Long-term exposure to air pollution and prevalent nonalcoholic fatty liver disease

Abstract

Background:

Nonalcoholic fatty liver disease (NAFLD) is a disease characterized by lipid accumulation within hepatocytes, ranging from simple steatosis to steatohepatitis, in the absence of secondary causes of hepatic fat accumulation. Although air pollution (AP) has been associated with several conditions related to NAFLD (e.g., metabolic syndrome, type 2 diabetes mellitus), few studies have explored an association between AP and NAFLD. The aim of the study was to investigate whether exposure to AP is associated with NAFLD prevalence.

Methods:

We used baseline cross-sectional data (2000–2003) of the Heinz–Nixdorf–Recall cohort study in Germany (baseline n = 4,814), a prospective population-based cohort study in the urbanized Ruhr Area. Mean annual exposure to size-fractioned particulate matter (PM₁₀, PM_2.5, PM_coarse, and PM_2.5abs), nitrogen dioxide, and particle number was assessed using two different exposure models: a chemistry transport dispersion model, which captures urban background AP exposure on a 1 km² grid at participant’s residential addresses, and a land use regression model, which captures point-specific AP exposure at participant’s residential addresses. NAFLD was assessed with the fatty liver index (n = 4,065), with NAFLD defined as fatty liver index ≥60. We estimated ORs of NAFLD per interquartile range of exposure using logistic regression, adjusted for socio-demographic and lifestyle variables.

Results:

We observed a NAFLD prevalence of 31.7% (n = 1,288). All air pollutants were positively associated with NAFLD prevalence, with an OR per interquartile range for PM_2.5 of 1.11 (95% confidence interval [CI] = 1.00, 1.24) using chemistry transport model, and 1.06 (95% CI = 0.94, 1.19) using the land use regression model, respectively.

Conclusion:

There was a positive association between long-term AP exposure and NAFLD.

What this study adds

Nonalcoholic fatty liver disease (NAFLD) is affecting about 25% of the global population and air pollution is a severe environmental health risk. We conducted an epidemiological study on the potential effect of exposure to air pollution on the risk of NAFLD. We found an association between several air pollutants and NAFLD prevalence with the most consistent associations found for PM_2.5 (particulate matter, aerodynamic diameter <2.5 µm) and NAFLD prevalence. Further, local traffic-specific PM exposure is more strongly associated with NAFLD than industry-specific PM exposure.

Introduction

Nonalcoholic fatty liver disease (NAFLD) is an underestimated public health problem, with an estimated prevalence of 25% in the global general population.¹ NAFLD is characterized by lipid accumulation within hepatocytes and is associated with obesity, insulin resistance, type 2 diabetes mellitus (T2DM), hypertension, and the metabolic syndrome.¹ NAFLD is an umbrella term, incorporating both hepatic steatosis (HS) and nonalcoholic steatohepatitis (NASH). NAFLD increases the susceptibility of the liver to acute liver injury, potentially contributing to the progression of NASH to cirrhosis, hepatocellular carcinoma, and the need for liver transplantation.^1,2 For the diagnosis of NAFLD, there must be: (1) evidence of HS and (2) lack of secondary causes of hepatic fat accumulation such as significant alcohol consumption, viral infection, or hereditary diseases.³

Because NAFLD increases a person’s risk for many noncommunicable diseases, prior studies aimed at identifying modifiable risk factors for NAFLD.^3,4 Environmental exposures, mainly metals, such as lead and arsenic and soil heavy metals have been investigated as modifiable environmental risk factors to develop NAFLD in epidemiological studies.^5–7 Environmental exposures, such as air pollution (AP), another potential modifiable risk factor for NAFLD, has been investigated primarily in toxicological studies, even though several epidemiological studies describe an association between AP and a wide range of metabolic health outcomes related to NAFLD.^8–10 Although initial studies mainly focused on the association of AP with cardiopulmonary diseases and mortality, recent epidemiological studies also suggested that exposure to major air pollutants increases the risk of metabolic diseases such as T2DM and insulin resistance.^9,11–13 Moreover, some epidemiological studies showed an association between long-term AP and metabolic syndrome, but to our knowledge, few epidemiological studies looked at an association between AP and NAFLD.^8,14–22

The aim of this study was to investigate whether long-term exposure to AP is associated with NAFLD prevalence in the general adult population, using data from the population-based prospective Heinz–Nixdorf–Recall (HNR) cohort study in Germany, an ongoing prospective population-based cohort study.

Materials and methods

Study design

This study was based on cross-sectional data from the baseline examination (t₀: 2000–2003) of the well-characterized “Heinz–Nixdorf–Recall Study” (risk factors, evaluation of coronary calcium and lifestyle), a prospective population-based cohort study located in three adjacent cities (Bochum, Essen, and Mülheim/Ruhr) in the highly urbanized German Ruhr Area.²³ The rationale and design of the study have been described in detail elsewhere.²³ In short, a sample of individuals aged 45–75 years were identified through a random selection process using local residency registries and a total of 4,814 participants were enrolled into the HNR study between December 2000 and August 2003 (recruitment efficacy proportion: 55.8%). Assessment included a self-administered questionnaire, face-to-face interviews for personal risk factor assessment, clinical examinations, and comprehensive laboratory tests following standard protocols. All examinations were conducted in accordance with the recommendations for research on human subjects by the 18th World Medical Assembly’s revised Declaration of Helsinki and were approved by the institutional ethics committees of the University of Duisburg-Essen and the University Hospital of Essen and adhered to strict internal and external quality assurance protocols. All participants were extensively informed about the goals of the study and gave their written informed consent.

