Biopsychosocial factors associated with health-related quality of life in long COVID: a matched case–control study in primary care

Abstract

Background

Long COVID is associated with persistent symptoms, functional impairment, and reduced health-related quality of life (HRQoL), but the factors most strongly associated with poorer HRQoL remain incompletely characterized. This study compared adults with long COVID and recovered controls and identified variables independently associated with worse HRQoL.

Methods

We conducted a case–control analysis of the ARALONGCOV 2022 dataset including 170 adults with prior COVID-19: 85 with long COVID and 85 recovered controls without persistent symptoms. HRQoL was assessed with the 12-item Short Form Health Survey (SF-12) total score. Additional measures included depressive symptoms (Patient Health Questionnaire-9, PHQ-9), fatigue severity (Fatigue Severity Scale, FSS), sleep quality (Pittsburgh Sleep Quality Index, PSQI), pain catastrophizing, post-COVID functional status (PCFS), physical activity, and distance completed during the six-minute walk test (6MWT-D). Between-group comparisons used Mann–Whitney U tests for continuous variables and chi-squared or Fisher exact tests for categorical variables. Univariable and multivariable linear regression models were fitted within long COVID cases to identify factors independently associated with SF-12 total score.

Results

Compared with recovered controls, participants with long COVID had markedly worse HRQoL (median SF-12 28.75 [IQR 21.25–39.17] vs 73.96 [62.81–77.50]; p < 0.001), more depressive symptoms (PHQ-9: 12.00 [7.00–18.00] vs 2.00 [0.00–6.00]; p < 0.001), greater fatigue (FSS: 57.00 [50.00–62.00] vs 13.00 [9.00–33.00]; p < 0.001), worse sleep quality (PSQI: 14.00 [9.00–16.00] vs 6.00 [3.00–11.00]; p < 0.001), and shorter 6MWT-D (481 [406–560] m vs 556 [491–609] m; p < 0.001). Long COVID cases were also less frequently employed (38.8% vs 81.2%; p < 0.001) and had more post-COVID comorbidity (88.2% vs 60.0%; p < 0.001). In the multivariable model within long COVID cases (n = 85), active employment was independently associated with better HRQoL (B 5.315, 95% CI 0.610 to 10.021; p = 0.027), whereas depressive symptoms (B − 1.124, 95% CI − 1.500 to − 0.748; p < 0.001) and post-COVID comorbidity count (B − 1.252, 95% CI − 2.484 to − 0.021; p = 0.046) were independently associated with worse HRQoL. Fatigue severity showed a borderline association in the same direction (B − 0.171, 95% CI − 0.343 to + 0.001; p = 0.051). The model explained 48.9% of the variance in SF-12 score within long COVID cases (R² = 0.489; adjusted R² = 0.463). Exploratory clustering did not identify discrete phenotypes; the burden distribution was more consistent with a continuous severity gradient. For descriptive purposes, a lower-burden stratum (33/78, 42.3%) and a higher-burden stratum (45/78, 57.7%) were operationalized and differed across symptom load, fatigue, mood, functional status, exercise capacity, and HRQoL.

Conclusions

Long COVID was associated with profound impairment in HRQoL compared with recovered controls. Depressive symptoms, post-COVID comorbidity burden, and lower active employment were the strongest independent correlates of poorer HRQoL; fatigue severity showed a borderline association in the same direction. These findings support multidisciplinary long COVID care models that integrate symptom control, mental health assessment, and social and occupational reintegration. Routine assessment of fatigue, depressive symptoms, post-COVID multimorbidity, and occupational status may help identify long COVID patients with the most severe concurrent HRQoL impairment, though longitudinal confirmation is needed to establish predictive utility.

Trial registration

Trial registration: ISRCTN registry, identifier ISRCTN27312680.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12875-026-03445-9.

Background

Post-COVID-19 condition, commonly termed long COVID, is a heterogeneous syndrome characterized by persistent or relapsing symptoms after SARS-CoV-2 infection. The World Health Organization case definition describes symptoms that usually begin within three months of infection, last at least two months, and cannot be explained by an alternative diagnosis. [1, 2] Contemporary reviews emphasize that long COVID is multisystemic, can affect nearly every organ system, and has become a major public health and policy challenge because of its scale, functional consequences, and uncertainty regarding mechanisms and treatment [3–6].

Health-related quality of life (HRQoL) is a particularly relevant outcome in long COVID because the syndrome affects daily functioning, mood, cognition, sleep, social participation, and the ability to work. Recent HRQoL research has shown persistent reductions in both physical and mental domains after SARS-CoV-2 infection, especially among individuals who develop post-COVID conditions [7, 8]. In a 2025 systematic review and meta-analysis of EQ-5D studies, pooled HRQoL values remained below population norms, reinforcing the lasting burden of COVID-19 on patient-reported health status [9]. Longitudinal data from primary care cohorts have similarly shown that post-COVID condition is associated with worse physical and mental HRQoL over time [10].

Long COVID symptom patterns are also clinically heterogeneous. Fatigue, cognitive complaints, dyspnea, pain, sleep disturbance, anxiety, and depressive symptoms commonly overlap and may not affect all patients equally. Recent reviews on neurocognitive and mental health outcomes describe long COVID as a biopsychosocial condition in which neuropsychiatric symptoms have major implications for functioning and societal participation [11]. In parallel, phenotype studies have suggested the existence of clinically distinct subgroups with different levels of symptom burden, HRQoL impairment, and work disability [12–16]. This heterogeneity makes a purely symptom-counting approach insufficient for clinical triage.

A previous publication from the parent ARALONGCOV cohort [17] demonstrated that long COVID cases had worse functional status, physical activity, fatigue, and HRQoL than recovered controls, and that functional status was the strongest predictor of quality of life. However, that analysis did not model depressive symptoms, sleep quality, pain cognitions, or employment—all established determinants of HRQoL in chronic conditions. The present analysis addresses this gap by placing HRQoL as the primary endpoint with a broader set of biopsychosocial factors.

We therefore addressed two primary aims and one secondary, exploratory aim. The primary aims were to (1) quantify the magnitude of HRQoL impairment in long COVID compared with matched recovered controls, and (2) identify the sociodemographic, clinical, functional, and psychological factors most strongly associated with worse HRQoL. As a secondary, exploratory aim, we (3) examined whether the heterogeneity of long COVID burden was better described by discrete, clinically interpretable subgroups or by a continuous severity gradient, using unsupervised clustering. We hypothesized that fatigue and emotional distress would be among the factors most strongly and independently associated with worse HRQoL. Because the existence of discrete long COVID phenotypes remains uncertain, the clustering aim was pre-specified as exploratory and hypothesis-generating rather than as a confirmatory test of distinct subgroups.

