Association between the cardiometabolic index and non-alcoholic fatty liver disease: insights from a general population

NAGALA is a cohort study based on the general population, which has been going on since 1994. It aims to evaluate the risk factors of common chronic diseases in the general population, and provides useful research data for the early prevention and treatment of chronic diseases. The design scheme of NAGALA has been described in detail previously [26]. This study was a post hoc analysis of NAGALA, and a cross-sectional design was adopted according to the new research hypothesis, in which the available research data have been uploaded to the DRYAD database by Hamaguchi et al. [27]. This study extracted data from a general population of 20,944 people recruited in the NAGALA program who took a comprehensive health examination in Murakami Memorial Hospital from 2004 to 2015. According to the purpose of the study, subjects with alcoholic/viral hepatitis, diabetes, impaired fasting blood glucose, incomplete covariates and excessive drinking were excluded, and 14,251 subjects were included in the final analysis of this study. Since NAGALA's research scheme has been previously approved by the Murakami Memorial Hospital institutional review committee (IRB number: 2018-09-01), and the subject's identification information has been replaced with a check code, there is no needless for separate ethical approval and informed consent in this study.

Clinical characteristic parameters such as sex, height, weight, WC, age, smoking/drinking status, systolic/diastolic blood pressure (S/DBP) and habit of exercise were evaluated and recorded by trained medical workers with a standard questionnaire. The subjects were classified according to their weekly alcohol consumption and smoking status, and the drinking status was divided into three categories: non or less drinking (< 40 g/w), light drinking (40–139 g/w) and moderate drinking (140–209 g/w) [28]. Smoking status can be divided into three categories: non-smoking, past smoking and current smoking. Habit of exercise was divided into groups according to the subjects' weekly exercise, in which regular participation in any type of exercise at least once a week was considered to have exercise habits. Hematological samples used to analyze biochemical parameters were collected at least 8 h after fasting, and an automatic analyzer was used to analyze and determine gamma-glutamyl transferase (GGT), hemoglobin A1c (HbA1c), aspartate aminotransferase (AST), HDL-C, total cholesterol (TC), TG, fasting blood glucose (FPG) and alanine aminotransferase (ALT) according to the standard method.

As previously described [26], the diagnosis of NAFLD in the NAGALA project was carried out by ultrasound. Ultrasound examinations were performed by professional technicians, and ultrasound images were evaluated by experienced gastroenterologists without knowing the subjects' other examination and diagnostic information, based on the echo of the hepatorenal echo contrast, liver brightness, deep attenuation and vascular fuzzy four criteria for the score of 0 to 4 points, a comprehensive score greater than 2 points in the diagnosis of NAFLD [29].

The statistical software Empower Stats (version 2.0) and R language (version 3.4.3) were used for data analysis. To further understand the correlation between CMI and NAFLD, this study took CMI as a continuous variable and a categorical variable for correlation analysis. The baseline characteristics of all patients were stratified by CMI quartile, and categorical variables and continuous variables were described as percentages or mean (standard deviation: SD) or median (interquartile range) according to the type and distribution pattern of variables. Chi-square test or one-way ANOVA or rank-sum test were used for inter-group comparison, and linear regression was used to check the association between baseline variables and CMI. Univariate and multivariate logistic regression models were then used to determine whether CMI was independently associated with NAFLD, and the results show that the odds ratio (OR) and the corresponding 95% confidence interval (CI). In the multivariate regression analysis, we ran a total of three models. According to the STROBE statement [30], we adopted different variable adjustment strategies in the three models. Model one, as a fine-tuning model, only adjusted the non-collinear general demographic variables (Additional file 1: Table S1), including sex, age, body mass index (BMI), drinking status and smoking status. In model two, the non-collinear covariates (that affecting the matching risk of CMI and NAFLD by more than 10% were regarded as confounding variables) were adjusted [31]. Model three further adjusted the non-collinear covariates that were statistically related to NAFLD in univariate analysis based on Model two. Additionally, we also conducted an exploratory stratified analysis among people of different ages, genders, BMI and whether or not to preserve exercise habits, and checked whether the effect of CMI on NAFLD was different among different subgroups by the likelihood ratio test. The receiver operating characteristic (ROC) curve was constructed to evaluate the accuracy of WC, height, TG, HDL-C and CMI in determining NAFLD, and the best cut-off value of CMI was calculated. In addition, we further used restricted cubic splines to flexibly model the association between the continuous variable CMI and NAFLD.

