Long-term chest CT abnormalities up to 4.5 years after COVID-19 hospitalization and their association with dyspnea and fatigue: a prospective cohort study

Abstract

Background

Long-term pulmonary sequelae following hospitalization for COVID-19 remain incompletely characterized, and few studies have reported radiological outcomes beyond 2 years. This study investigated the persistence of chest CT abnormalities up to 4.5 years after hospitalization for COVID-19 and their association with current dyspnea and fatigue.

Methods

This longitudinal cohort study included patients hospitalized for COVID-19 in 2020. Chest CT was performed 3 months after hospital admission, with follow-up imaging after 12 months and 4.5 years in patients with persisting abnormalities. CT images were evaluated using a CT severity score (CSS) on a 0–12 scale, which was subsequently recoded to an abbreviated CSS (aCSS) on a 0–3 scale. Trends in CT abnormalities and their determinants were analyzed using multivariable mixed effects ordinal regression.

Results

In total, 187 patients underwent chest CT at 3 months. Among patients with persistent abnormalities, 99 completed follow-up CT at 12 months and 54 at 4.5 years. Dyspnea (33%) and fatigue (61%) remained common at 4.5 years; however, correlations between current symptoms and aCSS were weak. Greater severity of the initial disease and older age were strong predictors of CT abnormalities. Regression analysis showed a gradual reduction in ground glass opacities (GGO) severity over time, whereas parenchymal bands increased and were associated with initial severity as assessed by the WHO 8-point ordinal scale for clinical improvement. Mosaic attenuation decreased during follow-up, but remained associated with the severity of the initial COVID-19 episode.

Conclusions

From 3 months to 4.5 years after hospitalization, GGO and mosaic attenuation decreased, whereas parenchymal bands increased, with minimal change beyond 12 months. CT abnormalities were associated with initial disease severity and age but showed little association with dyspnea or fatigue.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12931-026-03771-7.

Introduction

During the first wave of the COVID-19 pandemic, hospitalized patients frequently presented with marked abnormalities on chest CT, most commonly ground-glass opacities (GGO), parenchymal bands, consolidations, or other inflammatory changes [1, 2]. During the first year following COVID-19, GGOs generally resolved completely or partially [3, 4], whereas fibrotic-like abnormalities showed little change in prevalence over time [5].

Concerns have been raised regarding the long-term consequences of COVID-19, including persistent symptoms collectively referred to as long COVID, as well as the potential development or progression of structural pulmonary abnormalities [6, 7]. Reviews and meta-analyses of CT findings 12–24 months after COVID have reported considerable variation in the prevalence of residual abnormalities. Reported pooled prevalence estimates include fibrotic-like changes in 21–38% of patients, GGO in 21–34%, and reticulation in 8–28% [5, 8–10]. Residual chest CT abnormalities have also been associated with the severity of the initial disease [5, 8]. The substantial variation between studies likely reflects heterogeneity in study populations, definitions of radiological abnormalities, and imaging assessment methods [5, 9].

Only a limited number of studies have reported systematic chest CT follow-up beyond 2 years after COVID-19 [11, 12]. In one study, the proportion of hospitalized COVID-19 survivors with radiological lung abnormalities decreased from 46% at 6 months to 36% at 3 years. Non-fibrotic lung abnormalities gradually resolved, whereas fibrotic-like changes remained largely stable [11]. Persistent abnormalities at 3 years were also more common among patients with greater initial disease severity [11]. Similarly, another study with 3-year follow-up reported that most participants had stable fibrotic-like abnormalities over time [12].

We previously reported CT findings at 3 and 12 months in a cohort of patients hospitalized for COVID-19 in 2020. In that analysis, GGO and mosaicism decreased from 3 to 12 months, whereas parenchymal bands increased. Persistent GGO were associated with greater initial disease severity and age ≥ 60 years [13].

This study reports extended follow-up after 4.5 years (54 months) in patients from 3 of the 6 hospitals included in this cohort who had persistent CT abnormalities at 12 months. We hypothesized that GGO and mosaic attenuation would continue to decrease over time, whereas parenchymal bands would remain stable or increase. In addition, we determined the association of persistent CT abnormalities with the severity of the initial COVID-19 and current self-reported dyspnea or fatigue.

