Associations of hiatus hernia with CT-based interstitial lung changes: the MESA Lung Study

Abstract

Background

Hiatus hernia (HH) is prevalent in adults with pulmonary fibrosis. We hypothesised that HH would be associated with markers of lung inflammation and fibrosis among community-dwelling adults and stronger among MUC5B (rs35705950) risk allele carriers.

Methods

In the Multi-Ethnic Study of Atherosclerosis, HH was assessed from cardiac and full-lung computed tomography (CT) scans performed at Exam 1 (2000–2002, n=3342) and Exam 5 (2010–2012, n=3091), respectively. Percentage of high attenuation areas (HAAs; percentage of voxels with attenuation between −600 and −250 HU) was measured from cardiac and lung scans. Interstitial lung abnormalities (ILAs) were examined from Exam 5 scans (n=2380). Regression models were used to examine the associations of HH with HAAs, ILAs and serum matrix metalloproteinase-7 (MMP-7), and adjusted for age, sex, race/ethnicity, educational attainment, smoking, height, weight and scanner parameters for HAA analysis.

Results

HH detected from Exam 5 scans was associated with a mean percentage difference in HAAs of 2.23% (95% CI 0.57–3.93%) and an increase of 0.48% (95% CI 0.07–0.89%) per year, particularly in MUC5B risk allele carriers (p-value for interaction=0.02). HH was associated with ILAs among those <80 years of age (OR for ILAs 1.78, 95% CI 1.14–2.80) and higher serum MMP-7 level among smokers (p-value for smoking interaction=0.04).

Conclusions

HH was associated with more HAAs over time, particularly among MUC5B risk allele carriers, and ILAs in younger adults, and may be a risk factor in the early stages of interstitial lung disease.

Shareable abstract (@ERSpublications)

In a population cohort, hiatus hernia detected by computed tomography (CT) was associated with increased lung densities on CT over time, especially among carriers of the MUC5B (rs35705950) risk allele, and may be a risk factor in interstitial lung disease https://bit.ly/3xakyeW

Introduction

Repetitive injury to the lung parenchyma, coupled with aberrant wound healing, has been postulated to lead to interstitial lung disease (ILD) [1]. Hiatus hernia (HH), which is the protrusion of gastric contents through the lower oesophageal opening, promotes gastro-oesophageal reflux (GOR) which may injure the lung as HH is more prevalent among adults with ILD compared with other chronic lung diseases and associated with disease progression and worse mortality in idiopathic pulmonary fibrosis (IPF) [2-4]. It has been postulated that HH in a genetically predisposed individual (i.e. MUC5B (rs35705950) risk allele carrier) may increase the risk of pulmonary fibrosis [5, 6].

Investigators have leveraged computed tomography (CT) data to develop methods that capture lung parenchymal changes that may indicate inflammation and fibrosis, including include high attenuation areas (HAAs) and interstitial lung abnormalities (ILAs), with each associated with higher mortality, worse lung function and risk factors for ILD [7, 8]. Since standard approaches to identifying HH (i.e. upper endoscopy) are invasive and not feasible for large population-based cohort studies, we and others have developed an approach to define HH based on CT scans [2, 3, 9].

We tested the hypothesis that the presence of radiologically detected HH would be associated with more HAAs among community-dwelling adults. Secondarily, we examined associations between HH and ILAs, circulating biomarkers of lung injury and fibrogenesis (i.e. surfactant protein-A (SP-A) and matrix metalloproteinase-7 (MMP-7)), and mortality [10, 11].

Methods

Study participants

The Multi-Ethnic Study of Atherosclerosis (MESA) is a National Heart, Lung, and Blood Institute-sponsored longitudinal cohort study with the original intent of examining subclinical cardiovascular disease among community-dwelling adults [12]. There were 6814 adults between the ages of 45 and 84 years without clinically diagnosed cardiovascular disease who were originally recruited to MESA at Exam 1 (2000–2002). An additional 67 participants were enrolled as part of the MESA Air Study [9]. All MESA participants provided written informed consent and institutional review board approval was obtained by all study sites.

