In this study, patients who self-reported Long COVID were compared with those who reported recovering from symptoms within 3 months of the acute phase of COVID-19. The results suggest a range of impairments across different QoL domains are associated with a lack of recovery from COVID-19 within 3 months, and these impairments can be detected at a median of 2 years after initial infection. Among the Long COVID group, a minority (15.8%) reported being fully recovered from COVID-19 at time of survey, with a median reported recovery time of 7.9 months (Recovered Long COVID). While acute infection severity (approximated with total symptom count) was found to be associated with Long COVID risk, > 80% of this sample were not hospitalized during the acute phase of infection, indicating that mild-to-moderately severe infection still confers measurable risk for concerning long-term sequelae. Beyond physical complaints, including fatigue and pain, the Long COVID group showed significantly decreased social function and increased anxiety, depression, and subjective cognitive decline.
Our findings indicate that recovery from Long COVID occurs in some domains. The prevalence of moderate-to-severe fatigue, pain, depression, and both physical and social functional impairment were all significantly lower in the Recovered Long COVID group than in the Current Long COVID group. Of concern, however, the prevalence of cognitive decline, fatigue, and pain remained elevated in the Recovered Long COVID group compared with the Without Long COVID group. This may indicate that some domains of Long COVID-related impairment are more amenable to recovery than others. Alternatively, those reporting recovery from Long COVID may not subjectively attribute current functional impairment to COVID-19, or their functional impairments could have preceded COVID-19. Investigating the interactions between patient-reported outcome domains and their longitudinal trajectories could help advance the understanding of Long COVID and support recovery.
Quality of Life and Subjective Cognitive Decline
The QoL impairments detected in this study are similar to pooled prevalences calculated from 12 published studies of Long COVID in which all participants were hospitalized.(16) The impact on social function associated with Long COVID was prominent – 33% had severe impairment – and is consistent with qualitative work that identified Long COVID symptomatology as a barrier to social well-being.(38) Similarly, the prevalence of severe anxiety was 37% among Long COVID patients with anxiety, and among the overall Long COVID group, more than 10% reported severe anxiety. While moderate-to-severe pain and physical function impairment were more prevalent in the Long COVID group than among those Without Long COVID, the prevalences of severe impairment among those with at least moderate impairment in these domains were not significantly different. These findings underscore the importance of cognitive, mental health and social function impairments in Long COVID, which may increase in prevalence over time relative to other sequelae.(7, 12, 39)
The high proportion of cognitive decline, even among patients Without Long COVID, could indicate underrecognized long-term cognitive impacts of COVID-19. Early report of perceived cognitive deficit has been associated with later post-COVID condition.(20) The CCI-12 cut-offs used to operationalize cognitive decline in the present study have not undergone extensive validation. However, 93% of this sample with a CCI-12 score ≥ 20 reported concern about memory changes, whereas only 17% of those with a score of < 20 reported concern, suggesting the threshold measures differences relevant for patients. The proportion of subjective cognitive decline in this sample is notably higher than population-based and meta-analytic prevalence estimates from data collected before the COVID-19 pandemic, but different methodologies among studies prevents a clear comparison.(40, 41) Neurocognitive symptoms of Long COVID have been associated with decreased likelihood of working full-time and have been qualitatively connected to social isolation and work discrimination.(38, 42) A large medical record study of post-acute sequelae at 2 years post-infection found a greater 2-year cumulative disability burden from both neurological and mental health symptoms than from fatigue, pulmonary symptoms, or musculoskeletal symptoms.(18) Since subjective cognitive decline has been associated with more severe future cognitive impairment in patients without Long COVID-19,(25) the currently reported memory concerns suggest the clinical importance of carefully monitoring the cognitive trajectory of those who do not recover from COVID-19 within 3 months. Future studies examining the longitudinal correlates of subjective cognitive decline in the context of Long COVID would be valuable.
Long COVID Risk Factors
The regression model results suggest that a higher level of education attainment is associated with recovery from COVID-19, adding to existing published evidence indicating education and other social determinants of health (SDOH) are associated with Long COVID risk.(43) Participants without a bachelor’s degree had approximately 53% greater unadjusted Long COVID risk than those with an advanced degree and 33% greater with control for comorbidities and infection severity among other variables. This disparity could be connected to a differential capacity to utilize socially-determined resources (for example, paid time off from work) to aid recovery.(44, 45) Post-COVID conditions have been associated with greater odds of unemployment and lower odds of full-time employment, effects that were in turn strongly associated with both education and gender but not other SDOH variables.(42) Losing or lacking full-time employment could affect health insurance coverage and limit access to care. Discrimination and stigma experienced by those with Long COVID may create barriers in healthcare, occupational, and institutional settings that are more difficult to circumvent for those with less education or other marginalized identities.(38) Though female gender was not statistically significant in our multivariate results, the lower bound of the 95% confidence interval for its effect estimate was 0.996, indicating it may be associated with Long COVID risk. Rigorous, intersectional investigation of how education, employment, gender, and other SDOH variables affect COVID-19 recovery within multilayered, interlocking social structures is crucial to identify how to mitigate inequitable long-term effects.(46–48)
Table 3
Bivariate and Multivariable Relative Risk Regression Model Results Predicting Self-Reported Long COVID.