Material and methods

Study design, population and procedures

This longitudinal observational cohort study included patients ≥ 18 years of age recruited from six Norwegian hospitals during the first wave of COVID-19 pandemic as part of the Patient-Reported Outcomes and Lung Function After Hospitalization for COVID-19 (PROLUN) study [14, 15]. Consecutive patients discharged from hospital between 11 March 2020 and 1 June 2020 were invited to participate, and informed consent was obtained 4–8 weeks after discharge. Patients living outside the hospitals’ catchment areas, unable to provide informed consent due to cognitive impairment, or already participating in another clinical trial were excluded [14].

All participants provided written informed consent either via paper forms or through a secure web-based application. The study was approved by the Regional Ethics Committee for South-Eastern Norway (no. 2020/125384) and the Data Protection Officer at each participating hospital. The study is registered with ClinicalTrials.gov (NCT 04535154).

Baseline variables during acute COVID-19 was collected from electronic medical records, and comorbidity was scored using the Charlson comorbidity index with the original weights [16]. Disease severity was assessed using the WHO 8-point ordinal scale for clinical improvement [17]: 3, hospitalized, no oxygen therapy; 4, oxygen by mask or nasal prongs; 5, non-invasive ventilation or high-flow oxygen; 6, intubation and mechanical ventilation; 7, ventilation with additional organ support. For analysis, we used three categories: 3, 4, and 5–7.

All participants were invited to ambulatory follow-up visits at 3 and 12 months after hospital admission, including pulmonary function tests, blood sampling, and chest CT. Those with chest CT abnormalities compatible with COVID-19 sequelae after 3 months had a repeat CT at 12 months.

For the present 4.5-year sub-study, we invited all participants in 3 of the 6 hospitals of the PROLUN study (Akershus University Hospital, Oslo University Hospital Ullevål, and Østfold Hospital Kalnes) who had chest CT abnormalities at 12 months. These hospitals had recruited 206 of the 262 participants (79%) in the PROLUN study (Fig. 1). In total, 75 patients with persistent CT abnormalities after 12 months were eligible. After excluding 2 deceased patients, 7 who were lost to follow-up or could not be reached, and 12 who declined participation, 54 patients underwent chest CT approximately 4.5 years after hospitalization and responded to a brief online questionnaire distributed by their local hospital.

Fig. 1.

Patient inclusion and attrition throughout the study

Chest CT protocol, image review, and scoring

All three hospitals used a standardized volumetric CT protocol across different CT systems at all study time points, with reconstruction of thin Sects. (0.9–1.25 mm). CT scans were acquired in both supine and prone positions during deep inspiratory breath-holding. Details about the imaging protocol has been described previously [13].

Three thoracic radiologists (T.M.A, C.M., J.R.R.), each with more than 10 years of experience, reviewed all CT scans in consensus and in a random order, without access to clinical data. They registered the presence, extent, and distribution of CT features using nomenclature from the Fleischner Society [18], including GGO, parenchymal bands, reticular pattern, airspace consolidation, interlobular septal thickening, bronchiectasis and/or bronchiolectasis, and mosaicism. Reticular pattern, indicative of fibrotic coarseness, was graded as follows: 1, fine intra-lobular fibrosis without evident cysts; 2, predominantly microcystic reticular pattern involving air spaces smaller than or equal to 4 mm in diameter; and 3, a predominantly macrocystic reticular pattern with air spaces larger than 4 mm in diameter, typically corresponding to honeycombing [13].

Scoring procedures followed those described previously [13]. The distribution and extent of abnormalities were assessed in four lung zones: above the aortic arch, between the aortic arch and the level of the carina, between the level of the carina and the level of the left inferior pulmonary vein, and below. Each zone comprised both lungs.

For GGO and reticulation the extent of involvement in each zone was graded based on the percentage of affected parenchyma: 0 (none), 1 (1–2%; minimal), 2 (3–10%; moderate), 3 (11–60%, severe). For mosaic attenuation and parenchymal bands, severity was scored as 0 (none); 1 (minimal); 2 (moderate); 3 (severe). For each CT feature, zone scores were summed to generate an overall ordinal chest CT severity score (CSS) ranging from 0 (none) to 12 (maximum) [13], representing the total extent of involvement across both lungs.

To facilitate visualization and regression analysis of the most prevalent abnormalities (with > 20%), the CSS was recoded into an abbreviated CT severity score (aCSS) on a 0–3 scale: 0 (none), CSS score of 0, 1 (mild, CSS 1–3), 2 (moderate, CSS 4–7), and 3 (severe, CSS 8–12) [13]. Subsequent analyses focused on the four most prevalent CT features with the aCSS: GGO, parenchymal bands, mosaicism, and grade 1 reticular pattern.

