Long-term lung function trajectories in Swedish COVID-19 ICU survivors: a two-year follow-up study

ABSTRACT

Introduction

Longitudinal studies of lung function following COVID-19 remain limited. This study examined lung function in COVID-19 patients for two years after discharge from an intensive care unit (ICU).

Methods

Patients treated for COVID-19 in Uppsala ICU (mean 11.1 days) were assessed at four months, one year, and two years post-discharge. Lung function tests included spirometry, diffusing capacity for carbon monoxide (D_Lco), and body plethysmography. Logistic regression adjusted for age, sex, and body mass index (BMI) assessed the association between impaired D_Lco and patients’ characteristics.

Result

A total of 104 patients (32% female, with a mean age of 60 years) participated in four months follow-up, 40 in one year follow-up, and 21 in two years follow-up after discharge. Impaired D_Lco was observed in 50%, 58%, and 33% of patients at four months, one year, and two years, respectively. A comparison showed that D_Lco% predicted declined from four months to one year (mean 79.0, SD 14.7, to 74.3, SD 15.7; p< 0.001), then improved between one and two years (p< 0.001). Forced vital capacity (FVC) % predicted improved between four months and one year (p < 0.001) and between one and two years (p = 0.004). Forced expiratory volume during the first second (FEV₁) % predicted improved only between four months and one year (p = 0.002). Total lung capacity improved between the one- and two-year follow-ups (p = 0.006). Impaired D_Lco at four months was significantly associated with age ≥60 years (adjusted odds ratios, 95% confidence interval: 6.73 (2.64-17.12), mechanical ventilation (4.74 (1.82-12.34), longer ICU stay (5.84 (2.13-16.01), minimum lymphocyte count at ICU (0.19 (0.04-0.83), and FVC % predicted (0.93 (0.89-0.96)) and FEV₁ % predicted (0.93 (0.89-0.97)).

Conclusion

Despite recovery between one and two years, one-third of patients exhibited impaired D_Lco two years after critical COVID-19, highlighting the need for pulmonary follow-up to address persistent lung function deficits.

Introduction

The coronavirus disease 2019 (COVID-19), caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), is no longer considered a global threat for most individuals, as the pandemic has ended. However, it has left lasting consequences for survivors and continues to challenge healthcare systems worldwide [1]. Persistent symptoms that either continue or develop beyond 12 weeks after acute COVID-19 are categorized as post-acute COVID-19 syndrome, provided they cannot be attributed to an alternative diagnosis [2]. Severe and critical cases of COVID-19 frequently result in acute and serious lung injury due to the infection [3]. This has raised concerns about lung function recovery post-infection and whether any resulting impairment may represent a lifelong burden [4].

Meta-analyses assessing pooled lung function data across patients with varying COVID-19 severity have reported prevalence rates for impaired diffusing capacity for carbon monoxide (D_Lco) of 39% and 31% at 6 and 12 months, respectively, with restrictive pulmonary dysfunction observed in 13% and 5% at the same timepoints [5]. Another meta-analysis evaluating lung function one year post-COVID-19 across all severities found reduced DLco, reduced forced expiratory volume in the first second (FEV₁), reduced ratio of FEV₁ to forced vital capacity (FVC), and reduced total lung capacity (TLC) in 30.5%, 9.1%, 5.1%, and 8.6% of cases, respectively. The prevalence of lung function impairment correlated with the severity of the acute infection [6].

Most studies of hospitalized COVID-19 survivors have shown improved lung function within the first year, particularly regarding D_Lco. However, results beyond one year vary: while several reports continue to show improvement in spirometry and D_Lco [7–10], others report stable or even declining values over time [11]. A 5-year follow-up study in survivors of acute respiratory distress syndrome showed a gradual recovery of pulmonary function, particularly D_Lco [12]. However, whether this pattern applies to post-COVID-19 remains unclear. More long-term studies are therefore necessary.

In our previous study of a smaller patient population treated in an intensive care unit (ICU) for severe COVID-19, we found impaired lung function in 52% of cases, four months post-infection, typically characterized by reduced D_Lco. Abnormal lung function was associated with older age, mechanical ventilation, longer ICU stay, and lower levels of C-reactive protein (CRP) at admission [13]. In this study, we aimed to evaluate lung function trajectories, including spirometry, lung volumes, and D_Lco, over a two-year period in ICU-hospitalized COVID-19 patients. Furthermore, we investigated potential sociodemographic and clinical factors influencing lung function development. A subset of patients and lung function data were shared between the two studies.

