Imaging, Pulmonary Function, and Histopathologic Findings of Persistent Fibrosis in a Longitudinal Cohort 3-Years after COVID-19

Abstract

Rationale:

Survivors of severe COVID-19 frequently have persistent radiologic abnormalities beyond one year. Scant data exists for long-term outcomes of COVID-19.

Objectives:

To characterize a longitudinal multiethnic cohort of COVID-19 survivors 3-years after infection, to identify clinical factors associated with post-COVID-19 fibrotic-like abnormalities, to describe changes in radiologic abnormalities at 4-months, 15-months, and 3-years, and to describe histopathological features of lung parenchyma from participants with fibrotic-like abnormalities at 3-years.

Methods:

102 survivors of severe or critical COVID-19 (50% mechanically ventilated, all required oxygen supplementation) from a single-center, prospective, longitudinal, multi-ethnic cohort completed inspiratory and expiratory high resolution chest imaging, pulmonary function testing, and physical performance testing 3-years after hospitalization. Over 70% participated in earlier follow-up visits at 4-months and/or 15-months. Factors associated with persistent fibrotic-like abnormalities were examined using multivariable logistic regression with covariate balanced propensity scores to estimate adjusted associations. For subjects with more than one imaging study, changes in ground glass opacities, reticulations, and traction bronchiectasis were semi-quantitatively analyzed and qualitatively assessed. Five participants with post-COVID-19 fibrosis scores in the top quartile underwent transbronchial biopsy for histopathologic analysis.

Results:

Fibrotic-like abnormalities, including reticulations and traction bronchiectasis, were present in 61% of survivors of severe or critical COVID-19. In adjusted analyses, fibrotic-like abnormalities were positively associated with male sex, lower BMI, shorter leukocyte telomere length, increased severity of illness, and mechanical ventilation, and negatively associated with Black race. Participants with fibrotic-like abnormalities were more likely to have reduced diffusion capacity and reduced 6-minute walk distance. Reticulations, as assessed by semi-quantitative analysis, modestly improved across all time points, even between 15-months and 3-years. Qualitatively, most participants had stable fibrotic-like abnormalities across all time points with 9% improving from 15-months to 3-years and none worsening. Lung parenchyma from transbronchial biopsies of five individuals with elevated fibrotic scores showed small airways histopathology, consistent with air trapping during expiration, and infrequent interstitial thickening and fibrosis.

Conclusions:

Despite modest improvements in radiologic fibrotic-like abnormalities 3-years after hospitalization, their continued presence and their association with reduced diffusion capacity and reduced walk distance highlight the long-term consequences of severe COVID-19, which may require further monitoring.

INTRODUCTION

Survivors of COVID-19 can have persistent respiratory symptoms and radiologic abnormalities years after initial infection. Large cohort studies from the UK(1) and China(2) have reported that 11 to 36% have residual lung abnormalities following COVID-19. Radiologic imaging typically shows attenuation of inflammatory findings, including ground glass and consolidation, during the first year. However, fibrotic-like abnormalities of reticulation and traction bronchiectasis can persist beyond one year, especially for survivors of severe or critical COVID-19(2, 3). These residual fibrotic changes are associated with restriction and reduced diffusion capacity(2, 3).

Most patients experience gradual improvement in exercise tolerance and functional status after initial COVID-infection despite the persistence of fibrotic-like changes and spirometric abnormalities(2, 4). Despite these improvements, survivors of COVID-19 frequently report symptoms of fatigue and dyspnea(5). Few studies have identified clinical factors associated with persistent fibrotic abnormalities after 3 years, few have examined longitudinal radiologic changes over time, and few have sought to examine lung histopathology of post-COVID-19 fibrotic-like changes.

To address these knowledge gaps, we have conducted a longitudinal multi-ethnic cohort study of 102 severe COVID-19 survivors from New York City who underwent a battery of radiologic, physiologic, functional and symptomatic assessments at 4-months, 15-months, and 3-years. We sought to identify factors associated with persistent fibrotic abnormalities, to identify longitudinal radiologic changes over time, and to describe the histopathological analysis of transbronchial lung biopsies for participants within the highest quartile of post-COVID-19 fibrotic-like changes.

METHODS

Study Design and Participants

We conducted a single-center prospective study of adults aged 21 years and older hospitalized between March 1, 2020 and May 15, 2020 with positive SARS-CoV-2 RT-PCR nasopharyngeal swab. All patients required oxygen during their hospitalization, fulfilling criteria for severe disease(6) or WHO severity scores(2) of 4 (hospitalized patients requiring supplemental oxygen via nasal cannula or mask), 5 (hospitalized patients requiring high-flow nasal cannula or non-invasive mechanical ventilation), or 6 or 7 (hospitalized patients requiring mechanical ventilation, extracorporeal membrane oxygenation (ECMO), including those who also required vasopressors and renal replacement therapy). Patients were included if they were living independently in the community prior to hospitalization. Patients with a history of interstitial lung disease or lung transplantation were excluded. The study was approved by the Columbia University Irving Medical Center (CUIMC) Institutional Review Board (AAAT5605). Patients signed written consent forms; Spanish language forms and interpreter services were available.