Assessment of dyspnea and fatigue

Dyspnea was assessed using the modified Medical Research Council Questionnaire (mMRC) dyspnea scale [19]. The mMRC is a widely used self-completed 5-point ordinal scale ranging from 0 (no breathlessness except on strenuous exercise) to 4 (too breathless to leave the house, or breathless when dressing or undressing) [20]. Scores ≥ 2 indicate clinically relevant breathlessness [21].

Fatigue was assessed using the Chalder Fatigue Questionnaire (CFQ-11) [22], which contains 11 items scored on a 0–3 scale. Items are summed to produce a total score ranging from 0 to 33, with higher scores indicating greater fatigue. An alternative bimodal scoring method dichotomizes responses (0–1 = 0; 2–3 = 1), producing and score ranging from 0 to 11. We defined fatigue case-status using this scale with a cut-off of ≥ 4 [22].

Statistical analysis

We present descriptive statistics with the mean (SD), median (25th–75th percentile), or number (%). Independent groups were compared using the Kruskal–Wallis test or Fisher’s exact test, as appropriate. Trends for CSS and aCSS were determined using a non-parametric test for trend.

We determined correlations of the aCSS for the four CT features with dyspnea and fatigue scores using Spearman’s rank correlation. In addition, we compared aCSS according to the presence of current dyspnea (mMRC 2–4) with no dyspnea (mMRC 0–1) and fatigue (CFQ-11 bimodal score 4–11) vs. no fatigue (0–3) in crude analysis using Mann–Whitney U test. In a supplementary analysis, we adjusted for sex and age (above/below 60 years) using ordinal logistic regression analysis. In this analysis, because of few cases with level 3, we collapsed the aCSS to three levels, by combining levels 2 and 3. The proportional odds assumption was acceptable, as determined by the Brant test. These results are presented as the odds ratio (OR) for being in a higher category on the ordinal scale.

We assessed determinants of the aCSS for four CT features (GGO, parenchymal bands, mosaicism, and grade 1 reticular pattern, each scored on a 0–3 scale) using multivariable mixed effects ordinal regression analysis, with random effect for patient and fixed effects for other variables. For mosaicism and grade 1 reticular pattern, we collapsed the scale to three levels by combining level 2 and 3. As independent variables, we used WHO severity rating (3, 4, 5–7), age (< 60 vs ≥ 60 years), sex, and time of CT assessment (3 months, 12 months, or 4.5 years), as used previously [13]. The number and choice of independent variables in the models were decided prior to the analysis, without statistical variable selection. The results are presented as OR for being in a higher category of the aCSS dimension score.

We used Stata SE version 18.0 (StataCorp. LLC, College Station, TX, USA) for analysis and chose a significance level of 5%, using two-sided test.

Results

Chest CT examinations after 4.5 years were completed median 55.0 (25th–75th percentile 54.5–55.4, range 53.4–57.9) months post-admission. Participants with 2 or 3 follow-up chest CT examinations (n = 99) were older than those with CT imaging only at 3 months (n = 88), and those with multiple scans had more severe initial COVID-19 disease, as indicated by higher WHO severity rating, higher body temperature and serum C-reactive protein levels, longer hospital stays, and a larger proportion with ventilator treatment (Table 1).

Table 1.

Descriptives for participants during hospitalization for acute COVID-19, overall and according to number of chest CT exams during follow-up. Mean (SD) or number (%), unless stated otherwise