Environmental exposures

Two different exposure models were used for exposure assessment, the European air pollution dispersion (EURAD) chemistry transport model (CTM) and the European study of cohorts for air pollution effects (ESCAPE) land use regression (LUR) model.

For both exposure models, particulate matter (PM) was categorized into different sizes, PM₁₀ (aerodynamic diameter <10 µm) and PM_2.5 (aerodynamic diameter <2.5 µm). For the ESCAPE-LUR model, PM was further divided into PM_coarse (2.5< aerodynamic diameter <10 µm), and a measurement of the blackness of PM_2.5 filters (PM_2.5abs) as a proxy for soot and black carbon.^8,24 For EURAD-CTM, particle number (PN) was used as a proxy for ultrafine particles. In addition, for EURAD-CTM source-specific AP exposure PM_10IND, PM_2.5IND, PM_10TRA, and PM_2.5TRA was used, IND stands for industry and TRA stands for traffic.

The EURAD-CTM model is a validated, time-dependent, three-dimensional chemistry transport model used to predict daily concentrations of air pollutants on a horizontal grid resolution of 1 km².^25,26 Input variables include transport and industry, and employs four sequential nesting grid sizes (125, 25, 5, and 1 km). All participants of HNR were assigned the daily mean PM, PN, and NO₂ concentrations of the 1 km² grid cell corresponding to his/her given residential address.²⁵ From these daily values, concentrations for longer exposure periods were calculated. For this study, mean exposure concentrations for the years 2001–2003 (3-year average) were used.

We assessed traffic-specific PM in a two-step process: First, we set the contribution of local traffic source to zero and conducted model runs without the contribution of local traffic (PM_noTRA). Second, we calculated the traffic-specific contribution by subtracting the results of the model run without traffic sources (PM_noTRA) from the results of the model run with all contributions (total PM or PM_ALL): PM_ALL − PM_noTRA = PM_TRA.^25,27 The same method was used for industry-specific PM (PM_ALL − PM_noIND = PM_IND).

As a second approach, we used the LUR model according to the ESCAPE standardized procedure.^24,28 In short, PM of varying sizes (PM₁₀, PM_2.5, and PM_coarse) and PM_2.5abs was measured at 20 sites, and NO₂ at 40 sites in three separate 2-week periods, to cover different seasons, over 1 year (October 2008–October 2009).²⁴ The resulting LUR models were applied to estimate long-term exposure concentrations at the baseline year of the study.^24,29 The annual averages of the measured pollutant concentrations at the monitoring sites and predictor variables were used to develop the study-specific LUR model and to predict concentrations at each participant’s address.⁸

Outcome

Recently, a new nomenclature and definition for NAFLD has been established—“metabolic dysfunction-associated steatotic liver disease” (MASLD), since many experts had argued that NAFLD and its diagnostic criteria must be updated, and a new set of diagnostic criteria was proposed.^30–32 The criteria included evidence of HS in the presence of at least one of five cardiometabolic risk factors.³⁰ In this study, however, we will still focus on the definition for NAFLD.

We assessed NAFLD prevalence at baseline using the fatty liver index (FLI, main analysis) and the fibrosis-4 (FIB-4) index (secondary analysis). Exclusion criteria for the analysis were (1) participants with an ongoing or likely ongoing HBV and/or HCV infection, (2) participants with a diagnosis of HIV, and (3) participants with “significant alcohol consumption,” defined as >20 g of alcohol/day for women and >30 g of alcohol/day for men.

The FLI is a combination of anthropometric measurements (body mass index, waist circumference) and laboratory results (triglycerides [TG], gamma-glutamyl transferase). It was developed in the general population using a stepwise logistic regression analysis to obtain a final prediction model.³³ It varies between 0 and 100. FLI <30 rules out HS and FLI ≥60 is defined as HS. For this study, we use FLI as a surrogate marker for the diagnosis NAFLD. NAFLD was defined as FLI ≥60.

$$\text{FLI=}\frac{\left(e^{0.953\times \text{ln }\left(\text{TG}\left[\frac{\text{mg}}{\text{dL}}\right]\right)+0.139\times \text{BMI+0}\text{.718}\times \text{ln }\left(\text{GGT}\left[\frac{\text{U}}{\text{l}}\right]\right)+0.053\times (\text{Waist circumference)}-15.754}\right)}{\left(1+e^{0.953\times \text{ln }\left(\text{TG}\left[\frac{\text{mg}}{\text{dL}}\right]\right)+0.139\times \text{BMI+0}\text{.718}\times \text{ln }\left(\text{GGT}\left[\frac{\text{U}}{\text{l}}\right]\right)+0.053\times (\text{Waist circumference)}-15.754}\right)}\times 100.$$

FIB-4: The FIB-4 index was developed as a noninvasive panel to stage liver disease in patients with HCV. It was also shown to be an important noninvasive predictor for the development of liver cancer, and risk prediction for advanced fibrosis in patients with NAFLD.^34,35 FIB-4 is a formula used to estimate fibrosis stage based on four parameters (aspartate transaminase, alanine aminotransferase, platelet count, and age), and can be used for risk prediction in patients with NAFLD.³⁴ For this study, an increased risk of advanced fibrosis was defined as (1) FLI ≥60 and FIB-4 ≥1.3 when participants were <65 years or (2) FLI ≥60 and FIB-4 ≥2.0 when participants were ≥65 years.

$$\text{FIB-4=}\frac{\text{Age [years] }\times \text{ AST level [U/L) }}{\text{Platelet count [109/L] }\times \sqrt{\text{ALT level [U/L]}}}$$