| Unadjusted models | Multivariable model* |
| RR (95% CI) | p-value | Adj. RR (95% CI) | p-value |
Age | | | | |
20–34 | Reference | – | Reference | – |
35–49 | 1.28 (0.89, 1.85) | 0.18 | 1.47 (1.01, 2.13) | 0.05 |
50–64 | 1.39 (0.99, 1.96) | 0.06 | 1.52 (1.06, 2.18) | 0.03 |
65+ | 1.16 (0.81, 1.68) | 0.42 | 1.48 (1.01, 2.18) | 0.05 |
Gender | | | | |
Male | Reference | – | Reference | – |
Female | 1.25 (1.01, 1.53) | 0.03 | 1.23 (1.00, 1.51) | 0.06 |
Education | | | | |
Advanced Degree | Reference | – | Reference | – |
Bachelor’s Degree | 1.22 (0.93, 1.60) | 0.17 | 1.12 (0.87, 1.46) | 0.38 |
No bachelor’s Degree | 1.53 (1.18, 1.99) | < 0.01 | 1.33 (1.03, 1.71) | 0.03 |
Race | | | | |
White | Reference | – | Reference | – |
Asian | 1.08 (0.74, 1.59) | 0.66 | 1.12 (0.75, 1.67) | 0.59 |
Black | 1.00 (0.67, 1.51) | 0.96 | 0.90 (0.62, 1.29) | 0.57 |
American Indian/Alaska Native | 1.56 (1.03, 2.37) | 0.03 | 1.09 (0.67, 1.78) | 0.73 |
Other | 1.25 (0.67, 1.88) | 0.23 | 1.06 (0.66, 1.71) | 0.81 |
Hispanic Ethnicity | | | | |
Not Hispanic | Reference | – | – | – |
Hispanic | 1.16 (0.86, 1.56) | 0.32 | – | – |
Pre-Infection Comorbidities | | | | |
No Comorbidities | Reference | – | Reference | – |
1 | 1.02 (0.80, 1.30) | 0.87 | 0.93 (0.74, 1.18) | 0.56 |
2 | 1.25 (0.96, 1.64) | 0.12 | 1.15 (0.89, 1.50) | 0.29 |
3 or more | 1.58 (1.21, 2.07) | < 0.01 | 1.45 (1.11, 1.90) | < 0.01 |
Total Number of Acute Symptoms | 1.15 (1.11, 1.19) | < 0.01 | 1.14 (1.10, 1.18) | < 0.01 |
* A total of 424 participants were included in the final model. “Missing” categories for age (n = 3), gender (n = 2), and education (n = 4), and the Other category for gender (n = 3) were dropped from the analysis due to group size of < 5. Not displayed: “Missing” categories for race (n = 12) and ethnicity (n = 10). RR Relative Risk, CI Confidence Interval |
Strengths and Limitations
The strengths of this analysis include the long follow-up time after confirmed diagnosis, the Without Long COVID comparison group, and the sub-grouping of Long COVID into those with Current Long COVID and Recovered Long COVID. Patient-reported recovery and symptom measures of Long COVID provide data that are not easily captured in electronic health records, as ICD codes for symptoms are not consistently used. However, since these data are cross-sectional and were reported retrospectively, reverse causation and recall bias cannot be ruled out. The QoL impairments detected with PROMIS measures may have preceded COVID, and worse current health and current cognitive decline could affect recall of past comorbidities, infection severity, and COVID-19 recovery. Furthermore, we were unable to account for the effect of repeat infections. Given the low response rate (6.0%), this sample may not be representative of the larger pool of eligible patients. Survey data could not be linked to EHR data to compare the survey responders to non-responders in terms of demographics and clinical characteristics. The proportion of Long COVID was substantially higher in this sample than prevalence of Long COVID estimated in in population-based studies,(49) which may be due to response bias. Patients who attribute current health impairments to COVID-19 could have been more likely to provide consent and respond to survey, increasing their proportion in the sample relative to those who do not attribute current health to COVID-19 or who do not have current impairments. Those who died in the interim time between SARS-CoV-2 infection and the study could not be included, so their experience of Long COVID was not captured. The analysis of social factors was limited to gender, education, race, and ethnicity, all of which were self-classified based on United States Census categories.
Of note, 77% of participants were first diagnosed with COVID-19 in 2020, when either wild-type or the alpha variant were prominent and before the wide dissemination of COVID-19 vaccines. While this feature precludes analyzing the effect of vaccination and limits generalizability of findings to more recent variants, it furthers our understanding of Long COVID among people infected early in the pandemic – roughly 94 million globally and 20 million in the United States by the end of 2020, according to WHO estimates.(50)