We contacted and enrolled prospective participants meeting eligibility criteria based on their admission date by telephone, with sampling weighted to include approximately 50% survivors with WHO severity grade severity grade 6 or 7 (required mechanically ventilation). Individuals who had previously participated in either the 4- and 15-month post-COVID research study were first contacted for enrollment. Additional individuals were enrolled using the same inclusion criteria to increase the study number to ~100 participants (Figure S1). Patients with the highest quartile of fibrosis scores (described below) were invited to undergo transbronchial biopsies ~4-years after initial infection.

Electronic Medical Record Measurements

Clinical data was extracted from the electronic medical record. Chest imaging prior to COVID-19 hospitalization, which included CT chest scans and Chest radiographs for 19 (19%) and 49 (48%) participants, respectively, was reviewed to ensure that none of the participants had pre-existing interstitial lung disease (ILD). Patient data included admission vital signs, ventilator setting and flowsheet data, laboratory tests, procedures, administered medications, and procedures. Clinical and laboratory values from the first 24 hours of admission were used to calculate a sequential organ failure assessment (SOFA) score(7).

Chest Computed Tomography

All participants underwent a non-contrast high resolution CT scan at full inspiration and on exhalation using a GE-Revolution 256 ES CT Scanner. CT scans were analyzed with two thoracic radiologists (MMS, BMD) using a semi-quantitative method validated by ARDSnet(8) and previously utilized for the 4-month and 15-month post-COVID studies(4, 9). All follow-up scans for participants were examined by two thoracic radiologists together and consensus scores were generated. Images were analyzed for the presence of reticulation, honeycombing, ground glass opacities, traction bronchiectasis, diffuse centrilobular nodules, non-emphysematous cysts, and intraparenchymal opacities. Abnormalities affecting >5% of a lung slice quadrant were scored; those affecting <5% of a lung slice quadrant were not scored. Traction bronchiectasis, reticulations, and honeycombing were categorized as fibrotic; other abnormalities were categorized as non-fibrotic. For patients with serial scans, radiologists subjectively scored the series as demonstrating progression, regression, or stability. The radiologists noted the presence or absence of air trapping after review of both the inspiratory and expiratory scans.

Clinical Measurements

Pulmonary function testing with spirometry and DLCO measurements were performed using an EasyOne spirometer and interpreted based on established guidelines.(10) Cough and dyspnea were assessed utilizing the 100 mm cough visual analogue scale, modified medical research council dyspnea score and the UCSD shortness of breath questionnaire. Six-minute walk distance was obtained and analyzed utilizing establish guidelines.(11) A Columbia University Biobank post-COVID symptom survey was used to assess persistent post-COVID symptoms. Comorbid conditions were assessed by questionnaire at the study visit.

Frailty Measurements

At each follow up visit we prospectively assessed the five Fried frailty domains: gait-speed using a 4.57m walk test(12), grip-strength using a hand dynameter(12), weight loss, low physical activity using the Duke Activity Status Index Score(13, 14), and exhaustion using a validated questionnaire(12). Weight loss was defined as >10lb. decrease from hospitalization admission weight to follow-up weight. Participants were assigned a point for each frailty domain they met. Consistent with Cardiovascular Health Study methodology(12), patients were categorized as prefrail if 1 or 2 domains were abnormal and were categorized as frail if ≥3 domains were abnormal.

Bronchoscopy with Transbronchial Lung Biopsy

Participants who underwent bronchoscopy with transbronchial lung biopsy were specifically consented for this procedure. Patients within the highest quartile of reticulation (>4.2) and traction bronchiectasis (>1.0) scores who did not have a safety contraindication (defined as heart failure or reduced RV systolic function on echocardiogram, need for hemodialysis, age ≥70) were invited to undergo bronchoscopic lung biopsy under conscious sedation using fentanyl and midazolam. Individuals were approached in order of descending reticulation scores; five of eight participants agreed to participate. Between 10–16 biopsies were taken from regions of the lungs that appeared to have the highest burdens of reticulations on CT scans. Biopsy samples fixed in formalin were analyzed by pathology (AS).

Genomic Measurements

Blood was drawn from each participant at each follow-up visit (4 months, 15-months, and 3 years). DNA was isolated from blood leukocytes utilizing the Gentra Puregene Blood kit (Qiagen, Valencia CA). Leukocyte telomere length (LTL) was measured utilizing a quantitative PCR assay with the Rotogene rtPCR system (Qiagen).(15) The LTL was expressed as a log-transformed ratio of telomere to single-copy gene [ln(T/S)] and this value was compared to LTL from normal control subjects (n=201 unrelated multiethnic individuals from Dallas, TX, ranging in age from 19 to 89 years) to estimate an age-adjusted LTL percentile. The MUC5B risk allele rs35705950 genotype was determined by Sanger sequencing.

Missing Data

The Glasgow Coma Score, a component of the SOFA score, was missing for nearly all non-ICU patients, so we imputed a score of 15 for these participants based on previous literature(16). Three participants with tracheostomy were not liberated from mechanical ventilation prior to discharge. We assessed their ventilator days from intubation date until hospital discharge date. Hospital admission BMI was missing for one participant with fibrotic-like abnormalities; weight loss was imputed as none for the frailty assessment. Administration of steroids and tocilizumab was missing for one individual without fibrotic-like abnormalities, who was imputed as not having received these medications since the participant was not critically ill. One individual declined to perform PFTs at the 3-year visit. Five individuals did not complete six-minute walk testing because of study protocol contraindications in place to ensure patient safety: use of a walker or cane, or chest pain within the prior month. Results are reported as complete case analyses.

