Natural history of long-COVID in a nationwide, population cohort study

Understanding the scale and natural history of long-COVID is essential to planning health and social care. Most studies report the prevalence of long-COVID at a single timepoint after infection, sometimes adjusting for pre-existing symptoms. Less is known about changes over time. Previous longitudinal studies often focused on selected groups (e.g., hospitalised patients, older patients, or veterans with breakthrough infections) or specific outcomes (e.g., mental health), or lacked an uninfected comparison group, making it difficult to disentangle persistent or late-onset symptoms of long-COVID from symptoms that would have occurred regardless of SARS-CoV-2 infection. Hospital cohorts have variably reported no change, improvement, or deterioration over time, and some studies found no change in recovery between 5 months and 1 year post-discharge. A meta-analysis suggested declines in depression prevalence over time without significant differences from comparison groups beyond 2 months. The present study aims to characterise the natural history and trajectories of long-COVID in a nationwide, general population cohort with a contemporaneous never-infected comparison group to distinguish disease-related changes from background trends.

Several studies indicate mixed trajectories post-COVID-19. In a cohort of 1276 hospitalised patients, the proportion with at least one symptom decreased from 68% at 6 months to 49% at 12 months, but breathlessness and anxiety/depression increased over time. Among veterans with breakthrough infections, there was increased risk of new sequelae at 3 months. Other hospital cohorts reported no change in quality of life but improvements in functional tests, or stable proportions reporting full recovery from 5 months to 1 year. Meta-analyses of long-term outcomes report variable prevalence of persistent symptoms (e.g., fatigue, dyspnea, anosmia, depression, cognitive difficulties), with wide ranges across studies (approximately 28% to 77%), underscoring heterogeneity in populations, follow-up, and outcomes measured. These findings highlight both the potential for symptom resolution (e.g., smell/taste) and the possibility of late-onset sequelae (e.g., respiratory symptoms), and the need for robust comparison groups.

Study design and participants: The Long-COVID in Scotland Study (Long-CISS/Long-COS) is an ambidirectional, nationwide general population cohort. Adults (≥16 years) in Scotland with a positive SARS-CoV-2 PCR test from April 2020 were identified via the NHS Scotland notification platform and invited through automated SMS messages. A contemporaneous comparison cohort comprised adults who had a negative PCR test and never a subsequent positive test (“never infected”). Participation was voluntary; consenting participants allowed record linkage to ascertain infection status and covariates. Data collection: Serial self-completed online questionnaires captured pre-existing conditions and current health at approximately 6, 12, and 18 months after the index PCR test. Linkage provided age, sex, postcode-derived socioeconomic status (Scottish Index of Multiple Deprivation, SIMD), vaccination status (0/1/≥2 doses at index), hospitalisation and ICD-10 U07.1 for severe infection, and variant period (dominant defined as ≥95% of genotyped cases in a week). Pre-existing conditions were identified via self-report and linkage to hospitalisations and dispensed prescriptions; definitions included respiratory disease (ICD10 J40–J47, J98.2, J98.3 or relevant respiratory medications), coronary heart disease (ICD10 I10, I11.0, I13.2, I20–I25 excluding I24.1, I50, T32.2, T95.5), depression/anxiety (ICD10 F30–F33 or relevant medications), and diabetes (ICD10 E10–E14 and related codes). Total comorbidity count was categorised as 0, 1–2, or ≥3. Outcomes: Self-reported recovery status (full, partial, none) at each timepoint; change in status categorised as improvement (none→partial/full or partial→full), deterioration (full→partial/none or partial→none), or constant (unchanged). Presence of 26 individual symptoms was recorded at each timepoint. Health-related quality of life was measured using EQ-5D-5L VAS. Statistical analysis: Characteristics were summarised using medians/IQRs and frequencies, with Mann–Whitney U and chi-squared tests as appropriate. Changes from 6 to 12 months and from 6 to 18 months were analysed separately. Binary logistic regression models assessed factors associated with improvement and deterioration in recovery status among previously symptomatic infected participants, first unadjusted and then adjusted for age, sex, SIMD quintile, ethnic group, individual long-term conditions, total comorbidity, vaccination status, and variant period. McNemar’s tests compared within-person changes in symptom prevalence between timepoints. For each symptom at 12 (and 18) months, logistic regression (adjusted for infection status, presence of the symptom at 6 months, and confounders above) estimated odds ratios contrasting previously infected with never infected and assessed other predictors. EQ-5D VAS change was analysed using Poisson regression at 12 months adjusting for baseline (6-month) score, infection status, and confounders. All tests were two-sided; analyses used Stata v16.