Methods

Study design and setting

This manuscript reports an analysis of the ARALONGCOV project, an observational matched case–control study conducted in primary care in Aragon, Spain. Participants were recruited from health centers and volunteer contacts in the public health system. The project includes clinical, functional, psychosocial, and immunological data; the cohort and its design have been described in detail in the published study protocol [18] and in the primary cohort publication [17]. The present analysis was planned to address HRQoL determinants more comprehensively than the earlier cohort publication and is reported in accordance with STROBE recommendations [19]. Recruitment and exclusion figures from the parent cohort are reported in the study protocol [18] and primary cohort publication [17].

Eligible participants were aged 18 years or older and had confirmed SARS-CoV-2 infection (documented by polymerase chain reaction, antigen testing, or serology) between 2020 and 2021. Cases fulfilled the World Health Organization criteria for long COVID at the time of study assessment. Controls had a resolved acute infection lasting no more than four weeks and did not report persistent post-COVID symptoms. A total of 170 participants were included: 85 cases with long COVID and 85 recovered controls. Groups were matched by age, sex and date of SARS-CoV-2 infection.

Measures

The primary outcome for this secondary analysis was HRQoL measured using the Short Form-12 Health Survey (SF-12) [20]. In the study database, SF-12 responses were summarized as a global score ranging from 0 to 100, with higher scores indicating better perceived HRQoL. Because this cohort had already used the global SF-12 scoring approach in its earlier publication, the same operationalization was retained to preserve comparability within the dataset. Although this global score limits direct comparison with studies reporting conventional physical and mental component summaries, its use is consistent with published practice in observational cohorts employing composite HRQoL indices, and the magnitude of the between-group differences observed here supports the robustness of findings to this operationalization. The global index was computed as a linear transformation of the summed SF-12 item responses onto a 0–100 scale, with higher values indicating better perceived health, following the scoring used in the parent cohort publication [17]. The analytic dataset retained only this composite index rather than the individual item-level responses required to derive the conventional physical (PCS) and mental (MCS) component summaries; a PCS/MCS sensitivity analysis was therefore not feasible, and the resulting limited comparability with component-based studies is acknowledged as a limitation.

Psychological and functional correlates included depressive symptoms measured with the Patient Health Questionnaire-9 (PHQ-9) [21], fatigue severity measured with the Fatigue Severity Scale (FSS) [22], sleep quality assessed with the Pittsburgh Sleep Quality Index (PSQI) [23], pain catastrophizing assessed with the Pain Catastrophizing Scale [24], post-COVID functional limitation assessed with the Post-COVID-19 Functional Status scale (PCFS) [25], and exercise capacity assessed with the six-minute walk test [26, 27]. For descriptive purposes, PCFS grade 4 (severe functional limitation) was used as a binary threshold. Sociodemographic and clinical variables included employment status (defined as currently working—employed, self-employed, or equivalent—as opposed to unemployed, on sick leave, retired, or otherwise economically inactive), medication burden (number of daily medications and number of post-COVID medications), vaccination status, comorbidity burden before and after COVID-19, acute-phase severity, pneumonia, and COVID-related sick leave. Symptom burden in long COVID cases was described across major domains including fatigue, cognitive symptoms, dyspnea, headache, myalgia, neurological, cardiological, digestive, and sleep symptoms.

Statistical analysis

Continuous variables were summarized as means with standard deviations or medians with interquartile ranges, depending on distribution. Categorical variables were summarized as counts and percentages. The original recruitment applied individual matching by age, sex, and date of acute infection to ensure comparability between groups, but pairing identifiers were not retained in the consolidated database. We therefore analysed the case–control comparison as two independent groups, using the Mann–Whitney U test for continuous variables and the chi-squared test (or Fisher's exact test where expected counts were below five) for categorical variables. Matching variables (age, sex, and date of infection) were included as covariates in the multivariable models. Statistical significance was set at p < 0.05. All analyses were conducted in Python version 3.10 with scikit-learn version 1.7.2 and statsmodels version 0.14, using two-tailed tests. Sentinel integer values (− 2,147,483,648, representing missing or not-applicable entries in the database) were recoded as missing before analysis.

To identify correlates of HRQoL within the long COVID group, univariable ordinary least squares regression models were first fitted among long COVID cases (n = 85) with SF-12 total score as the dependent variable. A multivariable model was then fitted using clinically relevant predictors selected on the basis of univariable associations, theoretical relevance, and limited inter-predictor collinearity. Variables with Spearman correlation r > 0.50 with a retained predictor were excluded to avoid redundancy. The final model included active employment, post-COVID comorbidity count, FSS total score, and PHQ-9 total score. All four predictors had complete data across all 85 long COVID cases. Model fit was assessed using R2 and adjusted R2, and multicollinearity was evaluated with variance inflation factors (VIFs). Residual normality was assessed with the Shapiro–Wilk test (W = 0.985, p = 0.504). As a post-hoc sensitivity analysis, the model was refitted after adding any pre-COVID comorbidity as a covariate to assess whether post-COVID comorbidity effects were confounded by pre-existing disease. As a supplementary analysis, the model was additionally run in the pooled sample (n = 169) with a binary long COVID indicator included as a covariate, to estimate the incremental variance explained by the clinical predictors beyond the between-group separation. An exploratory FSS × PHQ-9 interaction term was tested and was not retained.

Exploratory subgroup analysis was restricted to participants with long COVID. To identify clinically interpretable subgroups, we performed unsupervised clustering using six variables selected a priori to capture symptom burden, psychological burden, functional limitation, exercise capacity, and post-COVID multimorbidity, while avoiding inclusion of HRQoL itself in cluster construction. The clustering variables were: (1) persistent symptom burden, operationalized as the count of eight persistent symptom domains (digestive, fatigue, myalgia, headache, dyspnea, cardiologic, neurologic, and cognitive); (2) Patient Health Questionnaire-9 (PHQ-9) total score; (3) Fatigue Severity Scale (FSS) total score; (4) post-COVID comorbidity burden, defined as the count of 29 binary indicators covering cardiovascular, metabolic, respiratory, neurological, rheumatological, and other conditions first recorded or diagnosed after the acute COVID-19 episode (see Measures); (5) Post-COVID Functional Status (PCFS) final grade; and (6) six-minute walk distance (6MWT-D, meters).