Overall 14,251 subjects were enrolled in this study, and the study population was quartered according to the CMI quartile. Table 1 summarizes the clinical and biochemical characteristics of the study population. The prevalence rates of NAFLD in the four groups were 1.26%, 5.96%, 17.56% and 45.52%, respectively. There were significant differences in parameters and indicators among the CMI groups (all P < 0.0001). Compared with the group with low CMI, the group with high CMI was older, had more people with smoking and drinking status, and had higher anthropometric parameters such as height, weight, BMI, WC and blood pressure, while the serological parameters except HDL-C increased gradually with the increase of CMI.

The correlation between CMI and baseline variables was evaluated using linear regression, and the results are summarized in Table 2. It is obvious from the table that all the baseline variables have a linear correlation with CMI, in which habit of exercise, HDL-C has a negative correlation with CMI, while other variables have a positive correlation with CMI. It is worth noting that the related variables of blood glucose and blood lipid metabolism were strongly correlated with CMI, and the correlation between WHtR and CMI is the strongest. This finding hints at a possibility that CMI has the value as an index of metabolic diseases; Additionally, these variables significantly correlated with CMI may be cofactors for the association between CMI and NAFLD.

Table 3 shows the results of univariate logical regression analysis. In univariate analysis, except for drinking status, all the covariables were significantly correlated with NAFLD, and the correlation between CMI and NAFLD was the highest (OR 19.64, 95% CI 16.97–22.73); These covariates related to NAFLD will be incorporated into the multivariable model as confounding factors. In order to be consistent with the results of previous studies [22‐25], CMI was also converted into Z scores for analysis in this study. In the multivariate regression analysis (Table 4), after adjusting the general demographic variables, model one found that CMI was positively correlated with NAFLD (OR 1.67 per SD increase, 95% CI 1.58–1.77), and the NAFLD risk gradually increased among the CMI quartiles [OR (Q1:1, Q2:4.95, Q3:16.64, Q4:65.29, P for trend < 0.0001). Model two adjusted for non-collinear covariates that altered the matching risk of CMI and NAFLD by more than 10%, the positive correlation weakened slightly (OR 1.28 per SD increase, 95% CI 1.20–1.37, P for trend < 0.0001). After further adjusting the non-collinear covariates related to NAFLD in univariate analysis (Model three: OR 1.28 per SD increase, 95% CI 1.19–1.37, P for trend < 0.0001), the positive correlation between them remained stable. In short, CMI is an independent risk factor for NAFLD.

In the subgroup analysis, the researchers further evaluated whether CMI and NAFLD differed between different populations by stratification analysis and interactivity tests. The age and BMI were stratified according to the commonly used clinical cutoff points, and the results of stratified analysis and interactive test suggested that (Table 5): there were significant differences between CMI and NAFLD in different ages and genders (P for interaction < 0.05), but no significant statistical differences were found in BMI and habit of exercise stratification (P for interaction > 0.05). Of note, this study found that the CMI-related NAFLD risk of young people was significantly higher than that of other age groups (OR = 2.63 per SD increase for young people, OR = 1.38 per SD increase for young and middle-aged people, OR = 1.18 per SD increase for middle-aged and elderly people; OR = 1.14 per SD increase for elderly people, P for interaction = 0.0010), and the CMI-related NAFLD risk of women was significantly higher than that of men (OR = 1.58 per SD increase for women, OR = 1.26 per SD increase for men, P for interaction = 0.0045).

ROC analysis was performed to calculate WC, height, TG, HDL-C and CMI for determining the area under the curve of NAFLD. The results showed that among these parameters, the accuracy of WC in determining NAFLD was the highest, followed by CMI, and the best cutoff value of CMI in determining NAFLD was 0.3241 (Table 6). In order to further verify whether the cutoff value used by CMI to determine NAFLD was stable, we continue to use restricted cubic splines to simulate the association between CMI and NAFLD. As shown in Fig. 1, when CMI was about 0.35, it may be a potential threshold point for NAFLD risk.