Materials and methods

Study design

This was a prospective longitudinal cohort study of patients hospitalized due to acute COVID-19 between March 2020 and February 2021 at the ICU of Uppsala University Hospital, Sweden. Follow-ups were done at 4 months (baseline), 1 year, and 2 years after discharge at Uppsala University Hospital. All patients who survived ICU treatment for COVID-19 during the study period were invited to participate in the 4-month follow-up visit, independent of the presence of persistent symptoms. Those who did not attend the 4-month follow-up were not invited to the later follow-up visits at 1 and 2 years. At all three visits, patients underwent spirometry and D_Lco. Lung volume measurements were performed using body plethysmography at 1 and 2 years. Patient characteristics were extracted from electronic medical records from the time of hospitalization.

The Swedish Ethical Review Authority approved the study (Dnr 2020–02697 with application amendments 2020–03629, 2020–05758, 2021–02205, and 2022–01115-02). The study was pre-registered at clinicaltrials.gov (NCT04474249). Written informed consent was obtained from all patients.

This study was performed in accordance with the recommendations of the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [14]. The study adhered to the principles of the Declaration of Helsinki [15].

Study patients

ICU-treated COVID-19 patients included in the study were followed up at approximately 4 months after discharge. Sixty-one patients from the first wave of COVID-19 were admitted between 10 March 2020, and 31 July 2020, as described elsewhere [13], and 43 patients from the second wave were admitted between 1 August 2020, and 28 February 2021. A total of 40 patients from the first wave were followed up 1 year after discharge, and 21 patients were followed up 2 years after discharge (Figure 1).

Figure 1.

Flowchart of study patients.

D_Lco: diffusing capacity for carbon monoxide.

Lung function

Spirometry and D_Lco were performed using a Jaeger MasterScreen PFT (Vyaire, Mettawa, IL, US). In spirometry, the following parameters were assessed: FEV₁, FVC, and the ratio of FEV₁ to FVC. A minimum of three acceptable and reproducible maneuvers were performed to measure all lung function parameters in accordance with American Thoracic Society/European Respiratory Society (ATS/ERS) guidelines [16]. D_Lco was also assessed in accordance with ATS/ERS guidelines [17].

TLC was determined through body plethysmography based on two measurements [18]. There were daily calibrations of pressure, volume, and flow.

Interpretation of pulmonary function tests was done in accordance with the ATS/ERS guidelines [19]. Determination of FEV₁, FVC, TLC, and D_Lco below the lower limit of normal (<LLN) was based on a z-score of <−1.645, using the reference values from the Global Lung Function Initiative [20].

Other covariates

This study included information on age (categorized as <60 years or ≥60 years), sex, and smoking status (classified as never, ex, or current), body mass index (BMI) in kg/m² (categorized as underweight < 18.5 kg/m², normal 18.5–25 kg/m², overweight 25–30 kg/m², obese 30–35 kg/m², or obese ≥35 kg/m²). BMI was measured at the time of ICU hospitalization. Data on pre-existing comorbidities (lung disease, including asthma and chronic obstructive pulmonary disorder, hypertension, ischemic heart disease, cancer, diabetes mellitus type 2, and psychiatric illness) and prior treatment with anticoagulants or platelet inhibitors were collected from electronic records.

Data on ICU stay (presented as mean days or ≥10 days), mechanical ventilation, use of dexamethasone, and secondary infection (based on bacteria or fungal cultures from the nasopharynx, blood, urine, pleural effusion, or tracheal or wound secretions) were also collected from electronic records. The waves of COVID-19 were defined as wave 1, patients infected between 10 March 2020, and 31 July 2020, and wave 2, patients infected between 1 August 2020, and 28 February 2021. In addition, data on laboratory parameters, such as the maximum and minimum values of CRP in mg/L and the maximum and minimum values of leukocytes and lymphocytes in cells per microliter (10⁹/L), were extracted from electronic records. The severity of COVID-19 at onset was classified in accordance with the World Health Organization (WHO) Clinical Progression Scale (0–10, where 0 represents uninfected and 10 represents death) [21].

