in Long COVID after covariate adjustment.

3.2. State and Metropolitan Area Disparities in COVID-19 Disease Severity

An estimated 15.2 million (13.8%) US adults aged ≥18 with COVID infection reported experiencing severe COVID symptoms from June to October 2022 (Table 2). The prevalence of severe COVID symptoms ranged from 19.9% in Mississippi and 19.4% in Alabama to 9.8% in DC and 7.0% in Vermont. Adults with COVID-19 diagnosis in Alabama, Arkansas, Colorado, Iowa, Kentucky, Mississippi, Montana, North Dakota, South Dakota, Texas, Utah, Virginia, Wyoming, and West Virginia had 2.0-2.5 times higher adjusted odds of experiencing severe COVID symptoms than their counterparts in Vermont. State maps on unadjusted and adjusted prevalence show higher rates of severe COVID symptoms in the Southeast, Southwest, and Northern Plains states and lower rates of severe COVID symptoms in states along the East and West coasts of the US (Figure 3). The correlation coefficient between state rates of Long COVID and severe COVID symptoms was 0.71.

Disparities in the prevalence of severe COVID symptoms among the 15 metropolitan areas were pronounced, with the prevalence ranging from 8.9% in San Francisco to 17.5% in Riverside, California (Figure 2). Adults with COVID-19 diagnosis in New York, Detroit, Phoenix, Atlanta, Houston, Riverside, and Dallas metropolitan areas had 50­77% higher adjusted odds of experiencing severe COVID symptoms than their counterparts in the San Francisco Metropolitan Area (Figure 2).

State disparities in severe COVID symptoms, as measured by CV(RMSD), were reduced from 101.0% for the unadjusted prevalence to 73.7% for the adjusted prevalence, indicating a 27% reduction in state disparity after the adjustment for the individual-level covariates (Table 2). Individual- level sociodemographic factors explained 13% of the metropolitan-area disparity in severe COVID symptoms (data not shown).

4.1. Limitations

This study has some limitations. HPS does not collect information on many factors that could help explain geographic disparities in Long COVID and disease severity among COVID patients/survivors, including social and physical isolation, lack of social support or connectedness, loss of self-identity and sense of control, unhealthy diet, lack of physical inactivity, smoking, alcohol and other substance use, lack of or limited access to care, and comorbidities.²² Although comorbidities such as hypertension, smoking, diabetes, obesity and other pre-existing conditions are important risk factors for COVID-19 infection, Long COVID, and disease severity, they may not fully explain geographic disparities in Long COVID shown here.^18,24,37 In fact, Long COVID can develop among people who have had no or mild symptoms, although people with more severe COVID symptoms are at the greatest risk of developing Long COVID and long-term impairments.^18,24

Second, data on specific clinically diagnosed COVID symptoms were not available in HPS. Moreover, the self-reported cases of Long COVID are probably underestimated as has been the case with official estimates of COVID-19 infections, hospitalizations, and deaths.³⁸ Many of the COVID patients whose symptoms have persisted for longer than three months since or sometime after the initial COVID-19 diagnosis may not be aware that the painful symptoms or health conditions they currently experience might actually be due to Long COVID. However, self-reported nature of the HPS data on Long COVID symptoms and lack of clinical assessment in some cases might have erroneously been attributed to Long COVID, resulting in an overestimation of the prevalence.³⁹

Third, due to confidentiality concerns, public use HPS files do not provide geographic data below the state level (e.g., county level).^12,28,29 Because of significant heterogeneity in socioeconomic, health status, and health care characteristics across US counties, actual geographic disparities in the prevalence of Long COVID and severe COVID symptoms, as captured by state-level differences, are likely underestimated.^9,35

Fourth, our study did not analyze geographic patterns in Long COVID and disease severity separately for men and women and by race/ethnicity. We found a 45% higher risk of Long COVID among women than men and 47-75% higher risk of Long COVID among Hispanics, Blacks, and non-Hispanic Whites than Asians (data not shown). Future studies might benefit from examining whether geographic patterns differ by gender and among racial/ethnic minorities.

Fifth, the respondents in HSP are more likely to be women and non-Hispanic Whites and have higher education, compared with the American Community Survey.⁴⁰ This might have resulted in an underestimate of the magnitude of disparities in COVID-related outcomes. However, we addressed disproportionate sampling of demographic characteristics by using survey weights, which rakes the demographics of the interviewed persons to education attainment/sex/age distributions and ethnicity/race/sex/age population distributions.²⁹

Compliance with Ethical Standards