CT scan

Full-lung noncontrast CT scans were completed in MESA using the MESA Lung protocol at Exams 5 (2010–2012) and 6 (2016–2018) [13]. Assessments for HH in MESA have been previously described [9]. CT scans were analysed using Horos, an open-source medical image viewer (https://horosproject.org). The presence of HH was defined as the gastric fold extending >2 cm above the diaphragm on axial images and/or reformations measured using electronic callipers. Classification of each HH (type I–IV) was performed based on previously published criteria, and the distance between the gastric fold and diaphragm was measured in Exam 5 scans [3]. HH was also assessed from Exam 1 cardiac scans. Three independent reviewers (J.S.K., X.Y. and M.R.P.) reviewed 1031 scans from Exam 5 and 393 scans from Exam 1. The remaining scans were reviewed by a single reader (J.S.K.). Interobserver disagreement was addressed by majority opinion. Fleiss’ κ for the presence of HH from Exam 1 and 5 scans was 0.88 and 0.86, respectively [9]. The primary exposure of interest was the presence of HH on CT scan at Exam 5. Given the possibility that a sliding HH may be missed on radiological assessment, we assigned the presence of HH among those who had HH at Exam 1 and absent at Exam 5. HH was not assessed from any other exams in MESA.

HAAs were defined as the percentage of voxels having attenuation values between −600 and −250 HU [8]. The Pulmonary Analysis Software Suite (PASS) at the University of Iowa’s Advanced Pulmonary Physiomic Imaging Laboratory (Iowa City, IA, USA) was used by a trained technician for semiautomatic segmentation and correction of the lung [14]. HAAs were measured from full-lung scans performed at Exams 5 and 6. HAAs were also measured from cardiac CT scans performed at Exam 1 (2000–2002), Exam 2 (2002–2004), Exam 3 (2004–2005), Exam 4 (2005–2008) and Exam 5 (2010–2012). Cardiac scans capture approximately >60% of the lung parenchyma and density measures are highly correlated with those from full-lung scans (r=0.93) [15].

ILA assessments were performed from Exam 5 full-lung CT scans by one of five trained radiologists [8]. ILA was defined as the presence of nondependent lung changes including ground-glass and/or reticular abnormalities, nonemphysematous cysts, honeycombing, and traction bronchiectasis involving >5% of a lung zone [16]. Fibrotic ILA was defined as the presence of traction bronchiectasis and/or honeycombing. Participants with an indeterminate status for ILAs were excluded from the analysis [8].

Spirometry

In accordance with guidelines, spirometry was performed at Exams 5 and 6 (2016–2018) as part of the MESA Lung Ancillary Study [17, 18].

MMP-7 and SP-A

MMP-7 and SP-A were measured in stored serum samples from Exam 1 at the MESA Core Laboratory at the University of Vermont’s Laboratory for Clinical Biochemistry Research (Burlington, VT, USA) [8]. Quantitative sandwich ELISA (R&D Systems, Minneapolis, MN, USA) was used to measure these serum biomarkers. MMP-7 and SP-A were not measured again in the majority of participants at subsequent MESA exams.

Genotyping of MUC5B (rs35705950)

The Affymetrix Human SNP array 6.0 (Affymetrix, Santa Clara, CA, USA) was used to perform genome-wide genotyping. Quality control of the genome-wide genotypes was carried out as previously described [19]. Genotypes for the MUC5B promoter polymorphism (rs35705950) in MESA were obtained through genome-wide imputation of our Affy 6.0 genotypes using the Michigan Imputation Server (https://imputationserver.sph.umich.edu) with the TOPMed reference panel. Imputation quality for the rs35705950 was good, with imputation R²-values of 0.89, 0.67, 0.85 and 0.91 in MESA White, Asian, African American and Hispanic participants, respectively.

Mortality

Each MESA participant or family member was contacted by the research staff every 9–12 months to determine the vital status. To supplement this review and ensure complete follow-up, the National Death Index was also used and mortality was adjudicated until 31 December 2018.

Statistical analysis

The primary analysis of this study was to examine associations of HH detected from Exam 5 full-lung scans with HAAs. Directed acyclic graphs (DAGs) were used to identify confounders and conceptualise the potential causal pathway (supplementary figure E1). We used linear and logistic regression models to examine cross-sectional associations of HH with HAAs and ILAs from Exam 5 full-lung scans, respectively. Models were adjusted for age, sex, race/ethnicity, smoking status (never, former and current), cigarette pack-years, height, weight and educational attainment (marker of socioeconomic status). We adjusted for socioeconomic status based on the DAG we constructed as it may influence other factors related to HH and reflux and be a confounder [20, 21]. HAAs models were also adjusted for CT-related factors which included study site, lung volume imaged, radiation dose, scanner model and percent emphysema (defined as proportion of the lung < −950 HU). In a sensitivity analysis, we also adjusted for different measures related to body habitus (body mass index (BMI) and waist circumference). Lung volume imaged from full-lung scans was strongly correlated with forced vital capacity (FVC) (Spearman’s correlation coefficient 0.80). Given that HH can reduce lung volumes, we examined whether lung volumes significantly mediated the association between HH and HAAs using the PROC CAUSALMED procedure from SAS version 9.4 (SAS Institute, Cary, NC, USA) [22]. We performed a sensitivity analysis in which HH was not re-adjudicated based on prior Exam 1 scan.