Statistical Analyses

We tested associations between clinical characteristics, genomic biomarkers and radiologic patterns using ANOVA, chi-square, Mann-Whitney, Wilcoxon rank sum, and Fischer exact tests where appropriate. McNemar’s and Wilcoxon signed-rank tests were utilized for longitudinal comparisons between these variables at 4-month, 15-month, and 3-year follow-up visits. We calculated Spearman’s rank correlation coefficients between HRCT scores and clinical measurements.

In adjusted analyses, we included covariables that were previously found to be associated with fibrotic-like abnormalities(4, 9) or that were biologically plausible, such as sex(17), ancestry(17), smoking status(17), and asthma(18). To control for confounding while minimizing model overparameterization, we generated covariate balance propensity scores (CBPS) using the CBPS package in R by regressing potential confounders on the independent variable of interest. The CBPS included age, sex, race, BMI, admission SOFA score, telomere length percentile, ventilator days, history of asthma, and smoking history. Generalized additive logistic models (GAMs) were generated for each independent variable to assess for non-linear associations between risk factors and fibrosis. In the GAM analyses the CBPS was considered balanced if the spearman correlation coefficient between the covariable and variable of interest was <[0.3]. Odds ratios were generated via logistic regression if there was no evidence of non-linearity in the GAM analyses. Age and SOFA showed evidence of non-linearity; hence values were divided into tertiles to generate tertile specific odds ratios and a P-for-trend is reported to test the significance of the relationship between tertiles (see Table 2).

Table 2.

Adjusted Associations between Clinical Variables and Fibrotic-Like Abnormalities 3 years after COVID-19

Variable	Odds Ratio	95% CI	P-Value
Age at hospital admission			0.38*
Tertile 1: 27–51 (ref)	1.00
Tertile 2: 52–62	0.66	(0.23
Tertile 3: 63–78	1.96	(0.64, 5.99
Male Sex	6.61	(1.68, 26.05)	0.007
Age-Adjusted Telomere Length Percentile†	0.82	(0.66, 0.9)	0.01
BMI on Admission	0.92	(0.86, 0.98)	0.01
Ventilator Days	1.1	(1.05, 1.16)	<0.001
SOFA Score on Admission			<0.001*
Tertile 1: 0–2 (ref)	1.00
Tertile 2: 3–4	3.38	(1.13, 10.14
Tertile 3: 5–11	4.07	(3.75, 74.43)
Black or African American Race‡	0.34	(0.12, 0.93)	0.04
Current or Former Smoker	0.58	(0.22, 1.52)	0.27
History of Asthma	0.38	(0.11, 1.3)	0.12

Given that the study sample size would not support a single multivariable model, individual models were created for each independent variable listed in the table. Covariate balanced propensity scores specific to each independent variable were generated from the other potentially confounding covariables, including age, sex, race, BMI, admission SOFA score, telomere length percentile, ventilator days, history of asthma, and smoking history. This model compares 61 patients with fibrotic-like abnormalities (excluding one patient with missing BMI data) against 33 patients with normal CT scans.

Abbreviations: OR, odds ratio; CI, confidence interval; BMI, body mass index; SOFA, sequential organ failure assessment.

^*P-for-trend between tertiles

^†per 10% increase in telomere length

^‡as compared to White, Asian, or Other Race

For the frailty and functional features with significant univariable associations with fibrosis (reduced and percent-predicted 6MWD, physical activity, slow walk speed, prefrail status, and combined frail or prefrail status, see Table 3), a CBPS including age, sex, hospital length of stay, maximal supplemental oxygen requirement, and SOFA scores was generated to reduce potential confounding, then regressed on the variables of interest.

Table 3.

Pulmonary function, Physical Function, and Symptoms 3 years after COVID-19

	Number with Data, N	Total, (N=102)	Fibrotic-Like Abnormalities* (N=62)	Non-Fibrotic Abnormalities* (N=7)	Normal CT scan (N=33)	P-value†	Adjusted P-value†
FVC %Predicted, mean (SD)	101‡	86.9 (17.7)	86.6 (18.3)	87.6 (12.2)	87.5 (18.2)	0.81
Reduced FVC, N (%)	101‡	27 (26.7%)	16 (25.8%)	2 (28.6%)	9 (28.1%)	0.81
DLCO %Predicted, mean (SD)	101‡	81.2 (20.2)	77.7 (20.2)	79.1 (21.1)	88.4 (18.6)	0.014
Reduced DLCO, N (%)§	101‡	41 (40.6%)	32 (51.6%)	2 (28.6%)	7 (21.9%)	0.006
6MWD %Predicted, mean (SD)	97∥	73.1 (18.4)	70.1 (17.8)	70.7 (12.4)	78.8 (19.6)	0.035	0.011
Reduced 6MWD, N (%)#	97∥	60 (61.9%)	42 (72.4%)	6 (85.7%)	12 (37.5%)	0.001
Weak Grip, N (%)♠	102	41 (40.2%)	26 (41.9%)	3 (42.9%)	12 (36.4%)	0.60
Decreased Activity, N (%)♠	102	10 (9.8%)	9 (14.5%)	0	1 (3.0%)	0.082	0.48
Exhaustion, N (%)♠	102	32 (31.4%)	22 (35.5%)	3 (42.9%)	7 (21.2%)	0.15
Slow 4.57 M Walk Speed, N (%)♠	102	29 (28.4%)	23 (37.1%)	0	6 (18.2%)	0.057	0.86
Weight Loss > 10 lbs, N (%)♠	101¶	19 (18.6%)	11 (17.7%)	3 (42.9%)	5 (15.2%)	0.75
Prefrail, N (%)♥	102	60 (58.8%)	42 (67.7%)	4 (57.1%)	14 (42.4%)	0.017	0.12
Frail, N (%)♥	102	12 (11.8%)	8 (12.9%)	1 (14.3%)	3 (9.1%)	0.58
Frail or Prefrail, N (%)	102	72 (70.6%)	50 (80.7%)	5 (71.4%)	17 (51.5%)	0.003	0.27
UCSD SOBQ Score, median (IQR)	102	30 (9–55)	37 (13–58)	34 (4–67)	25 (6–42)	0.057
MMRC Dyspnea Score, mean (SD)	102	1.2 (1.1)	1.2 (1.1)	1.4 (1.3)	1.1 (1.2)	0.68
Cough Severity ≥30, N (%)	102	38 (37.3%)	24 (38.7%)	3 (42.9%)	11 (33.3%)	0.61

