Waist-to-height ratio and non-alcoholic fatty liver disease in adults

Our study population was from a large-scale health examination programme called ‘human dock’ in Japan that aims to promote public health and assesses common chronic diseases and their risk factors through physical examination. The research data have been uploaded to Dryad public database by Okamura et al. According to the terms of service of Drayad database, we can use this data for secondary data analysis based on different assumptions [20]. In previous studies, research ethics was approved by the Murakami Memorial Hospital Ethics Committee and informed consent was obtained from all subjects; Additionally, in order to protect the privacy of patients, the effective identification ID of all subjects in the study was replaced by health codes, and the whole study process followed the Declaration of Helsinki. Since this study was a secondary analysis of previous studies, there was no need to apply for separate ethical approval.

In previous studies, the researchers used a longitudinal design and analysed the risk factors for diabetes [20]. In this study, we adopted a cross-sectional design that included participants in the ‘human dock’ physical examination programme from 2004 to 2015. We have added some exclusion criteria to the previous research design: (a) male alcohol consumption ≥ 210 g/w and female alcohol consumption ≥ 140 g/w [21]; (b) viral hepatitis or diabetes diagnosed at baseline; (c) on medication at baseline; (d) impaired fasting blood glucose (FPG); (e) age < 18 years and (e) missing covariant data.

As mentioned previously [20], the clinical baseline information in this study was collected through a standardised self-administered questionnaire, including age, sex, height, weight, WC, smoking status, drinking status, systolic blood pressure (SBP), diastolic blood pressure (DBP) and habit of exercise. Information such as height, weight and WC came from the patient's self-report. The habit of exercise was defined as participating in any exercise more than once a week; Smoking status was divided into nonsmokers, former smokers, and current smokers by asking about smoking history when baseline data were collected. Drinking status was divided into non-drinking or small drinking (< 40 g/w), light drinking (40–139 g/w) and moderate drinking (140–209 g/w) depending on the amount of alcohol consumed. BMI was calculated as weight/height², and WHtR as WC/height. Haematological indicators were tested by sampling venous blood after a night of fasting and included gamma-glutamyl transferase (GGT), alanine aminotransferase (ALT), high-density lipoprotein cholesterol (HDL-C), aspartate aminotransferase (AST), total cholesterol (TC), haemoglobin A1c (HBA1c), triglyceride (TG), and FPG.

NAFLD was assessed by abdominal ultrasound, and experienced gastroenterologists reviewed the ultrasound images without knowing the participants’ personal information. According to the results of four types of ultrasound, such as liver brightness, hepatorenal echo contrast, vascular blurring, and deep attenuation, the evaluation was made and a final diagnosis was made [22].

All the analyses in this study were performed using R (version 3.4.3) and Empower (R) (version 2.0) statistical software. To understand the association between WHtR and other variables more intuitively, we grouped WHtR into quintiles, and the linear trend of baseline characteristics was tested by logical regression or linear regression. Additionally, we used a multivariate logical regression model to examine the NAFLD risk corresponding to each WHtR quintile. As a sensitivity analysis, we treated the WHtR quintile as a continuous variable and examined the WHtR quintile and NAFLD risk trend. In multivariate analysis, the adjustment rules of variables were followed according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement [23]. The adjustment results of different degrees were shown in different models, where model 1 was the unadjusted model, model 2 adjusted the basic demographic variables (adjusted for age and sex), model 3 adjusted for variables whose impact on the WHtR and NAFLD matching risk was greater than 10% (adjusted for BMI, ALT, HDL-C, TG and height) [24], and model 4 adjusts all non-collinear variables (collinearity diagnosis discriminates according to variance inflation factor [25], Additional file 1: Table S1). We also checked whether a non-linear relationship existed between WHtR and NAFLD using the generalised additive model (GAM, spline smoothing function). Finally, to further explore the association between WHtR and NAFLD, we used the likelihood ratio test to examine the interaction between the pre-determined subgroup (sex, age, BMI, habit of exercise, and drinking status) and WHtR.

In this study, the distribution pattern of the quantitative variables was judged by the QQ plot, and the expression was summarised as medians (quartile interval) or means ± standard deviation (SD). The frequency of qualitative variables was reported as a percentage, and chi-squared test evaluated the differences between groups.

In total, 20,944 participants were enrolled, among which 6693 participants were excluded because they did not meet the screening criteria (736 participants with viral hepatitis or diabetes, 2321 participants were on medication at baseline, 808 participants had impaired FPG, 1952 participants were heavy drinkers, 10 participants for unknown reasons and 863 participants with missing covariates), leaving 14,251 adults for analysis, including 7411 male and 6840 female, with an average age of 43.53 ± 8.89 years.