		No. of chest CT examinations (time)
	Total	1 (3 months)	2 (3 and 12 months)	3 (3,12, 54 months)	p-value
n	187	88	45	54
Age at discharge (years)	57.9 (14.2)	53.8 (13.3)	61 (16.5)	62.2 (11.6)	< 0.001
Education, highest attained
Primary (< 11 years)	33 (19)	12 (14)	8 (19)	13 (25)	0.41
Secondary (11–13 years)	57 (32)	23 (27)	17 (41)	17 (33)
University (> 13 years)	26 (20)	21 (25)	7 (17)	8 (15)
University (> 16 years)	52 (29)	28 (33)	10 (24)	14 (27)
Sex
Female	82 (44)	46 (52)	14 (31)	22 (41)	0.058
Male	105 (56)	42 (48)	31 (69)	32 (59)
Norwegian origin
No	54 (32)	28 (34)	12 (29)	14 (30)	0.85
Yes	116 (68)	54 (66)	29 (71)	33 (70)
Body mass index (kg/m2)	27.7 (4.9)	27.8 (4.1)	27.8 (7.0)	27.6 (4.5)	0.52
Smoking status
Never daily smoker	99 (59)	49 (61)	26 (63)	24 (51)	0.76
Previous daily smoker	66 (39)	30 (37)	14 (34)	22 (47)
Current daily smoker	4 (2)	2 (2)	1 (2)	4 (2)
Comorbidity
Chronic obstructive pulmonary disease	9 (5)	3 (3)	2 (4)	4 (7)	0.56
Asthma	36 (18)	24 (27)	6 (13)	6 (11)	0.039
Sarcoidosis or other interstitial lung disease	0 (0)	1 (2)	0 (0)	1 (0.5)	0.24
Diabetes	18 (10)	5 (6)	3 (7)	10 (19)	0.045
Myocardial infarction	14 (7)	4 (5)	4 (9)	4 (11)	0.30
Congestive heart failure	7 (4)	3 (3)	1(2)	3 (6)	0.70
Hypertension (self-report)	60 (34)	28 (33)	12 (29)	20 (38)	0.66
Charlson comorbidity index
0	134 (72)	67 (76)	33 (73)	34 (63)	0.16
1	34 (18)	14 (16)	5 (11)	15 (28)
≥ 2	19 (10)	7 (8)	7 (16)	5 (9)
During hospitalization for acute COVID-19
Systolic blood pressure (mmHg)	135 (18)	133 (16)	136 (20)	135 (18)	0.91
Temperature (°C)	37.7 (1.1)	37.5 (1.1)	37.5 (1.0)	38.1 (1.2)	0.002
Pulse rate (per min)	86 (19)	86 (20)	83 (19)	88 (15)	0.28
B-Hemoglobin (Hb) (g/dL)	14.0 (1.4)	14.1 (1.2)	14.0 (1.4)	13.9 (1.6)	0.82
C-Reactive protein (mg/L)	54 (20–120)	34 (9–90)	86 (30–128)	88.5 (46–160)	< 0.001
P-Creatinine (µmol/L)	84 (48)	80.1 (38.3)	82 (22)	92 (72)	0.53
B-Leukocytes (× 10^9/L)	8.7 (4.2)	7.6 (3.4)	9.4 (5.1)	9.8 (4.1)	< 0.001
B-Lymphycytes (× 10^9/L)	1.1 (0.6)	1.2 (0.6)	1.2 (0.8)	0.9 (0.5)	< 0.001
Ventilator with intubation
No	162 (88)	82 (94)	40 (89)	40 (76)	0.006
Yes	23 (12)	5 (6)	5 (11)	13 (24)
Who severity rating, grouped
3	66 (35)	43 (49)	11 (24)	12 (22)	< 0.001
4	89 (48)	37 (42)	27 (60)	25 (46)
5–7	32 (17)	8 (9)	7 (16)	17 (32)
Length of stay (days), median (25th–75th percentile)	6 (3–11)	4 (3–7)	7 (5–13)	10.5 (4–17)	< 0.001
PCR-verified COVID-19 (throat/nasopharynx swab)
No	5 (3)	2 (2)	0 (0)	3 (6)	0.33
Yes, before hospital admission	92 (49)	48 (55)	20 (44)	24 (44)
Yes, during the hospital stay	90 (48)	38 (43)	25 (56)	27 (50)
Hospital
Akershus University Hospital	97 (52)	44 (50)	22 (49)	31 (57)	0.13
Oslo University Hospital (Ullevål)	62 (33)	34 (39)	17 (38)	11 (20)
Østfold Hospital Kalnes	28 (15)	10 (11)	6 (13)	12 (22)

Chest CT severity scores and abbreviated scores

Using the full CSS score (0–12 scale), there was an improvement in GGO over time (p = 0.012), but non-significant trends for the other six chest CT features (Table S1). When applying the aCSS score (0–3 scale) to the four most prevalent features, the distribution of scores over time showed an improvement in GGO and an increase in parenchymal bands, but little change in mosaicism and reticulation grade 1 (Fig. 2).

Fig. 2.

Distribution of abbreviated CT severity scores (0–3 scale with percentages) 3 months, 12 months and 4.5 years after initial COVID-19 (n = 54): a Ground glass opacities, b Parenchymal bands, c Mosaicism, d Reticulation grade 1

Association of CT findings with current dyspnea and fatigue

At 4.5 years after the index hospitalization for COVID-19, 16 patients (33%) reported dyspnea mMRC grade 2–4, and 30 (61%) reported fatigue (CFQ-11 score 4–11) (Table 2).

Table 2.