All clustering variables were standardized as z scores before model fitting. Because 6MWT-D was missing in seven long-COVID cases, the primary clustering analysis used a complete-case dataset. The seven excluded participants had higher median FSS (63 vs 57) and lower median SF-12 (20.4 vs 30.2) than included cases, suggesting they may have been systematically more severely affected; this may slightly underrepresent the most burdened patients in the primary clustering solution. As a sensitivity analysis, missing 6MWT-D values were replaced by the sample median and the clustering procedure was repeated in the full long-COVID sample. We fitted k-means models for 2, 3, and 4 clusters using 100 random initializations and a fixed random seed of 42. Candidate solutions were compared using silhouette width, the Calinski-Harabasz index, and the Davies-Bouldin index. We selected the final solution on the basis of internal validity, stability across sensitivity analyses, and clinical interpretability.

After cluster assignment, the resulting strata were compared on the variables used to build the clusters and on external validation variables that were not entered into the clustering algorithm, including SF-12 total score, Pittsburgh Sleep Quality Index (PSQI) total score, Pain Catastrophizing Scale (PCS) total score, employment status, and self-perceived quality-of-life worsening. Continuous variables are presented as median and interquartile range (IQR) and were compared using the Mann–Whitney U test. Categorical variables are presented as counts and percentages and were compared using Fisher’s exact test. Because this subgroup analysis was exploratory, p values were interpreted descriptively and were not adjusted for multiple comparisons.

As an exploratory sensitivity analysis, we examined whether adding routine inflammation/coagulation markers altered the cluster structure. The six primary clinical clustering variables were augmented with five prespecified laboratory markers available in the dataset: C-reactive protein, erythrocyte sedimentation rate, ferritin, fibrinogen derivative, and D-dimer. All 11 variables were standardized as z scores before clustering. Because laboratory completeness varied across participants, this sensitivity analysis used complete cases for all 11 variables. Candidate k-means solutions for 2, 3, and 4 clusters were compared using silhouette width, the Calinski-Harabasz index, and the Davies-Bouldin index, and were compared secondarily with hierarchical clustering using Ward linkage. This analysis was prespecified as exploratory and was intended to assess whether a clinical-inflammatory cluster structure emerged beyond the primary clinical solution.

Results

Participant characteristics

The study included 170 adults, with 85 participants in the long COVID group and 85 age-sex-date of infection-matched recovered controls. Mean age was similar between groups (47.33 ± 10.00 years in the long COVID group vs 48.12 ± 9.71 years in recovered controls; p = 0.432), as was the proportion of women (80.0% vs 77.6%; p = 0.707). Compared with recovered controls, participants with long COVID more frequently lived in housing without a terrace, balcony, or garden (24.7% vs 9.4%; p = 0.001), were substantially less likely to be in active employment (38.8% vs 81.2%; p < 0.001), and more often had pre-COVID comorbidity (81.2% vs 57.6%; p = 0.001) and post-COVID comorbidity (88.2% vs 60.0%; p < 0.001). COVID-19 pneumonia was also more frequent in long COVID cases (15.3% vs 3.5%; p = 0.009). Vaccination rates were similar between groups (85.9% vs 90.6%; p = 0.341). Long COVID participants also had a substantially higher daily medication burden (median 2.00 [IQR 1.00–4.00] vs 0.00 [0.00–2.00]; p < 0.001) and a higher number of post-COVID medications (3.00 [1.00–4.00] vs 0.00 [0.00–0.00]; p < 0.001). COVID-related sick leave was more common among long COVID participants (85.9% vs 64.7%; p = 0.002), and its duration was substantially longer (median 5.00 [IQR 1.00–10.00] months vs 0.50 [0.00–1.00] months; p < 0.001). (Table 1).

Table 1.

Selected participant characteristics by study group

Characteristic	Recovered controls (n = 85)	Long COVID (n = 85)	p value
Age, mean ± SD, years	48.12 ± 9.71	47.33 ± 10.00	0.432
Female sex, n/N (%)	66/85 (77.6)	68/85 (80.0)	0.707
Housing without terrace/balcony/garden, n/N (%)	8/85 (9.4)	21/85 (24.7)	0.001
Active employment, n/N (%)	69/85 (81.2)	33/85 (38.8)	< 0.001
Any pre-COVID comorbidity, n/N (%)	49/85 (57.6)	69/85 (81.2)	< 0.001
Any post-COVID comorbidity, n/N (%)	51/85 (60.0)	75/85 (88.2)	< 0.001
COVID-19 pneumonia, n/N (%)	3/85 (3.5)	13/85 (15.3)	0.009
Vaccinated, n/N (%)	77/85 (90.6)	73/85 (85.9)	0.341
COVID-related sick leave, n/N (%)	55/85 (64.7)	73/85 (85.9)	0.002
Sick-leave duration, median [IQR], months*	0.50 [0.00–1.00]	5.00 [1.00–10.00]	< 0.001
Daily medications, median [IQR]	0.00 [0.00–2.00]	2.00 [1.00–4.00]	< 0.001
Post-COVID medications, median [IQR]	0.00 [0.00–0.00]	3.00 [1.00–4.00]	< 0.001

Abbreviations: SD Standard deviation, IQR Interquartile range

^*Available cases: recovered controls n = 75, long COVID n = 83

HRQoL, symptom burden, and functional status

Participants with long COVID showed marked impairment across all major clinical outcomes. Median SF-12 total score was substantially lower in the long COVID group than in recovered controls (28.75 [21.25–39.17] vs 73.96 [62.81–77.50]; p < 0.001). Long COVID cases also had higher depressive symptom burden (PHQ-9: 12.00 [7.00–18.00] vs 2.00 [0.00–6.00]; p < 0.001), greater fatigue severity (FSS: 57.00 [50.00–62.00] vs 13.00 [9.00–33.00]; p < 0.001), poorer sleep quality (PSQI: 14.00 [9.00–16.00] vs 6.00 [3.00–11.00]; p < 0.001), and higher pain catastrophizing scores (12.00 [8.00–25.00] vs 3.00 [0.00–8.00]; p < 0.001). Severe functional limitation on the PCFS was present in 67.1% of long COVID cases compared with 3.5% of recovered controls (p < 0.001), and moderate-severe or severe depressive symptoms (PHQ-9 ≥ 15) were also more frequent (36.5% vs 3.6%; p < 0.001). Long COVID participants more often had low post-COVID physical activity (27.1% vs 7.1%; p < 0.001), walked shorter distances on the 6MWT-D (481 [406–560] m vs 556 [491–609] m; p < 0.001), and more frequently reported a notable worsening in quality of life after COVID-19 (78.8% vs 4.7%; p < 0.001). (Table 2).

Table 2.

