Lean Nonalcoholic Fatty Liver Disease and Risk of Incident Diabetes in a Euglycemic Population Receiving Health Checkups: A Cohort Study

Background: Although recent evidence suggests that nonalcoholic fatty liver disease (NAFLD) is associated with insulin resistance and an increased risk of diabetes, the association between lean NAFLD and incident diabetes is unclear. This study aimed to investigate whether lean NAFLD and overweight/obese NAFLD have similar or dissimilar effects on the risk of new-onset diabetes. Methods: A longitudinal study was performed in 14,482 euglycemic adults who participated in a health check-up program. Fatty liver was diagnosed by abdominal ultrasonography. The outcome of interest was incident diabetes.Cox proportional hazards regression models were applied to calculate HRs with 95% CIs for future diabetes risk. Results: During the median 6.0 years of follow-up, 356 cases of diabetes occurred. Despite a low probability of hepatic brosis indicated by the BAAT score, lean NAFLD was positively associated with an increased risk of diabetes. Moreover, after adjusting for sociodemographic and potential confounders, the fullyadjusted HRs (95% CIs) for incident diabetes between lean NAFLD and overweight/obese NAFLD to the reference (lean without NAFLD) were 2.58 (95% CI 1.68 to 3.97) and 2.52 (95% CI 1.79 to 3.55), respectively. In post hoc analysis, the HR (95% CI) for diabetes comparing lean NAFLD to obese/overweight NAFLD was 1.02 (95% CI 0.68 to 1.54, p = 0.909). The results were robust to challenges in multiple subgroup analyses and appeared to be more pronounced for female participants (p for interaction = 0.005). Conclusions: In this cohort study, lean patients with NAFLD had a risk of incident type 2 diabetes similar to that of overweight/obese ones with NAFLD. These ndings suggest that lean NAFLD is not a benign condition. Further investigations are needed to gain a better understanding of the pathogenesis and natural history of NAFLD in lean subjects.

Lei Chen (  chl1221@hotmail.com ) First a lated Hospital of Xi'an Jiaotong University ones with NAFLD. All these factors deliver a key message that lean NAFLD is a distinctive phenotype rather than abyproduct of BMI. [12] The liver is a key organ that plays critical roles in the regulation of systemic glucose and lipid metabolism. [13] A recent meta-analysis involving 19 observational studies with 296,439 participants convincingly demonstrated that NAFLD is signi cantly associated with a twofold increased risk of newonset diabetes (random-effects hazard ratio 2.22, 95% CI 1.84-2.60; I2 = 79.2%). [14]However, data describing the association between lean NAFLD and type 2 diabetes (T2DM) risk are scarce. Therefore, we conducted a longitudinal cohort study to investigate the effect of lean NAFLD and its severity on the risk of incident diabetes.

Date sources and study population
This is a observational cohort study performed by using participants' date from the DRYAD public database (https://datadryad.org). Initially, the records of 20,944 participants who attended a comprehensive medical examination program at Murakami Memorial Hospital between 2004 and 2015 were extracted by Okamura T et.al. [15,16] The details of the medical health check-up programme were described previously. [15,17]Individuals were excluded at baseline for the following reasons: 1) no available records for abdominal ultrasonography and important variables including age, sex, BMI, waist circumference (WC), blood pressure, or fasting plasma glucose (FPG), triglycerides (TG), and glycated hemoglobin (HbA1c) levels; 2) other known chronic liver diseases, such as liver cirrhosis (history or ndings on ultrasound), viral hepatitis (de ned by serum positive serological markers for HBV or HCV), or alcoholic fatty liver disease (mean alcohol consumption at least 60 g for males and 40 g for females per day); 3) use of any medication; or 4) diagnosed with diabetes or impaired fasting glucose (IFG) at baseline. To avoid reverse association, individuals with a follow-up period < 1 yearwere excluded, Moreover, participants with an unde ned diabetes status at the follow-up visit were also excluded. Some participants met more than one exclusion criterion. Ultimately, 14,482 subjects were selected for further analysis in the present study ( Figure 1).
This study conformed to the Declaration of Helsinki.Given that our date were obtained from the public database, no prior ethical approval was required. The requirement for informed consent was also waived as the data were anonymous.

