Study population
This study population was composed of over 500,000 participants from an ongoing large-scale prospective cohort, UK Biobank (UKB). Briefly, participants ranging from 37 to 73 years of age from 22 assessment centers across England, Wales, and Scotland were enrolled between 2006 and 2010. All participants completed baseline questionnaires with anthropometric assessments and reported medical conditions. Details of UKB design were described elsewhere [18]. The study was approved by North West Multicenter Research Ethical Committee, and all participants signed written informed content.
Participants who were free of IBS with an available non-alcoholic fatty liver index at enrollment were included in this study. Those who already had cancer, inflammatory bowel disease (IBD), alcoholic liver disease (ALD), or coeliac disease diagnosis at enrollment were excluded. All diagnoses were identified through International Classification of Disease-10 (ICD-10) codes (Additional file 1: Table S1). Additionally, 1 participant withdrawal from UKB was excluded. Therefore, a total of 396,838 participants were included in the final analysis. Flowchart of participant selection was listed in Fig. 1.
Assessment of baseline non-alcoholic fatty liver degree and NAFLD
As no imaging, ultrasonography, or histological data regarding fatty liver was available in the large-scale UKB cohort, we used a well-established index, fatty liver index (FLI), to measure the degree of non-alcoholic fatty liver [19]. Briefly, FLI was calculated through four variables including BMI, waist circumstance (WC), triglycerides (TG), and gamma-glutamyltransferase (GGT) using a previously published and validated regression model [19]. It has been proved to be a reliable index with good discrimination performance of liver ultrasonography-determined NAFLD [area under the receiver operator curve, AUROC = 0.85 (95%CI: 0.81–0.88)] and transient elastography-determined NAFLD (AUROC = 0.85), which has been externally validated and widely accepted in a population-based study [19,20,21]. Meanwhile, the weighted percent-agreement between FLI and transient elastography was as high as 75.11% (95%CI: 75.10%-75.12%) when validated in a nationally representative sample of the western general population rather than a clinical population [21]. We classified FLI according to quartile distribution with the lowest quartile group as the reference group and the other three quartile groups as exposure groups. Moreover, we also used NAFLD diagnosis or not according to a predefined cutoff, with FLI ≥ 60 as an indicator of NAFLD [19]. Participants who had baseline FLI < 60 were considered in the non-exposure group (non-NAFLD group), while others who were diagnosed as NAFLD were considered in the exposure group (NAFLD group). Further, NAFLD patients with BMI < 25 kg/m2 and ≥ 25 kg/m2 would be defined as lean and non-lean NAFLD, respectively. Accordingly, NAFLD patients with BMI ≥ 30 kg/m2 and < 30 kg/m2 would be considered as obese and non-obese NAFLD, separately [22, 23]. Besides, in order to examine the impact of fatty liver measurement on our findings, another well-established index, hepatic steatosis index (HSI), was used to define NAFLD in sensitivity analyses. HSI could be calculated as 8* (serum alanine aminotransferase (ALT)/aspartate aminotransferase (AST) ratio) + BMI (+ 2, if female; + 2, if type 2 diabetes) [24]. An HSI > 36 was defined as an indicator of NAFLD [24].
Ascertainment of outcome
Primary endpoint was incident IBS, which was determined via ICD-10 codes (K58, Additional file 1: Table S1). IBS diagnosis was based on self-report or linkage to primary care and/or hospital admission data with a censoring date of June 2021.
Covariates
Based on epidemiological evidence, some sociodemographic characteristics, health behaviors, and comorbidities at baseline were adjusted as covariates [1, 4, 16, 17]. Potential confounders included age (continuous variable), gender (male or female), ethnicity (white or nonwhite), socioeconomic status, education level, smoking status (never, current, or previous), alcohol drinking (never, current, or previous), type 2 diabetes (Yes or No) and physical activity. Socioeconomic status was based on the Townsend deprivation index, which was calculated immediately prior to participants joining UKB using preceding national census output areas [25]. Townsend deprivation index for socioeconomic deprivation was divided into four quartiles. Education was based on self-report of the highest qualification achieved and classified as university or non-university. Physical activity was self-reported and divided into three levels (high, moderate, and low) based on IPAQ (International Physical Activity Questionnaire).
Statistical analysis
Incidence rate with 95% confidence interval (CI) of IBS was calculated as a number of events per 1000 person-years through Poisson regression. The 12-year cumulative incidence of IBS was calculated by the Kaplan–Meier method. Cox proportional hazard model was conducted to examine the association between fatty liver and incident IBS. The follow-up period started from baseline to the date of first IBS diagnosis or censored at end of the study (June 2021), date of death, or lost-to-follow-up for participants who did not develop IBS. Considering a very small percentage of missing values (0.1–1.2% for all covariates were missing), missing indicators were used.
For FLI quartiles, per standard deviation (SD) change of FLI and diagnosis of NAFLD or not according to predefined cutoff, three multivariable models in addition to univariable analysis were accomplished: model 1, adjusted for age and gender; model 2, additionally adjusted for Townsend deprivation index, education level, ethnicity, smoking status, and alcohol drinking; model 3, additionally adjusted for physical activity and type 2 diabetes. Moreover, restricted cubic spline analysis was conducted to examine the potential non-linear association between baseline FLI and incident IBS, with knots placed at 10th, 50th, and 90th percentiles and the median value of baseline FLI (46.55) as a reference point. Furthermore, subgroup analysis was performed to investigate whether the association between the degree of non-alcoholic fatty liver as well as NAFLD and IBS varied by age (< 45 years, 45-64 years, ≥ 65 years), gender, alcohol drinking, and smoking status. Effect modification was also detected by adding interaction terms of each stratified variable (i.e., age, gender, alcohol drinking, smoking status) and non-alcoholic fatty liver exposure (i.e., FLI quartiles, per SD change of FLI, diagnosis of NAFLD or not). Further analyses were conducted to investigate the association between NAFLD type (lean/non-lean, non-obese/obese NAFLD) and risk of IBS.
In order to assess the robustness of the results, several sensitivity analyses were conducted. Firstly, we excluded participants who had an IBS diagnosis within 1 or 2 years after recruitment respectively, in order to avoid detection bias. Secondly, to rule out the influence of alcohol intake on the non-alcoholic fatty liver during the whole follow-up period, incident ALD cases after baseline were further excluded. Thirdly, the competing risk model by considering lost-to-follow-up and death as competing events were conducted, since those participants might develop IBS thereafter. Fourthly, participants who had hepatitis B/C virus seropositivity were excluded. Fifthly, we additionally adjusted psychologic disorder including depression and anxiety as confounders. Finally, an age-matched (1:1 matching, ± 2 years) cohort was generated as the new dataset to further investigate the association between NAFLD and IBS.
Additionally, sensitivity analyses were conducted by using HSI [diagnosis of NAFLD or not according to predefined cutoff (HSI > 36), per SD change] via adjusted model 3, with additional similar analyses by excluding incident IBS cases within 1 or 2 years after baseline, excluding incident ALD cases, excluding participants with hepatitis B/C virus seropositivity or performing competing risk model.
A two-tailed P value < 0.05 was considered to be statistically significant. All analyses were conducted using SAS software Version 9.4 and R version 4.0.2 (forestplot, tableone, ggplot2, and survival packages).