Dyspnea and fatigue scores at 4-year follow up (n = 49), number (%) or median (25th–75th percentile)

Dyspnea, mMRC grade
0	16 (33)
1	17 (35)
2	10 (20)
3	5 (10)
4	1 (2)
Dyspnea, mMRC dichotomized
0–1	33 (67)
2–4	16 (33)
Fatigue, CFQ-11 score (0–33 scale)	15 (12–22)
Fatigue, CFQ-11 bimodal score (0–11 scale)	4 (0–9)
Fatigue, CFQ-11 bimodal score dichotomized
0–3	19 (39)
4–11	30 (61)

The Spearman’s rank correlations (rho) between aCSS features and dyspnea score were generally low, ranging from −0.27 to 0.09, and three of the associations were negative. The correlations between aCSS features and fatigue were even smaller, with rho ranging from −0.16 to 0.00, i.e., none were positive (Table S2).

When comparing participants with dyspnea (mMRC grade ≥ 2) (n = 15) with those without (n = 31), the aCSS was lower for parenchymal bands in dyspnea than without (Fig. 3). There were no group differences for the other chest CT features. There was no difference in aCSS scores for any of the four CT features when comparing participants with and without fatigue (Fig. 4).

Fig. 3.

aCSS at 4.5 years according to current dyspnea (mMRC 2–4, n = 15) vs. no dyspnea (mMRC 0–1, n = 31) for patients with persistent chest CT changes 12 months after COVID-19 (n = 49): a Ground glass opacities, b Parenchymal bands, c Mosaicism, d Reticulation grade 1

Fig. 4.

aCSS at 4.5 years according to current fatigue (CFQ-11 4–11, n = vs. 0–3) for patients with persistent chest CT changes 12 months after COVID-19 (n = 49): a Ground glass opacities, b Parenchymal bands, c Mosaicism, d Reticulation grade 1

A supplementary analysis, adjusting for sex and age, showed a lower OR for having a higher aCSS score for parenchymal bands in dyspnea vs. no dyspnea, supporting the unadjusted findings (Table S3). The results for dyspnea in relation to the other aCSS features, as well as for fatigue to all aCSS features, also were in line with the unadjusted results.

Determinants of abbreviated CT severity scores

In multivariable mixed effects ordinal regression of aCSS scores for GGO, the odds of being in a more severe aCSS category at 4.5 years compared to 3 months were 0.13 (95%CI 00.05 to 0.33, p < 0.001), after adjustment for age, sex, and WHO severity, and the odds at 4.5 years were little different from those at 12 months (Table 3). Higher initial WHO severity was associated with GGO severity, with an OR of 21.99 (2.77 to 174.64, p = 0.03) for WHO severity categories 5–7 compared to category 3. For age, the OR of being in a higher aCSS category was 2.60 (1.55 to 4.36, p < 0.001) per 10 years.

Table 3.

Odds ratios for being in a higher category of abbreviated CT severity score (0–3 scale), multivariable mixed ordinal regression (340 cases, 187 patients)

		Ground-glass opacities			Parenchymal bands			Mosaicism			Reticular patterns grade 1
	n	Odds ratio	95% Conf. Interval	p-value	Odds ratio	95% Conf. Interval	p-value	Odds ratio	95% Conf. Interval	p-value	Odds ratio	95% Conf. Interval	p-value
Who severity rating
3* (no oxygen)	101	1			1			1			1
4 (oxygen)	166	7.31	(1.58 to 33.78)	0.011	2.00	(0.54 to 7.35)	0.30	2.35	(0.69 to 7.98)	0.17	1.28	(0.10 to 16.04)	0.85
5–7 (High-flow/NIV/Ventilator/ECMO)	73	21.99	(2.77 to 174.64)	0.003	21.77	(3.89 to 121.75)	< 0.001	22.38	(4.19 to 119.41)	< 0.001	17.53	(0.68 to 451.96)	0.084
Age, per decade	340	2.60	(1.55 to 4.36)	< 0.001	1.75	(1.13 to 2.72)	0.012	1.22	(0.83 to 1.79)	0.31	2.38	(0.92 to 6.17)	0.075
Sex
Female*	140	1			1			1			1
Male	200	1.88	(0.49 to 7.29)	0.36	0.91	(0.29 to 2.85)	0.88	1.51	(0.52 to 4.39)	0.45	15.21	(1.17 to 197.22)	0.037
Months
3*	187	1			1			1			1
12	99	0.15	(0.07 to 0.32)	< 0.001	4.70	(2.25 to 9.81)	< 0.001	0.42	(0.20 to 0.90)	0.025	0.64	(0.18 to 2.21)	0.48
54	54	0.13	(0.05 to 0.33)	< 0.001	8.28	(3.42 to 20.04)	< 0.001	0.34	(0.13 to 0.89)	0.028	0.44	(0.08 to 2.37)	0.34
*Reference category
ICC subject			0.809 (0.693 to 0.888)			0.697 (0.537 to 0.821)			0.630 (0.439 to 0.788)			0.934 (0.734 to 0.986)
AIC			708.0			501.1			493.0			260.2