Abbreviations: FVC, forced vital capacity; FEV1, forced expiratory volume in 1 second; DLCO, diffusion capacity for carbon monoxide; 6MWD, six-minute walk distance; VAS, visual analogue scale from 0 – 100.

^*Fibrotic-like abnormalities defined as the presence of reticulations, traction bronchiectasis, or honeycombing; non-fibrotic abnormalities defined as the presence of ground glass opacities, intraparenchymal opacities, nonemphysematous opacities or centrilobular nodules in the absence of fibrotic-like abnormalities.

^†Univariable and adjusted analysis performed using fibrotic-like abnormalities (N=62) vs Normal CT scan (N=33) as binary outcome.

^‡Data missing for one individual with a normal CT scan

^§Per 2017 ATS/ERS guidelines

^∥Data missing for 5 individuals who met patient safety protocol contraindications for not performing the 6MWD (see Methods). Four had fibrotic-like abnormalities, and one had a normal CT scan

^¶One individual with missing admission weight was excluded from the univariable analysis. No weight loss was imputed for this individual when evaluating Fried Frailty phenotype domains (see Methods).

^#<80% predicted based on sex, age, height, and weight, per 2017 ATS/ERS guidelines

^♠Included as a Fried Frailty Phenotype measure. Weight loss >10 lb based on follow-up visit weight.

^♥Pre-frail: 1–2 positive Fried Frailty phenotype domains, Frail 3–5 positive Fried Frailty phenotype domains.

RESULTS

Participants had a mean(SD) age of 60(12) years; 55(54%) were male, 62(61%) were non-White, and 50(49%) required mechanical ventilation (Table 1). Over 70% previously participated in our 4-month and/or 15-month post-COVID studies (Figure S1)(4, 9). While there was no difference in demographics or clinical variables between those who did or did not follow up after the 15- month study, the 27 participants who were newly recruited to participate in the 3-year study were slightly older and reported decreased physical activity (Table S1 and Table S2). 67% of participants had some type of radiologic abnormality, including 61% with fibrotic-like patterns (Figure 1A, Table S3). Most fibrotic-like abnormalities were characterized as reticulations (94%) and bronchiectasis (74%) (Table S3). 42 of 62 participants (68%) with fibrotic-like abnormalities had both reticulations and traction bronchiectasis; 26% had reticulations only and 6% had traction bronchiectasis only. The one participant with honeycombing also had reticulations and traction bronchiectasis.

Table 1.

Demographic and clinical features of severe COVID-19 Survivors 3-years from initial Illness