Data acquisition
As described in the previous study, [15] at each visit to the health check center, the participants' demographic data, including age, sex, smoking status, drinking status, exercise habits and medication history, were acquired from a standardized questionnaire by the same trained team of interviewers.
Smoking status was categorized as non, ex-, or current smoker. [18] Individuals who performed any type of physical activity at least once a week on a regular basis were considered regular exercisers. [19] Average alcohol consumption per week was calculated by asking the frequency and amount of alcoholic beverage during the prior month. Grade of alcohol consumption was de ned as follows: no or minimal (< 40 g/wk); light (40 -140 g/wk); moderate (>140 g/wk). [20] Physical parameters, including height, weight, waist circumference, and blood pressure, were measured by trained investigators under standardized conditions according to a standard protocol. BMI was calculated as weight in kilograms divided by height in meters squared, and the result is expressed in units of kg/m 2 .
Venous blood specimens drawn from the antecubital vein were obtained after an overnight fast of at least 8 h. Laboratory analyses, including TGs, total cholesterol (TC), high-density lipoprotein cholesterol (HDL-c), gamma glutamyltransferase (GGT), HbA1c, FPG, aspartate aminotransferase (AST), and alanine aminotransferase (ALT) levels, were carried out in accordance with relevant guidelines and regulations using a Modular Analytics system (Hitachi High-Technologies Corp., Ltd., Tokyo, Japan), which is widely applied for biochemical analysis in Japan.
Abdominal ultrasonographic examinations were performed by trained sonographers using an Aloka SSD-650CL ultrasound machine (Aloka Co., Ltd., Tokyo, Japan) at baseline, and all ultrasonographic images were stored as photocopies. [18] Fatty liver was diagnosed by gastroenterologists according to the following four known criteria: hepatorenal echo contrast, liver brightness, gradual attenuation of far-eld, and vascular blurring. [21] The clinicianswere all blinded to the clinical data of the participants.

Endpoint and de nitions
The endpoint was the occurrence of incident diabetes during the follow-up period. According to the diagnostic criteria of the American Diabetes Association, [22] diabetes was de ned as FPG level ≥ 7.00 mmol/L and/or HbA1c ≥ 6.5% and/or self-reported diabetes that was previously diagnosed by a physician and/or current use of anti-hyperglycemic agents.
NAFLD was de ned as the presence of fatty liver in the absence of excessive alcohol consumption. There is no standard de nition of lean NAFLD; however, numerous studies recommended a cut-off point of BMI 23 kg/m 2 for Asian populations, [23][24][25] and the following four groups were assessed: lean without NAFLD, lean with NAFLD, overweight/obese without NAFLD, and overweight/obese with NAFLD. For further NAFLD categorization, a representativenoninvasive score was used to assess the severity of brosis. BAAT scores consist of the sum of the following categorical variables: BMI (≥ 28 kg/m 2 = 1), age (≥ 50 years=1), ALT [≥ 2UNL (male ≥ 60 IU/L; female ≥ 40 IU/L) =1], and triglycerides (≥ 1.7 mmol/L= 1). Patients with NAFLD were further categorized into two groups: low (BAAT < 2) and high (BAAT ≥2) probability of advanced brosis. [26] Visceral-fat obesity was de ned as a waist circumference≥90 cm in males or ≥ 80 cm in females. [27] Statistical analysis Categorical variables are presented as counts (percentages), and continuous data are expressed as means (standard deviations, SD). The characteristics of the study participants at baseline are summarized in Table 1, and signi cant differences among the four groups were analyzed by one-way analysis of variance (ANOVA) followed by the LSD post hoc test and Chi-square test for continuous variables and categorical variables, respectively. < 0.001 < 0.001 † p values using ANOVA for comparisons among four groups; ‡ p value using LSD-t as the post hoc analysis for comparing lean NAFLD group and overweight /obese with NAFLD group.
Data are means (s.d.) and counts (percentages); ALT, alanine aminotransferase; AST, aspartate aminotransferase; BMI, body mass index; BP, blood pressure; FPG, fasting plasma glucose; GGT, gamma-glutamyltransferase, HbA1c, hemoglobin A1c; HDL, high-density lipoprotein; NAFLD, nonalcoholic fatty liver disease; TC, total cholesterol. Cox proportional hazards regression analyses were performed to calculate hazard ratios (HRs) with 95% con dence intervals (CIs) for incident diabetes according to different phenotypes, with patients with a lean status and without NAFLD de ned as the reference group. We used three models with progressive adjustments: model 1 was adjusted for age and sex; model 2 was further adjusted for smoking status (non-, ex-, or current), grade of alcohol consumption (no/minimal, light, or moderate), and regular exerciser; model 3 was further adjusted forvariables associated with metabolic syndrome,including visceral-fat obesity (presence or absence), blood pressure (SBP and DBP), triglycerides, total cholesterol, HDL-c, and HbA1c. Potential confounders in multivariable models were selected based on their associations with the outcome or a change in effect estimate of more than 10%.
Strati ed analyses were conducted in various subgroups, and their interactions were also tested. Each strati cation adjusted for all the factors (age, sex, smoking status, grade of alcohol consumption, regular exerciser, visceral-fat obesity, blood pressure, triglycerides, total cholesterol, HDL-c, and HbA1c), except for the strati cation factor itself. Moreover, we evaluated the effect of the severity of NAFLD on incident diabetes using Cox proportional hazards modeling, in which models were not adjusted for age or triglycerides, as these factors were included in the calculation of BAAT scores. Statistical analyses were conducted in IBM Statistical Package for the Social Sciences (SPSS) 21 (version 21.0, Armonk, NY). A two-tailed p-value < 0.05 was considered statistically signi cant.