ICC intraclass correlation coefficient, AIC Akaike's information criterion, NIV non-invasive ventilation, ECMO extra-corporeal membrane oxygenation

Parenchymal bands were more marked at both 12 months and 4.5 years compared to 3 months in multivariable analysis, with OR 4.70 (95%CI 2.25 to 9.81, p < 0.001) and OR 8.28 (95%CI 3.42 to 20.04, p < 0.001), respectively (Table 3). WHO severity category 5–7 was also strongly associated with higher aCSS for parenchymal bands compared to category 3, with OR 21.77 (95%CI 3.89 to 121.75, p < 0.001).

For mosaicism, aCSS was lower at 12 months and 4.5 years than at 3 months but increased with increasing WHO severity scores (Table 3). Reticular pattern grade 1 had a low number of events, and male sex was the only significant determinant.

Discussion

This study provides chest CT follow-up approximately 4.5 years after hospitalization for COVID-19, representing, to our knowledge, one of the longest longitudinal imaging studies of COVID-19 survivors. The principal findings were that GGOs gradually decreased over time, whereas parenchymal bands increased modestly, and mosaic attenuation showed early improvement with subsequent stability. Importantly, most radiological changes occurred during the first year after hospitalization, with minimal progression thereafter.

Persistent CT abnormalities were strongly associated with the severity of the initial illness and with older age. However, these imaging findings showed little association with long-term symptoms, including dyspnea and fatigue. This suggests that residual structural abnormalities detected on chest CT may have limited clinical relevance for persistent symptom burden in the chronic phase after COVID-19.

Evolution of CT abnormalities

Using the full CT severity score (0–12 scale), only GGO showed significant improvement over time, supporting the concept that inflammatory or partially reversible parenchymal abnormalities may continue to resolve several years after infection. Similar trends have been reported in other longitudinal studies [11, 23–25]. A recent meta-analysis reported that the pooled GGO prevalence decreased from 32% at 6 months to 20% at 36 months after discharge [26]. Some authors have suggested that GGO after COVID-19 may gradually resolve completely [27, 28] or almost completely [25]. For example, the latter study reported a decline in GGO from 56.6% at 3 months after discharge to 0.9% at 4 years [25]. In our study; however, there was little difference at approximately 4.5 years compared with 12 months, indicating that resolution primarily occurs within the first year.

In contrast, fibrotic-like abnormalities—including reticulation and parenchymal bands showed little improvement. These features likely reflect residual architectural distortion rather than active inflammation. Importantly, findings suggestive of established pulmonary fibrosis, such as honeycombing or traction bronchiectasis, were rare, indicating no clear evidence of progressive fibrotic lung disease in this cohort. Also, it is important to differ between manifest fibrosis and post-infectious scarring without clinical significance, and to differentiate post-COVID-19 abnormalities from unrelated interstitial lung abnormalities (ILA) [29]. ILA is incidental, not related to COVID-19, and may represent an early feature of interstitial lung disease [30].

There is variability in the definition of fibrosis and fibrotic-like abnormalities across studies, and therefore, heterogeneity in the reported prevalences. Our findings are, however, consistent with recent meta-regression analyses of longitudinal chest CT features up to 3 years after COVID-19 infection, which reported no significant temporal trend in the polled proportion of fibrosis or parenchymal bands, suggesting a stable but non-progressive pattern [26]. The findings also support a recent report of non-progressive fibrotic-like abnormalities without clinical relevance approximately 4 years after discharge [25].

Mosaic attenuation, which may represent obliterative small-airways disease or occlusive vascular disease [18], decreased at 12 months compared to 3 months, but showed little change beyond 12 months. Mosaic attenuation is recommended as a specific descriptor when reporting chest CT abnormalities after COVID-19-pneumonia [29], but there are heterogeneity and inconsistency in the reported proportions. In recent 3-year and 4-year studies, the prevalence of mosaicism, air trapping, or other indicators of small-airway disease was not reported [11, 25, 26].