Statistical analysis

The basic characteristics of the study patients are presented as means with standard deviations (±SDs) for continuous variables or as numbers and percentages for categorical variables. Paired t-tests were performed to compare lung function variables between follow-up points. To identify associations between studied factors and impaired D_Lco, the chi-2 test or Fisher’s test (when the expected frequency in any contingency table cell was <5) with 95% confidence intervals (CI) was performed. Logistic regression, adjusted for age, age and sex, or age, sex, and BMI, was performed to evaluate factors associated with impaired D_Lco at 4 months and 1 year. This was presented as odds ratios (ORs) with 95% CI. All missing data was reported. A sensitivity analysis was performed to identify any differences between patients who attended follow-up and those who were lost to follow-up. A p-value < 0.05 was chosen to represent statistical significance. Statistical analysis was performed using STATA version SE 17 (StataCorp, College Station, TX, USA). Supplementary Figure S1 displays the trajectories of D_Lco over a 2-year period, generated using SankeyMATIC (https://sankeymatic.com).

Results

Baseline characteristics

This study initially included 104 patients assessed four months after ICU discharge. The flowchart of the study patients is shown in Figure 1, and patient characteristics are summarized in Table 1. The mean age of patients was 60 years (SD: ±12), with 32% being female. Most patients (73%) were never smokers. The mean BMI was 30.3 kg/m² (SD: ±5.6), with the majority classified as having a high BMI. The average ICU stay was 11.1 days (SD: ±3.7). Forty-seven percent of patients received dexamethasone during ICU care, while 15% were treated with anticoagulants and platelet inhibitors.

Table 1.

Baseline characteristics of the study patients (n = 104) at four months post-discharge from the intensive care unit (ICU) following critical COVID-19.

Baseline characteristics	N = 104
≥60 years old, n (%)	55 (52)
Female sex, n (%)	33 (32)
Smoking status, n (%)
Never	76 (73)
Current	5 (5)
Ex	23 (22)
BMI kg/m2 groups, n (%)
Underweight <18.5	0
Normal 18.5–25	14 (14)
Overweight 25–30	43 (43)
Obesity 30–35	24 (24)
Obesity ≥35	19 (19)
Comorbidities or treatment before COVID-19, n (%)
Asthma	16 (15)
Chronic obstructive pulmonary disorder	6 (6)
Other pulmonary disease	2 (2)
Hypertension	53 (51)
Ischemic heart disease	9 (9)
Cancer	7 (7)
Diabetes mellitus type 2	22 (21)
Psychiatric illness	5 (5)
Anticoagulants or platelet inhibitors	16 (15)
COVID-19 wave
1	61 (59)
2	43 (41)
Other ICU stay-related, n (%)
ICU stay ≥ 10 days	39 (8)
Mechanical ventilation	49 (48)
Treatment with dexamethasone	48 (47)
Secondary infection	46 (46)
WHO Clinical Progression Scale score, n (%)
6	53 (52)
8	40 (40)
9	8 (8)
Laboratory indices during ICU stay, mean (±SD)
Maximum CRP (mg/L)	246 (±124.9)
Minimum CRP (mg/L)	51.76 (±54.78)
Maximum leukocytes (109/L)	14 (±6.05)
Minimum leukocytes (109/L)	6.43 (±2.23)
Maximum lymphocytes (109/L)	1.8 (±0.87)
Minimum lymphocytes (109/L)	0.83 (±0.44)

Missing variables/abbreviations: BMI (Body mass index) in n=4, ICU stay in n=1, mechanical.

ventilation in n=1; WHO Progression Scale score in n= 3, secondary infection in n= 4.

maximum and minimum CRP (C-reactive protein) in n=6, maximum and minimum leukocytes

in n=6, and maximum and minimum lymphocytes in n=8.

Lung function

D_Lco < LLN was observed in 50%, 57.5%, and 33% of patients at 4 months, 1 year, and 2 years, respectively (Table 2). Figure 2 illustrates the trajectory of D_Lco % predicted over time. Supplementary Figure S1 compares the proportion of patients with normal D_Lco and D_Lco < LLN. Paired comparisons showed a significant decline in D_Lco between four months and one year, followed by significant improvement between one and two years (Table 2). There were 15% (n = 6) who showed improvement from abnormal to normal D_LCO, while 82% (n = 32) remained stable and 3% (n = 1) experienced deterioration from normal to abnormal D_LCO over the course of 4 months to 1 year. The average change in D_LCO z-score between 4 months and 1 year was −0.40 ± 0.65. There were 28 of 39 (72%) patients who showed decreased D_LCO, while 11 (28%) demonstrated improvement in D_LCO between 4 months and 1 year. No statistically significant differences were observed between the two groups (Supplementary Table S2).

Table 2.

