Antibiotic use and development of nonalcoholic fatty liver disease: A population‐based case–control study

Antibiotics affect the gut microbiome. Preclinical studies suggest a role of gut dysbiosis in the development of nonalcoholic fatty liver disease (NAFLD), but data from large cohorts with liver histology are lacking.


| INTRODUC TI ON
2][3][4] NAFLD has traditionally been recognized as a complication of the metabolic syndrome. 5However, there is a high interindividual variability in the natural history of the disease. 6[13] Many studies show that patients with NAFLD feature distinctive compositional alterations of gut microbiota. 14,15Gut microbiota produce a variety of small molecules and metabolites, including shortchain fatty acids, bile acids and ethanol, all of which have direct and indirect effects on the liver. 16Recently, the role of gut microbiota and microbial metabolites in NAFLD has attracted more attention, as there seems to be significant dysbiosis-related production of ethanol to induce steatosis and inflammation in the liver of some individuals with NAFLD. 17Against this background, some suggest that gut dysbiosis may be a causal factor in the pathogenesis and progression of NAFLD 18 -perhaps even more so in individuals without classic risk factors of the metabolic syndrome. 19tibiotics are known to cause significant short-term and, to some extent, even long-term taxonomic changes in gut microbiota that may impact the natural history of several diseases. 20However, no study has investigated whether previous antibiotic treatment increases the risk of developing NAFLD.
Thus, this study aims to help answer whether previous antibiotic treatment is linked to the development of NAFLD by using a large population-based cohort with biopsy-confirmed early stages of NAFLD.

| Study design and data sources
This was a population-based case-control study using the Epidemiology Strengthened by histoPathology Reports in Sweden (ESPRESSO) cohort. 21The ESPRESSO cohort encompasses all liver histopathology data from all 28 Swedish pathology departments from 1965 to 2017.In Sweden, all residents have a unique personal identity number assigned at birth or when legal permanent residence is established. 22Using this identity number, we could link the ESPRESSO cohort to additional validated, nationwide registers: the Total Population Register, 23 the National Patient Register, 24 the Prescribed Drug Register, 25 the Cause of Death Register 26 and the longitudinal integrated database for health insurance and labour market studies. 27Sweden provides tax-funded free and equal health care to all its legal residents.Registration in population healthcare histologic stages (p > .05).The highest risk of NAFLD was observed after treatment with fluoroquinolones (aOR 1.38; 95% CI = 1.17-1.59).Associations remained robust when patients were compared with their full siblings (aOR 1.29; 95% CI = 1.08-1.55).
Conclusions: Antibiotic use may be a risk factor for incident NAFLD, especially in individuals without the metabolic syndrome.The risk was highest for fluoroquinolones and remained robust in sibling comparisons with whom individuals share genetic and early environmental susceptibilities.

K E Y W O R D S
antibiotics, dysbiosis, gut-liver axis, metabolic syndrome, microbiome, nonalcoholic fatty liver disease

Key points
Nonalcoholic fatty liver disease (NAFLD) is the most common chronic liver disease worldwide, and several studies suggest that an altered gut microbiome composition may be a key contributing factor in the development of the disease.Antibiotics are known to profoundly perturb the microbiome composition even in the long term.In this nationwide Swedish study, previous antibiotic use was associated with an increased risk of developing NAFLD, and this association was strongest for antibiotics that are known to disrupt the gut microbiome and was stronger the higher the exposure to antibiotics.registers is mandatory.This study followed the Strengthening The Reporting of OBservational studies in Epidemiology (STROBE) reporting guideline. 28