Linear mixed effects models were used to examine the association of HH assessed from Exam 5 full-lung scans with longitudinal change in HAAs from full-lung scans between Exams 5 and 6 (2010–2018) with a random intercept. Models were adjusted for scanner model, radiation dose, voxel size, lung volume imaged, percent emphysema, and time-variant covariates age, height, weight and cigarettes smoked per day. Sex, race/ethnicity, baseline smoking status, educational attainment and each of their interactions with “time since initial HAAs assessment” were also adjusted for in the model. The β-coefficient of “HH×time since initial HAAs assessment” was interpreted as the association between HH and longitudinal change in HAAs. Positive and negative estimates indicate more rapid and slower progression in HAAs over time, respectively. Secondarily, we used the same approach in secondary analyses to examine associations of HH assessed from Exam 1 cardiac scans with HAAs (supplementary material).

A secondary analysis was performed that examined associations of HH type and distance between the gastric fold and diaphragm with HAAs and ILAs (description in supplementary material).

Associations of changes in the prevalence of HH from Exam 1 to Exam 5 with longitudinal changes in HAAs and Exam 5 ILAs were examined with regression models that are described in the supplementary material. Participants without HH at both Exams 1 and 5 scans were considered to have no HH and formed the reference group. Participants with HH at Exam 5 were considered to have HH present. There were six participants who had HH at Exam 1 and none detected on their Exam 5 scan. These participants were grouped into having HH at Exam 5 because there was no evidence of surgical correction.

We examined associations of HH with overall death by MUC5B (rs35705950) risk allele carrier status using Kaplan–Meier curves and Cox regression models as described in the supplementary material.

Linear mixed effects models were used to examine the association of HH assessed from Exam 5 full-lung scans with longitudinal change in FVC between Exams 5 and 6 (2010–2018) with a random intercept (supplementary material).

We used linear regression models to examine cross-sectional associations of Exam 1 HH with serum MMP-7 and SP-A (supplementary material). Inverse probability was used to account for the selection of participants for biomarker measurements [23].

Motivated by prior hypotheses that individuals with a higher genetic risk for ILD and concomitant HH may be particularly vulnerable to lung injury and fibrosis, we examined whether the MUC5B (rs35705950) risk allele (T) modified the associations of HH with HAAs and ILAs with additional adjustment for principal components of genetic ancestry [5, 6, 24]. We performed a priori stratified analyses based on factors that may modify the association between HH and our outcomes: smoking status (ever versus never), sex, BMI (<25, 25–30 and ⩾30 kg·m⁻²) and histamine type 2 (H2) receptor antagonist/proton pump inhibitor use [21, 25]. We used the log-likelihood ratio test with and without interaction terms to determine effect modification for our cross-sectional analyses. We used the F-test of the term “HH×effect modifier×time since HAAs assessment” for our longitudinal HAAs analysis. We included participants with complete covariate data for our analysis. HAAs, MMP-7 and SP-A were log-transformed in our models. SAS version 9.4 and R version 4.0.4 (R Foundation for Statistical Computing Platform, Vienna, Austria) were used for the statistical analysis.

Results

The flowchart of participants with HH assessments and HAAs and ILAs measures is presented in figure 1. Less than 1% of the cohort had missing covariate data. There were 3091 participants with HH and HAA assessments from full-lung scans at Exam 5. Among this group, 2380 had ILA assessments. Baseline characteristics of participants are shown in table 1. The prevalence of HH from Exam 5 scans was 10.6% and the majority were type I. There were no type II HH. Those with HH were older, had higher BMI and HAAs, and a higher prevalence of ILAs. Lung volumes from CT, forced expiratory volume in 1 s and FVC were each lower among those with HH. Characteristics of participants with Exam 1 HH assessments are shown in supplementary table E1. The prevalence of HH from Exam 1 scans was 7.1%.

