Abstract

Aim

To evaluate adults with previously documented severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection who presented to the emergency department with new or persistent symptoms at least 4 weeks after acute coronavirus disease 2019 (COVID-19).

Materials and Methods

We examined presenting symptoms; demographic and clinical characteristics, laboratory and imaging findings; and factors concurrently associated with all-cause hospital admission during the index post-acute emergency department encounter. Medical records of 760 patients with documented previous SARS-CoV-2 infection who presented to the emergency department between January 2020 and November 2022 were reviewed. Patients with ongoing polymerase chain reaction positivity (n=54), with complaints clearly unrelated to the post-acute presentation (n=74), with incomplete laboratory data (n=84), or aged <18 years (n=48) were excluded. The final cohort comprised 500 patients who presented at least 4 weeks after the documented acute COVID-19 diagnosis.

Results

Of the 500 patients, 275 (55.0%) were female and 225 (45.0%) were male; the median age was 52 years (range, 18-90). Thirty patients had an all-cause hospital admission during the index post-acute emergency department encounter. Admission was more frequent among patients who had required hospital-based care during acute COVID-19 than among those managed as outpatients (19.3% vs. 2.3%; p<0.001). After Holm adjustment, white blood cell count, neutrophil count, blood urea nitrogen, C-reactive protein, procalcitonin (PCT), N-terminal pro-B-type natriuretic peptide, and D-dimer levels were significantly higher among hospitalized patients. In the post hoc exploratory Firth-penalized multivariable model, older age, higher PCT and D-dimer levels, and acute-period hospitalization were associated with same-encounter admission after adjustment.

Conclusion

Older age, higher concurrent PCT and D-dimer levels, and hospitalization during acute COVID-19 were associated with all-cause admission at the index post-acute emergency department encounter. These exploratory same-encounter associations do not establish prediction, causation, or hospitalization attributable to a post-COVID condition, and they require external validation.

Introduction

Coronavirus disease 2019 (COVID-19), caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), became a global pandemic with substantial morbidity and mortality (1). Persistent, recurrent, or new symptoms after acute infection remain an important clinical problem and may impair daily functioning (2, 3).

The post-acute spectrum is referred to as Long COVID, post-COVID-19 condition, post-acute sequelae of SARS-CoV-2 infection, ongoing symptomatic COVID-19, or post-COVID-19 syndrome. Definitions differ mainly by duration: WHO describes symptoms beginning within 3 months, lasting at least 2 months, and unexplained by another diagnosis (4); CDC describes a chronic condition lasting at least 3 months (2); and NICE uses Long COVID as an umbrella term for symptoms persisting beyond 4 weeks (5).

Long COVID may follow infection of any severity and fluctuate across organ systems (6). Common complaints include dyspnea, fatigue, and pain (7); proposed mechanisms include immune dysregulation, inflammation, endothelial and microvascular injury, and persistent organ damage (7, 8). The diagnosis, treatment, and management of Long COVID are multidisciplinary and symptom-oriented. In addition to routine treatment methods, patient-specific adjunctive therapies are required (8, 9).

Long COVID denotes the study's operational threshold of at least 4 weeks after documented infection under the NICE framework (10).

This study examined symptomatic adults who presented to the emergency department at least 4 weeks after documented COVID-19, consistent with the broader NICE framework. We described symptoms, clinical and laboratory findings, and factors that were concurrently associated with all-cause admission during the index post-acute emergency department encounter.

Materials and Methods

This study was approved by the Scientific Research and Publication Ethics Committee of İnönü University, the Republic of Türkiye, under decision number 2022/3020 at the session held on September 20, 2022. The study was a retrospective observational investigation conducted in the emergency department of a tertiary university hospital.

Adults with documented SARS-CoV-2 infection who presented to the emergency department of a tertiary university hospital between January 2020 and November 2022 with new or persistent symptoms at least 4 weeks after diagnosis were eligible. The interval was calculated from diagnosis because recovery dates were inconsistently recorded. Exclusion criteria were ongoing polymerase chain reaction positivity; age <18 years; missing required laboratory measurements, and complaints unrelated to the post-acute presentation (trauma, pregnancy, or forensic cases). Symptom onset and duration were not uniformly recorded; therefore, the cohort was not reclassified using the stricter WHO definition.

Statistical Analysis

SPSS 26.0 was used for descriptive and univariable analyses. Continuous variables were summarized as mean ± standard deviation or median (interquartile range [IQR]), and categorical variables were summarized as n (%). Two groups were compared using independent-samples t-tests or Mann-Whitney U tests; comparisons involving more than two groups used the Kruskal-Wallis test when appropriate. Categorical data were analyzed using the chi-square test, the two-sided Fisher exact test, or the Fisher-Freeman-Halton exact test. Laboratory p values were Holm-adjusted; rank-biserial correlations and 95% confidence intervals (CIs) (5,000 stratified bootstrap resamples) were used to quantify effect size.