Comparison of lung function indices between follow-ups after intensive care unit treatment for COVID-19 using paired t-tests, showing statistical significance.

Lung function indices	4 months vs 1 year, for n=39		1 year vs 2 years, for n=16		4 months vs 2 years, for n=21
	4 months	1 year	1 year	2 years	4 months	2 years
FVC, L	3.82 ± 1.03	3.93 ± 1.03	3.76 ± 0.92	3.91 ± 1.00	3.63 ± 0.90	3.91 ± 0.94
p-value	0.003		0.04		0.004
FVC, % predicted	90.37 ± 14.83	93.67 ± 14.75	96.23 ± 15.87	99.95 ± 17.07	88.37 ± 17.88	97.30 ± 17.37
p-value	0.0007		0.004		0.0002
FEV1, L	3.03 ± 0.81	3.10 ± 0.82	2.95 ± 0.70	2.98 ± 0.75	2.88 ± 0.67	2.99 ± 0.72
p-value	0.017		0.52		0.08
FEV1, % predicted	92.44 ± 14.07	95.40 ± 14.87	96.23 ± 15.88	99.18 ± 17.55	90.42 ± 16.77	96.38 ± 17.82
p-value	0.002		0.10		0.005
TLC, L	No data	5.39 ± 2.11	*5.84 ± 1.21	*5.95 ± 1.25	No data	5.89 ± 1.28
p-value	NA		0.004		NA
TLC, % predicted	No data	89.54 ± 13.87	*89.93 ± 9.82	*92.82 ± 11.29	No data	92.82 ± 11.84
p-value			0.006
DLco, mmol/min/kPa/L	6.74 ± 1.93	6.37 ± 1.85	5.88 ± 1.64	6.61 ± 1.90	6.26 ± 1.68	6.59 ± 1.87
p-value	0.0007		0.006		0.09
DLco % predicted	78.95 ± 14.73	74.33 ± 15.69	71.38 ± 15.10	81.11 ± 14.93	73.35 ± 13.84	78.91 ± 15.23
p-value	<0.0001		<0.0001		0.01

*For n=15.

Abbreviations: D_Lco: diffusing capacity for carbon monoxide, FVC: forced vital capacity, FEV₁: forced expiratory volume in one second, TLC: total lung capacity, NA: not applicable.

Figure 2.

Trajectory of diffusing capacity for carbon monoxide (D_Lco) % predicted for each study patient.

FVC % predicted showed significant improvement between four months and one year and between one and two years (Table 2). Regarding FEV1% predicted, there was a significant difference between four months and one year, and between four months and two years, but not between one and two years.

Among all patients, regardless of when they attended the follow-up at four months post-discharge, 13% had FEV₁ < LLN (n = 13) and 22% had FVC < LLN (n = 22). At one year, these proportions had decreased to 5% for both FEV₁ < LLN and FVC < LLN (n = 2), whereas at two years, they were 10% for both parameters (n = 2) (Supplementary Table S1).

Given the significant patient dropout across follow-ups, sensitivity analyses were performed to assess potential bias. No significant differences were found in mean % predicted D_Lco at four months and one year between patients who attended subsequent follow-ups and those who did not. Similar findings were observed between one and two years (Supplementary Table S3).

Factors affecting D_Lco

Impaired D_Lco was significantly associated with several factors throughout the follow-up period. The main factors are listed in Table 3, with additional factors detailed in Supplementary Table S4. At four months, impaired D_Lco was associated with higher age (≥60 years), mechanical ventilation, ICU stay ≥10 days, lower mean FVC, lower mean FEV₁% predicted, secondary infection, and minimum lymphocytes during the ICU stay, as well as higher WHO Clinical Progression Scale scores for COVID-19. At four months and one year, patients with normal D_Lco were shown to have significantly higher mean BMI than patients with impaired D_Lco. Among these, there were fewer subjects with a normal BMI and more subjects with a BMI over 30 kg/m², particularly those over 35 kg/m² (Supplementary Table S4). The trends in mean BMI persisted at the one-year follow-up. At one-year follow-up, impaired D_Lco remained associated with higher age (≥60 years), lower mean FVC% predicted, and higher BMI in those with normal D_Lco. At two years, impaired D_Lco was primarily associated with extended ICU stays (≥10 days), minimum leukocytes during ICU hospitalization, and lower mean FVC% % predicted. No significant differences were observed in BMI or age. ICU hospitalization during waves 1 and 2 of COVID-19 was not associated with the risk of abnormal D_Lco at four months. None of the patients with impaired D_Lco had a known, pre-existing lung disease. All patients who were followed up further were from wave 1.