	Number with Data	Total	Fibrotic-Like Abnormalities*	Non-Fibrotic Abnormalities*	Normal CT Scan
Number	102	102	62 (60.8%)	7 (6.9%)	33 (32.4%)
DEMOGRAPHICS
Age, mean (SD)	102	55.9 (12.3)	57.2 (13.0)	50.8 (14.8)	54.6 (10.1)
Age at Visit	102	58.4 (12.5)	59.3 (14.9)	54.1 (14.8)	57.7(10.3)
Male	102	55 (53.9%)	43 (69.4%)	3 (42.9%)	9 (27.3%)
Race
White	40	40 (39.2%)	25 (40.3%)	2 (28.6%)	13 (39.4%)
Black	25	25 (24.5%)	11 (10.8%)	1 (14.3%)	13 (39.4%)
Asian	1	1 (1.0%)	1 (1.6%)	0	0
Other	36	36 (35.3%)	25 (40.3%)	4 (57.1%)	7 (21.2%)
Hispanic Ethnicity	102	70 (68.6%)	46 (74.2%)	4 (57.1%)	20 (60.6%)
BMI on Admission	101†	32.6 (7.4)	31.4 (6.3)	29.2 (4.7)	35.6 (8.8)
BMI at Visit	102	33.3 (8.5)	32.1 (6.6)	28.8 (3.5)	36.8 (11.2)
GENOMIC FACTORS
Leukocyte Telomere Length Percentile, Median (IQR)	102	50 (46–66)	50 (37–50)	49 (43–50)	56 (50–86)
LTL <10th percentile	102	11 (10.8%)	7 (11.3%)	1 (14.3%)	3 (9.1%)
MUC5B homozygous or heterozygous variant carriers	102	7 (6.9%)	3 (4.8%)	1 (14.3%)	3 (9.1%)
COMORBIDITIES
Hypertension	102	61 (59.8%)	36 (58.1%)	5 (71.4%)	20 (60.6%)
Diabetes	102	29 (28.4%)	17 (27.4%)	1 (14.3%)	11 (33.3%)
COPD	102	5 (4.9%)	3 (4.8%)	0	2 (6.1%)
Asthma	102	22 (21.6%)	6 (9.7%)	5 (71.4%)	11 (33.3%)
Heart Disease	102	4 (3.9%)	2 (3.2%)	0	2 (6.1.%)
Former Smoker	102	37 (36.3%)	18 (29.0%)	2 (28.6%)	17 (51.5%)
Active Smoker	102	2 (2.0%)	0	0	2 (6.1%)
CLINICAL FACTORS
Admission SOFA, mean (SD)	102	3.7 (2.3)	4.5 (2.1)	2.5 (1.9)	2.6 (2.1)
Received Steroids	102‡	47 (46.1%)	38 (61.3%)	3 (42.9%)	6 (18.2%)
Received Anti IL-6R Blocker	102‡	23 (22.6%)	20 (32.3%)	1 (14.3%)	2 (6.1%)
Maximum Oxygen Requirement
Nasal Canula	102	32 (31.4%)	9 (14.5%)	3 (42.9%)	20 (60.6%)
Non-Rebreather	102	18 (17.6%)	10 (16.1%)	2 (28.6%)	6 (18.2%)
NIPPV or HFNC	102	2 (2.0%)	2 (3.2%)	0	0
Mechanical Ventilation	102	50 (49.0%)	41 (66.1%)	2 (28.6%)	7 (21.2%)
Ventilator Days, Median (IQR)	50§	35 (17–46)	39 (20–49)	32 (6–58)	12 (9–15)
Hospital Days, Median (IQR)	102	23 (7–46)	38 (21–55)	15 (2–18)	7 (4–13)
OUTCOMES
FVC, % Predicted, mean (SD)	101∥	86.9 (17.7)	86.6 (18.3)	87.6 (12.2)	87.5 (18.2)
Reduced FVC at 3 years	101∥	27 (26.7%)	16 (25.8%)	2 (28.6%)	9 (28.1%)
DLCO, % Predicted, mean (SD)	101∥	81.2 (20.2)	77.7 (20.2)	79.1(21.1)	88.4 (18.6)
Reduced DLCO at 3 years	101∥	41 (40.6%)	32 (51.6%)	2 (28.6%)	7 (21.9%)
6MWD, % Predicted, mean (SD)	97¶	73.1 (18.4)	70.1 (17.8)	70.7 (12.4)	78.8 (19.6)
Reduced 6MWD at 3 Years	97¶	60 (61.9%)	42 (72.4%)	6 (85.7%)	12 (37.5%)
Change in BMI from admission, median (IQR)	101†	0.7 (−.9–2.3)	0.4 (−.9–2.4)	−0.3 (−3.1 – 1.5)	1.4 (−0.3–2.3)
Cough VAS, median (IQR)	102	17 (0–50)	18 (0–50)	29 (0–50)	16 (0–48)
Cough VAS >30	102	38 (37%)	24 (38.7%)	3 (42.8%)	11 (33%)

^*Fibrotic-like abnormalities defined as the presence of reticulations, traction bronchiectasis, or honeycombing; non-fibrotic abnormalities defined as the presence of ground glass opacities, intraparenchymal opacities, nonemphysematous opacities or centrilobular nodules in the absence of reticulations, traction bronchiectasis, and honeycombing.

^†Data missing for one individual with fibrotic-like abnormalities

^‡‡Reflects total including imputed data. The Glasgow Coma Score, a component of the SOFA score, was missing for nearly all non-ICU patients, wo we imputed a score of 15 for these participants (see Methods).

^‡Reflects total including imputed data, which was missing for one individual with a normal CT scan.

^§Of the 50 individuals who were mechanically ventilated, N=41 had fibrotic-like abnormalities, N=2 had non-fibrotic abnormalities, and N=7 had normal CT scans

^∥Data missing for one individual with a normal CT scan.

^¶Data missing for 5 individuals who met patient safety protocol contraindications for not performing the 6MWD (see Methods). Four had fibrotic-like abnormalities and one had a normal CT scan

Figure 1. Chest imaging patterns, trends over time, and lung transbronchial histopathology of COVID-19 survivors.