Assessment of outcome variables and covariates

FLI and FIB-4

Anthropometric measurements (height, weight, and waist circumference) were determined according to standard protocols at baseline. TGs and gamma-glutamyl transferases were measured at the baseline examination with participants being advised to fast before the examination visit, and information on the time since the last meal was collected at time of blood draw. All above-mentioned analyses were performed in the central laboratory of the University Hospital of Essen following a standard procedure.³⁶

Covariates

Potential confounding factors were individual-level characteristics such as age, sex, socioeconomic status (SES), and lifestyle variables (smoking, diet, physical activity, and alcohol consumption), and were assessed in standardized interviews and self-administered questionnaires. Individual SES was defined as years of formal education according to the “International Standard Classification of Education” (UNESCO 1997) as total years of formal education. Smoking status was defined as current, former (>1 year since quitting), or never smoker. Cumulative smoking was assessed using pack-years for current and former smokers, whereas environmental tobacco smoke (Yes/No) reflected any passive exposure to smoke at home, work, or other location. Diet was calculated via a nutrition index, using a qualitative food frequency questionnaire with information on the consumption frequency of 13 food items.³⁷ Frequency of the consumption of the food items was categorized into “hardly ever/never,” “1–3 times/month,” “1–3 times/week,” “4–6 times/week,” and “daily,” and then given points between 1 and 5. By summing up the points for these food items, a score (dietary pattern index [Ernährungsmusterindex]) was calculated with a range from 0 to 26, with a high score indicating better accordance to the recommendations of the “German Society of Nutrition.”³⁷ Two different physical activity measures were used, one according to self-reported activity (sport—Yes/No), and one using self-reported units of physical activity in a typical week (U/wk). Alcohol consumption was measured in grams of pure alcohol per day (g/d), according to information on self-reported alcohol consumption. To capture well-known effects of neighborhood conditions on disease risk, neighborhood SES was assessed as neighborhood unemployment rate (%). For this, the study area was divided into 106 neighborhoods with a median size of 11,263 inhabitants.³⁸

Statistical analysis

We excluded participants with missing information on exposure, outcome, or covariate data, and participants with exclusion criteria for the outcome definition for NAFLD (see also Supplementary Table 1; http://links.lww.com/EE/A238). We calculated descriptive statistics on the main study population and compared population characteristics of participants with versus without NAFLD, and included versus excluded participants. We also calculated Spearman correlation coefficients to examine correlations between the different air pollutants.

We analyzed the associations between AP exposure (PM₁₀, PM_2.5, PN, NO₂, PM_10IND, PM_10TRA, PM_2.5IND, and PM_2.5TRA) and prevalent NAFLD using binary logistic regression. We controlled for potential confounding during the analysis of the data with incrementally adjusted models. One model (model 1 directed acyclic graph [DAG]) was identified based on the concept of causality via a DAG that enabled the identification of a minimal sufficient adjustment set, and two models (model 2 and 3) were based on knowledge from similar studies.³⁹ In model 1 (DAG), we adjusted for individual and neighborhood SES. In model 2, we additionally adjusted for sex and age. In model 3 (main model), we additionally adjusted for lifestyle variables, including smoking status, pack-years, environmental tobacco smoke, diet (Dietary Pattern Index [Ernährungsmusterindex]), physical activity, and alcohol consumption. All models were conducted for each air pollutant exposure.

In secondary analyses, we analyzed the association between AP exposure and FIB-4 (increased risk of advanced fibrosis). We calculated Cook’s distance to find the most influential observations (outliers) and removed the observations that had a cook’s distance greater than four times the mean.⁴⁰

Effect estimates for the air pollutants PM₁₀, PM_2.5, PN, and NO₂ were expressed in OR per interquartile range increase in AP exposure and corresponding 95% confidence interval (CI), with the air pollutants as continuous exposure variables. For the source-specific air pollutants, effect estimates were expressed in OR per 1 µg/m³ increase in AP exposure.

Sensitivity analyses

For FLI as an outcome, we also included one analysis using the log-binomial regression instead of the logistic regression to prevent overestimation of the relative risk with the odds ratio, given a frequent outcome. Further, we made one analysis using FLI as a continuous outcome, using a beta-binomial regression model with a logit link function to account for the non-normal distribution of the outcome variable, where values are probabilities and bounded between 0% and 100%. In further sensitivity analyses, we added distance to major road as a potential confounding factor to model 3 for EURAD-CTM to account for differences of AP exposure within the 1 km² grid cell. In another sensitivity analyses, we added noise as a potential confounding factor to model 3, since chronic noise exposure may represent another environmental risk factor that has common sources with AP (e.g., traffic).

Further, we analyzed potential outcome misclassification. First, we repeated the analysis, using a stricter cutpoint for “significant alcohol consumption,” excluding participants with an alcohol intake of >10 g/day for women and >20 g/day for men. We then excluded participants taking lipid-lowering drugs in one analysis, and excluded participants taking drugs that potentially cause drug-induced liver injury in another analysis. Since TGs are dependent on participants’ fasting state, we excluded nonfasting participants in one analysis. Further, since Bedogni et al.³³ state, that FLI ≥60 rules in fatty liver and that FLI <30 rules out fatty liver, we conducted one analysis excluding participants with a medium FLI score that does not reliably inform about the existence of NAFLD (30≤ FLI <60).

Since HS might diminish with increasing fibrosis, some participants with risk of advanced fibrosis (high FIB-4) might not have been identified with the FLI ≥60 criterion. We therefore carried out one analysis reducing the outcome definition to: (1) FIB-4 ≥1.3 when participants were <65 years or (2) FIB-4 ≥2.0 when participants were ≥65 years.

All statistical analyses and processing of the data were conducted in RStudio version 4.0.3.