FIGURE 1.

Flowchart of participants with Multi-Ethnic Study of Atherosclerosis (MESA) Exam 5 hiatus hernia (HH) and a) high attenuation area (HAA) and b) interstitial lung abnormality (ILA) assessments from Exam 5 full-lung scans. c) Participants with HH assessments from Exam 1 scans and repeat assessment from Exam 5 scans.

TABLE 1.

Characteristics of participants at Multi-Ethnic Study of Atherosclerosis Exam 5 (2010–2012)

	No HH (n=2764)	HH (n=327)
Age (years)	69±9	74±8
Female	50	66
Race/ethnicity
Non-Hispanic White	38	47
Asian	14	4
African American	27	25
Hispanic	21	24
Height (cm)	166±10	162±10
Weight (kg)	78±18	79±16
Body mass index (kg·m−2)	28±5	30±5
Smoking status
Never-smoker	46	47
Former smoker	46	50
Current smoker	8	3
Cigarette pack-years	11±21	12±21
MUC5B (rs35705950) #
GG	88	85
GT	11	14
TT	1	1
High attenuation area (%)
Exam 5 (2010–2012)	5.0±2.4	5.3±2.1
Exam 6 (2016–2018)	4.6±1.6	5.3±2.2
Lung volume imaged (mL)
Exam 5 (2010–2012)	4825±1288	4548±1249
Exam 6 (2016–2018)	4670±1200	4312±1143
ILAs ¶	11	17
Forced expiratory volume in 1 s (L) +	2.33±0.72	2.04±0.67
Forced vital capacity (L) +	3.16±0.94	2.78±0.91
Distance between gastric fold and diaphragm (cm)	0.04±0.25	3.65±2.13
Type of HH
I		71
III		29
IV		0

Data are presented as mean±sd or %. HH: hiatus hernia; ILA: interstitial lung abnormality.

^#:there were 2876 participants with MUC5B genotype assessment

^¶:there were 2380 participants with ILA assessments at Exam 5

^+:there were 2728 participants with spirometry assessments at Exam 5.

HAAs and ILAs

The presence of HH on full-lung CT was associated with 2.23% (95% CI 0.57–3.93%) higher HAAs (figure 2a) and was not mediated by lung volumes (supplementary table E2). The MUC5B (rs35705950) risk allele did not modify associations of HH with Exam 5 HAAs. Compared with those without HH, type I and III HH were associated with 3.06% (95% CI 1.08–5.07%) and 0.39% (95% CI −2.50–3.37%) higher HAAs, respectively (figure 2a). The distance between the gastric fold and diaphragm was not associated with Exam 5 HAAs.

FIGURE 2.

Forest plots showing associations of Multi-Ethnic Study of Atherosclerosis Exam 5 hiatus hernia (HH), HH types, distance between the gastric fold and diaphragm, and serial assessments of HH with a) Exam 5 high attenuation areas (HAAs) from full-lung scans and b) longitudinal changes in HAAs. Results are reported for every 20% increase in distance for gastric fold and diaphragm distance analysis. p-values for MUC5B interaction were a) 0.18 and b) 0.02. Square data points represent effect estimates and horizontal lines represent 95% confidence intervals.

Of the 3091 participants with Exam 5 HH and HAAs assessments, 1503 had a HH assessment from Exam 1 scans, of whom 220 (15%) had HH detected at Exam 5. HH detected at Exam 5 was associated with 2.87% (95% CI 0.72–5.07%) higher HAAs compared with those without HH on Exam 1 and 5 scans (figure 2a).

There were 1374 participants with a repeat full-lung scan at Exam 6. The presence of HH on full-lung scan at Exam 5 was associated with a 0.48% (95% CI 0.07–0.89%) increase in HAAs per year (figure 2b). HH was associated with more HAAs over time among participants that carried the MUC5B (rs35705950) risk allele (T) (p-value for MUC5B interaction=0.02). HH was associated with a 1.31% (95% CI 0.19–2.45%) increase in HAAs per year among carriers of the MUC5B risk allele compared with a 0.29% (95% CI −0.16–0.74%) increase among noncarriers. Type I and III HH were associated with a 0.69% (95% CI 0.20–1.17%) and 0.06% (95% CI −0.67–0.78) increase in HAAs per year compared with those without HH, respectively (figure 2b). A 20% increase in distance between the gastric fold and diaphragm was associated with a 0.06% (95% CI 0.01–0.11%) increase in HAAs per year.