Table 3.

Comparison of study patients with normal diffusing capacity for carbon monoxide (D_Lco) and D_Lco < lower limit of normal (LLN) at three follow-ups.

Patients’ characteristics	4 months, n=102			1 year, n=40			2 years, n=21
	Normal DLco N=51	DLco < LLN N=51	p-value (95% CI)	Normal DLco N=17	DLco < LLN N=23	p-value (95% CI)	Normal DLco N=14	DLco < LLN N=7	p-value (95% CI)
Age in years, mean ± SD	56.69 ± 10.20	62.98 ± 12.98	0.008	57.47 ± 9.47	63.13 ± 11.35	0.10 (−12.52, 1.20)	61.79 ± 8.94	59.71 ± 17.24	0.71 (−9.74, 13.88)
Age groups <60 years ≥60 years	36 (71) 15 (30)	14 (27) 37 (73)	<0.001 (−0.61,-0.26)	12 (71) 5 (29)	9 (39) 14 (60)	0.051 (−0.63,0.01)	6 (43) 8 (57)	2 (29) 5 (71)	0.55 (0–0.63,0.35)
Sex, n (%) Male Female	32 (63) 19 (37)	38 (75) 13 (25)	0.20 (−0.07, 0.34)	14 (82) 3 (18)	14 (61) 9 (39	0.18 (−56, 0.96)	11 (75) 3 (25)	5 (78) 2 (22)	0.71 (−0.57, 0.40)
BMI kg/m2, mean ± SD	31.72 ± 5.90	28.92 ± 5.01	0.013 (10.88, 1.71)	31.75 ± 4.86	28.37 ± 4.68	0.039 (0.17, 6.58)	30.53 ± 6.38	29.99 ± 2.97	0.56 (−3.86, 6.95)
Mechanical ventilation, n (%)	16 (31)	33 (65)	<0.001 (−0.53, −0.16)	8 (47)	13 (57)	0.55 (−0.40, 0.21)	6 (43)	8 (88)	0.19 (−0.65. 0.08)
≥10 days in ICU, n (%)	11 (22)	28 (55)	0.001 (−0.54, −0.17)	6 (35)	12 (52)	0.29 (−0.47, 0.14)	6 (42)	7 (89)	0.01 (−0.81, −0.27)
Preexisting pulmonary disease, n (%)	13 (25)	10 (20)	0.48 (−0.15, 0.31)	4 (24)	7 (30)	0.73 (−0.42, 0,25)	5 (42)	2 (22)	0.74 (−0.35, 0.49)
FVC, % predicted, mean ± SD	94.34 ± 11.87	82.93 ± 53.53	<0.001 (5.87, 16.95)	99.36 ± 15.54	89.66 ± 12.62	0.036 (0.69, 18.72)	102.91 ± 14.06	86.08 ± 18.90	0.03 (1.58, 32.09)
FEV1 % predicted, mean ± SD	102.43 ± 17.79	90.89 ± 16.75	0.001 (3.73, 14.39)	98.27 ± 15.30	94.32 ± 15.15	0.42 (−5.91, 13.80)	101.53 ± 14.88	86.09 ± 19.84	0.06 (−0.66, 31.53)
TLC % predicted, mean ± SD	No data	No data	No data	94.05 ± 17.06	85.98 ±9.35	0.09 (17.84, 29.08)	96.79 ± 11.65	82.89 ± 4.13	1.00 (−0.36, 0.36)
Secondary infection, n (%)	16 (31)	28 (55)	0.016 (0.05, 0.43)	9 (53)	11 (48)	0.75 (−0.36, 0.26)	7 (50)	5 (72)	0.35 (−0.19,0.58)
Minimum leukocytes, mean ± SD	6.59 ± 2.17	6.20 ± 2.25	0.39 (−0.51, 1.29)	6.19 ± 2.29	6.01 ± 2.35	0.81 (−1.35, 1.72)	6.79 ± 1.95	4.67 ± 1.62	0.02 (0.32, 3.92)
Minimum lymphocytes, mean ± SD	0.91 ± 0.45	0.65 ± 0.31	<0.01 (0.11, 0.42)	0.98 ± 0.62	0.69 ± 0.39	0.09 (−0.04, 0.62)	0.88 ± 0.27	0.73 ± 0.31	0.30 (−0.14, 0.44)
Treatment with dexamethasone during ICU, n (%)	22 (43)	26 (51)	0.43 (−0.27, 0.12)	3 (18)	2 (9)	0.40 (−0.26, 0.66)	3 (21)	1 (13)	0.69 (−0.38, 0.58)
COVID-19 wave, n (%) 1 2	32 (63) 19 (37)	28 (55) 23 (45)	0.42 (−0.12, 0.28)	17 (100) 0	23 (100) 0	NA	14 (100) 0	7 (100) 0	NA

