ABSTRACT
As the coronavirus (COVID-19) pandemic approaches the three-year mark and the number of confirmed COVID-19 cases now exceeds 100 million cases in the United States and continues to climb, there have been increasing concerns raised over the future public health and economic burden of long COVID including disability and concomitant declines in labor force participation. Only a handful of US studies have explored sociodemographic and socioeconomic characteristics that may put people at risk of long COVID or have investigated the possible mental health and socioeconomic sequelae of long COVID. In this paper, I report findings from the largest multivariable analysis to date using US nationally-representative population-based data on 152,812 adults including 19,871 adults with long COVID to explore key predictors and sequelae of long COVID. An estimated 14.0% of US adults aged 18-84 years (34.55 million adults; 95% CI = 33.63 to 35.47 million adults) and 15.4% of US working-aged adults aged 18-64 years (31.14 million adults; 95% CI = 30.26 to 32.02 million adults) had developed long COVID by November 2022. I identify several sociodemographic and socioeconomic factors as predictors of long COVID including lower household income levels, being Hispanic, female, gay/lesbian or bisexual, and divorced or separated. Even after accounting for such factors, having long COVID was linked to higher risks of recent unemployment, financial hardship, and anxiety and depressive symptomatology, with evidence of dose-response relationships. Overall, an estimated 27.3 million US adults with long COVID were at risk of adverse socioeconomic and health outcomes including anxiety and depression. These preliminary estimates highlight the substantial public health and economic implications of long COVID among Americans and should prompt further inquiry and intervention.
INTRODUCTION
As the coronavirus (COVID-19) pandemic approaches the three-year mark and the number of confirmed COVID-19 cases now exceeds 100 million cases in the United States and continues to climb, there have been increasing concerns raised over the future public health and economic burden of long COVID including disability and concomitant declines in labor force participation. Long COVID (also known as post-COVID-19 syndrome) is a condition characterized by long-term health problems persisting beyond the typical recovery period for COVID-19 (2). By yielding symptoms affecting daily functioning, long COVID can plausibly affect one’s ability to work, thereby leading to loss of employment income and inducing financial strain. Through affecting physical and mental functioning, long COVID can also contribute to mental health disorders including anxiety and depression. Yet only a handful of population-based studies exist of sociodemographic and socioeconomic predictors of long COVID as well as its potential economic and health sequelae, and remain limited to relatively small samples (3-8).
The present study was undertaken to use nationally-representative data from the US Census Bureau’s Household Pulse Survey (HPS) on 152,812 adults including 19,871 adults with long COVID to investigate key sociodemographic and socioeconomic characteristics as possible risk factors for long COVID among American adults, and to explore the associations of long COVID with socioeconomic and health sequelae including unemployment and financial hardship and negative mental health outcomes.
METHODS
Study population
Repeated cross-sectional individual-level data were pooled from nationally-representative HPS surveys in data collection Phase 3.6 administered in 3 waves from September 14—November 14, 2022 (9). The study population consisted of US adults aged 18–84 years. Analyses of employment and financial hardship as outcomes were limited to participants of working age (aged 18-64 years). For the samples analyzed in multivariable models, data were available on 152,812 adults including 118,771 working-aged adults (representative of 202 million American adults). The HPS used the U.S. Census Bureau’s Master Address File as the source of sampled housing units (HUs). The sampling frame was a systematic sample of all eligible HUs. The HPS was conducted online by Qualtrics as the data collection platform. In this data collection phase, survey response rates ranged from 3.9% to 5.6% (10).
Outcomes
The first set of models examined the risks of 1) long COVID in the full sample; 2) long COVID among those with COVID-19 infection; and 3) COVID-19 infection in the full sample. The risk of long COVID is a function of the risk of becoming infected with COVID and the risk of progressing from COVID-19 infection to long COVID. Long COVID was defined in the HPS as reporting any symptoms lasting 3 months or longer that were not present prior to having COVID-19 including fatigue, difficulty concentrating and forgetfulness, “brain fog”, shortness of breath, joint or muscle pain, heart palpitations, changes to taste/smell, and inability to exercise (10).