| Study population
We identified all adult patients aged ≥18 years in whom liver biopsy was performed between 1 January 2007 and 30 April 2017, confirming the diagnosis of early-stage NAFLD histology without cirrhosis or any other competing liver disease.Using a validated International Classification of Diseases (ICD) algorithm, 29 we excluded patients with prior alcohol abuse/misuse, recorded other aetiology of acute or chronic liver disease, prior liver transplantation or emigration from Sweden before the liver biopsy date (i.e. the index date) (Table S1 presents relevant ICD codes).
Patients meeting the criteria for NAFLD were categorized into three histological subgroups using SNOMED definitions 30 for coherent nationwide histopathology reporting in Sweden: simple steatosis, nonalcoholic steatohepatitis (NASH) without fibrosis and NAFLD with non-cirrhotic fibrosis (Table S2 gives detailed definitions). 29At the time of ESPRESSO inclusion, each NAFLD patient was matched to up to five general population reference individuals (controls) without recorded NAFLD, according to age at the index date, sex, calendar year of the index date and county of residence.The general population controls were derived from the Total Population Register, 23 and identical exclusion criteria were applied, ensuring that controls did neither have a diagnosis of any other liver disease at or before baseline.

| Antibiotic exposure
Antibiotic exposure was assessed using the Prescribed Drug Register, which was initiated on 1 July 2005 and has since collected information on all medications prescribed to the Swedish population, including WHO Anatomical Therapeutic Chemical (ATC) codes, date of redemption, amount dispensed and dosage. 25This register allows complete investigations of antibiotic use, as antibiotics are available by prescription only and cannot be purchased over the counter in Sweden.Moreover, the register is not based on declarations of physicians' prescriptions but on automated reports on drugs that were effectively dispensed at any authorized Swedish pharmacy.Antibiotic exposure was identified via ATC codes for the therapeutic class of antibacterial drugs approved for systemic use (J01A-X).We further collected information about the number of prescription medication dispensations (categorized into 0, 1, 2 and 3 or more), as well as information on the cumulative defined daily doses (cDDDs) of prescribed antibiotics as defined by the World Health Organization. 31 secondary analyses, to further define the relationship between antimicrobial coverage and risk of NAFLD, we evaluated whether unique ATC classes of systemic antibiotics (penicillins, cephalosporins, macrolides, fluoroquinolones, tetracyclines, sulfonamides and other antibacterials) or the spectrum of coverage (broad vs. narrow; definitions are listed in Table S3) influenced the risk of disease.To account for the possibility of reverse causation because of the assumed higher risk of infections in patients with NAFLD, we did not count antibiotic dispensations in the year leading up to the histologic diagnosis of NAFLD.To ensure adequate at-risk exposure time of at least 6 months for each individual we excluded patients diagnosed with NAFLD within the first 18 months from the study baseline or initiation of the Prescribed Drug Register (Figure S1).

| Statistical analysis
We performed conditional logistic regression to evaluate the association between previous antibiotic therapy and NAFLD.We estimated crude and multivariable-adjusted odds ratios (aORs) and their 95% confidence intervals (CIs) conditioned on matching factors (age, sex, calendar year, county of residence) and further adjusted for potential confounding factors: education (≤9 years, 10-12 years, ≥13 years, missing), country of birth, parameters of metabolic syndrome (diabetes, obesity, dyslipidaemia, hypertension), number of inpatient and outpatient encounters 18 to 6 months before the index/matching date (a proxy for health care use), number of non-antibiotic drug dispensations 18 months to 6 months before the index/matching date and the Charlson comorbidity index. 32 conducted several sensitivity analyses to test the robustness of our results.First, to minimize potential reverse causation we increased the lead-in period from 1 year in our primary analysis to a more conservative 3 years and even 5 years preceding the index date.Second, we compared patients to their full siblings to account for the possible confounding of genetic predisposition and childhood exposures.Subgroup analyses were performed stratified by sex, age category (<40 years, 40-59 years, ≥60 years), residence, level of education, presence of metabolic syndrome and diabetes mellitus.Furthermore, we studied the association between cumulative antibiotic exposure as assessed by cDDDs and the risk of developing NAFLD.Statistical analyses were conducted using Stata version 17.0 (StataCorp LLC).A two-sided p < .05 was considered statistically significant.