- Cohort and follow-up: Of 4,409,590 invitations, 345,673 questionnaires were completed by 288,173 individuals; 257,341 (89%) consented to linkage. After exclusions (53,530 self-reported positives not in PCR database; 5,687 asymptomatic infections; 37,343 recruited beyond 6 months), 160,781 were eligible: 80,332 previously symptomatic, PCR-confirmed infections and 80,449 never infected. Among the infected cohort, 12,947 completed both 6- and 12-month questionnaires; 4,196 completed 6- and 18-month questionnaires. Among never infected, 11,026 and 1,711 completed 6- and 12-, and 6- and 18-month follow-up, respectively. - Recovery trajectories: Beyond 6 months, there was no significant overall change in recovery status at the population level, but individual heterogeneity was evident. Between 6 and 12 months, approximately 12% improved and 12% deteriorated; between 6 and 18 months, ~15–16% deteriorated and ~15% improved (Table 1). Among those partially recovered at 6 months, 21% improved by 12 months; among those with no recovery at 6 months, 45% showed some improvement and 3% fully recovered by 12 months. Deterioration by 12 months occurred in 16% of those fully recovered at 6 months. - Risk factors for trajectory: Prior depression and socioeconomic deprivation were more common among those who deteriorated. Adjusted models showed that improvement by 12 months was less likely with older age and prior depression, and more likely among the most affluent; deterioration was less likely with older age and among the most affluent, and more likely with prior depression. - Symptoms over time: Among previously symptomatic infected participants, the overall proportion reporting at least one of 26 symptoms was relatively stable from 6 to 12 and to 18 months, whereas this proportion increased over time among never infected comparators. Specific symptom changes differed by infection status: • Improvements: Altered taste and smell and confusion (brain fog) decreased significantly between 6 and 12 months in the previously infected group, with greater reductions than in never infected after adjustment (e.g., adjusted ORs at 12 months vs never infected: altered taste 0.22 [95% CI 0.14–0.34]; altered smell 0.20 [0.12–0.34]; confusion 0.43 [0.35–0.54]). Improvements were concentrated among those whose recovery status improved. • Late-onset/persisting symptoms: Dyspnea and cough (dry and productive) increased between 6 and 12 months in both groups, but increases were significantly greater among previously infected after adjustment (e.g., dry cough OR 1.41 [1.28–1.56]; cough with phlegm OR 1.15 [1.05–1.27]). Pre-existing respiratory disease was associated with higher prevalence of productive cough. • Hearing problems: Prevalence increased from 6 to 18 months in both groups, but increases were significantly greater among previously infected (adjusted analyses). Hearing problems were associated with socioeconomic deprivation, greater infection severity, and more pre-existing conditions (including depression/anxiety). - Quality of life: EQ-5D VAS declined slightly from 6 to 12 months in both groups (infected median 77 to 74; never infected 80 to 77). In adjusted Poisson models, symptomatic infection was associated with a marginally larger fall vs never infected (IRR 0.98, 95% CI 0.98–0.98).

In a nationwide, general population cohort with a contemporaneous never-infected comparison group, long-COVID showed overall stability beyond 6 months, but with meaningful individual-level heterogeneity: about one in eight improved and a similar proportion deteriorated between 6 and 12 months. The drivers of these trajectories differed by symptom: resolution of altered taste/smell and confusion contributed to improvement, whereas late-onset or persisting respiratory symptoms (dry/productive cough, dyspnea) and hearing problems contributed to deterioration. These patterns were not explained by background temporal trends or measured confounders, underscoring the value of serial measurement and a comparison group. Prior depression and socioeconomic deprivation were associated with worse trajectories, suggesting psychosocial and clinical vulnerabilities influence recovery. Quality of life declined modestly over time in both groups, with a slightly greater decline among those previously infected, indicating a small but detectable longer-term impact. Overall, the findings argue for monitoring individual symptom domains rather than only composite outcomes, and for targeted support for at-risk subgroups.

Long-COVID appears stable for many people beyond 6 months, yet a substantial minority experience improvement or deterioration over time. Reassuringly, altered taste, smell, and confusion tended to resolve. Conversely, increased prevalence of late-onset cough and hearing problems up to a year post-infection, independent of background trends and confounders, warrants further investigation. The study’s nationwide scope, linkage-confirmed infection status, and inclusion of a contemporaneous never-infected group strengthen causal inference regarding symptom trajectories. Future research should elucidate mechanisms underlying late-onset respiratory and auditory symptoms, evaluate interventions to mitigate deterioration—particularly among socioeconomically deprived individuals and those with prior depression—and extend follow-up beyond 18 months to define longer-term trajectories.

- Possible misclassification of infection status in the comparison group due to undetected or asymptomatic infections despite efforts to exclude self-reported positives without PCR confirmation. - Inability to assess differential attrition by exposure status under data sharing constraints (participants who later converted from negative to positive could not be identified), raising potential selection bias. - Residual confounding inherent to observational designs due to unknown or unmeasured factors. - Voluntary participation may introduce response bias; hospitalised and specific subgroups may be over- or under-represented. - Some data inconsistencies and non-disclosure of small cell counts may limit precision of certain estimates.