The current manuscript is based on data from the CIBERESUCICOVID study,13 which is an observational, pragmatic, multicenter, ambispective study including critically ill COVID-19 patients admitted to the ICUs of 55 Spanish hospitals. CIBERESUCICOVID was registered in ClinicalTrials.gov with the identifier NCT04457505. The study collected retrospective data from patients admitted to participating ICUs before May 2020 and prospective data from then onward. CIBERESUCICOVID included a follow-up within the first year after hospital discharge of the maximum number of patients that the pandemic situation allowed in each participating hospital, without a specific protocol and regardless of whether the patients presented symptoms or not.

The data for the current analyses correspond to consecutive COVID-19 patients admitted to 52 Spanish ICUs from March 2020 to August 2021. All included patients had a confirmed COVID-19 diagnosis (positive nasopharyngeal swab polymerase chain reaction (PCR) test for SARS-CoV-2) and were admitted to the ICU. Patients not surviving the hospital stay or patients transferred to other hospitals during or after ICU admission were not considered eligible. Patients lacking a follow-up visit with lung function test after discharge was excluded from the analyses. Additionally, patients receiving palliative care or with severe mental disability precluding pulmonary function tests after discharge were also excluded. The study flow chart is shown in Fig. 1.

Fig. 1.

Flowchart of the study.

(0.14MB).

Additionally, an external cohort consisting of 200 critically ill COVID-19 patients participating in the Post-COVID study5 held at the University Hospital Arnau de Vilanova and Santa Maria in Lleida, Spain, was used as a validation cohort.

Sociodemographic, anthropometric, comorbidity and lifestyle variables were collected using a large predetermined checklist at hospital admission (see Comorbidity checklist in the online supplement). Detailed information from the ICU stay included arterial blood gas test and complete blood test (at ICU admission and through ICU stay), including estimated glomerular filtration rate (EGFR) by means of the 2021 Chronic Kidney Disease Epidemiology Collaboration creatinine equation14; medical procedures before and during ICU stay, including ventilatory support; pharmacological treatment; and in-hospital complications such as ARDS, infections, thrombotic events or acute organ failure. Sequelae were objectively assessed at a follow-up visit by means of a thoracic CT scan (persistent infiltrates, emphysema and fibrotic lesions) and a lung function test (forced expiratory volume in 1 second (FEV1), forced vital capacity (FVC), and DLCO).

The primary outcome of this study was DLCO as measured at a follow-up visit. Secondary outcomes were also assessed at the same visit and included CT scan findings and other parameters of the lung function test (FEV1 and FVC).

The participating hospitals obtained ethical approval from the corresponding governing board. Study number: HCB/2020/0370; date of approval: 14/05/2020; original project title: “Factores de riesgo, pronóstico personalizados y seguimiento a un año de los enfermos ingresados en las unidades de Cuidados Intensivos Españolas infectados por el virus COVID19: CIBERESUCICOVID” (Risk factors, personalized prognosis and one-year follow-up of patients admitted to Spanish Intensive Care units infected with the COVID19 virus: CIBERESUCICOVID); governing board granting approval: Clinical Research Ethics Committee of Hospital Clínic de Barcelona. Participants or their relatives provided informed consent when possible or, when unfeasible, an informed consent waiver was authorized by the ethics board. Procedures were followed in accordance with the ethical standards of the Clinical Research Ethics Committee of Hospital Clínic de Barcelona and with the Helsinki Declaration of 1975 and its most recent amendments. Data were pseudonymized and stored in a REDCap database hosted in the Centro de Investigación Biomédica en Red (CIBER) premises in Madrid, Spain. The study complied with national and international law on data protection.