In this study, the WHtR values ranged from 0.28 to 0.79, and Table 1 evaluates the association between the baseline data and WHtR quintiles. With the gradual increase in WHtR, the proportion of male participants gradually increased, the age, height, weight, BMI, WC, SBP, and DBP of all participants showed a gradual increase, the NAFLD patients, smokers and drinkers, and the levels of ALT, AST, GGT, TC, TG, HbA1c and FPG were also on the increased (all P_trend < 0.05). Notably, when the WHtR was elevated, we found that the number of individuals who maintained the habit of exercise gradually decreased, and the HDL-C levels were also decreased (P_trend < 0.05).

Table 2 shows the baseline characteristics of the participants diagnosed with NAFLD and non-NAFLD. Among the 14,251 participants, the prevalence of NAFLD was 17.59%, mainly male patients (80.93%). In the NAFLD group, the general clinical indexes (age, height, weight, BMI, WC, WHtR, SBP, DBP, smoking status, and drinking status) of the participants were higher than those in the non-NAFLD group (P < 0.001). Additionally, among the participants with confirmed NAFLD, significantly fewer participants had a habit of exercise (15.04%). Furthermore, except for HDL-C, the haematological indicators of the NAFLD participants were all higher than those of the non-NAFLD participants (P < 0.001).

In the multivariate logistic regression model, WHtR was significantly positively correlated with NAFLD, and this association remained unchanged in the unadjusted model (model 1), the basic adjusted model (model 2), the model adjusted according to the method recommended by the STROBE statement (model 3), and the model adjusted for all non-collinear variables (model 4). In the model that adjusted all the non-collinear variables, we observed a 66% increase in NAFLD risk for per SD increase in WHtR (adjusted odds ratio (OR):1.66; 95% confidence interval (CI): 1.45, 1.89). Additionally, among the quintile groups of WHtR, participants in quintile 2, quintile 3, quintile 4, and quintile 5 showed 3.62-fold, 5.98-fold, 9.55-fold, and 11.08-fold increased risks of NAFLD, respectively, compared with that in quintile 1 (P trend < 0.0001) (Table 3). These results suggest that individuals with higher WHtR are more likely to develop NAFLD than those with lower WHtR.

We continued to use GAM to model the association between WHtR and NAFLD. The regression spline curve indicated that the relationship between WHtR and NAFLD was non-linear, in which a WHtR of approximately 0.4 might be the threshold effect of NAFLD risk, 0.6 might be the saturation effect of NAFLD risk, and a value between 0.4 and 0.6 might demonstrate linear relationship with the risk of NAFLD (Fig. 1).

To further reveal the deep association between WHtR and NAFLD, interaction tests were performed in the pre-defined subgroups (Table 4). Among these subgroups, we found a significant interaction between WHtR and BMI (P _interaction < 0.01). Among them, this risk association was significantly higher in non-obese individuals than in those with BMI ≥ 24 kg/m² (adjusted OR: 2.73 vs 1.94). Additionally, the interaction tests for age, sex, habit of exercise, and drinking status were not significant (P_interaction > 0.05).

This study’s advantage is that, based on a large sample size, a standard questionnaire was used to evaluate the habit of exercise as an important risk factor, and a rigorous statistical adjustment was made to explore the association between WHtR and NAFLD deeply. Additionally, this study is the first to confirm that the association between WHtR and NAFLD is non-linear and provides a reference range of WHtR associated with NAFLD risk.

This study’s limitations are mainly due to the following aspects: (a) a cross-sectional design was adopted; thus, the causal association between WHtR and NAFLD could not be determined. However, due to the retrospective nature of the study, the issue of observational bias was avoided. Additionally, the data analysis in this study was based on a large sample, and the conclusions drawn could be considered relatively reliable; (b) IR was not evaluated; thus, regarding the association between WHtR and NAFLD, we can only speculate that the mechanism of IR may be involved in one of the important links; (c) the diagnosis of NAFLD was evaluated by abdominal colour ultrasound, and the true prevalence of NAFLD may be underestimated compared with liver biopsies—that is, the effect of WHtR on NAFLD may also be underestimated. However, abdominal colour ultrasound has reduced the economic burden and physical damage of physical examination in healthy individuals; with the improvement in ultrasonic detection technology, the sensitivity and specificity of abdominal colour ultrasound in detecting NAFLD have reached a high level [39]; (d) although we have adjusted a large range of known risk factors, many risk factors may exist that we have not yet discovered or cannot measure, causing inevitable residual confusion; (e) since this study is based on a secondary analysis of previous research data [20], diabetes patients are not included in the data package, so it may cause a certain selection bias. But from another point of view, the positive correlation between diabetes and NAFLD has been proved many times in many previous studies [1, 2, 6], but this study still found that there is a positive correlation between WHtR and NAFLD even excluding patients with diabetes, so it can be considered that this association is relatively reliable.