HRQoL and major symptom-related outcomes by study group

Outcome	Recovered controls (n = 85)	Long COVID (n = 85)	p value
SF-12 total score, median [IQR]*	73.96 [62.81–77.50]	28.75 [21.25–39.17]	< 0.001
PHQ-9 total score, median [IQR]*	2.00 [0.00–6.00]	12.00 [7.00–18.00]	< 0.001
FSS total score, median [IQR]	13.00 [9.00–33.00]	57.00 [50.00–62.00]	< 0.001
PSQI total score, median [IQR]*	6.00 [3.00–11.00]	14.00 [9.00–16.00]	< 0.001
Pain catastrophizing score, median [IQR]	3.00 [0.00–8.00]	12.00 [8.00–25.00]	< 0.001
Severe functional limitation (PCFS), n/N (%)	3/85 (3.5)	57/85 (67.1)	< 0.001
Moderate-severe/severe depression (PHQ-9 ≥ 15), n/N (%)†	3/84 (3.6)	31/85 (36.5)	< 0.001
Low post-COVID physical activity, n/N (%)	6/85 (7.1)	23/85 (27.1)	< 0.001
Six-minute walk distance, median [IQR], m‡	556 [491–609]	481 [406–560]	< 0.001
Self-reported worsening in quality of life after COVID-19, n/N (%)	4/85 (4.7)	67/85 (78.8)	< 0.001

Abbreviations: FSS Fatigue Severity Scale, IQR Interquartile range, PCFS Post-COVID-19 Functional Status, PHQ-9 Patient Health Questionnaire-9, PSQI Pittsburgh Sleep Quality Index, QoL Quality of life

^*Available cases: recovered controls n = 84, long COVID n = 85

^†One PHQ-9 value was missing in the recovered control group

^‡Available cases: recovered controls n = 79, long COVID n = 78

P values are from Mann–Whitney U tests (continuous variables) and chi-squared or Fisher exact tests (binary variables) comparing the two independent groups. Effect sizes for continuous outcomes (Cohen’s d) ranged from 0.67 (six-minute walk distance) to 2.70 (SF-12 total score), with fatigue severity (d = 2.13), depressive symptoms (d = 1.69), sleep quality (d = 1.26), and pain catastrophizing (d = 1.05) all exceeding conventional thresholds for large effects [28]. Among binary outcomes, Cohen’s h ranged from 0.56 (low physical activity) to 1.75 (self-reported quality-of-life worsening), indicating uniformly large between-group separations

Within the long COVID group, the most frequent persistent symptoms were fatigue (96.5%), cognitive symptoms (88.2%), dyspnea (80.0%), myalgia (77.6%), neurological symptoms (77.6%), headache (76.5%), digestive symptoms (67.1%), and cardiological symptoms (60.0%).

Multivariable correlates of HRQoL

Prior to model fitting, pairwise Spearman correlations among candidate predictors confirmed substantial overlap between PSQI and FSS (r = 0.560), PSQI and PHQ-9 (r = 0.692), PCFS and FSS (r = 0.754), PCFS and PHQ-9 (r = 0.757), and pain catastrophizing and both FSS (r = 0.594) and PHQ-9 (r = 0.664). These variables were excluded from the multivariable model to avoid redundancy.

In univariable regressions within the long COVID group, older age and female sex were not significantly associated with SF-12 score. Active employment (B = 8.04, 95% CI 2.35 to 13.72; p = 0.007), post-COVID comorbidity count (B = − 2.23, 95% CI − 3.99 to − 0.46; p = 0.016), FSS total score (B = − 0.43, 95% CI − 0.61 to − 0.25; p < 0.001), and PHQ-9 total score (B = − 1.29, 95% CI − 1.68 to − 0.90; p < 0.001) were all associated with SF-12 in univariable analyses. Full univariable results for all 18 candidate predictors are reported in Supplementary Table S4.

The final multivariable model was fitted within long COVID cases with complete data on all four predictors (n = 85). PHQ-9 total score was the strongest independent correlate of worse HRQoL (B = − 1.124, 95% CI − 1.500 to − 0.748; standardised β = − 0.505; p < 0.001), followed by post-COVID comorbidity count (B = − 1.252, 95% CI − 2.484 to − 0.021; standardised β = − 0.176; p = 0.046). Active employment remained independently associated with better HRQoL (B = 5.315, 95% CI 0.610 to 10.021; standardised β = 0.193; p = 0.027). FSS total score showed a borderline association with worse HRQoL in the same direction as the univariable estimate (B = − 0.171, 95% CI − 0.343 to + 0.001; standardised β = − 0.180; p = 0.051). The model explained 48.9% of the variance in SF-12 score within the long COVID group (R² = 0.489; adjusted R² = 0.463; F(4,80) = 19.11; p < 0.001)(Table 3). Bootstrap resampling (10,000 iterations) confirmed the stability of this estimate, with a 95% confidence interval for R² of 0.373 to 0.650 (adjusted R²: 0.342–0.633). The exploratory FSS × PHQ-9 interaction term was not significant and was not retained.

Table 3.

Final multivariable linear regression model with SF-12 as the dependent variable

Predictor	B	95% CI	Standardized β	p value	VIF
Active employment	5.315	0.610 to 10.021	0.193	0.027	1.12
Post-COVID comorbidity count	−1.252	−2.484 to −0.021	−0.176	0.046	1.11
FSS total score	−0.171	−0.343 to + 0.001	−0.180	0.051	1.27
PHQ-9 total score	−1.124	−1.500 to −0.748	−0.505	< 0.001	1.13

Abbreviations: CI Confidence interval, FSS Fatigue Severity Scale, PHQ-9 Patient Health Questionnaire-9, VIF variance inflation factor. Dependent variable: SF-12 total score. Model statistics: n = 85 long COVID cases; R² = 0.489 (bootstrap 95% CI: 0.373–0.650); adjusted R² = 0.463 (95% CI: 0.342–0.633); F (4,80) = 19.11; p < 0.001. Shapiro–Wilk test on residuals: p = 0.504. Interaction test: the exploratory FSS × PHQ-9 interaction was not significant and was not retained

In the sensitivity analysis adding pre-COVID comorbidity to the model, the coefficient for pre-COVID comorbidity was not significant (B = − 0.42, 95% CI − 6.72 to + 5.88; p = 0.897), and the remaining coefficients were unchanged in direction and magnitude, supporting the robustness of the primary model to pre-existing disease confounding. Because the multivariable model excluded several clinically relevant but collinear constructs (PCFS, sleep quality, and pain catastrophizing; Spearman r > 0.50 with a retained variable), two further pre-specified sensitivity models are reported in the Supplement to show how the choice of variables influences the result. Replacing fatigue severity with the Post-COVID-19 Functional Status scale showed that functional limitation was independently associated with worse HRQoL (B = − 3.41; 95% CI − 6.60 to − 0.23; p = 0.036), with depressive symptoms remaining the strongest correlate; this indicates that the apparent prominence of any single domain partly reflects the model architecture rather than a unique causal role. When the primary model was additionally adjusted for the matching variables (age, sex, and date of infection), the associations of active employment (p = 0.083) and post-COVID comorbidity count (p = 0.126) attenuated to non-significance, whereas depressive symptoms and, after this adjustment, fatigue severity remained independently associated with HRQoL (Supplementary Tables S5–S6). A further sensitivity analysis restricting recovered controls to a strict definition (no persistent symptoms and no active immunosuppression; n = 63) left all between-group differences essentially unchanged (Supplementary Table S7).