Abbreviations:BMI: body mass index, ICU: intensive care unit. FVC: forced vital capacity, FEV1: forced expiratory volume in one second, TLC: total lung capacity, NA: not applicable.

For the 4-month follow-up (n = 102), we performed logistic regression to assess the association between D_Lco < LLN and relevant patient characteristics, adjusting for age, sex, and BMI (Table 4). This analysis showed that older age (≥60 years), mechanical ventilation, and an ICU stay of ≥10 days were strongly associated with higher odds of impaired D_Lco. Additionally, lower FVC and FEV₁ % predicted at 4-months’ follow-up, as well as lower minimum lymphocyte counts during the ICU visit, were independently associated with higher odds of impaired D_Lco. The use of dexamethasone during ICU did not affect D_LCO. For the 1-year follow-up (n = 40), D_Lco < LLN was significantly associated only with lower FVC % predicted at 1 year (Supplementary Table 5). The analysis was not conducted for a two-year follow-up due to a low sample size (n = 21).

Table 4.

Association between diffusing capacity for carbon monoxide (D_Lco) < lower limit of normal (LLN) and other variables at 4 months follow-up. Logistic regression analysis presented as odds ratio (OR) with 95% confidence intervals (CI). Age and BMI as continuous variables in models.

Patients’ characteristics	OR (95% CI) Adjusted for age	OR (95% CI) Adjusted for age, sex	OR (95% CI) Adjusted for age, sex, BMI
Age (≥60 years)	6.34 (2.68–15.00)	6.34 (2.66–15.10)	6.73 (2.64–17.12)
Age (per 1 year)	1.05 (1.01–1.09)	1.05 (1.01–1.09)	1.05 (1.01–1.10)
Sex(female vs male)	0.55 (0.23–1.33)	0.55 (0.23–1.33)	0.51 (0.20–1.30)
BMI kg/m2 (continues)	0.93 (0.85–1.01)	0.93 (0.85- 1.01)	0.93 (0.85- 1.01)
Mechanical ventilation	4.15 (1.76–9.80)	4.03 (1.70–9.57)	4.74 (1.82–12.34)
≥10 days in ICU	4.64 (1.88–11.45)	4.51 (1.82–11.20)	5.84 (2.13–16.01)
Preexisting pulmonary disease	0.65 (0.24–1.72)	0.73 (0.27–1.97)	0.72 (0.25–2.02)
FVC (per 1% predicted)	0.93 (0.90–0.96)	0.93 (0.90–0.96)	0.93 (0.89–0.96)
FEV1 (per 1% predicted)	0.93 (0.90–0.97)	0.93 (0.90–0.97)	0.93 (0.89–0.97)
Secondary infection	0.41 (0.18–0.95)	0.42 (0.18–0.98)	0.41 (0.16–1.02)
Minimum leukocytes (per 109/L)	0.92 (0.76–1.11)	0.92 (0.76–1.11)	0.93 (0.76–1.14)
Minimum lymphocytes (per 109/L)	0.26 (0.05–1.41)	0.28 (0.05–1.56)	0.18 (0.02–1.49)
Treatment with dexamethasone during ICU	0.44 (0.06–3.11)	0.36 (0.05–2.88)	0.62 (0.07–5.38)
COVID-19 wave (1 vs 2)	0.78 (0.35–1.78)	0.75 (0.33–1.73)	0.74 (0.31–1.80)

Abbreviations: BMI: body mass index, ICU: intensive care unit. FVC: forced vital capacity, FEV1: forced expiratory volume in one second.

Discussion

This study presents a two-year follow-up evaluation of lung function in critically ill COVID-19 patients hospitalized in the ICU of a university hospital. Key findings include that impaired D_Lco was observed in 50%, 58%, and 33% of the examined patients at four months, one and two years post-discharge, respectively. In paired comparisons, D_Lco (expressed as % predicted) declined significantly between four months and one year, and thereafter improved significantly between one and two years. FVC (% predicted) improved significantly at each time point, whereas FEV₁ (% predicted) showed significant improvement between four months and one year, as well as between four months and two years, but not between one and two years. TLC (% predicted) improved significantly between one and two years. At four months, D_Lco < LLN was significantly associated with older age (≥60 years), mechanical ventilation, and an ICU stay of ≥10 days, as well as lower FVC and FEV₁ % predicted at four months, and minimum lymphocyte counts during ICU.