Similarly, the Post-COVID study, used as a validation cohort, was approved by the Clinical Research Ethics Committee of the University Hospital Arnau de Vilanova and Santa Maria (ref.: CEIC/2273; Title: “Factores de riesgo, pronósticos personalizados y seguimiento a un año de los enfermos ingresados en las unidades de Cuidados Intensivos Españolas infectados por el virus COVID19: ESTUDIO CIBERESUCICOVID” (Risk factors, personalized prognoses and one-year follow-up of patients admitted to Spanish Intensive Care units infected with the COVID19 virus: CIBERESUCICOVID STUDY); date: 02/06/2020); which was conducted in accordance with the ethical standards of the Clinical Research Ethics Committee of Hospital Arnau de Vilanova and Santa Maria and with the Helsinki Declaration of 1975 and its most recent amendments, and complied with national and international law on data protection.

Descriptive statistics were used to summarize the characteristics of the study population. Absolute and relative frequencies were used for qualitative data. The means (SD) or medians (25–75th percentile) were estimated for quantitative variables with normal and non-normal distributions, respectively. Normal distributions were assessed by the Shapiro–Wilk test.

Clinical data during the hospital stay were compared between surviving patients with and without follow-up using a t test (or Wilcoxon signed-rank test for variables with nonnormal distribution) for continuous variables and a chi-squared test (or Fisher exact test when the expected frequencies were less than 5 in some cells) for qualitative variables.

Hospital factors were compared to DLCO severity (categorized as: DLCO<80%, 60%<DLCO<80% and DLCO<60%) using Mantel–Haenszel test of trend for categorical factors and Pearson test (or Spearman test in non-normal distribution) for continuous variables.

The missingness mechanism was assumed to be missing at random (MAR). In multivariable analyses, missing values were handled with multiple imputation by chained equations (MICE).15 Primary and secondary outcomes were included in the imputation models but were not imputed. The MICE procedure created 10 complete datasets. A minimum threshold of absolute correlation of 0.15 was used to select predictors in the imputation models. The predictors included in final multivariable model for each outcomes were selected using multiple imputation grouped adaptive least absolute shrinkage and selection operator (LASSO).16 The lambda value was set to the sparsest model within one standard error of the minimum 5-fold cross-validation error. Finally, final multivariable models were based on a logistic model (or linear model for continuous outcomes) for each outcome with variables selected (in LASSO regression) as predictors. The results across the multiply imputed datasets were combined using Rubin's rules.17 Additionally, a sensitivity analysis was carried out by fitting the final multivariable models in the population of complete cases.

The individual association between the primary outcome and the selected important variables was represented by violin plots or bar charts for dichotomous outcomes, and representations of generalized additive models for continuous outcomes.

Odds ratios were estimated to assess the direction and magnitude of the associations between the selected factors and the primary outcome in an independent cohort. This was used as a validation of the main results.

R statistical software, version 4.0.1 (R Project for Statistical Computing), was used for all analyses.

From a total of 3860 severe COVID-19 patients discharged from participating hospitals, 1327 had DLCO measured in a follow-up visit after discharge and were included in the current analyses (Fig. 1). A comparison of patients with and without follow-up visits is shown in the online supplement (eTable 1), showing that both groups of patients were similar and had no striking or clinically relevant differences a part from the higher use of hydroxychloroquine among the included patients. The included patients had a median [p25; p75] age of 60 [53; 67] years, 27.8% were women, and 59.9% were never smokers. The most common comorbidities were hypertension (44.8%), obesity (35.8%) and diabetes mellitus (17.6%). The median hospital stay was 27 [17; 43] days. The median [p25; p75] time from hospital discharge to the first follow-up visit was 3.57 [2.77; 4.92] months (eFigure 1).

The mean (SD) percentage of predicted diffusion capacity at the follow-up visit was 72.02 (18.33)% predicted. A total of 877 (66%) patients showed an impairment of diffusion capacity (DLCO<80%), and 318 (24%) had moderate to severe impairment (DLCO<60%).

A univariate analysis was performed to evaluate the dose–response association between hospital factors and DLCO impairment using p for trend. Briefly, age, comorbidities (except obesity), duration of IMV, and most hospital complications increased according to the level of DLCO impairment (Tables 1 and 2).