| Patient characteristics
We identified 2584 individuals aged ≥18 years in the ESPRESSO cohort who underwent liver biopsy and were diagnosed with earlystage NAFLD between 1 January 2007 and 30 April 2017.Most patients had simple steatosis (n = 1435; 55.5%), while 383 (14.8%) had NASH without fibrosis, and 766 (29.7%) had NAFLD with noncirrhotic fibrosis (Figure 1).The baseline characteristics of all patients and controls are summarized in Table 1.For NAFLD patients, the average age at index biopsy was 56 years, and 51% were male.NAFLD patients were more likely than controls to have a diagnosis linked to the metabolic syndrome, such as diabetes mellitus, hypertension or dyslipidaemia (Table 1).

| Antibiotic use and risk of nonalcoholic fatty liver disease
Previous antibiotic use was seen in 1748 (68%) NAFLD patients and 7001 (55%) controls, corresponding to a 1. 35  2).Analyses on the association between cDDDs and risk of NAFLD revealed that exposure to any antibiotic medication for less than or equal to 7 days was already associated with an aOR of 1.63 (95% CI, 1.27-2.09).The risks increased with longer duration of exposure to antibiotics in a dose-dependent manner (p for trend < .001)with highest risks in individuals with exposure to antibiotics for more than 6 weeks (≥43 cDDDs; aOR 1.72; 95% CI, 1.49-1.99;Table S4).In addition, the relationship between prior antibiotic use and the devel-

18,115
Patients with NAFLD-compatible histopathological features (histological features and diagnostic impression) Note: All variables are reported as mean (SD) or %, unless indicated otherwise.The Supplemental Appendix provides definitions of the NAFLD histological groups and all covariates.
Abbreviations: NAFLD, nonalcoholic fatty liver disease; NASH, nonalcoholic steatohepatitis.a Education categories based on compulsory school, high school, and college.
b Components of metabolic syndrome include obesity, diabetes mellitus, dyslipidaemia and hypertension.

| Antibiotic classes and spectrum of coverage
Analyses by the spectrum of antibiotic coverage showed no difference in the risk among patients with prior use of broad-spectrum antibiotics (aOR 1.

| Sibling comparison
To clarify genetic predisposition to disease development and to account in part for shared but unspecified childhood exposures we compared antibiotic use in patients with biopsy-proven NAFLD to their full siblings.Some 2837 siblings of 1491 patients with at least one living sibling were identified.Mean age, county of residence and level of education were all similar between patients and their unaffected siblings; however, patients had higher rates of diabetes mellitus and hypertension (Table S7).When siblings were used as the reference group, NAFLD risk estimates were almost similar to general population controls, with a multivariable aOR of 1.29 (95% CI, 1.08-1.55;p = .005;Table S8).