Association of CT abnormalities with disease severity

Initial disease severity and age were powerful predictors of persistent GGO in multivariable analysis: WHO severity 5–7 conferred 22-fold higher odds of more severe GGOs compared with WHO 3, and the odds increased 2.6-fold per decade of age. This is consistent with prior findings that severe acute illness and older age predispose for persistent post-COVID chest CT abnormalities and diffusion impairment [25, 31–33]. In addition, parenchymal bands and mosaicism were associated with initial disease severity.

Previous studies at about 1 year after COVID-19 have reported no association with initial severity for any chest CT feature [34], or an association for a global CT score and GGO, but not for other CT features [35]. The latter study also conducted multivariable analysis, showing that increasing age was associated with an abnormal chest CT score [35], but disease severity was not examined.

In a meta-analysis, residual CT abnormalities at 1 year were higher in severe/critical cases than in mild/moderate cases [5], and in a meta-regression of 14 studies, the proportion of people with severe or critical COVID-19 or relevant subject characteristics, such as age, sex, smoking status, hypertension or diabetes, were not associated with the 1-year prevalence of any residual lung abnormalities [9].

Association of CT abnormalities with current dyspnea or fatigue

At 4.5 years after hospitalization, 61% of participants reported fatigue, and 33% reported dyspnea scores corresponding to clinical important dyspnea. In a large general population sample one would expect about 8.5–11% to have clinically important dyspnea [21], although there are large variations between populations [36]. The persistent dyspnea and fatigue over time indicate a substantial and persistent symptom burden more than 4 years after acute infection. Notably, that there was no positive association between aCSS and symptoms of either dyspnea or fatigue after 4.5 years. This suggests that persistent chest CT findings here have little clinical relevance for long-time symptom burden, and that perceived impairments may persist independent of findings on inspiratory CT.

The lack of association of symptoms of dyspnea and fatigue with chest CT findings supports previous reports [25, 31, 37–39]. Persistent symptoms may therefore reflect multifactorial mechanisms, including deconditioning, autonomic dysfunction, systemic inflammation, or other components of post-acute COVID-19 syndrome rather than irreversible pulmonary damage.

Strengths and limitations

Several different scoring and reporting systems for chest CT were developed during the pandemic for standardization and consistency in reporting and prediction of disease severity and outcomes [40–42]. However, most existing studies have analysed on the prevalence of dichotomized radiologic features, i.e., yes/no, and not used more granular severity rating. Therefore, a strength of the present study is the analysis using ordinal severity scores as outcome and appropriate multilevel ordinal regression models for analysis [43].

There is no universally accepted optimal approach for the reading and interpretation of CT scans, particularly in the context of post-COVID-19 lung disease [44]. The potential of AI-based quantitative CT analysis to identify structural and functional abnormalities relevant to long COVID symptom burden has been increasingly recognized [45]. In our study, however, all estimates of affected lung area were made by visual assessment without the assistance of computer-based quantification methods.

Semi-quantitative scoring systems rely on visual assessment of predefined patterns and their estimated extent, which introduces a degree of subjectivity and the possibility of inter-observer variability. As an alternative, quantitative AI-based tools have been proposed for assessing lung involvement, particularly in settings with high patient volumes or limited access to expert chest radiologists [44, 46, 47]. Several studies suggest that machine-learning based quantitative CT can detect small airways disease and identify pulmonary imaging biomarkers associated with long COVID symptom burden [45, 48], offering advantages in terms of reproducibility and reduced dependence on reader experience [49, 50]. However, quantitative, AI-based tools can miss subtle nuances best recognized by expert radiologists, and higher diagnostic accuracy of semi-quantitative visual scoring compared with quantitative assessment has been reported [51]. Other studies have demonstrated high concordance and similar accuracy between visual and quantitative methods [44].

Overall, AI-based tools represent a promising complement to radiologist interpretation, with potential to improve efficiency and reduce variability. In our study, experienced chest radiologists performed a structured, consensus-based visual assessment, systematically evaluating anatomical distribution. Baseline and follow-up CT scans were reviewed side-by-side at each anatomical level to ensure consistent comparison over time. This approach likely reduced inter-observer variability and strengthened the reliability of visual scoring.

All CT images were reviewed at one hospital, with a limited number of readers in consensus, which should facilitate standardization. The protocol included both prone and supine acquisitions, which enabled us to avoid misinterpreting random gravity-dependent atelectasis as GGO or reticular opacities.