As an exploratory and explicitly non-causal analysis, we examined the extent to which the fatigue–HRQoL association was shared with depressive symptoms [29, 30]. The univariable FSS–SF-12 association (B = − 0.431) was reduced by 34.7% (to B = − 0.281) after adjustment for depressive symptoms alone, indicating that a substantial part of the apparent fatigue–HRQoL association is shared variance with depressive symptomatology rather than an independent fatigue effect. Because the design is cross-sectional, the direction of this relationship cannot be established; the full variance decomposition, including path-by-path estimates (indirect effect a × b = − 0.149; bootstrap 95% CI − 0.256 to − 0.055) and its assumptions, is reported in the Supplementary Methods and should be read as hypothesis-generating only. In a supplementary pooled-sample model (n = 169) including a long COVID indicator alongside the four clinical variables, the model explained 82.7% of the variance (R² = 0.827; adjusted R² = 0.821); the long COVID indicator alone explained 64.8% (R² = 0.648), and the four clinical variables explained an additional 17.9% beyond group membership (ΔR² = 0.179). Within this pooled model, long COVID status (B = − 15.33, 95% CI − 20.00 to − 10.67; p < 0.001), PHQ-9 (B = − 1.28; p < 0.001), FSS (B = − 0.22; p < 0.001), active employment (B = 3.69; p = 0.037), and post-COVID comorbidity count (B = − 1.40; p = 0.004) were all independently associated with SF-12 score (Supplementary Table S3).

Exploratory severity strata in long COVID

Exploratory subgroup analysis was performed among long-COVID cases only. Because this analysis was hypothesis-generating, p-values are reported without adjustment for multiple comparisons and should be interpreted descriptively. The primary complete-case analysis included 78 of the 85 cases, because 6MWT-D was missing in seven participants. Across candidate k-means solutions, the 2-cluster model showed the best overall separation, with the highest silhouette width and Calinski-Harabasz index, whereas the 3- and 4-cluster solutions yielded weaker separation and less clinically coherent partitions (Table 4). Silhouette width values were interpreted against published thresholds: values below 0.25 indicate no substantial cluster structure, 0.25–0.50 indicate weak structure, and values above 0.50 indicate reasonable to strong structure [31]. The Calinski-Harabasz index also favored 2 clusters (25.43 vs 20.54 and 18.53, respectively). Although the Davies-Bouldin index decreased slightly with additional partitioning, the 2-cluster solution provided the most consistent pattern across the two principal separation criteria and the clearest clinical interpretability.

Table 4.

Internal validation metrics for candidate clustering solutions among long-COVID cases

Analysis	k	Silhouette	Calinski-Harabasz	Davies-Bouldin
Complete-case k-means (n = 78)	2	0.231	25.43	1.636
Complete-case k-means (n = 78)	3	0.206	20.54	1.621
Complete-case k-means (n = 78)	4	0.185	18.53	1.552
Complete-case Ward hierarchical (n = 78)	2	0.218	23.75	1.730
Complete-case Ward hierarchical (n = 78)	3	0.186	18.65	1.601
Complete-case Ward hierarchical (n = 78)	4	0.176	16.20	1.687
Median-imputed k-means (n = 85)	2	0.232	26.99	1.690
Median-imputed k-means (n = 85)	3	0.210	22.35	1.591
Median-imputed k-means (n = 85)	4	0.190	19.80	1.518

Variables entered into clustering were persistent symptom burden, PHQ-9 total score, FSS total score, post-COVID comorbidity burden, PCFS final grade, and six-minute walk distance. The final solution was selected by considering internal validity, reproducibility across sensitivity analyses, and clinical interpretability. Concordance between the complete-case and imputed 2-cluster k-means solutions was high (adjusted Rand index 0.949). Silhouette values below 0.25 indicate no substantial cluster structure [31]; all solutions in this dataset fall below this threshold, consistent with a continuous severity gradient rather than discrete subgroups

Ward hierarchical clustering, and a sensitivity analysis using median imputation for the seven missing 6MWT-D values, supported the same 2-cluster structure, with high concordance between the complete-case and imputed k-means solutions (adjusted Rand index 0.949); the corresponding internal-validity indices are reported in Table 4.

As noted above, the silhouette width of the 2-cluster solution fell below the 0.25 threshold that indicates weak cluster structure, consistent with a continuous rather than discrete distribution of burden. The data therefore do not support the existence of two discrete syndromic phenotypes; rather, the structure is better interpreted as a biopsychosocial severity gradient. For descriptive purposes, we operationalized this gradient as two strata, labelled post hoc as a lower-burden stratum (33/78, 42.3%) and a higher-burden stratum (45/78, 57.7%). Compared with the lower-burden stratum, the higher-burden stratum had greater persistent symptom burden (median 8.0 [IQR 7.0–8.0] vs 5.0 [4.0–6.0]; p < 0.001), more depressive symptoms (PHQ-9 16.0 [10.0–19.0] vs 9.0 [6.0–11.0]; p < 0.001), greater fatigue (FSS 60.0 [56.0–62.0] vs 48.0 [35.0–59.0]; p < 0.001), higher post-COVID comorbidity burden (3.0 [2.0–5.0] vs 2.0 [1.0–2.0]; p < 0.001), worse functional status (PCFS 4.0 [4.0–4.0] vs 3.0 [3.0–4.0]; p < 0.001), and shorter 6MWT-D (455.0 m [358.0–540.0] vs 522.0 m [446.0–570.0]; p = 0.015) (Table 5).

Table 5.

