Persistent COVID-19 symptoms in a community study of 606,434 people in England

The UK experienced a large COVID-19 epidemic, but the natural history beyond acute illness and the long-term sequelae (Long COVID) remain incompletely understood. Persistent symptoms reported after the acute phase include severe fatigue, breathlessness, chest pain, palpitations, cognitive impairment, loss of smell/taste, rashes, and joint problems. Reported prevalence and duration vary widely, reflecting heterogeneous designs and definitions, and Long COVID likely encompasses disparate conditions. Prior work suggests persistent symptoms relate to acute disease severity, with higher risk among those hospitalised, and associations with the number of acute symptoms, older age, and sex. This study uses the community-based REACT-2 program to estimate the prevalence of persistent symptoms lasting at least 12 weeks after suspected or confirmed COVID-19, characterise symptom co-occurrence, and identify sociodemographic and clinical risk factors for persistence.

The paper situates its work within evidence showing broad and variable estimates of persistent post-COVID symptoms across hospitalised and community cohorts, with reported persistence ranging from weeks to months. Prior studies link Long COVID risk to acute severity, number of acute symptoms, age, and sex, though findings differ on which sex is at higher risk. Existing studies often focus on hospitalised cohorts, limiting generalisability. Reviews and cohort reports have highlighted a wide spectrum of symptoms and potential subgroups, motivating community-based, representative assessments and exploration of symptom clusters.

Design and setting: REACT-2 is a repeated cross-sectional, community survey of adults (≥18 years) in England, sampling from the NHS patient list across 315 lower-tier local authority areas. Rounds 3–5 were conducted 15–28 Sep 2020, 27 Oct–10 Nov 2020, and 15 Jan–8 Feb 2021 (n = 508,707). A replication round (round 6) was conducted in May 2021 (n = 97,717; excluding an additional booster sample ≥55 years). Participants registered online or by phone, self-administered an antibody lateral flow test, and completed a questionnaire on demographics, comorbidities, suspected/confirmed COVID-19, symptoms, and symptom duration. Primary outcome: persistent symptoms lasting ≥12 weeks after initial symptom onset among those reporting suspected or PCR-confirmed COVID-19. Rounds 3–5 surveyed 29 symptoms; round 6 surveyed 35, of which 27 were common with rounds 3–5. Free-text responses in rounds 3–5 were analysed to identify additional symptoms (e.g., brain fog, palpitations, hair loss) later included in round 6. Weighting: Prevalence estimates were weighted by sex, age, ethnicity, LTLA population, and deprivation to obtain estimates representative of the adult population of England. Statistical analysis: - Prevalence over time: Persistence of any and multiple symptoms up to 150 days after onset was described, with key cut points at 4 and 12 weeks. - Risk factors: Age- and sex-adjusted and mutually adjusted logistic regression assessed associations between sociodemographic, lifestyle, and clinical variables and persistence ≥12 weeks (one or more symptoms). Variables included sex, age, BMI, smoking, vaping, deprivation, household income, healthcare/care home worker status, and prior hospitalisation for COVID-19; multivariable variable selection was used for ranking predictors. Generalised additive models (GAMs) modelled persistence probability as a smooth function of age by sex. Replication analysis repeated modelling in round 6. - Clustering: Among participants still symptomatic at 12 weeks in rounds 3–5 (n = 20,240), symptom profiles at 12 weeks were clustered using CLARA (PAM) with Hamming distance to identify stable clusters (validated by bootstrap stability and alternative clustering methods/distances; latent class analysis also performed). Cluster characteristics and outcomes (e.g., hospitalisation rates, severity, care seeking) were compared. - Sensitivity analyses: (i) Using a reduced set of 15 symptoms; (ii) restricting to those self-reporting COVID-19 with positive antibody LFA; (iii) background prevalence among PCR-negative adults from REACT-1 (rounds 2–14) for 26 symptoms lasting ≥11 days (n = 1,879,842). Ethics: Approved by South Central–Berkshire Research Ethics Committee (ID: 738739); informed consent obtained.