A. High resolution Computed Tomography (HRCT) chest scores for 102 post-COVID-19 survivors at 3 years. The patterns are segregated into non-fibrotic and fibrotic patterns. B. Change in individual CT pattern scores for 75 individuals 3-years after COVID-19 who at least one prior scan. This includes 41 participants with scans at all time points (4-month, 15-month, 3-year), 72 individuals with scans for only the last two time points, and 3 individuals with scans for only the first and last time points. Each line represents an individual. C. Representative example of one study participant with persistent radiologic abnormalities. HRCT reticulation scores at 4-month, 1-, and 3-year were 8.4, 7.8, and 6.8, respectively. Traction bronchiectasis scores at 4-month, 1-, and 3-year were 5, 3, and 4, respectively. D. Lung transbronchial biopsies of two participants whose fibrosis scores were in the top quartile. The top row shows the lung histopathology of the same participant whose HRCT scan is shown in C. The lung shows evidence of small airways disease with peribronchiolar metaplasia; H&E (left) and Masson trichome (right) stained slides are shown. The bottom row shows the lung histopathology of another participant (HRCT scans shown in Figure S2 panel E) with alveolar interstitial thickening by fibromyxoid tissue and type 2 cell hyperplasia; H&E (left) and Movat (right) stained slides are shown. Bars represent 100 microns.

In adjusted analyses, male sex, lower BMI, shorter telomere length, SOFA score, and days of mechanical ventilation were independently associated with increased odds of fibrotic-like abnormalities (Table 2), consistent with our 4-months and 15-month studies(4, 9). Black race was negatively associated in adjusted analyses, a novel finding. In sensitivity analyses with fibrotic-like abnormalities defined as only traction bronchiectasis and honeycombing, BMI, ventilator days, SOFA scores, and age-adjusted telomere lengths remained associated with fibrotic changes, and Black race remained negatively associated with fibrotic changes (Table S4).

Forced vital capacity (FVC) and diffusion capacity (DLCO) measurements are shown in Figure 2. We did not find that the FVC or DLCO significantly changed over time. Fibrotic-like abnormalities were associated with lower DLCO, lower 6-minute walk distance (6MWD), lower physical activity, slower gait-speed, and higher prevalence of pre-frailty in unadjusted univariable analyses (Table 3). Lower percent predicted 6MWD, but not the other markers of frailty, were associated with fibrotic-like abnormalities after adjusting for severity of illness. Findings were robust to sensitivity analyses with fibrotic-like abnormalities defined as only traction bronchiectasis and honeycombing (Table S5). Reticulation and traction bronchiectasis scores were weakly correlated with DLCO and dyspnea (Table S6). Change in ground glass opacity scores from 4-months to 3-years was inversely correlated with change in FVC, while changes in reticulation scores over this time were weakly correlated with changes in DLCO (Table S7).

Figure 2. Pulmonary function test trends over time of COVID-19 survivors.

A. Forced vital capacity (FVC) and Diffusion capacity of carbon monoxide (DLCO) for 41 individuals who had measurements at all time points (4-month, 15-month, 3-yr). B. FVC and DLCO for 72 individuals who had measurements for two time points (15-month, 3-year). C. FVC and DLCO for 102 individuals who had measurements 3-years after hospitalization for severe COVID-19.

Fibrotic-like abnormalities modestly improved over time (Table 4, Figure 1B, 1C) but in no case completely resolved. Semi-quantified reticulation scores improved between 4-months and 15-months, 4-months and 3-years, and 15-months and 3-years. Traction bronchiectasis scores improved between 4-months and 15-months as well as between 4-months and 3-years. As expected, the improvement in ground glass opacity scores were most notable early on, from 4-months to 15-months. Subjectively, thoracic radiologists noted an improvement in 29% participants with imaging studies at 4-months and 3-years, with the remainder (71%) having stable findings (Table 5). For 77 participants with imaging at 15- months and 3-years, 9% showed subjective improvement and the rest (91%) had stable findings. No one had worsening fibrosis.

Table 4.

Pairwise comparisons of participants with HRCT scans with non-zero HRCT scores

	Ground Glass Opacities			Reticulations			Traction Bronchiectasis
	N	Score, Median (IQR)	P-value	N	Score, Median (IQR)	P-value	N	Score, Median (IQR)	P-value
Individuals with non-zero scores at 4-mo and imaging at two time points:
At 4-mo post-COVID	10	2.9 (1–7.4)	0.02	28	3.4 (1.6–5.9)	<0.001	23	3 (1–5)	0.001
At 15-mo post-COVID	10	2.8 (0.2–4.2)		28	2.3 (1.1–4.9)		23	2 (1–5)
At 4-mo post-COVID	11	2.2 (0.4–7.4)	0.04	31	3.6 (1.6– 6.0)	<0.001	24	3 (1.5–5)	0.002
At 3-yr post-COVID	11	1.8 (0.2–5.0)		31	1.8 (1.0–5.2)		24	2.5 (1–3.5)
At 15-yr post-COVID	10	2.8 (0.2–4.2)	0.68	28	2.3 (1.1–4.9)	0.03	23	2 (1–5)	0.92
At 3-yr post-COVID	10	2.1 (0.2–5)		28	2.1 (1.1–4.7)		23	2 (1–4)
Individuals with non-zero scores at 15-mo and imaging at two time points
At 4-mo post-COVID	8	4.6 (1.6– 7.9)	0.04	27	3.6 (1.6–6.0)	0.0004	18	4 (2–5)	0.02
At 15-mo post-COVID	8	3.8 (1.2– 4.2)		27	2.6 (1.2–5.0)		18	3 (2–5)
At 4-mo post-COVID	8	4.6 (0.4–7.9)	0.079	27	3.6 (1.6–6.0)	<0.001	18	4 (2–5)	0.006
At 3-yr post-COVID	8	3.1 (1–5.1)		27	2.4 (1.2–5.2)		18	3 (2–4)
At 15-mo post-COVID	13	2 (1.2–4)	0.38	46	3.0 (1.2–6.6)	0.005	28	2 (1–4)	0.06
At 3-yr post-COVID	13	1.8 (0.2–3.8)		46	2.6 (1.2–5.8)		28	2 (1–3.5)
Individuals with non-zero scores at 3-yr and imaging at two time points:
At 4-mo post-COVID	10	2.9 (0.4–7.4)	0.02	28	3.4 (1.6–5.9)	0.0002	24	3 (1–5)	0.001
At 15-mo post-COVID	10	2.8 (0.2–4.2)		28	2.3 (1.1–4.9)		24	2 (0.5–4)
At 4-mo post-COVID	10	2.9 (0.4–7.4)	0.13	31	3.6 (1.6–6.0)	<0.001	25	3 (1–5)	0.006
At 3-yr post-COVID	10	2.1 (0.4–5)		31	1.8 (1.0–5.2)		25	2 (1–3)
At 15-mo post-COVID	14	1.9 (0.2–4.0)	0.39	46	3.0 (1.0–6.6)	0.02	36	2.0 (1–3)	0.91
At 3-yr post-COVID	14	1.8 (0.4–3.8)		46	2.6 (1.2–5.8)		36	1.0 (1–3)