Characteristics of the two exploratory long-COVID severity strata

Variable	Lower burden stratum (n = 33)	Higher burden stratum (n = 45)	p value
Variables used for clustering
Persistent symptom burden, median [IQR]	5.0 [4.0–6.0]	8.0 [7.0–8.0]	< 0.001
PHQ-9 total, median [IQR]	9.0 [6.0–11.0]	16.0 [10.0–19.0]	< 0.001
FSS total, median [IQR]	48.0 [35.0–59.0]	60.0 [56.0–62.0]	< 0.001
Post-COVID comorbidity burden, median [IQR]	2.0 [1.0–2.0]	3.0 [2.0–5.0]	< 0.001
PCFS final grade, median [IQR]	3.0 [3.0–4.0]	4.0 [4.0–4.0]	< 0.001
6MWT-D, m, median [IQR]	522.0 [446.0–570.0]	455.0 [358.0–540.0]	0.015
External validation variables not used for clustering
SF-12 total, median [IQR]	37.92 [31.67–45.00]	24.17 [18.75–30.00]	< 0.001
PSQI total, median [IQR]	10.0 [7.0–13.0]	15.0 [13.0–18.0]	< 0.001
PCS total, median [IQR]	10.0 [5.0–19.0]	16.0 [9.0–29.0]	0.016
Active employment, n (%)	22 (66.7)	10 (22.2)	< 0.001
Marked self-perceived QoL worsening, n (%)	22 (66.7)	40 (88.9)	0.023
Persistent symptom domains
Fatigue, n (%)	31 (93.9)	44 (97.8)	0.571
Cognitive symptoms, n (%)	25 (75.8)	45 (100.0)	< 0.001
Dyspnea, n (%)	22 (66.7)	42 (93.3)	0.006
Headache, n (%)	20 (60.6)	41 (91.1)	0.002
Myalgia, n (%)	20 (60.6)	41 (91.1)	0.002
Neurologic symptoms, n (%)	19 (57.6)	41 (91.1)	< 0.001
Cardiologic symptoms, n (%)	12 (36.4)	34 (75.6)	< 0.001
Digestive symptoms, n (%)	18 (54.5)	36 (80.0)	0.025
Demographic variables
Age, years, median [IQR]	45.0 [41.0–50.0]	48.0 [41.0–56.0]	0.354
Female sex, n (%)	25 (75.8)	38 (84.4)	0.391

The primary clustering analysis used complete cases only (n = 78). Persistent symptom burden was defined as the count of eight persistent symptom domains. Post-COVID comorbidity burden was defined as the count of 29 binary indicators covering cardiovascular, metabolic, respiratory, neurological, rheumatological, and other conditions first recorded or diagnosed after the acute COVID-19 episode. Marked self-perceived QoL worsening was defined as a decline of at least 2 points in the retrospective quality-of-life change variable. Continuous variables were compared using the Mann–Whitney U test and categorical variables using Fisher’s exact test

External variables not used in clustering confirmed worse outcomes in the higher-burden stratum. SF-12 total score was markedly lower (24.17 [18.75–30.00] vs 37.92 [31.67–45.00]; p < 0.001), sleep quality was poorer (PSQI 15.0 [13.0–18.0] vs 10.0 [7.0–13.0]; p < 0.001), pain catastrophizing was higher (PCS 16.0 [9.0–29.0] vs 10.0 [5.0–19.0]; p = 0.016), active employment was less frequent (22.2% vs 66.7%; p < 0.001), and self-perceived quality-of-life worsening was more common (88.9% vs 66.7%; p = 0.023).

When persistent symptom domains were compared between strata, the higher-burden stratum showed a higher prevalence of cognitive symptoms (100.0% vs 75.8%; p < 0.001), dyspnea (93.3% vs 66.7%; p = 0.006), myalgia (91.1% vs 60.6%; p = 0.002), neurologic symptoms (91.1% vs 57.6%; p < 0.001), headache (91.1% vs 60.6%; p = 0.002), and cardiologic symptoms (75.6% vs 36.4%; p < 0.001). Digestive symptoms were also more common in the higher-burden stratum (80.0% vs 54.5%; p = 0.025). Fatigue prevalence was high in both strata and did not differ significantly (97.8% vs 93.9%; p = 0.571). No clear differences were observed between strata in age or sex distribution (Table 5).

Taken together, these findings indicate that heterogeneity in this cohort was better captured by a continuous severity gradient than by discrete syndromic subtypes. Figure 1 shows the standardized profile of the two strata across the six variables used for clustering and Fig. 2 shows the PCA-based two-dimensional projection of the final 2-cluster solution. Although some overlap was present, the plot showed a coherent separation of lower-burden and higher-burden cases along the first principal component, consistent with the interpretation of the cluster structure as a biopsychosocial severity gradient rather than as sharply separated syndromic subtypes.

Fig. 1.

Standardized centroid profiles of the two exploratory long-COVID severity strata. Line plot showing the standardized cluster centroids (z scores) from the final 2-cluster k-means solution for persistent symptom burden, PHQ-9 total score, FSS total score, post-COVID comorbidity burden, PCFS final grade, and six-minute walk distance. Positive values indicate above-sample-average burden, except for six-minute walk distance, where lower standardized values indicate poorer exercise capacity

Fig. 2.

Scatter plot of long-COVID cases included in the complete-case clustering analysis (n = 78), projected onto the first two principal components derived from the six standardized clustering variables: persistent symptom burden, PHQ-9 total score, FSS total score, post-COVID comorbidity burden, PCFS final grade, and six-minute walk distance. Points represent individual participants, symbols marked with “X” represent cluster centroids from the final 2-cluster k-means solution, and ellipses represent the 95% covariance ellipses for each stratum. PC1 explained 36.4% of total variance and PC2 explained 16.6%

Exploratory clinico-inflammatory sensitivity analysis

As a further exploratory sensitivity analysis, clustering was repeated after adding five routine inflammation/coagulation markers (C-reactive protein, erythrocyte sedimentation rate, ferritin, fibrinogen derivative, and D-dimer) to the six clinical variables (77 complete cases). This produced only a small inflammation-enriched subgroup (9/77, 11.7%) that had higher laboratory markers but did not differ from the remainder in symptom burden, fatigue, post-COVID comorbidity, functional status, HRQoL, sleep, pain catastrophizing, employment, or six-minute walk distance; the partition was imbalanced and did not improve clinical stratification (Supplementary Tables S1 and S2). The six-variable clinical 2-cluster solution was therefore retained.

Discussion

This analysis of a matched primary care cohort documented marked HRQoL impairment in long COVID compared with recovered controls and showed that greater fatigue severity and higher depressive symptom burden were independently associated with worse SF-12 scores, whereas active employment was independently associated with better HRQoL. The direction of this association cannot be established in a cross-sectional design since better-functioning patients may be more able to remain employed (reverse causation), or employment may have a protective role through routine, social participation, and financial stability. Longitudinal data are needed to distinguish these pathways. More broadly, cross-sectional associations cannot establish causal direction. As with employment, worse HRQoL may itself amplify reported fatigue and depressive symptoms through shared measurement variance and the bidirectional relationship between functional status and mood. The between-group separation was substantial across psychosocial and functional domains, indicating that long COVID burden in this cohort extends beyond isolated residual symptoms after acute infection and is better understood as a multidimensional syndrome involving physical limitation, emotional distress, and disrupted social participation. Within a biopsychosocial framework, the present findings identify psychological burden (depressive symptoms), accumulated post-acute disease burden (post-COVID comorbidity count), and social participation (active employment) as the dominant independent correlates of HRQoL, with fatigue severity showing a borderline contribution that overlaps with depressive symptomatology. Notably, the measured biological indicators captured by routine clinical and laboratory assessments did not yield a clinically informative cluster structure in the clustering analyses, suggesting that the patient-reported and functional domains carry the bulk of HRQoL-relevant information in this cohort.