Six previously published longitudinal studies assessed lung function in hospitalized COVID-19 survivors at three time points over two years [7–11,22], whereas two additional studies included three or four follow-ups over three years [23,24]. Across all these studies, D_Lco was consistently the most affected lung function parameter, though it improved between the first follow-up and the follow-up after two years [7–10,23,24]. For example, in a Spanish study that also focused on critically ill COVID-19 patients, 46% of patients had impaired D_Lco at 24 months [10], a somewhat higher proportion than in our study (33%). In contrast to our study, which showed a decline in D_Lco between four months and 1 year, followed by improvement between one and two years, the Spanish study on ICU survivors [10] and two three-year studies from China demonstrated a gradual recovery of D_Lco at each follow-up [23,24], with lung function eventually becoming comparable to that in matched community controls without SARS-CoV-2 infection [23]. Two other studies revealed differing trends. A study from China observed an improvement in D_Lco between six months and one year, followed by a decline between one and two years [11]. A study from Brazil reported no significant changes in D_Lco over two years [22].

The reduction in D_Lco among ICU survivors likely reflects persistent microvascular and interstitial abnormalities after severe lung injury. Severe COVID-19 is characterized by a dysregulated host inflammatory response with endothelial injury, and coagulopathy, which provides a biological basis for impaired gas transfer [25,26]. Consistently, follow-up HRCT studies report that lower D_Lco correlates with residual ground-glass opacities, reticulation, and fibrotic-like changes, in line with diffuse alveolar damage and post-inflammatory remodeling [27]. These findings suggest that impaired gas transfer after critical COVID-19 May primarily result from fibrotic remodeling and persistent endothelial dysfunction.

Like this study, previous studies [7–10,22] have demonstrated progressive improvement in mean FEV₁, FVC, and TLC % predicted over two years. Additionally, two studies that followed discharged patients over three years reported gradual recovery in spirometry and lung volumes [23,24]. However, a study from China observed an initial increase in FVC between six months and one year, followed by a decline at a two-year follow-up [11].

In line with our findings, older age, mechanical ventilation, prolonged ICU stay, and greater severity of COVID-19 at onset have been found to be linked to impaired DLco [8,10,22,28], along with concurrent impairments in other lung function parameters [10]. Other studies have highlighted additional contributing factors, such as fibrotic interstitial abnormalities o computed tomography chest scans [7], pre-existing type 2 diabetes [24], and post-acute COVID-19 syndrome [23].

We showed that abnormal DL_CO after two years was significantly associated with lower minimum leukocyte levels in the ICU. Although acute COVID-19 infection generally causes leukocytosis, severe infections, as seen in our study patients, could result in leukocytopenia due to immune system exhaustion [29]. In addition, abnormal DL_CO at four months was significantly associated with lower minimum lymphocyte levels. Both lung involvement and lymphocytopenia are independent early indicators of more severe COVID-19 [30]. We did not establish a causal relationship between these factors, which most likely share common risk factors, including increasing age, male sex, and comorbidities [31]. One of the underlying mechanisms of lymphocytopenia is immune system dysregulation, which may lead to secondary infections [32]. A previous study showed that need of ICU stay was linked to a higher risk of secondary infections [33]. In our study, we demonstrated that secondary infections during ICU hospitalization were significantly correlated with impaired DL_CO at four months. This may suggest a link beyond that of more severe COVID-19 in itself. However, secondary infections did not remain a significant risk factor for impaired DL_CO at the two-year follow-up.

Notably, we showed that patients with impaired DLco had a lower obesity prevalence (BMI > 30 kg/m²). Previous studies have shown conflicting results, with some indicating a decrease of DLco in obese individuals [34], others an increase [35], and some showing no association [36] between obesity and DLco. However, a meta-analysis indicated that obesity was associated with a higher rate of hospitalization and ICU admission [37]. Different mechanisms may contribute to severe COVID-19 in individuals with lung function impairment and in those with higher BMI. For example, the need for high levels of care in individuals with obesity might be due to impaired ventilation caused by increased body mass or comorbidities related to obesity. In contrast, individuals with normal BMI may experience higher inflammatory engagement in the lungs, explaining the more impaired lung function over time.