| DISCUSS ION
This nationwide, population-based case-control study, including more than 2500 biopsy-proven cases with early histological stages of NAFLD, had three key findings.First, patients with antecedent antibiotic treatment had a 35% increased relative risk of developing NAFLD, a risk that seemed to increase with each subsequent dispensation and with cumulative defined daily doses of antibiotics.Second, while some antibiotic classes did not affect the risk of NAFLD, there was a marked risk increase for antibiotics known to perturb the gut microbiome, with the highest risk following exposure to fluoroquinolones.Finally, the link between antibiotic use and later NAFLD was mainly evident in patients without features of the metabolic syndrome, potentially indicative of an alternative mechanism in the development of NAFLD in these individuals.
To our knowledge, this is the first study to examine the association between antibiotic use and the future risk of NAFLD.Overall, estimates appeared to be robust to adjustment for age, sex, location, calendar year, level of education, degree of health care and medication use, features of the metabolic syndrome and multiple comorbidities.Moreover, the study was designed to minimize risks of reverse causation.We did not allow dispensations of antibiotics to accrue in the year before the histologic diagnosis of NAFLD.
Additionally, in sensitivity analyses the lead-in time was prolonged to 3 years and even 5 years before NAFLD diagnosis, yielding comparable estimates, which confirms the positive association between antibiotic use and risk of NAFLD.In addition, patients with cirrhosis were excluded from the study because, on the one hand, the diagnosis of cirrhosis is often delayed for years and remains obscured until complications arise, such as decompensation or hepatocellular carcinoma. 33On the other hand, patients with cirrhosis are at a much higher risk of recurrent infections and thus are more likely to receive antibiotic treatment. 34In contrast, over half of the patients in our study (56%) had simple steatosis at the time of diagnosis.While a stronger association between antibiotics and progressive histological signs of liver inflammation and fibrosis could be assumed, we found similar risks across all histologic stages, from simple steatosis to NASH with or without fibrosis.Thus, it appears that antibiotics might rather be involved in the development of NAFLD than in the progression of the disease.
We performed a sensitivity analysis comparing NAFLD patients with their full siblings with whom they share genetic predispositions for the development of NAFLD.Importantly, in the sibling analyses the risk increase in NAFLD after antibiotic exposure was comparable to that in general population controls.Because sibling analyses partly account for heritable predispositions, this finding provides further evidence that antibiotic use might be a relevant contributing environmental exposure in developing NAFLD that has not been sufficiently acknowledged in clinical research, so far.
However, the underlying mechanisms of how antibiotics may put a person at risk of NAFLD remain elusive.Antibiotic use has been associated with a higher risk of inflammatory bowel disease (IBD) 35,36 and asthma. 37For both disorders, the mediator between antibiotics and the development of the disease was linked to dysbiosis of gut microbiota composition. 38,391][42] Gut dysbiosis has been linked to an impaired gut barrier that enhances the translocation of various antigens, endotoxins, metabolites, and even gut microbes that may drive liver and systemic inflammation, eventually leading to NAFLD. 16,43Against this background, the gut microbiome is increasingly being investigated as a future therapeutic target for treating NAFLD. 44,45A recent metaanalysis demonstrated that modulation of the gut microbiome composition through probiotics or synbiotics could improve liver-specific markers of hepatic inflammation, liver stiffness and steatosis in individuals with NAFLD. 46oking at antibiotic classes, we found a high heterogeneity of the associated risks, with fluoroquinolones showing the highest risk of future NAFLD.This finding may strengthen the hypothesis that, just as with asthma and IBD, antibiotic perturbation-induced gut dysbiosis may also be the underlying mechanism of NAFLD.
Fluoroquinolones have a broad antibacterial spectrum, a high oral bioavailability and a high tissue penetration, which is why they are among the most prescribed antibiotics worldwide. 47Despite these favourable aspects, fluoroquinolones are among the most potent inducers of alterations in gut microbiome composition. 48Acute severe complications of their strong effects to perturb the gut microbiome are well known, as observed in several large Clostridioides difficile outbreaks, which in retrospect could be related to the use of fluoroquinolones. 49,50Yet, data on long-term complications of their use are scarce.Over 10 years ago, Dethlefsen et al performed a unique study, repeatedly analysing the gut microbial composition using extensive pyrosequencing of stool samples of healthy individuals over a period of 10 months after two courses of ciprofloxacin. 51They found a rapid and profound loss of microbial diversity already occurring within 3-4 days of exposure.However, while the composition of the gut microbiome had stabilized thereafter, it was permanently altered from its original state leading to long-term dysbiosis.The authors concluded that the full consequences of this phenomenon 'remained unknown'.Our findings suggest that one long-term complication of this fluoroquinolone-induced dysbiosis may be a higher risk of NAFLD and we demonstrate that there is a dose-response relationship with the number of dispensations as well as number of cumulative defined daily doses exposed to antibiotics.
In addition, in our study the association between antibiotic use and NAFLD was only evident in patients without features of the metabolic syndrome.While NAFLD is widely acknowledged as a complication of the metabolic syndrome, 52 a significant subset of patients is lean and possesses a relatively favourable metabolic profile-an entity termed 'lean NAFLD'. 19  Nevertheless, the present data must be interpreted in the context of the study design.First, given that we did not have data on in-hospital intravenous antibiotic treatments, our results apply only to oral antibiotics dispensed from pharmacies.However, while intravenous administration of antibiotics has been shown to impact the gut microbiome in the short term similarly, evidence suggests that it does not lead to long-term dysbiosis to the same extent as orally administered antibiotics. 53Second, although we had data on actual dispensations of antibiotics from pharmacies, real-world adherence could not be confirmed.Third, although we adjusted for a variety of covariates including comorbidities, healthcare and medication use, residual confounding cannot be entirely excluded, for instance we lacked detailed data on body mass index or waist circumference.
Last, due to the fact that we did not have stool specimen of included individuals, we were not able to perform mechanistic analyses on the gut microbial compositions.
In conclusion, this nationwide case-control study of individuals with biopsy-confirmed NAFLD provides evidence of a higher long-term risk of NAFLD after antibiotic exposure with positive dose-response.This risk was most evident after fluoroquinolone -fold increased risk of NAFLD (95% CI, 1.21-1.51).The median time from first antibiotic exposure to NAFLD diagnosis was 4.7 years (interquartile range 2.7-7.3 years) and the median duration of antibiotic usage as measured in cDDDs equalled to 15 days (IQR 0-42.5 days) in individuals with NAFLD compared to 10 days (IQR 0-30 days) in matched controls.There was a frequency-dependent relationship between increased antibiotic prescriptions and risk of NALFD with an aOR of 1.30 (95% CI, 1.13-1.49)for one dispensation, 1.38 (95% CI, 1.18-1.63)for two and 1.39 (95% CI, 1.21-1.60)for three or more (p for trend < .001;Table opment of NAFLD was observed among all histologic stages of NAFLD: simple steatosis (aOR 1.33 [95% CI, 1.15-1.55]),NASH F I G U R E 1 Study flow chart.*Patients diagnosed with NAFLD in the first 18 months from the initiation of the Swedish Prescribed Drug Register were excluded to ensure at least 6 months of at-risk exposure time and a lead-in period of 12 months.ESPRESSO, Epidemiology Strengthened by histoPathology Reports in Sweden; NAFLD, nonalcoholic fatty liver disease; NASH, nonalcoholic steatohepatitis.