The second set of models examined 4 recent outcomes in the full sample and among those still reporting long COVID symptoms: whether the respondent was employed in the past week; experienced financial hardship, defined as household difficulty (not at all/a little/somewhat/very difficult) to pay for usual household expenses including food, rent/mortgage, and loans, within the previous week; and the frequency of experiencing anxiety symptoms and depressive symptoms (using the 2-item Generalized Anxiety Disorder-2 (GAD-2) and Patient Health Questionnaire-2 (PHQ-2)) over the previous 2 weeks, with scores of 3–6 vs 0–2 being used previously to screen for anxiety and depressive disorders (10). Based on a cut-off score of ≥3, the GAD-2 demonstrated optimal sensitivity and specificity (sensitivity = 0.71, specificity = 0.69), and the PHQ-2 exhibited peak sensitivity and adequate specificity (sensitivity = 0.64, specificity = 0.85). Internal consistency reliability has been demonstrated to be high for both measures (Cronbach’s α = 0.81 and 0.83 for the GAD-2 and PHQ-2, respectively) (11).
Predictors
The first set of models included the following predictors: age group, gender (male, female, transgender, other), sexual orientation (straight, gay/lesbian, bisexual, other), race/ethnicity (Hispanic, White non-Hispanic, Black non-Hispanic, Asian non-Hispanic, Other non-Hispanic), marital status, education, household income in 2021, household size, child in the household, receipt of a COVID-19 primary vaccine or booster, and health insurance status (10).
In the second set of models, the main predictors were long COVID status (yes, no) in the full sample and the reported extent to which current long COVID symptoms reduced the ability to carry out day-to-day activities (a lot, a little, not at all) compared with before having COVID-19.
All models further controlled for self-reported non-receipt of the COVID-19 booster due to a COVID-19 infection and state of residence.
Statistical analysis
Multivariable modified Poisson regression models (that can approximate yet circumvent issues with log-binomial models) were fit to estimate adjusted prevalence ratios (PR) from generalized estimating equations that accounted for repeated measures within individuals and person-level survey weights and provided robust standard errors based on sandwich estimators (12).
Multiple imputation (using 25 multiply imputed datasets and the MCMC algorithm without rounding under a missing at random assumption (13)) was used to handle missing data. All analyses were conducted using SAS version 9.4 (SAS Institute Inc., Cary, NC).
RESULTS
Risk of long COVID in the full sample (n = 152,812)
In multivariable-adjusted models, compared to those aged 18-29 years, adults aged 30-39 years and 40-49 years were at 13% higher risk of long COVID, while adults aged 65+ years were at 40-53% lower risk (Fig. 1). Higher risks of long COVID were observed among those with any (vs. no) children in the household, Hispanic and other race/ethnicity adults (vs. White non-Hispanic adults), adults of lower income, with a larger household size, who were divorced/separated (vs. married), gay/lesbian or bisexual (vs. straight), and female, transgender, or other gender (vs. male) (all PRs ranging from 1.07 to 1.73; nearly all P<0.05). Black and Asian adults and never married adults were at 20-34% lower risk and 8% lower risk of long COVID than White non-Hispanic adults and married adults, respectively (all P<0.05) (Fig. 1). None of the vaccinated groups exhibited lower risks of long COVID than the unvaccinated group.
Risk of long COVID among those with COVID-19 infection (n = 72,198)
Modeling the risk of developing long COVID among those with COVID-19 infection showed similar patterns (Fig. 2), with a few exceptions: Having a child in the household was not associated with long COVID risk, while inverse associations between educational attainment and long COVID were more consistently observed. Being widowed (vs. married) was linked to a 14% higher risk of long COVID, and no association was observed for never being married. Those who completed their primary vaccination series but did not receive the vaccine booster had a higher risk of long COVID than the unvaccinated (Fig. 2).
Risk of COVID-19 infection in the full sample (n = 152,812)
Modeling the risk of COVID-19 infection showed several similar patterns as for the risk of long COVID in the full sample, including for having a child in the household and for racial/ethnic groups (Fig. 3). However, older age more consistently predicted a lower risk of COVID-19 infection, and lower education and income were associated with lower rather than higher risks of COVID-19 infection. For gender, only females showed a higher risk of COVID-19 infection compared to males. Furthermore, all vaccinated groups had 8-18% lower risks than unvaccinated adults (all P<0.05) (Fig. 3).