HRCT: high resolution computed tomography. Longitudinal comparison testing performed utilizing Wilcoxon signed-rank test.

Table 5.

Qualitative assessment of HRCT changes after COVID-19

	Between 4-mo and 15-mo (N=51)	Between 15-mo and 3-yr (N=77)	Between 4-mo and 3-yr (N=56)
Improvement, N (%)	10 (19.6%)	7 (9.1%)	16 (28.6%)
Stability, N (%)	40 (78.4%)	70 (90.9%)	40 (71.4%)
Worsening, N (%)	1 (1.9%)*	0 (0%)	0 (0%)

HRCT: high resolution computed tomography.

^*The one participant who had worsened HRCt between 4-months and 15-months was diagnosed with sarcoidosis(26), with radiologists noting an increase in nodular densities without worsening of reticulations or traction bronchiectasis. After treatment of the sarcoidosis, the radiologists thought his imaging between 4-months to 3-years and between 15-months to 3-years was improved.

HRCT: high resolution computed tomography.

Due to the limited quality of some expiratory imaging studies, air trapping could only be assessed for 35 participants (45%) who had imaging studies both at 15-months and 3-years. There was more air trapping at 3-years (88.7% vs. 48.6%, p<0.001, Table S8). Air trapping was associated with self-reported asthma but was not associated with fibrotic-like abnormalities (Table S9). We did not find that cough or any other self-reported post-COVID symptom was specifically associated with fibrotic-like abnormalities (Table S10).

Five participants with fibrosis scores in the top quartile (Table S11, Figure S2) agreed to undergo bronchoscopy with transbronchial lung biopsy. All five of these participants had evidence of air trapping on their 3-year imaging study. Evidence of small airways disease, primarily peribronchiolar metaplasia, was found in four of four biopsies containing small airways (Figure 1D). All five biopsies contained lung parenchyma but only one showed increased interstitial thickness and fibrosis.

DISCUSSION

In this single-center, New York City-based, Black and Hispanic-predominant cohort of adult survivors of severe COVID-19, three years from the initial wave, we find a high prevalence (61%) of fibrotic-like abnormalities, characterized primarily by reticulations and traction bronchiectasis. We leverage this cohort to identify clinical and genomic factors predictive of post-COVID-19 fibrosis, confirming prior associations with increased illness severity, longer hospitalization and shorter blood leukocyte telomere length. We report the intriguing finding that Black race is negatively associated with post-COVID fibrotic-like abnormalities. For individuals with more than one imaging study at 4-months, 15-months, or 3-years after COVID-19, we find evidence of modest improvement of reticulation scores across all time points, with no participant demonstrating complete resolution or worsening fibrotic-like abnormalities.

Risk factors for fibrotic-like abnormalities at 3 years are similar to those described by others, including greater initial illness severity(1, 3, 19), extended hospitalization(3), ventilatory support requirement(3), and shorter leukocyte telomere lengths(20). We identify male sex as a risk factor for post-COVID fibrosis, consistent with its risk for Idiopathic Pulmonary Fibrosis (IPF). This is concordant with an interim analysis of the UKILD post-COVID study, in which male survivors had increased persistent interstitial changes(1), and may reflect the increased illness severity and mortality noted in men with COVID-ARDS(21, 22).

Higher BMI was also noted to be a protective factor against development of fibrotic-like abnormality at 3-years. In this study, the median BMI for participants with and without residual fibrotic changes exceeded the CDC cutoff for obesity (>30). Obesity is a risk factor for mechanical ventilation and death in patients with COVID-ARDS(23). Survivor bias may explain the lower BMI noted in patients with fibrotic-like changes, as they were more likely to survive despite higher illness severity.