There were 1555 participants with HH assessments from Exams 1 and 5 scans and repeated HAAs measurements between 2000 and 2018. There were 1331 (86%) participants without HH from Exams 1 and 5 scans and 224 (14%) with HH detected at Exam 5. Compared with those without HH, the presence of HH between Exams 1 and 5 was associated with a 0.28% (95% CI 0.04–0.53%) change in HAAs per year (figure 2b).

HH was associated with an OR for ILAs of 1.69 (95% CI 1.18–2.41) in an unadjusted model (supplementary table E3). This association was attenuated after adjustment for covariates. Specifically, the adjustment for age appeared to attenuate this association (supplementary table E3). When we stratified by age cut-offs, HH was associated with a higher OR for ILAs among those <80 years of age (OR 1.78, 95% CI 1.14–2.80) compared with those ⩾80 years (OR 0.85, 95% CI 0.45–1.59) (p-value for age interaction=0.06) (supplementary table E4). Longer gastric fold distance from the diaphragm was associated with fibrotic ILAs (OR 1.15, 95% CI 1.02–1.29) (table 2). There were 1190 participants with HH assessments from Exam 1 and 5 scans who had Exam 5 ILA assessments. HH detected between Exams 1 and 5 scans was associated with an OR for fibrotic ILAs of 3.51 (95% CI 0.98–12.52) compared with those without HH (table 2).

TABLE 2.

Associations of hiatus hernia (HH) with interstitial lung abnormalities (ILAs)

Model	Participants (n)	Exam 5 ILAs		Exam 5 fibrotic ILAs
		OR (95% CI)	p-value	OR (95% CI)	p-value
HH at Exam 5
No HH	2139	1.00 (reference)		1.00 (reference)
HH present	241	1.16 (0.79–1.69)	0.46	1.98 (0.64–6.06)	0.23
HH type at Exam 5
No HH	2147	1.00 (reference)		1.00 (reference)
Type I HH present	170	1.15 (0.74–1.79)	0.54	1.87 (0.50–6.91)	0.35
Type III HH present	63	1.29 (0.66–2.50)	0.46	2.60 (0.48–14.03)	0.27
Distance between gastric fold and diaphragm at Exam 5 #	2380	1.02 (0.98–1.07)	0.30	1.15 (1.02–1.29)	0.02
Serial HH assessment between Exams 1 and 5 scans
No HH on Exams 1 and 5 scans	1026	1.00 (reference)		1.00 (reference)
HH present on Exam 5 scan	164	1.33 (0.84–2.12)	0.22	3.51 (0.98–12.52)	0.05

Models adjusted for Multi-Ethnic Study of Atherosclerosis Exam 5 age, sex, race/ethnicity, smoking status, cigarette pack-years, height and weight.

^#:results reported per 20% increase in distance between the gastric fold and diaphragm.

HAA and ILA findings were similar from a sensitivity analysis in which HH was not re-adjudicated based on prior Exam 1 scan (supplementary table E5). Smoking status, sex, BMI and GOR medication use did not modify associations of HH with HAAs and ILAs (supplementary tables E6 and E7). The MUC5B (rs35705950) risk allele did not modify associations between HH and ILAs (supplementary table E8).

Associations of HH with HAAs and ILAs were similar after adjustment for different parameters related to body habitus (height, weight, BMI and waist circumference) and for FVC (supplementary tables E9 and E10).

Results in secondary analyses using cardiac CT measures yielded qualitatively similar cross-sectional results. The presence of HH on cardiac CT scan was associated with 2.41% (95% CI −0.29–5.18%) greater HAAs after adjustment for covariates (supplementary table E11). HH was associated with more HAAs among H2-antagonist users compared with nonusers (p-value for use interaction=0.03). Associations between HH at Exam 1 and change in HAAs on cardiac CT were nonsignificant (supplementary table E12). Stratified analyses are shown in supplementary tables E11 and E12.