Risk of unemployment, financial hardship, and anxiety and depression among working-aged adults (n = 118,771) and among adults with current long COVID symptoms (n = 11,860) Those who reported having long COVID had a higher likelihood of not being employed, having financial hardship, and reporting anxiety or depressive symptoms recently (with PRs ranging from 1.04 to 1.52; all P<0.05 except for current employment; Fig. 4). Among those who still had long COVID symptoms, there was evidence of dose-response relationships, with those reporting a successively greater impact of long COVID symptoms on daily functioning having a stronger relationship with each outcome (e.g., for those reporting the highest impact on daily functioning, PR of depressive symptoms = 2.28; 95% = 1.95-2.67, p<0.001; Fig. 4).
DISCUSSION
To the author’s knowledge, this study represents the largest multivariable analysis to date using nationally-representative data to explore key sociodemographic and socioeconomic predictors and mental health and economic sequelae of long COVID in the United States. The large sample size enabled a comprehensive examination of potential risk factors including marital status and gender and reports novel associations of sexual orientation with long COVID. Several characteristics predicted greater risks of long COVID including lower education and household income levels, larger household size, and being Hispanic, female, gay/lesbian or bisexual, and divorced or separated. Receipt of a COVID-19 vaccine (with or without a booster) was linked to lower risks of COVID-19 infection.
Even after accounting for sociodemographic and socioeconomic factors, having long COVID predicted higher risks of recent unemployment, financial hardship, and anxiety and depressive symptomatology. For all of these outcomes, there were stronger relationships at successively higher levels of reported impacts of long COVID symptoms on daily functioning, compatible with causal relationships. Previous population-based US studies have likewise found long COVID to be associated with a lower odds of employment (7) and a higher risk of anxiety (8).
Limitations of this study include its cross-sectional and observational design, restricting the ability to make causal inferences due to bias from reverse causation or confounding by factors including comorbid conditions. While adjustment was made for receipt of a COVID-19 vaccine booster due to a COVID-19 infection, residual confounding due to indication could explain the null and positive associations between vaccination against COVID-19 and long COVID. Finally, while sampling weights accounted for non-response and there is evidence that weighting adjustments mitigated non-response bias due to the low response rates (14), such bias may not have been eliminated and could have led to selection bias either towards or away from the null.
Based on the study sample, an estimated 14.0% of US adults aged 18-84 years (34.55 million adults; 95% CI = 33.63 to 35.47 million adults) and 15.4% of US working-aged adults aged 18-64 years (31.14 million adults; 95% CI = 30.26 to 32.02 million adults) had developed long COVID by November 2022. The estimated percentages of those with long COVID still experiencing symptoms and reporting a modest impact or considerable impact on daily functioning were 57.2% and 21.8%, respectively. Assuming the same proportions in those who no longer experience long COVID symptoms, this translates into an estimated 27.3 million US adults (95% CI = 26.57 million to 28.02 million adults) aged 18-84 years and 24.6 million working-aged US adults (95% CI = 23.91 million to 25.30 million adults) with long COVID who had been or may still be at risk of adverse outcomes including anxiety and depression. Given that cases of COVID will continue to accrue from new infections and re-infections in the ongoing pandemic, these numbers of susceptible individuals will only escalate over time.
The preliminary findings from this study highlight the anticipated substantial health and economic burden of long COVID among American adults. Should these estimates be confirmed in longitudinal studies, future efforts should focus on identifying biological and social explanations and mediating pathways for associations, and on targeting risk reduction in those at higher risk of long COVID. Candidate interventions to mitigate adverse health and economic sequelae of long COVID could include enhanced uptake of COVID-19 vaccines by higher-risk groups to reduce the risk of acute COVID-19 infection and expanded access to paid sick leave and disability insurance (15).
Data Availability
All data produced in the present study are available upon reasonable request to the author.