- Sample and response: Rounds 3–5 included 508,707 respondents; round 6 included 97,717, with response rates ~29–30%. Weighted prevalence of prior COVID-19 with symptoms was ~19.1% in rounds 3–5 and ~17.7% in round 6. - Prevalence of persistent symptoms (≥12 weeks) among those with prior symptomatic COVID-19: • Rounds 3–5: 37.7% (95% CI: 37.4–38.1) had ≥1 symptom; 17.5% (17.2–17.7) had ≥3 symptoms. Weighted population prevalence: 5.80% (5.73–5.86) for ≥1 persistent symptom; 2.23% (2.19–2.27) for ≥3. • Round 6: 21.6% (20.9–22.3) had ≥1 symptom; 11.9% (11.4–12.5) had ≥3. Weighted population prevalence: 3.06% (2.98–3.14) for ≥1; 1.61% (1.56–1.67) for ≥3 (27 common symptoms). Using all 35 symptoms in round 6, these were 3.26% (3.18–3.34) and 1.86% (1.80–1.92). - Symptom dynamics: Rapid decline in symptom reporting by 4 weeks; further decline by 12 weeks; slower decline thereafter in both sexes. Women exhibited higher persistence than men. - Symptom profile: In rounds 3–5, tiredness was most prevalent at 12 weeks (16.8% [16.7–17.1]); in round 6, tiredness prevalence was lower (8.0% [7.5–8.6]). Of 27 common symptoms, most declined from rounds 3–5 to 6. - Risk factors for persistence (≥12 weeks; rounds 3–5): Female sex (OR 1.38, 95% CI: 1.32–1.45), increasing age, overweight/obesity, smoking, prior hospitalisation with COVID-19 (OR 3.45, 2.57–4.64), higher deprivation, low household income, and healthcare/care home worker status were associated with higher odds; Asian ethnicity had lower odds vs White (OR 0.84, 0.74–0.96). In GAMs, risk increased ~3.5 percentage points per decade; women had ~8 percentage points higher risk than men across ages. - Replication (round 6): Similar patterns, though smoking, vaping, and deprivation were not associated in multivariable models; variable selection did not retain healthcare/care home worker status, household income, or deprivation; Asian ethnicity was selected. - Clustering (rounds 3–5): Two stable clusters at 12 weeks among 20,240 symptomatic participants: • Cluster L1 (n = 15,799): High prevalence of tiredness with co-occurrence of muscle aches and sleep difficulty. • Cluster L2 (n = 4,441): Predominantly respiratory symptoms (shortness of breath, tight chest) and chest pain. In replication, a similar respiratory subset was identified; hospitalisation was higher in L2 (2.9% [2.5–3.5]) than L1 (1.1% [0.9–1.3]). - Sensitivity analyses: Using 15 of 29 symptoms reduced 12-week persistence by ~4 percentage points to 19.3% (16.3–23.3), with similar risk factors. Restricting to LFA-positive participants increased 12-week persistence to 42.4% (41.6–43.2). - Background prevalence in PCR-negative adults (REACT-1): 3.06% (3.04–3.09) reported any of 26 symptoms lasting ≥11 days, suggesting COVID-19-related persistent symptoms were approximately tenfold higher than background.

This large, representative community study estimates the burden and predictors of symptoms persisting ≥12 weeks after COVID-19, addressing gaps left by hospital-focused cohorts. The findings show substantial persistence, with a plateau after 12 weeks, indicating a sizable group with prolonged morbidity. Risk increases with age and is higher among women, those with higher BMI, smokers, individuals with prior COVID-19 hospitalisation, and those in more deprived areas, highlighting socio-economic inequities. Asian ethnicity was associated with lower risk compared to White ethnicity. The identification of two robust symptom clusters—particularly a respiratory-dominant cluster with higher severity and hospitalisation—supports the hypothesis of distinct Long COVID subgroups, which may require differentiated clinical pathways. The decline in persistent symptoms between autumn/winter 2020–21 and spring 2021 was driven largely by reductions in persistent tiredness; potential explanations include recovery over time, recall differences, changing social restrictions, and survey instrument changes. Background estimates from PCR-negative participants suggest observed persistence is not explained by non-COVID symptom persistence. Overall, the results underscore the clinical and public health importance of Long COVID and the need for targeted management and support.

In a nationally representative community sample, 5.8% of adults had or had experienced at least one persistent symptom ≥12 weeks after COVID-19 in late 2020–early 2021, decreasing to ~3.1% by May 2021, with clear risk gradients by age, sex, BMI, prior hospitalisation, and socio-economic factors. Two consistent symptom clusters were identified, including a respiratory-dominant subgroup with higher severity and care needs. These findings quantify the population burden of Long COVID and support tailored clinical assessment and rehabilitation pathways. Future research should elucidate the biological mechanisms underpinning symptom clusters, refine phenotypes, evaluate interventions (including rehabilitation and symptom-targeted treatments), and monitor long-term outcomes across diverse populations, with attention to health inequalities.

Strengths include very large sample size from a random community sample, weighting for population representativeness, and replication across survey rounds. The study asked about persistent symptoms rather than using the label "Long COVID" to reduce reporting bias and benchmarked against background symptom persistence in PCR-negative adults. Limitations include reliance on self-reported COVID-19 and symptoms (potential misclassification), retrospective reporting introducing recall bias, the non-specific nature of many symptoms, changes in the symptom list between rounds (potentially affecting comparability), and reduced statistical power in round 6 relative to rounds 3–5. Although response rates were moderate, weighting was applied to mitigate non-response bias.