The multivariate model, which included the most relevant variables associated with DLCO, was based on LASSO models. The most relevant variables associated with DLCO<80% in the multivariate analyses were age (OR[per 1 SD] (95% CI): 1.23 (1.07–1.41)), female sex (1.89 (1.42–2.51)), current smoking (2.19 (1.11–4.32)), duration of IMV (1.57 (1.30–1.89)), duration of NIMV (1.25 (1.05–1.49)), EGFR at ICU admission (0.75 (0.67–0.88)) and hospital infectious complications (1.27 (0.94–1.70)) (Fig. 2, Table 3, and eFig. 2). Regarding moderate/severe impairment diffusion capacity (DLCO<60%), the multivariate model included age (1.39 (1.18–1.63)), chronic lung disease (CLD) (1.86 (1.18–2.92)), duration of IMV (1.56 (1.37–1.77)), urea at ICU admission (1.16 (0.97–1.39)) and EGFR at ICU admission (0.88 (0.73–1.06)) (Fig. 3, Table 3, and eFig. 2). Furthermore, age and duration of IMV and EGFR at ICU admission showed a linear dose–response association with DLCO in a multivariate linear regression model (eFigs. 2 and 3, and Table 3). Most of these results were validated using an external cohort, with the exception of female sex, which had a different prognostic value in each of the two cohorts (eTable 2). Similarly, the inclusion of time since hospital discharge as a confounder in sensitivity analyses of the associations between clinical parameters at hospitalization and lung diffusion capacity impairment, CT scan findings and spirometry parameters at the follow-up visit did not affect the magnitude of the associations (eTable 3).

Fig. 2.

Hospital factors related to diffusion capacity impairment (DLCO<80%) at the follow-up visit. Logistic LASSO regression. Abbreviations: DLCO: lung diffusing capacity; NIMV, non-invasive mechanic ventilation; IMV, invasive mechanic ventilation; EGFR, Estimated Glomerular Filtration Rate; LASSO, least absolute shrinkage and selection operator.

(0.13MB).

Fig. 3.

Hospital factors related to diffusion capacity impairment (DLCO<60%) at the follow-up visit. Logistic LASSO regression. Abbreviations: DLCO: lung diffusing capacity; IMV, invasive mechanic ventilation; EGFR, Estimated Glomerular Filtration Rate; LASSO, least absolute shrinkage and selection operator.

(0.11MB).

The CT scan exploration at the follow-up visit showed prevalent lung damage in the cohort. The prevalence of the assessed CT scan abnormalities was as follows: persistent pulmonary infiltrate (n: 322 (32.59%) patients), fibrotic lesions (244 (24.69%)), and emphysema (56 (5.66%)).

Table 4 shows, on the one hand, the key variables associated with CT findings at the follow-up visit. Briefly, the duration of ventilatory support (NIMV (1.23 (1.06–1.42)) and IMV (1.21 (1.01–1.45)), bacterial pneumonia (1.62 (1.11–2.35)) and the duration of prone positioning (1.17 (0.98–1.39)) were associated with fibrotic lesions. Acute respiratory distress syndrome (OR: 1.93 (1.33–2.79)), need for a prone position (OR: 1.98 (1.44–2.71)) and antiviral treatment (1.66 (1.14–2.40)), APACHE score (1.13 (0.95–1.33)), neutrophil (1.13 (0.98–1.30)) and platelet count at ICU admission (1.17 (1.02–1.35)), and the partial pressure of carbon dioxide (1.09 (0.94–1.26)) were associated with pulmonary infiltrates. Finally, emphysema was only determined by smoking status (former: 7.04 (3.56–13.92); current: 3.20 (0.67–15.11)). On the other hand, Table 4 shows the key variables associated with lung function parameters (FEV1 and FVC) measured at the follow-up visit. In this regard, the combination of baseline chronic lung or renal disease, renal function at ICU admission and the duration of ventilatory procedures were the main determinants of FEV1 and FVC.