Adjusted associations with all-cause ward and/or intensive-care admission during the index post-acute emergency department encounter were examined using Firth-penalized logistic regression. For the revised analysis, a fixed, post hoc, exploratory nine-variable model was defined after a review of the available data and prior analyses, without automated stepwise selection: sex, age, acute-period hospitalization, thoracic involvement, malignancy, white blood cell count (WBC), C-reactive protein (CRP), procalcitonin (PCT), and D-dimer. Female sex and acute outpatient care were the reference categories. Odds ratios (ORs) were reported per 10 years of age, 1 ×10³/µL WBC, 1 mg/dL CRP, 0.1 ng/mL PCT, and 1 mg/L D-dimer. Profile penalized-likelihood 95% CIs and penalized-likelihood-ratio p-values were reported; variance inflation factors were used to assess collinearity.

Firth estimation was implemented using a custom Python function (Python 3.12; SciPy 1.17) that maximized the Jeffreys-prior-penalized log-likelihood, ℓF=ℓ+0.5 log|X′WX|. Adjusted-score Newton–Raphson updates used step-halving when ℓF decreased. Convergence required a maximum absolute adjusted-score component <1×10^-6 and a change in ℓF of (maximum 1,000 iterations). Penalized likelihood-ratio p-values were obtained using constrained refits in which the tested coefficient was fixed at zero. Each coefficient was fixed by profile limits; nuisance parameters were re-optimized with L-BFGS-B; and 2ΔℓF=χ²₁,0.95 was located. Area under the curve (AUC) (scikit-learn 1.8) was estimated using a stratified bootstrap 95% CI (5,000 resamples); calibration was assessed using intercept, slope, and Brier score. The optimism-adjusted AUC was estimated using 1,000 bootstrap resamples (seed 20260908), refitting the fixed model in each sample. CRP values <0.33 mg/dL were set to 0.33 mg/dL for analysis. PCT was standardized to ng/mL; eight misplaced decimal separators were corrected after review of the source data.

All tests were two-sided, with p<0.05. Complete laboratory and model data were available for all 500 included patients; 84 screened patients lacked at least one required measurement. It was not possible to determine whether missingness was related to severity or disposition because outcomes were incomplete for excluded records. Patient selection is shown in Figure 1.

Results

Among 500 patients, 275 (55.0%) were women and 225 (45.0%) were men; the median age was 52 years (range, 18-90). The median time from acute diagnosis to presentation was 54 days (IQR, 40-69; range, 30-139). Of these, 445 (89.0%) presented at 4-12 weeks and 55 (11.0%) thereafter. Acute care was outpatient for 391 patients (78.2%), ward-based for 76 (15.2%), and intensive care for 33 (6.6%). At least one comorbidity was present in 71.6% of patients, and thoracic involvement was present in 377 (75.4%) (Table 1).

Common symptoms were dyspnea (73.8%), weakness or fatigue (68.0%), exertional dyspnea (53.4%), loss of appetite (47.0%), cough (33.4%), chest pain (32.2%), myalgia (23.8%), sputum (20.4%), and brain fog (8.0%). Symptoms were not mutually exclusive.

Thirty patients were admitted at the index encounter: 19 to wards and 11 to intensive care. Recorded diagnoses that were not mutually exclusive included pulmonary embolism and pneumonia (n=11 each), residual fibrotic changes (n=6), heart failure and gastrointestinal bleeding (n=2 each), and myocardial infarction, total atelectasis, elevated liver enzymes, adrenal insufficiency, alveolar hemorrhage, and diabetes-related decompensation (n=1 each). Diagnoses and routine-care admission decisions were not independently adjudicated.

Acute and post-acute care levels were associated (Fisher-Freeman-Halton p<0.001). Specifically, 5/11 post-acute intensive-care admissions   required acute intensive care, and 12/19 post-acute ward admissions   received acute ward care (Table 2).

After Holm correction, admitted patients had higher WBC (9.87 vs 7.60 x10³/uL; adjusted p=0.012), neutrophils (5.97 vs 4.38 x10³/uL; p=0.019), blood urea nitrogen (18.45 vs 14.49 mg/dL; p=0.009), CRP (2.96 vs <0.33 mg/dL), PCT (0.099 vs 0.034 ng/mL), N-terminal pro-B-type natriuretic peptide (260.5 vs 52 pg/mL), and D-dimer (1.05 vs 0.33 mg/L FEU; the latter four p<0.001) (Table 3). Glucose and alanine aminotransferase were no longer significant after correction; other laboratory measures did not differ.