Based on our finding that HH was associated with ILAs in younger adults and carriers of the MUC5B risk allele appeared to have increased lung densities over time, we examined the association between Exam 1 HH and death by MUC5B risk allele carrier status among participants <80 years old. Participants with HH at Exam 1 and carriers of the MUC5B risk allele (T) had the highest event rate for death (27.5 per 1000 person-years) compared with other participants (table 3). Among participants with at least one copy of the (T) allele, HH was associated with a hazard ratio of 2.05 (95% CI 1.12–3.74) compared with 1.06 (95% CI 0.79–1.41) among noncarriers (p-value for MUC5B interaction=0.05) (figure 3 and table 3). Associations were much weaker in the overall cohort (supplementary table E13).

TABLE 3.

Associations of Exam 1 hiatus hernia (HH) with death by MUC5B (rs35705950) risk allele (T) carrier status

Carrier  status	Participants (n)	Total person-years	Events (n)	Event rate per 1000 person-years	Hazard ratio (95% CI)	p-value
GG
No HH	2421	36 999	674	18.2 (16.7–19.6)	1.00 (reference)
HH	169	2474	52	21.0 (15.9–27.4)	1.06 (0.79–1.41)	0.70
GT/TT
No HH	310	4830	71	14.7 (11.6–18.4)	1.00 (reference)
HH	33	472	13	27.5 (15.3–45.9)	2.05 (1.12–3.74)	0.02

Hazard ratios are reported among those with HH compared with those without HH (reference group). Cox regression model in which the primary exposure variable is HH. Adjusted for baseline sex, race/ethnicity, smoking status, cigarette pack-years, percent emphysema, education attainment, height, weight, statin use, blood pressure medication use, systolic and diastolic blood pressure, Agatston coronary artery calcium score, diabetes, total intentional exercise, history of cancer, study site, principal components of genetic ancestry, MUC5B (rs35705950) carrier status (GG versus GT/TT), and interaction term “HH×MUC5B carrier status”. p-value for MUC5B (rs35705950) interaction=0.05. Analysis restricted to participants <80 years old at baseline.

FIGURE 3.

Kaplan–Meier survival curves of Multi-Ethnic Study of Atherosclerosis Exam 1 hiatus hernia (HH) status and MUC5B (rs35705950) risk allele (T) carrier status.

FVC

There were 1501 participants with HH assessment at Exam 5 and FVC measured at both Exams 5 and 6. HH was associated with a decrease of 346.78 (95% CI 171.68–521.88) mL per year compared with those without HH (supplementary table E14). This association was attenuated after adjustment for covariates.

Serum MMP-7 and SP-A

There were 658 Exam 1 participants who had concurrent HH assessments and serum MMP-7 and SP-A measurements. The mean±sd MMP-7 levels were 5.1±3.5 and 4.4±2.9 ng·mL⁻¹ among those with and without HH, respectively. The mean±sd SP-A levels were 95.4±138.6 and 77.2±102.9 ng·mL⁻¹ among those with and without HH, respectively. HH was not associated MMP-7 after adjustment for covariates (supplementary table E15). HH was associated with a 13.5% (95% CI 0.4–28.2%) increment in MMP-7 among ever-smokers, while there was no significant association in never-smokers (−5.4%, 95% CI −4.7–16.7%) (p-value for smoking interaction=0.04). HH was not associated with SP-A.

Discussion

A radiologically detected HH was associated with more HAAs and their increase over time on full-lung CT scans in a general population-based sample of US adults. This association was strongest among carriers of the MUC5B (rs35705950) risk allele. Among younger adults, HH was associated with ILAs and death in carriers of the MUC5B risk allele and higher MMP-7 serum levels among ever-smokers.

Although studies suggest recurrent reflux of acidic and bile components can be harmful to the lungs, it has been alternatively hypothesised that pulmonary fibrosis can promote GOR due to impaired oesophageal sphincter function related to reduced elasticity and increased negative intrathoracic pressure [26, 27]. In the WRAP-IPF trial, 85% of participants with IPF randomised to the surgical group had HH [28]. While FVC has been used as a primary end-point in ILD-related clinical trials, and is a validated measurement of disease severity and progression, its utility in earlier stages of ILD remains unknown as HH was not significantly associated with FVC decline in our study. A potential reason may be that the thoracic space is occupied by gastrointestinal contents from the HH that reduces lung volumes rather than a restrictive process related to ILD. Therefore, FVC may not be a reliable outcome to examine the potential relationship between HH and earlier stages of ILD, which makes CT an appealing measure.