Results

Study population

A total of 4,065 participants were available for the cross-sectional analysis with a NAFLD prevalence of 31.7% (n = 1,288). Participants were excluded from the analysis if they fulfilled the exclusion criteria for the outcome definition or if exposure, outcome, or covariate data were missing (n = 749, Supplementary Figure 1; http://links.lww.com/EE/A238).

Mean (±SD) age of participants was 59.6 (±7.8), women were slightly more represented than men (women 53%) (Table 1). Participants with NAFLD (FLI ≥60) were more likely to be older, male, to have more pack-years of smoking, a higher body mass index and to drink more alcohol than participants without NAFLD (Table 1).

Table 1.

Baseline characteristics of the study population using FLI as an outcome stratified by FLI (low vs. high), Ruhr Area, Heinz–Nixdorf–Recall study, 2000–2003.

Variable	Total study population, n = 4,065	FLI
		<60 n = 2,777	≥60 n = 1,288
Age (years), mean ± SD	59.6 ± 7.8	59.1 ± 7.8	60.6 ± 7.6
Sex, women, n (%)	2,157 (53.1)	1,685 (60.7)	472 (36.7)
BMI, (Kg/m2), mean ± SD	27.9 ± 4.7	25.9 ± 3.1	32.3 ± 3.0
Neighborhood unemployment (%), mean ± SD	6.7 ± 2.1	6.7 ± 2.1	6.9 ± 2.2
Education, n (%)
<11 (years)	455 (11.2)	288 (10.371)	167 (12.966)
11–13 (years)	2,304 (56.7)	1,597 (57.508)	707 (54.891)
14–17 (years)	892 (21.9)	570 (20.526)	322 (25)
>17 (years)	414 (10.2)	322 (11.595)	92 (7.143)
Pack-years for former and current smokers, median (IQR)	20.5 (28.5)	18.6 (26.2)	25.9 (33.0)
Smoking status
Never smoker, n (%)	1,783 (43.9)	1,291 (46.5)	492 (38.2)
Former smoker, n (%)	1,371 (33.7)	852 (30.7)	519 (40.3)
Current smoker, n (%)	911 (22.4)	634 (22.8)	277 (21.5)
Exposure to ETS, n (%)	1,426 (35.1)	958 (34.5)	468 (36.3)
Physical activity (U/wk), median (IQR)	0.80 (3.0)	1.00 (3.5)	0.00 (2.3)
Sport, yes, n (%)	2,222 (54.7)	1,611 (58.0)	611 (47.4)
Nutrition index (EMI), mean ± SD	12.8 ± 3.1	12.9 ± 3.1	12.4 ± 3.1
Alcohol consumption (g/d), median (IQR)	1.1 (7.1)	1.0 (6.0)	2.0 (7.9)
Aspartate aminotransferasea mean ± SD	12.8 ± 4.3	12.1 ± 3.2	14.2 ± 5.8
Alanine aminotransferasea mean ± SD	16.1 ± 8.2	14.0 ± 6.2	20.5 ± 10.1

^aIncluded in secondary analyses not included in main analyses, missing AST n = 3, missing ALT n = 1.

ETS indicates environmental tobacco smoke; FLI, fatty liver index.

Participants were excluded from the analyses due to incomplete data (n = 139) or NAFLD exclusion criteria (n = 610), and differed from included participants (n = 4,065) in several ways: more men than women were excluded, they were more often prior or current smoker and less physical active (Supplementary Table 1; http://links.lww.com/EE/A238).

At baseline, mean PM₁₀, PM_2.5, and NO₂ exposures for EURAD-CTM were 21.3, 17.7, and 41.4 µg/m³, respectively (Table 2). The correlation coefficients for the AP of EURAD-CTM and the ESCAPE-LUR exposure model were all positively correlated, except PM_IND, who were slightly negatively associated with PM_coarse (Spearman correlation coefficients [r], −0.02 to −0.03) (Supplementary Table 2; http://links.lww.com/EE/A238). Within the EURAD exposure model, all AP were moderately (r = 0.48–0.62) or strongly (r = 0.71–0.97) correlated, with the exception of PM_TRA with PM_IND, where a weak correlation was observed (r = 0.27–0.28). Also within the ESCAPE-LUR model, APs were either moderately (r = 0.54–0.70) or strongly (r = 0.898–0.90) correlated. Between the exposure models (ESCAPE-LUR and EURAD-CTM), a weak to moderate correlated of air pollutants was seen, with the exception of PN and PM_2.5 (ESCAPE-LUR model), where a strong correlation was observed (r = 0.73).

Table 2.

Description of air pollution exposures for EURAD-CTM (2001–2003, 3-year mean) and ESCAPE-LUR (2008–2009, annual mean) of the study population using FLI as an outcome.

	IQR	Min	Q1	Median	Q3	Max	Mean ± SD
EURAD-CTM
PM10 (µg/m3)	4.2	17.3	18.9	21.4	23.0	29.5	21.3 ± 2.6
PM2.5 (µg/m3)	2.1	15.4	16.7	17.7	18.7	22.1	17.7 ± 1.3
PN (n/mL)	26,558	50,730	74,997	87,037	101,554	184,618	88,568 ± 18,513
NO2 (µg/m3)	5.1	29.5	38.7	41.5	43.9	53.8	41.4 ± 4.0
PM10IND (µg/m3)	2.2	0.69	1.33	2.07	3.50	8.08	2.46 ± 1.35
PM10TRA (µg/m3)	0.3	0.25	0.66	0.84	0.99	1.71	0.84 ± 0.24
PM2.5IND (µg/m3)	1.4	0.53	0.95	1.49	2.34	4.97	1.69 ± 0.87
PM2.5TRA (µg/m3)	0.3	0.26	0.67	0.84	0.98	1.69	0.85 ± 0.24
ESCAPE-LUR
PM10 (µg/m3)	2.1	23.9	26.6	27.5	28.7	34.7	27.8 ± 1.9
PM2.5 (µg/m3)	1.5	16.0	17.7	18.3	19.1	21.5	18.4 ± 1.1
PMcoarse (µg/m3)	1.9	0.8	9.2	10.1	11.1	15.0	10.0 ± 1.9
PM2.5abs  (0.0001/m)	0.4	1.0	1.4	1.5	1.7	5.4	1.6 ± 0.4
NO2 (µg/m3)	6.2	19.8	26.9	29.6	33.1	59.1	30.3 ± 4.9

Ruhr Area, Germany, Heinz–Nixdorf–Recall study, n = 4,065.