The depth of HRQoL impairment observed here is consistent with recent literature [32–35]. Likewise, a 2024 longitudinal study showed that post-COVID condition was associated with persistently worse physical and mental HRQoL over time [10]. Our findings extend that work by showing that, within a matched case–control design, the strongest correlates of HRQoL are not limited to acute severity markers or isolated symptom domains but include potentially modifiable biopsychosocial factors that are measurable in routine care. Effect sizes for all between-group comparisons were large (Cohen’s d 0.67–2.70 for continuous outcomes; Cohen’s h 0.56–1.75 for binary outcomes), underscoring that the burden of long COVID in this cohort is not only statistically significant but clinically substantial [28].

Fatigue emerged as one of the principal determinants of worse HRQoL, which accords with systematic review evidence showing persistent fatigue in approximately one-third of patients 12 weeks or more after infection [36]. More recent HRQoL data further suggest that fatigue has downstream effects on emotional recovery, sleep, and later HRQoL [37, 38]. In the present study, fatigue showed a strong univariable association with worse HRQoL and a borderline independent association after adjustment for depressive symptoms and post-COVID multimorbidity (p = 0.051). The univariable association is consistent with a central role of fatigue in the lived experience of long COVID, but the attenuation of the FSS coefficient in the multivariable model—together with the mediation analysis described below—indicates that a substantial portion of this association is shared with depressive symptoms. For clinicians, fatigue should still be assessed systematically as a major determinant of HRQoL, with the recognition that its association with HRQoL is closely entwined with depressive symptomatology and may best be addressed jointly with mental health.

Depressive symptoms showed the largest standardized association with worse HRQoL in the final model. This does not imply that long COVID is primarily psychiatric; rather, it indicates that mental health is closely linked to symptom burden, sleep, coping, participation, and perceived recovery. Contemporary reviews have emphasized that long COVID frequently includes cognitive and affective sequelae with major functional implications [11, 39–41]. In our data, depressive symptoms remained independently associated with worse HRQoL after adjustment for fatigue severity, employment status, and post-COVID comorbidity count, supporting care models that incorporate systematic mental health assessment alongside physical rehabilitation and symptom management. In a sensitivity analysis, adding any pre-COVID comorbidity to the model did not alter the primary coefficients and pre-COVID comorbidity itself was not independently associated with SF-12 (B = − 0.42; p = 0.897), suggesting that the association between post-COVID comorbidity and HRQoL is not substantially confounded by pre-existing disease in this cohort. An exploratory mediation analysis further suggested that depressive symptoms mayaccount for a substantial share of the variance in the fatigue–HRQoL association. The FSS coefficient was attenuated by 34.7% after adjustment for PHQ-9 alone, with a statistically significant indirect effect (Sobel Z = − 2.73, p = 0.006; bootstrap 95% CI excluding zero) — a pattern compatible with, but not diagnostic of, a pathway in which fatigue and emotional distress are intertwined contributors to HRQoL impairment. Because the design is cross-sectional, depressive symptoms could equally lie on a fatigue–HRQoL pathway, operate as a parallel correlate, or themselves be consequences of HRQoL impairment. This accords with emerging evidence that fatigue and depression interact bidirectionally in long COVID, with downstream effects on perceived recovery and daily functioning [37, 42]. These findings support care models in which psychological screening is integrated with fatigue management rather than addressed in isolation. The cross-sectional design precludes directional causal inference; longitudinal mediation studies in larger long COVID samples are needed to disentangle the temporal relationships among fatigue, depressive symptoms, and HRQoL.

Active employment also remained independently associated with better HRQoL. This is consistent with recent evidence linking long COVID to work impairment and adverse economic consequences [43–45]. From a health services perspective, HRQoL assessment in long COVID should therefore be interpreted within a broader social context that includes work ability and social disadvantage, rather than through symptom counts alone [46, 47]. Pain catastrophizing was also substantially elevated in long COVID cases relative to controls and was higher in the higher-burden stratum, consistent with a role for maladaptive pain cognitions in the maintenance of symptom burden and functional limitation in this population.

The clustering analysis did not reveal distinct phenotypic subtypes in this cohort. Silhouette values for all candidate solutions fell below 0.25, indicating that the distribution of long COVID burden is better characterised as a continuous severity gradient than as discrete syndromic categories. For descriptive purposes, we operationalised this gradient as two strata. The higher-burden stratum was characterised by greater persistent symptom load, more severe fatigue and depressive symptoms, worse functional status, lower exercise capacity, and markedly worse HRQoL, alongside lower employment participation. These differences across strata mirror the same variables that emerged as independent correlates of HRQoL in the regression model — fatigue, depressive symptoms, post-COVID comorbidities, and employment — which reinforces that long COVID burden in this cohort is continuously distributed and multidimensional. Notably, acute-phase severity and COVID-19 pneumonia, which were more frequent in the long COVID group, were not tested in the within-case regression because they were not hypothesised as direct determinants of chronic HRQoL. The finding that HRQoL correlates are predominantly biopsychosocial rather than acute-severity-related further supports a care model centred on modifiable factors. The practical clinical implication is not that patients belong to one of two types, but that greater accumulation of biopsychosocial burden at any point along the severity spectrum is associated with worse quality of life and greater need for multidisciplinary support [12–16]. Population-based data have further shown that somatic symptom persistence after COVID-19 is common and not fully explained by pre-existing illness [48, 49].

The clustering findings should nonetheless be interpreted cautiously. Cluster separation was modest, the sample size was limited, and the analysis was performed in a single cohort without external validation. In addition, one clustering variable, 6MWT-D, was missing in a small number of participants, although the sensitivity analysis yielded a highly concordant 2-cluster solution. Accordingly, these strata should be presented as exploratory and hypothesis-generating, not as definitive long-COVID subtypes. These strata mainly show that patients in this cohort can be meaningfully stratified into lower- and higher-burden groups that differ across multiple clinically relevant domains.