a
much more relevant mechanistic pathway for the development of NAFLD.This study has several strengths including the nationwide cohort with biopsy-proven NAFLD, allowing a clear distinction between the different histologic stages of the severity of NAFLD disease on a population-based level and high external validity.The Prescribed Drug Register allows to prospectively study all prescribed medications for all individuals in Sweden over time precluding selection or recall bias.Furthermore, our data were based on effectively dispensed drugs from all Swedish pharmacies.Therefore, according to our definition, we assume a high realworld validity of antibiotic use.In addition, our results were highly robust in several sensitivity analyses, even after comparison with full siblings that share many predisposing genetic and early environmental factors for the development of NAFLD.Preclinical and mechanistic clinical studies are warranted to uncover the underlying mechanisms in the interplay between antibiotic-induced gut dysbiosis and NAFLD development and to exploit targeted pharmaceutic modalities for future treatments.
treatment and was most notable in patients without the metabolic syndrome.Our findings highlight that antibiotic perturbation of the gut microbiome may be a relevant pathophysiological mechanism in the development of NAFLD in some individuals and that future research in this field is highly desired.While antibiotic stewardship has until now focused on slowing the global rise in antibiotic resistance, long-term risks for health conditions (such as NAFLD) should be acknowledged a public health threat and lead to a more reasonable use of antibiotics.AUTH O R CO NTR I B UTI O N SFahimEbrahimi contributed to conceptualization, methodology, statistics, writing-original draft.Tracey G. Simon, Hannes Hagström and Bjorn Roelstraete contributed to writing-review and editing.Jiangwei Sun, David Bergman and Anders Forss contributed to review and editing.Jonas F. Ludvigsson contributed to conceptualization, methodology, writing-original draft, resources and supervision.FU N D I N G I N FO R M ATI O N FE was supported by the Swiss National Science Foundation (P500PM_210866).CO N FLI C T O F I NTER E S T S TATEM ENT Dr Ebrahimi has served as an Advisory Board Member for Boehringer Ingelheim.Dr Ludvigsson has coordinated an unrelated study on behalf of the Swedish IBD quality register (SWIBREG).That study received funding from Janssen Corporation.Dr Ludvigsson has also received financial support from MSD to develop a paper reviewing national healthcare registers in China.Dr Hagström's institutions have received research grants from Astra Zeneca, EchoSens, Gilead, Intercept, MSD and Pfizer.Lars Engstrand receives a research grant from Ferring Pharmaceuticals to study the human microbiome in health and disease.The other authors have nothing to disclose.