The subjects initially recruited to the three hospitals in this study represent about 24% of surviving patients with COVID-19 in all Norwegian hospitals during the recruitment period in 2020 (882 admitted with COVID-19 as main diagnosis, and 91 COVID-19-associated deaths in hospitals) [52]. Therefore, the participants were reasonably representative of survivors after hospitalization for COVID-19 during the first wave in Norway. The proportion of patients admitted to the intensive care unit and the mortality and complication rates were low, probably reflecting that Norwegian hospitals had adequate capacity and admitted less severely ill patients than in some other countries.

The initial sample size was larger than some other longitudinal studies of CT after COVID-19, but due to radiation considerations we only invited those with abnormalities at each step for the next follow-up. This limited the number of patients with follow-up at 4.5 years.

This selection of only patients with chest CT abnormalities creates a selection bias, and in the context of mixed models creates a “missing not at random” (MNAR) problem, which challenges the assumptions for use of most statistical methods. In addition, it was the dependent variable that was missing. This might lead to biased estimates. However, for ethical reasons, we could not justify further follow-up with chest CT of those with normal findings. We think it is unlikely that they would have developed new radiological findings, but in this age group it is a possibility. Nevertheless, we do not think it would be fair to conduct a sensitivity analysis assuming that non-participants would have a negative chest CT at follow-up,”last observation carried forward”, either. Moreover, imputation of an ordinal dependent variable, and with an uncertain mechanism for missingness, would also be challenging. This selection might lead to less variability in the sample throughout the study, which could lead to reduced association between images and dyspnea.

The reported CT abnormalities or structural lung abnormalities here refer to the four most prevalent CT features identified from inspiratory scans only, as assessed by three radiologists across four lung zones, as previously described. The findings were based on a specific protocol with inspiration scans only and a particular reading and scoring procedure, and may therefore not apply to other settings with different populations or protocols.

The sample size also limited the number of independent variables in multivariable analysis. All patients were hospitalized during the first wave of COVID-19 in 2020; therefore, the findings might not apply to chest CT abnormalities for later corona virus variants, such as omicron [53].

Pre-existing interstitial lung abnormalities, including honeycombing and bronchial dilatation, are common [54–56], potentially predating the onset of COVID-19 infection. We did not have CT studies prior to infection for comparison, except for a handful of patients, which is an important limitation in this and similar studies. The protocol did not include routine expiratory scanning, limiting the assessment of mosaic attenuation as a marker of small-airway disease.

Conclusions

This study has shown that GGO and mosaicism decreased between 3 months and 4.5 years after hospitalization for COVID-19, whereas parenchymal bands increased modestly over time, as indicated by changes in ordinal aCSS in multivariable models. There was little change in CT abnormalities beyond 12 months. These findings suggest that routine long-term CT surveillance beyond the first year after COVID-19 hospitalization may be questioned in most patients, particularly in the absence of progressive symptoms.

Abbreviations

aCSS

Abbreviated CT severity score

CFQ

Chalder fatigue questionnaire

COVID-19

Coronavirus disease 2019

CSS

CT severity score

CT

Computed tomography

GGO

Ground glass opacities

ILA

Interstitial lung abnormalities

mMRC

Modified Medical Research Council

OR

Odds ratio

WHO

World Health Organization

Authors’ contributions

Conceptualization: KS, GE, OHS, TVL, TMA, HA Data curation: TVL, CM, JRR, TMA, JG, SRL Methodology: KS, TMA, GE Project administration: KS, GE, OHS, HA, TMA Resource: GE, HA, OHS, TMA, JG Formal analysis: KS, TMA Supervision: KS, GE, OHS, TMA Original draft: KS, TMA Writing – review and editing: KS, TMA, CM, JG, OHS, TVL, JRR, HA, SRL, GE All authors read and approved the final manuscript.

Funding

Open access funding provided by University of Oslo (incl Oslo University Hospital). The CT acquisitions were partially supported by an unrestricted grant to HA from Boehringer Ingelheim, Norway. The funder had no role in the conceptualization, design, data collection, analysis, decision to publish, or preparation of the manuscript.

Data availability

The datasets generated and analysed during the current study are not publicly available due to limitations in the regulations and ethics approval, but an anonymized limited dataset to reproduce the findings are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

All study participants provided informed consent through a paper form or a secure web-application. The study was approved by the Regional Ethics Committee for South-Eastern Norway (no. 2020/125384) and the Data Protection Officer at each participating hospital. The study is registered with ClinicalTrials.gov (NCT 04535154).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.