Exploratory incorporation of routine inflammation/coagulation markers did not improve the clinical interpretability of the clustering solution. Although a small subgroup with higher C-reactive protein, erythrocyte sedimentation rate, ferritin, fibrinogen derivative, and D-dimer emerged, this partition was highly imbalanced and was not accompanied by clearer differences in symptom burden, fatigue, functional status, sleep, pain catastrophizing, or HRQoL. This suggests that, in our cohort, laboratory variation may identify a subset with relatively higher inflammatory/coagulation burden, but it does not provide a more clinically useful stratification than the primary six-variable clinical model. Accordingly, the main cluster structure in this study is better understood as predominantly clinical/biopsychosocial rather than biomarker-defined.

The higher prevalence of housing without outdoor access in long COVID cases may reflect socioeconomic disadvantage or may have amplified symptom burden during the lockdown period, when outdoor space access was particularly restricted in Spain. This variable was not tested in the multivariable model because it was not hypothesised as a direct predictor of HRQoL; however, future work should explore how housing quality interacts with long COVID recovery trajectories. This study has several strengths. It used a matched control group, captured a broad range of patient-reported and functional measures, and focused directly on HRQoL, a clinically meaningful outcome. It also examined determinants of HRQoL beyond the physical-function variables emphasized in the earlier cohort publication, thereby providing a more HRQoL-centered interpretation of the same parent study. Several limitations should nevertheless be acknowledged. First, this was a cross-sectional observational analysis, and causal inferences are not warranted; recruitment was moreover confined to a single Spanish region (Aragon) through primary care and relied in part on volunteer contacts, so possible selection bias cannot be excluded and the transportability of these associations to other populations and health-care systems remains untested. Vaccination status was available in the database (approximately 89% of cases and 90% of controls were vaccinated) and did not differ meaningfully between groups; it was therefore not included as a predictor in the regression model, but its potential influence on HRQoL outcomes cannot be excluded. Second, the sample size of this secondary analysis (85 cases, 85 controls) was determined by the parent ARALONGCOV cohort. A formal power calculation published in the study protocol [18] estimated that approximately 155 participants per group were required to detect moderate associations at 90% power; the present analysis falls below that target, which may have reduced precision of regression coefficients and limited the power to detect smaller cluster separation indices, and clustering results should accordingly be considered exploratory. Third, several measures were self-reported and may be affected by recall or reporting bias, particularly given that the comparison between pre- and post-COVID states relies on participants’ retrospective recall. Fourth, the study used a global SF-12 score already established in the cohort database rather than the conventional physical and mental component summaries, which may limit direct comparison with some external studies. Fifth, individual matched-pair identifiers were not retained in the exported analytic dataset; the case–control comparison was therefore analysed as two independent groups (Mann–Whitney U and chi-squared or Fisher exact tests) rather than with paired tests, which may have reduced statistical efficiency relative to a fully paired analysis. Finally, because a prior publication from this cohort [17] has already reported overlapping variables, the present manuscript should be understood as a distinct secondary HRQoL-focused analysis rather than a de novo cohort description.

Overall, these findings support multidimensional assessment in long COVID care [50–54]. Screening only for residual respiratory or exercise-related problems will miss important determinants of HRQoL. In primary care and post-COVID pathways, fatigue severity, depressive symptoms, post-COVID comorbidity burden, and work participation appear particularly relevant for identifying patients with the most severe concurrent HRQoL impairment and for guiding rehabilitation, psychosocial support, and return-to-work strategies. Here, “primary care” denotes both the recruitment setting and the level of the health system best placed to deliver this assessment: the patient-reported and functional measures used are inexpensive and already familiar in primary care, and could be combined into a brief screen to flag a high-need subgroup—patients with concurrently elevated depressive-symptom scores, fatigue, post-COVID multimorbidity, and low work participation—for more structured, multidisciplinary evaluation. Throughout, depressive symptoms should be interpreted as a symptom-severity score (PHQ-9) used for screening, not as a formal psychiatric diagnosis.

Conclusions

Among adults with long COVID in primary care, depressive symptoms and post-COVID comorbidity burden were the strongest independent correlates of worse HRQoL, while active employment was associated with better HRQoL; fatigue severity showed a borderline association in the same direction.

These findings support routine multidimensional assessment in long COVID, particularly in primary care settings. Fatigue severity, depressive symptoms, post-COVID comorbidity burden, and work participation appear especially relevant for identifying patients with the most severe concurrent HRQoL impairment and for guiding rehabilitation, psychosocial support, and return-to-work strategies; longitudinal confirmation is needed to establish whether these variables also predict future HRQoL trajectories. These findings support a continuous biopsychosocial severity gradient in long COVID rather than discrete phenotypic subtypes, with multidimensional assessment needed across the full spectrum of burden.

Abbreviations

6MWT-D

Six-minute walk test distance

FSS

Fatigue Severity Scale

HRQoL

Health-related quality of life

IQR

Interquartile range

PCS

Pain Catastrophizing Scale

PC1/PC2

Principal component 1/2

PCFS

Post-COVID-19 Functional Status

PHQ-9

Patient Health Questionnaire-9

PSQI

Pittsburgh Sleep Quality Index

QoL

Quality of life

SF-12

Short Form-12 Health Survey

STROBE

Strengthening the Reporting of Observational Studies in Epidemiology

VIF

Variance inflation factor

Authors’ contributions

DLI and RMB conceived the present secondary analysis. DLI and FML curated the data. DLI and EPG performed the statistical analyses. DLI, RMB and EPG drafted the manuscript. CVG and IBG contributed to the interpretation of results and provided clinical and methodological input. All authors critically revised the manuscript, approved the final version, and agree to be accountable for all aspects of the work.

Funding

This study has been funded by the Instituto de Salud Carlos III through the project PI22/01070 and cofunded by the European Union. The funder had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Data availability

The dataset supporting the conclusions of this article is deposited in the Zenodo repository (https://doi.org/10.5281/zenodo.20599481; https://doi.org/10.5281/zenodo.20599481). Because the data consist of individual-level personal health information collected under the EU General Data Protection Regulation (2016/679) and Spanish Organic Law 3/2018 on Personal Data Protection, the files are held under controlled access and are not openly downloadable. The fully anonymized dataset is available from the corresponding author on reasonable request, subject to authorization by the Clinical Research Ethics Committee of Aragón (CEICA, protocol PI21/278) and consistent with the terms of participants' informed consent.

Declarations

Ethics approval and consent to participate

The ARALONGCOV project and this secondary analysis were conducted in accordance with the Declaration of Helsinki and were approved by the Ethics Committee for Clinical Research of Aragon (CEICA; reference PI21/278). All participants provided written informed consent before inclusion in the parent study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.