ESCAPE-LUR indicates European study of cohorts for air pollution effects land use regression; EURAD-CTM, European air pollution dispersion chemistry transport model; FLI, fatty liver index; PM, particulate matter; PN, particle number.

Main analyses

In general, air pollutants were positively associated with NAFLD, with more consistent and statistically significant associations observed for urban background AP modeled by EURAD-CTM (Tables 3 and 4). These associations weakened slightly upon increasing covariate adjustment. In model 3, all air pollutants except PM₁₀ (ESCAPE-LUR) showed positive effect estimates with ORs per interquartile range between 1.09 and 1.11 for EURAD-CTM and between 1.02 and 1.06 for ESCAPE-LUR.

Table 3.

Logistic regression analysis of the associations between air pollutants and prevalent NAFLD per IQR at baseline using EURAD-CTM and FLI as a binary outcome.

	IQR	Crudea OR (95% CI)	Model 1 (DAG)b OR (95% CI)	Model 2c OR (95% CI)	Model 3d OR (95% CI)
PM10 (µg/m3)	4.2	1.15 (1.04, 1.28)	1.13 (1.01, 1.25)	1.12 (1.00, 1.25)	1.11 (0.99, 1.24)
PM2.5 (µg/m3)	2.1	1.11 (1.01, 1.23)	1.12 (1.01, 1.24)	1.11 (1.00, 1.24)	1.11 (1.00, 1.24)
PN (n/mL)	26,558	1.18 (1.07, 1.29)	1.12 (1.01, 1.24)	1.11 (1.00, 1.23)	1.10 (1.00, 1.22)
NO2 (µg/m3)	5.1	1.12 (1.03, 1.22)	1.09 (0.99, 1.19)	1.10 (1.00, 1.20)	1.09 (1.00, 1.20)

Ruhr Area, Germany, Heinz–Nixdorf–Recall study, n = 4,065.

^a+Air pollutant.

^bCrude + individual and neighborhood socioeconomic status.

^cModel 1 + Age, sex.

^dModel 2 + Lifestyle variables (smoking status, cumulative smoking, environmental tobacco smoking, physical activity, nutrition, and alcohol consumption).

EURAD-CTM indicates European air pollution dispersion chemistry transport model; FLI, fatty liver index; NAFLD, nonalcoholic fatty liver disease; PM, particulate matter, PN, particle number.

Table 4.

Logistic regression analysis of the associations between air pollutants and prevalent NAFLD at baseline per IQR using ESCAPE-LUR and FLI as a binary outcome.

	IQR	Crudea OR (95% CI)	Model 1 (DAG)b OR (95% CI)	Model 2c OR (95% CI)	Model 3d OR (95% CI)
PM10 (µg/m3)	2.1	1.07 (0.99, 1.15)	1.00 (0.92, 1.09)	1.01 (0.93, 1.10)	1.00 (0.91, 1.09)
PM2.5 (µg/m3)	1.5	1.15 (1.05, 1.26)	1.06 (0.95, 1.19)	1.07 (0.96, 1.21)	1.06 (0.94, 1.19)
PMcoarse (µg/m3)	1.9	1.08 (1.01, 1.16)	1.04 (0.97, 1.12)	1.03 (0.96, 1.12)	1.04 (0.96, 1.12)
PM2.5 abs (0.0001/m)	0.4	1.07 (1.01, 1.14)	1.03 (0.96, 1.10)	1.03 (0.96, 1.11)	1.02 (0.95, 1.09)
NO2 (µg/m3)	6.2	1.11 (1.02, 1.20)	1.05 (0.96, 1.15)	1.04 (0.95, 1.14)	1.03 (0.93, 1.13)

Ruhr Area, Germany, Heinz–Nixdorf–Recall study, n = 4,065.

^a+Air pollutant.

^bCrude + individual and neighborhood socioeconomic status.

^cModel 1 + Age, sex.

^dModel 2 + Lifestyle variables (smoking status, cumulative smoking, environmental tobacco smoking, physical activity, nutrition, and alcohol consumption).

ESCAPE-LUR indicates European study of cohorts for air pollution effects land use regression; FLI, fatty liver index, NAFLD, nonalcoholic fatty liver disease; PM, particulate matter; PN, particle number.

The association between source-specific traffic PM per 1-µg/m³ increase (PM_10TRA and PM_2.5TRA) was stronger than for industry-specific PM (PM_10IND and PM_2.5IND), but confidence intervals were large and included the null (Figure 1 and Supplementary Table 3; http://links.lww.com/EE/A238). For example, for model 3 the OR per 1-µg/m³ increase for PM_2.5IND was 1.07 (95% CI: 0.98, 1.16) and for PM_2.5TRA was 1.28 (95% CI: 0.91, 1.79), respectively.

Figure 1.