While HAA histopathological correlation is unknown, HAAs are strongly associated with peripheral biomarkers related to pulmonary fibrosis, presence of ILAs, and ILD-related hospitalisation and death [8, 23, 29]. In addition to HH overall, type I HH and a longer distance between the gastric fold and diaphragm detected from Exam 5 full-lung scans were each associated with more HAAs and their progression over time. These findings suggest subgroups of individuals with certain characteristics of hernias that are promotive of reflux may be vulnerable to early repetitive lung injury.

Age was a significant confounder in our cross-sectional analysis of HH and ILAs. The MESA cohort is slightly younger than a previous study that did not find an association between HH and ILAs [30, 31]. Our findings suggest HH may be more relevant to the development of ILAs in younger adults prior to overlapping contributions by ageing-related factors. Although the detection of HH over time was associated with fibrotic ILAs, we acknowledge possible type I error and that this is speculative given the exploratory and post hoc nature of these findings after our initial ILA analysis.

We considered whether associations between HH and HAAs could be mediated by reduced lung volumes [32]. However, even with adjustments for total lung volume imaged, height and weight in our regression models, the association persisted. Associations between HH and HAAs from cardiac CT scans were attenuated compared with the main results using full-lung scans, which may be due to reduced accuracy of measurement of HAAs on cardiac CT [15]. Replication of our findings using full-lung CT in other independent cohorts and utilising more sophisticated, texture-based CT lung measures will be important [33, 34].

The association between HH and greater HAA progression over time was stronger among carriers of the MUC5B (rs35705950) risk allele (T) [35]. This finding supports prior hypotheses that a genetic predisposition to pulmonary fibrosis combined with external risk factors may lead to repetitive lung injury and aberrant healing [5, 6]. HH was also associated with worse survival among younger adults who carried the MUC5B risk allele. A possible explanation for this finding may be that younger individuals with a combination of HH and underlying genetic risk may be predisposed to develop ILD, which itself is associated with worse overall survival. We acknowledge that this is speculative and hypothesis generating. We had limited data that captured ILD-related clinical outcomes and this is a major knowledge gap in large, prospective cohorts that necessitates further research.

We also observed an association between HH and serum MMP-7 levels among those with a history of smoking. The combination of cigarette smoking and recurrent injury triggered by HH may lead to increased extracellular matrix turnover in the lung leading to higher serum MMP-7 levels. We caution that this interpretation is speculative as this was a subgroup analysis of a secondary outcome without repeated measurements of MMP-7 over time.

GOR medications may reduce acid secretion, but may not reduce nonacidic reflux, which may be injurious to the lung and may explain why studies that have examined their potential utility in pulmonary fibrosis have been conflicting [25, 36]. Our stratified analysis by GOR medication use does not suggest a protective role of these medications in adults with HH, although we did not have dosage or duration of medication information. Although H2-antagonist users with HH had more HAAs compared with nonusers in Exam 1, we caution over-interpretation because this was not observed in other HAA and ILA analyses.

This study has several limitations. The prevalence of radiologically detected HH in this study was expectedly lower compared with IPF cohorts, which may be due to the fact that MESA is a relatively healthy, population-based cohort of community-dwelling adults [2-4]. The prevalence of HH in MESA was also lower compared with the Age Gene/Environment Susceptibility (AGES)-Reykjavik study, which may be due to demographic differences (AGES-Reykjavik participants were older on average than MESA participants) and HH detection protocols [30]. Future studies using similar protocols will be important to validate our findings in independent cohorts. We did not have repeat assessments of ILAs available at the time of this analysis to examine associations with ILA progression. Participants did not undergo endoscopic procedures for research purposes and medical records to adjudicate prior endoscopies to determine the extent of the HH were not available. We did not have access to surgical records to confirm whether MESA participants underwent surgery for HH. We cannot completely rule out potential bias from single-reader interpretations of many scans, although interobserver agreement in a subset of scans was high among the three readers. Although we tried to account for lung volume in our statistical models, we cannot completely rule out its potential confounding effect on the associations we observed. Our study population was restricted to individuals older than 45 years, and further studies are needed to examine HH and interstitial lung changes in younger adults. We acknowledge our findings may be driven by type I error, particularly our mortality analysis, as prospective studies in population-based cohorts with a priori hypotheses related to HH is a future research objective.

In summary, HH is associated with more HAAs and progression on CT among community-dwelling adults and ILAs among younger adults. Further investigation is needed to determine the underlying biological pathways and possible effect of therapies for preventing the development and progression of fibrosing ILD.