The fixed post hoc nine-variable Firth model included all 500 patients and 30 events (Table 4). Admission was associated with age (adjusted OR per 10 years=1.45; 95% profile penalized-likelihood CI, 1.07-2.01; p=0.017), PCT (per 0.1 ng/mL=1.12; 1.02-1.24; p=0.013), D-dimer (per 1 mg/L=1.26; 1.10-1.49; p<0.001), and acute-period hospitalization (OR=3.45; 1.31-9.39; p=0.012). Concurrent biomarkers should not be interpreted as prospective predictors.

Sex, thoracic involvement, malignancy, and CRP were not significant after adjustment. WBC was borderline (adjusted OR per 1 ×10³/µL=1.117; 95% profile penalized-likelihood CI= 0.9999-1.2386; p=0.0502). Apparent AUC was 0.914 (bootstrap 95% CI, 0.870–0.951); optimism-adjusted AUC was 0.885. Calibration intercept, slope, and Brier score were −0.098, 1.078, and 0.039, respectively; VIFs ranged from 1.06 to 1.42.

Unadjusted analyses associated admission with male sex (OR=3.04; 95% CI, 1.36-6.78; p=0.007) and the presence of malignancy (OR=3.04; 1.08-8.53; p=0.044), but not with coronary artery disease, asthma/chronic obstructive pulmonary disease, diabetes, or hypertension (Table 5).

After excluding isolated myocardial infarction and gastrointestinal bleeding, 498 patients and 28 events remained. Estimates were similar: adjusted ORs were 1.38 per 10 years (95% profile CI, 1.0001-1.95; p=0.0499), 1.11 per 0.1 ng/mL PCT (1.02-1.23; p=0.015), 1.27 per 1 mg/L D-dimer (1.10-1.50; p<0.001), and 3.85 for acute hospitalization (1.41-11.00; p=0.009). The apparent AUC was 0.918 (bootstrap 95% CI, 0.873-0.955).

Discussion

Long COVID is thought to be associated with female sex, advanced age, obesity, diabetes, cardiovascular diseases, and other comorbidities. In our study population, however, male sex and the presence of malignancy were associated with hospitalization only in unadjusted analyses and were not significant after adjustment, highlighting the need for cautious interpretation (11, 12).

Literature reviews include studies showing that hospitalization during the acute phase is associated with hospitalization during the long COVID period (13, 14). Consistent with reports of a greater post-acute burden following severe acute illness, acute-phase hospitalization remained associated with hospitalization during the initial post-acute emergency department encounter (adjusted OR=3.45). This may reflect persistent frailty or healthcare needs; however, the observational design and the all-cause nature of the outcome preclude attributing the findings to causal relationships or specific post-COVID conditions.

Inflammation and endothelial dysfunction are factors potentially contributing to post-acute thromboembolic and multisystem symptoms; higher levels of inflammatory, cardiac stress, and coagulation markers have been observed in patients hospitalized following post-acute presentations (15, 16). In the exploratory model, advanced age, concurrent PCT and D-dimer levels, and acute-phase hospitalization showed a significant association with hospitalization during the prolonged period, while WBC count was of borderline significance. Laboratory parameters may reflect acute conditions or influence clinical decisions. Therefore, these biomarkers should be interpreted not as predictors of future deterioration, diagnostic biomarkers, or causal determinants of Long COVID severity, but rather as indicators associated concurrently with hospitalization.

Study Limitations

The retrospective, single-center design relied on routine records and included only symptomatic emergency patients, limiting generalizability and permitting selection, referral, measurement, and classification bias. Symptom onset, duration, and recovery dates were inconsistently documented; therefore, the cohort follows a broad NICE-based threshold. Excluding 84 records with incomplete laboratory data may have favored patients who were more extensively tested. Vaccination, variant, smoking, body mass index, acute treatments and respiratory support, length of stay, readmission, mortality, and socioeconomic status were unavailable or incomplete, thereby leaving residual confounding.

Only 30 events supported nine model variables. Firth penalization and bootstrap validation reduce but do not eliminate overfitting. Some CIs were wide, and external validation was unavailable. The model was specified post hoc during revision after review of the available data and prior analyses, although no automated stepwise selection was used. Refitting the fixed, selected model in each bootstrap sample estimates optimism conditional on that model, but does not fully account for optimism introduced by analyst-guided model specification. The all-cause outcome was based on routine clinician decisions without standardized criteria or independent adjudication of diagnoses. Alternative acute diagnoses and uncertainty regarding causation remain. Reports of pulmonary fibrosis reflected residual or preliminary imaging interpretations rather than standardized longitudinal diagnoses. Findings are exploratory, not prognostic.

Conclusion

Among symptomatic adults presenting at least 4 weeks after COVID-19, older age, concurrent PCT and D-dimer levels, and hospitalization during acute infection were associated after adjustment with all-cause admission at the index post-acute emergency department encounter. These exploratory same-encounter associations are neither causal nor specific to a post-COVID condition. Prospective multicenter studies with standardized definitions, adjudicated diagnoses, and longitudinal outcomes are needed.