Logistic regression analysis of the associations between source-specific air pollutants and prevalent NAFLD at baseline per 1 µg/m³ for PM industry (IND) and traffic (TRA) using EURAD-CTM and FLI as a binary outcome. Ruhr Area, Germany, Heinz–Nixdorf–Recall Study, n = 4,065. Crude—Air pollutant; Model 1—Crude + individual and neighborhood socioeconomic status, Model 2—Model 1 + Age, sex; Model 3—Model 2 + Lifestyle variables (smoking status, cumulative smoking, environmental tobacco smoking, physical activity, nutrition, and alcohol consumption). EURAD-CTM indicates European air pollution dispersion chemistry transport model; FLI, fatty liver index; NAFLD, nonalcoholic fatty liver disease; PM, particulate matter.

Secondary analysis

For FIB-4, 4,029 participants were available for the cross-sectional analyses with 54 (1.3%) participants having higher risk of advanced fibrosis (FLI ≥60 and FIB-4 ≥1.3 when participants were <65 years or FLI ≥60 and FIB-4 ≥2.0 when participants were ≥65 years). Descriptive statistics for the study population of the secondary analyses using FIB-4 as a binary outcome variable differed only marginally from the study population of the main outcome (data not shown).

We observed inconsistent associations of air pollutants with FIB-4. Although background AP modeled by the EURAD-CTM was not or inversely associated with higher risk of advanced fibrosis, air pollutants modeled by ESCAPE-LUR showed positive, though imprecise effect estimates (Tables 5 and 6). No major differences in effect estimates were observed before and after removing the most influential observations (See Supplementary Tables 4 and 5; http://links.lww.com/EE/A238 for the analyses excluding the most influential observations).

Table 5.

Regression analysis of the associations between air pollutants and FIB-4 at baseline per IQR using EURAD-CTM and FIB-4 as a binary outcome.

	IQR	Crudea OR (95% CI)	Model 1 (DAG)b OR (95% CI)	Model 2c OR (95% CI)	Model 3d OR (95% CI)
PM10 (µg/m3)	4.2	0.72 (0.45, 1.11)	0.68 (0.43, 1.06)	0.67 (0.43, 1.04)	0.68 (0.43, 1.05)
PM2.5 (µg/m3)	2.1	0.56 (0.35, 0.87)	0.54 (0.33, 0.86)	0.53 (0.33, 0.84)	0.53 (0.33, 0.85)
PN (n/mL)	26,557	0.92 (0.62, 1.35)	0.78 (0.50, 1.21)	0.78 (0.50, 1.20)	0.78 (0.49, 1.20)
NO2 (µg/m3)	5.1	0.92 (0.65, 1.30)	0.85 (0.58, 1.23)	0.86 (0.59, 1.24)	0.86 (0.58, 1.24)

Ruhr Area, Germany, Heinz–Nixdorf–Recall study, n = 4,029 (FIB-4 n = 54, no FIB-4 3,975).

^a+Air pollutant.

^bCrude + individual and neighborhood socioeconomic status

^cModel 1 + Age, sex.

^dModel 2 + Lifestyle variables (smoking status, cumulative smoking, environmental tobacco smoking, physical activity, nutrition, and alcohol consumption).

EURAD-CTM indicates European air pollution dispersion chemistry transport model; FIB-4, fibrosis-4; PM, particulate matter; PN, particle number.

Table 6.

Regression analysis of the associations between air pollutants and FIB-4 at baseline per IQR using the ESCAPE-LUR exposure model and FIB-4 as a binary outcome.

	IQR	Crudea OR (95% CI)	Model 1 (DAG)b OR (95% CI)	Model 2c OR (95% CI)	Model 3d OR (95% CI)
PM10 (µg/m3)	2.1	1.17 (0.87, 1.54)	1.08 (0.77, 1.48)	1.08 (0.72, 1.48)	1.07 (0.76, 1.47)
PM2.5 (µg/m3)	1.5	1.19 (0.82, 1.71)	1.03 (0.65, 1.61)	1.03 (0.66, 1.61)	1.03 (0.65, 1.61)
PMcoarse (µg/m3)	1.9	1.53 (1.11, 2.14)	1.49 (1.05, 2.18)	1.48 (1.04, 2.17)	1.48 (1.04, 2.18)
PM2.5 abs (0.0001/m)	0.4	1.11 (0.88, 1.33)	1.05 (0.79, 1.29)	1.06 (0.80, 1.32)	1.06 (0.79, 1.32)
NO2 (µg/m3]	6.2	1.03 (0.73, 1.41)	0.90 (0.61, 1.30)	0.90 (0.61, 1.30)	0.91 (0.61, 1.31)

Ruhr Area, Germany, Heinz–Nixdorf–Recall study, n = 4,029 (FIB-4 n = 54, no FIB-4 3,975).

^a+Air pollutant.

^bCrude + individual and neighborhood socioeconomic status.

^cModel 1 + Age, sex.

^dModel 2 + Lifestyle variables (smoking status, cumulative smoking, environmental tobacco smoking, physical activity, nutrition, and alcohol consumption).

ESCAPE-LUR indicates European study of cohorts for air pollution effects land use regression; FIB-4, fibrosis-4; PM, particulate matter.

Sensitivity analyses

The associations of the air pollutants PM₁₀, PM_2.5, and NO₂ with NAFLD prevalence were robust in sensitivity analyses and did not vary extensively (Supplementary Tables 6–9; http://links.lww.com/EE/A238). Results were attenuated slightly compared to the main analysis when the log-binomial regression was used. Including distance to major road as a covariate to the main model for the EURAD exposure model, did not change effect estimates. Including noise as a covariate to the main model attenuated results slightly compared to the main analysis. After excluding participants with lipid-lowering or potential hepatotoxic medication, the association between air pollutants and NAFLD increased slightly compared to the main analysis. For the urban background assessment using EURAD-CTM, results were also robust when excluding nonfasting participants from the analyses, except for PM₁₀ where estimates decreased to the null. For the ESCAPE-LUR exposure model, results weakened slightly compared to the main analysis when excluding nonfasting participants. When excluding participants in the medium range of FLI (30≤ FLI <60), all effect estimates increased compared to the main analysis (Supplementary Tables 8 and 9; http://links.lww.com/EE/A238). The null or inverse associations with FIB-4 did not change notably with a reduced outcome definition.