We report the intriguing finding that Black race appears protective against development of residual fibrotic changes when compared against other races in adjusted analysis. This is surprising, because a meta-analysis of 68 cross-sectional and cohort studies showed increased rates of COVID-19 test positivity and ICU admission in African American patients that were only partially explained by social determinants of health(24). Our finding may reflect the relatively small sample size of this cohort. However, variable presentations of post-acute COVID symptoms have been noted amongst racial groups. Interim analysis of the RECOVER program in New York City revealed that Black and Hispanic adults were more likely to experience headaches, chest pain, and joint pains but are less likely to have sleep disorders, brain fog, or fatigue than white participants(25). Although these are not respiratory symptoms, it may be that post-recovery organ dysfunction and symptoms differ by race.

Fibrotic-like abnormalities were associated with reduced DLCO, and fibrosis scores were weakly correlated with DLCO, a finding noted in other cohort studies(2, 3). Our observed high prevalence of fibrotic-like imaging abnormalities at 3-years likely reflects the high numbers of patients who underwent mechanical ventilation. In comparison, the Wuhan cohort reported an overall prevalence of fibrotic-like abnormalities of 36% and 47% for those requiring high-flow nasal oxygen or mechanical ventilation, respectively(2).

Although none of the participants had evidence of resolution of their fibrotic-like abnormalities, we did find evidence for modest improvement even at later time points. In contrast to the improvement of ground glass opacities scores between 4-months and 15-months, the semi-quantitative reticulation scores showed significant improvement across all time points, from 4- to 15-months, from 4-months to 3-years, and from 15-months to 3-years. We find that the semi-quantitative traction bronchiectasis scores also show improvement from 4- to 15-months and from 4-months to 3-years. Beside the one participant who was diagnosed with sarcoidosis and whose imaging study demonstrated increased nodular densities between 4-months to 15-months(26), no other participant had worsening radiologic abnormalities over the course of the study. Thus, the general trend of modest improvement of reticulations and traction bronchiectasis differentiates these post-COVID fibrotic-like abnormalities from progressive pulmonary fibrosis(27).

In contrast with other studies that categorize reticulations in the absence of architectural distortion (traction bronchiectasis or honeycombing) as non-fibrotic abnormalities(2), here we include reticulations as fibrotic-like abnormalities, which may overestimate the prevalence of post-COVID fibrosis. Yet, the associations of clinical factors and impairments in gas exchange and exercise that we identified in our primary analysis were robust in sensitivity analyses that defined fibrotic-like abnormalities as only traction bronchiectasis or honeycombing. The concordant improvement of both reticulations and traction bronchiectasis suggests that both have the potential to slowly improve over time as part of the lung’s normal healing process. Given the 68% co-occurrence of reticulations and traction bronchiectasis in this cohort and the persistence of reticulations over 3-years, we favor our strategy of aggregating reticulations, traction bronchiectasis, and honeycombing into a designation of “fibrotic-like abnormalities”. Continued follow-up of longitudinal post-COVID cohorts will be needed to assess the reversibility or irreversibility of lingering fibrotic-like abnormalities.

Unadjusted univariable associations were noted between fibrotic-like abnormalities and various frailty indices including reduced activity status, gait speed, and prefrailty, a phenotype that portends increased risk of progression to frailty, which has been linked to increased mortality, falls, and hospitalizations(12). After adjusting for indicators of illness severity, percent of predicted 6MWD remained associated with fibrotic-like changes, concordant with studies from Wuhan(5). Thus, COVID-19 survivors, like non-COVID ARDS survivors, are at elevated risk for functional impairments, indicating the need for ongoing monitoring.

Despite the small numbers of participants with paired inspiratory and expiratory imaging studies at both time points, we find evidence of increasing radiologic air-trapping, from 49% at 15-months to 89% at 3-years, consistent with other cohorts showing persistent air trapping after COVID-19 infection(28–30). Similarly, the histopathological analyses of transbronchial biopsies from participants with the highest fibrosis scores show a high degree of peribronchial metaplasia of the small respiratory bronchioles. Larger studies, including more detailed cellular and molecular studies, will be needed to fully understand the scope of large and small epithelial alterations, peribronchial inflammatory changes, and airway smooth muscle remodeling that persist after COVID-19 infection.

Our study has several limitations. We experienced 30% attrition from the 15-month to 3-year study, although this rate was less than the 40–65% attrition experienced by other US COVID-19 studies(31, 32). Despite the attrition, we find a similar prevalence of fibrotic-like abnormalities. Of the 41 individuals who participated in the initial 4-month study, 26 (63%) continue to have fibrotic-like abnormalities 3 years later, similar to the overall prevalence of 61%. Newly recruited participants had lower physical activity despite similar initial illness severity and demographics. The radiologic scoring system involves categorizing five discrete axials levels of CT scans, leaving room for variability in scores longitudinally and between readers. There is inherent risk for sampling error with transbronchial biopsy which may explain why histopathologic evidence of interstitial fibrosis was not found in all subjects. Other limitations include the relatively small numbers of participants, the lack of a replication cohort, the decreased number of imaging studies to rigorously evaluate air trapping, and the need to use propensity scores to adjust for covariates. Given the limited sample size, some of the observed associations should be considered preliminary and hypothesis generating.

In this Black and Hispanic-predominant cohort of severe and critical COVID-19 survivors, we note a 61% prevalence of persistent fibrotic-like abnormalities 3-years after initial infection, which are associated with reduced DLCO and reduced 6MWD. The persistence of fibrotic-like abnormalities and their association with reduced lung function and exercise capacity highlight the need for continued monitoring in certain cases.