Strengths and limitations

The main strength of our study is its longitudinal measurement of lung function at three time points over two years. Additionally, clinical data from the acute onset of critical COVID-19 provide a contrast to registry-based studies and studies of hospitalized patients with mixed disease severity. This enables a detailed description of lung function trajectories, specifically in patients with critical COVID-19. Only one previous study, like ours, has followed up on lung function in critically ill COVID-19 patients [10].

The main limitation of this study is the significant loss to follow-up, with nearly 80% of patients not attending the two-year follow-up. In addition, no patients from wave 2 were followed up at one and two years, which introduces nonresponse bias and limits the generalizability of our findings. However, sensitivity analyses revealed no significant difference in mean D_Lco% predicted at four months and one year between participants who continued and those who dropped out. This suggests that the results are not skewed by including solely the patients with the most severe or the mildest impairments. Other long-term follow-up studies on COVID-19 survivors have also reported substantial dropout rates.

Another limitation is that the study was conducted at a single university hospital, which may reduce its external validity. Additionally, not all examinations were available at all three timepoints – for instance, lung volume measurements were not performed at four months. Moreover, the absence of unexposed control data limits our ability to precisely quantify the impact of COVID-19. However, this limitation is common in nearly all respiratory studies examining the effects of COVID-19.

Pre-COVID-19 lung function data for the study patients were not available, which means that we cannot determine whether lung function was already reduced before infection. We did not have radiological findings or measured clinical consequences of abnormal D_Lco, such as dyspnea at follow-up, which additionally limits the assessment of long-term functional outcomes.

Furthermore, this study included patients from both the first and second waves of the pandemic at a four-month follow-up. As most study participants contracted COVID-19 before vaccines were widely available to the general population, we were unable to study the effects of vaccination.

Among our patients, approximately half received dexamethasone. At the 4-month follow-up, D_LCO did not differ significantly between the treated and untreated groups in either the descriptive analysis or the adjusted regression analysis. However, only a small subset was assessed at 1- and 2-year follow-up; therefore, long-term lung function outcomes – and comparisons between dexamethasone-treated and untreated patients – cannot be interpreted reliably. Furthermore, antiviral therapy was not part of routine care during the first wave, and only one of our study patients from that period received it. In contrast, almost all patients in the second wave were treated with antiviral agents. However, we observed no significant differences in long-term pulmonary outcomes between the first and subsequent waves, suggesting that unmeasured antiviral use is unlikely to have affected our main conclusions.

Additionally, we lacked data from computed tomography chest scans and information on patients’ self-reported persistent symptoms, quality of life post-infection, and whether impaired lung function was associated with these factors. Finally, since only survivors were assessed at each visit, patients who died earlier weren’t captured, risking immortal time. The absence of data on other reasons for non-participation further limits the interpretation of the follow-up results.

Conclusion

Despite some recovery between one and two years, one-third of patients still exhibited impaired D_Lco two years after critical COVID-19. These findings highlight the prolonged impact on lung function of critical COVID-19, though caution is warranted due to the progressively smaller sample size at each follow-up in this study. The future direction of this study will be a five-year follow-up to assess continued lung function trajectory, as well as to obtain registry data on mortality, allowing us to explore long-term survival and its association with post-COVID lung impairment.

Funding Statement

The study was funded by the SciLifeLab/Knut and Alice Wallenberg national COVID-19 research program [KAW 2020.0182, KAW 2020.0241], the Swedish Heart-Lung Foundation [20210089, 20190639, 20190637, 20230627, 20230732], the Swedish Research Council [2014–02569, 2014–07606], Swedish Society of Medicine (SLS-938101), and the Swedish Kidney Foundation [F2020-0054], Åke Wiberg Stiftelsen [stipendium M23-0133, M22-0119], Lars Hiertas Minne Stiftelsen [stipendium F02022-0098]. Funders had no role in the design of the study, data collection, interpretation, or in the writing of the manuscript.

Author contributions

All authors contributed to the study concept and design. All authors contributed to patient inclusion and data collection. M.A.K. and E.E. analyzed data and wrote the first draft. All authors edited the paper for intellectual content and approved the final version for publication.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

Data are available through the SciLifeLab data repository, after securing ethical permission and obtaining the appropriate data access agreements (https://doi.org/10.17044/scilifelab.14229410).

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/20018525.2026.2618849