Discussion

Overall, our results showed a positive association between long-term AP exposure and NAFLD, with the most consistent associations apparent between PM_2.5 and NAFLD. Our results also revealed that local traffic-specific PM exposure is more strongly associated with NAFLD than industry-specific PM exposure. We did not find consistent associations between AP exposure and higher risk of advanced fibrosis (FIB-4).

Our findings of a potential association between AP exposure and NAFLD are supported by several reports from controlled animal studies, which found positive associations between different AP exposures and liver-fat accumulation.^41,42 Further, other epidemiological studies also described the relationship between AP exposure and NAFLD. One study from Wang et al.²⁰ examined the association between short-term exposure to different air pollutants and the number of outpatient visits for MASLD, and found that an increase in short-term AP exposure was related to outpatients visit for MASLD.²⁰ Similar to our study, another epidemiological study with 90,086 participants looked with a cross-sectional design at long-term exposure to AP and MASLD and found that long-term exposure to ambient PM (PM₁₀, PM_2.5, and PM₁) and NO₂ was positively associated with an increased risk of MASLD and is therefore in line with our findings, even though HS was diagnosed via abdominal ultrasound and not with FLI.¹⁵ However, our effect estimates were weaker, probably due to the lower number of study participants. Another cross-sectional study from the United States found an association between higher ambient PM_2.5 exposure and increased odds of NAFLD among hospitalized patients.²¹ On the other hand, a study from Li et al.¹⁴ on proximity to major roads and PM_2.5 found no association between annual PM_2.5 exposure with HS diagnosed by computed tomography, although in a younger and healthier population than in our study. However, Li et al.¹⁴ did observe a higher prevalence of HS among participants who lived closer to major roadways. This supports our finding of traffic-associated AP being potentially more harmful than other AP sources. A stronger association between different diseases or biomarkers with traffic-related AP exposure has been shown in other studies, and in the HNR study population.^27,43,44

Our findings are further supported by recent epidemiological studies with longitudinal study designs. One study from Taiwan and another from China found an association between long-term exposure to PM_2.5 and incident NAFLD. Although the study with a Taiwanese study population also used FLI (and the HS index) as an outcome for NAFLD, the Chinese study used abdominal ultrasound as an outcome for NAFLD. Both studies found a positive association between exposure to PM_2.5 and incident NAFLD and further supports our findings.^19,22 Another longitudinal study from Li et al.¹⁸ not only looked at an association between exposure to different air pollutants and incident NAFLD, but also at exposure to AP with cirrhosis among UK residents. They found that exposure to higher concentrations of AP was linked to an increased risk of incident NAFLD, and cirrhosis. Associations of AP with liver cirrhosis have also been suggested in a large administrative cohort in Rome.⁴⁵ However, this study lacked lifestyle information including alcohol consumption and smoking, and instead adjusted comprehensively for area-level socioeconomic variables. In contrast to our study that examined an advanced risk for advanced fibrosis, the Rome cohort used the clinical diagnosis of cirrhosis as an outcome. Of note, the Rome administrative cohort used the same ESCAPE-LUR exposure modeling as our study and the observed associations were very similar to our FIB-4 results of Model 2 (including individual and neighborhood SES, age and sex, but without lifestyle variables), but more precise due to the large sample size of this large administrative cohort. The inverse association observed for urban background PM_2.5 in our study stands out and can most likely be explained by random variation.

The majority of pathophysiological hypotheses about the development of NAFLD are based on factors predisposing to obesity and insulin resistance.^46,47 Several reviews have described potential pathophysiological pathways that could explain an association between AP and metabolic diseases, to which NAFLD can be added.^48–51 Chronic inflammation and oxidative stress may play a key pathogenic role in the development of T2DM, insulin resistance, and obesity, it is as well known to occur with AP exposure, and may therefore represent a potential link between AP exposure and metabolic dysfunction.^48,50 The mechanisms underlying initiation of systemic inflammation and oxidative stress in response to AP is widely thought to originate primarily in the lung.⁴⁸

Four potential mechanisms have been described as a pathophysiological basis for the association of AP on metabolic diseases including NAFLD and risk factors after inhalation of air pollutants.^48–50: (1) activation of innate immune cytokines in the lungs, causing a release of proinflammatory mediators and a systemic “spill-over”; (2) uptake of air pollutants by macrophages resulting in adaptive immune response and an increase in systemic inflammatory responses and oxidative stress; (3) direct penetration of leachable components into the systemic vasculature; (4) perturbation of the systemic autonomic nervous system causing acute autonomic imbalance, favoring the sympathetic over the parasympathetic limb, and facilitating systemic inflammation. The inflammatory milieu and oxidative stress caused through the different pathways could contribute to a persistent autonomic imbalance and inflammation within insulin-sensitive tissue (e.g., adipocytes) causing insulin resistance, HS, and endoplasmic reticulum stress.^42,51

We observed differences in the association of APs with FLI, depending on the exposure model used (EURAD-CTM vs. ESCAPE-LUR), with generally stronger associations seen for the time-varying urban background model (EURAD-CTM) compared to the time-invariant ESCAPE-LUR model. Even though the exposure models had different modeling techniques, effect estimates were relatively consistent positively associated with FLI. For risk of advanced fibrosis as an outcome (FIB-4), effect estimates were contrary and inconsistent between the two exposure models. One possible explanation could be the stage of the disease, since it is suggested that individual factors, such as a harmful lifestyle, play a stronger role in advanced diseases (e.g., NASH), compared to preclinical/mild stages of diseases (e.g., NAFLD). Further, since the diagnosis of NASH seems to be more challenging without the use of liver biopsy, outcome misclassification with the noninvasive formular FIB-4 is a possibility.