12 646) NAFLD (n = 2584) Unadjusted OR a Multivariable aOR b p value c
Data are n (%) or OR (95% CI).Cumulative dispensations accrued from study baseline to 1 year before diagnosis or matching.Abbreviations: CI, confidence interval; NAFLD, nonalcoholic fatty liver disease; OR, odds ratio.a Conditional logistic regression matched for age at index time, sex, calendar year and county of residence.Conditional logistic regression matched for age at index time, sex, calendar year and county of residence, and further adjusted for education, country of birth, parameters of metabolic syndrome (diabetes, obesity, dyslipidaemia, hypertension), number of inpatient and outpatient encounters 18 to 6 months before the index date (or corresponding matching date), number of non-antibiotic drug dispensations 18 to 6 months before the index date (or corresponding matching date) and the Charlson's comorbidity index.Antibiotic use and risk of NAFLD according to histologic subgroups.
28 [95% CI, 1.09-1.49])comparedtotreatment with narrow-spectrum antibiotics (aOR 1.38 [95% CI, 1.23-1.56];p>.05;TableS6).Calculating risks according to the class of systemic antimicrobials as defined by WHO ATC categorization, we found heterogeneous estimates, with the strongest increase in the risk of NAFLD for the prior use of fluoroquinolones (aOR 1.38 [95% CI, 1.17-1.59])followed by cephalosporins (aOR 1.32 [95% CI, 1.03-1.61]).No association was found for sulfonamides (aOR 0.96 [95% TA B L E 2 Antibiotic use and risk of NAFLD in patients and matched general population controls.Controls (n = b c p-Values are shown for the multivariable-adjusted model.d p for trend.TA B L E 3 Note: Data are n (%) or OR (95% CI).Cumulative dispensations accrued from study baseline to 1 year before diagnosis or matching.Abbreviations: CI, confidence interval; NAFLD, nonalcoholic fatty liver disease; OR, odds ratio.a p-Values are shown for overall antibiotic use.b Conditional logistic regression matched for age at index time, sex, calendar year and county of residence.c CI, 0.77-1.25])or those classified as other antibiotics (aOR 1.01 [95% CI, 0.77-1.25];p > .05; Figure 2).
Risk of NAFLD according to antibiotic classes.Conditional logistic regression matched for age, sex, calendar year and county of residence and further adjusted for education, country of birth, parameters of metabolic syndrome (diabetes, obesity, dyslipidaemia, hypertension), number of inpatient and outpatient encounters, number of non-antibiotic drug dispensations and the Charlson comorbidity index.NAFLD, nonalcoholic fatty liver disease; NASH, nonalcoholic steatohepatitis.
Antibiotic Use and Risk of NAFLD in subgroups.Conditional logistic regression matched for age, sex, calendar year and county of residence and further adjusted for education, country of birth, parameters of metabolic syndrome (diabetes mellitus, obesity, dyslipidaemia, hypertension), number of inpatient and outpatient encounters, number of non-antibiotic drug dispensations and the Charlson comorbidity index.p-values for homogeneity were obtained by fitting interaction terms.NAFLD, nonalcoholic fatty liver disease.
drome already exhibit significant gut dysbiosis and that antibiotic use does not lead to materially additional perturbations in microbiome composition that would increase their risk of developing NAFLD.In contrast, these effects would be of greater relevance in patients without features of the metabolic syndrome, in whom antibiotic perturbation of the microbiome composition might beF I G U R E 3