When comparing AP reference values to our results, air pollutant concentrations were mostly below the current European Union (EU) annual limit values for PM_2.5 (EU Limit 25 µg/m³, PM_2.5 mean ± SD for EURAD-CTM 17.7 ± 1.3 µg/m³, for ESCAPE-LUR 18.4 ± 1.1 µg/m³). NO₂ (EU limit value 40 µg/m³) was above the limit value with EURAD-CTM (NO₂ mean ± SD 41.4 ± 4.0 µg/m³) and below for ESCAPE-LUR (NO₂ mean ± SD 30.3 ± 4.9 µg/m³). However, when looking at health-based AP guidelines from the World Health Organization (WHO), AP concentrations exceed current recommendations (WHO Air Quality Guidelines: PM_2.5 5 µg/m³, NO₂ 10 µg/m³). Since our study suggests a positive association between long-term AP exposure and NAFLD, it supports alignment of future air quality limit values with the WHO Air Quality Guidelines to protect the population.

There are several diagnostic tools to diagnose NAFLD and depending on the diagnostic tool used, NAFLD prevalence varies. We found a NAFLD prevalence of 31.7% and an increased risk for advanced fibrosis (high FIB-4) of 1.4% among participants having NAFLD and 3.2% in the population when a NAFLD diagnosis was not taken into account for the outcome definition. Our NAFLD prevalence is in line with data from other studies.^52,53 Further, the prevalence of NASH in the general population ranges from 3% to 5% globally.⁵³ Even though high FIB-4 prevalence cannot be equated completely with NASH prevalence, since it only predicts high risk of fibrosis, the high FIB-4 prevalence is consistent with this NASH prevalence.

Our study has several strengths. First, the HNR study is a deeply characterized population-based cohort with extensive covariate data, which enabled us to adjust for multiple potential confounders. Second, we were able to use two different outcome variables (FLI and FIB-4), and two comprehensive and complementary exposure assessment models. Third, we conducted extensive sensitivity analyses, including different outcome definitions and model specifications. Fourth, we conducted extensive analyses of selection bias by comparing participants with and without NAFLD (FLI <60 vs. FLI ≥60), and included and excluded participants. Fifth, we used different regression models with increasing covariate adjustment, including one DAG-based model. Sixth, our study population is based in a highly urbanized area within Germany, which may question the generalizability of our results. However, since demographic movements towards living in urbanized areas and big cities also occurs in Germany, we suggest that many people may be affected, in Germany and elsewhere. Further, in more rural areas AP concentrations can also be high due to increased industrialized agricultural practices, and wood stove use in private homes. Seventh, the gold standard to diagnose NAFLD and hepatic fibrosis is a liver biopsy. Since liver biopsies are not possible in large population-based studies, outcome misclassification is possible. We therefore carried out extensive sensitivity analyses to investigate the robustness of our estimations.

Our study also has several limitations. First, since FLI is a diagnostic tool for the diagnosis of HS, we may have missed participants with more advanced stages of NAFLD, where fat content is reduced. Second, it is known that many participants report a lower amount of alcohol intake due to reporting bias. Hence, since alcohol intake is a strong predictor for the outcome of NAFLD, misclassification is likely. Third, since we excluded several participants who turned out to be unhealthier in lifestyle variables than the included participants, and it is known that vulnerable groups are more susceptible to the adverse effects of AP, we may have weaker effect estimates in our study compared to the general population. Fourth, analyses were carried out cross-sectionally, hence we were not able to investigate temporal relations between AP and NAFLD. Fifth, nondifferential exposure misclassification most likely has contributed to a downward bias of the effect estimates.

In summary, our results suggest a positive association between long-term AP exposure and NAFLD, with the most consistent associations apparent between PM_2.5 and NAFLD. We did not find a consistent association between AP exposure and higher risk of advanced fibrosis, however. Future epidemiological studies using different diagnostic tools for NAFLD diagnosis, more longitudinal study designs, different study regions, and other potential harmful environmental factors such as noise, could help clarify mechanisms underlying the association between AP exposure and NAFLD.

Conflicts of interest statement

The authors declare that they have no conflicts of interest with regard to the content of this report.

Supported by the German Ministry of Education and Science (BMBF) and the German Aerospace Center (Deutsches Zentrum für Luft- und Raumfahrt, DLR), Bonn, Germany. We acknowledge the support of the Sarstedt AG & Co. (Nümbrecht, Germany) concerning laboratory equipment. This work was supported by the German Research Council (DFG; projects [SI236/8-1, SI236/9-1, HO3314/4-3, HO3314/2-3, HO3314/2-1]).

The data is due to data protection issues not available. The code used for this publication is available upon request to the authors.

ACKNOWLEDGMENTS

We thank the Heinz Nixdorf Foundation (chairman: M. Nixdorf; former chairman: jur. Schmidt) for the generous support of this study. We are indebted to the investigative group and the study personnel of the Heinz Nixdorf Recall study. We also thank the North Rhine-Westphalia State Agency for Nature, Environment and Consumer Protection for providing emission and land use data for North Rhine-Westphalia. We thank Anna Buschka for her support and data management.