Results
The basic characteristics and lifestyle of the participants are summarized in Table 1. The present study enrolled 14,482 individuals with a mean age of 43.7 years and a mean BMI of 22.1 kg/m 2 ; 54.5% was male (n = 7898). The proportions of lean without NAFLD, overweight/obese without NAFLD, lean with NAFLD, and overweight/obese with NAFLD were 61.5% (8900), 20.8%(3008), 3.5% (514), and 14.2% (2060), respectively. In comparison with NAFLD patients with obesity/overweight, NAFLD patients with a lean status appeared to have a better biochemical pro le. In detail, patients with NAFLD and obesity had higher levels of blood lipids (triglycerides and total cholesterol), elevated blood pressure (SBP and DBP) andincreased waist circumference and liver enzymes (ALT, AST, and GGT) (all p< 0.05, Table 1).
During the median 6.0 years of follow-up, there were 356 cases of T2DM (271 men and 85 women). The average follow-up period did not differ signi cantly among the four groups (p = 0.351, data not shown).
The cumulative incidence of diabetes in patients with NAFLD was almost 7.5 times higher than that in subjects without NAFLD (13.31 vs 1.82 per 1000 person-years, p for log rank test < 0.001, date not shown). Of the 2574 patients with NAFLD at baseline, the incidence densities of new-onset diabetes were 9.61 and 14.26 per 1000 person-years in individuals with a lean and obese/overweight status, respectively (p for log rank test = 0.032, Table 2). Additionally, we found that the overall incidence of diabetes in men was higher than in women (5.24 vs 2.08 per 1000 person-years, p for log rank test < 0.001, data not shown). Table 2 Hazard ratios (HRs) of incidence of diabetes in relation to non -alcoholic fatty liver disease (NAFLD) and body mass index (BMI) status in overall participants. CI, con dence interval; NAFLD non-alcoholic fatty liver disease.
The reference group was lean without NAFLD; a 95% CI that does not include 1.00 were considered statistically signi cant.
Crude model was unadjusted;Model 1 was adjusted for age and sex; Model 2 was adjusted for model 1 plus smoking status, grade of alcohol consumption, and regular exercise; Model 3 was further adjusted for model 2 plusvarious variables associated with metabolic syndrome including presence or absence visceral fat obesity, blood pressure levels, triglycerides, total cholesterol, HDL-c, and HbA1c at baseline.
Visceral fat obesity was de ned as waist circumference >90 cm in male and >80 cm in female.
The hazard ratios of incident T2DM in relation to the four phenotypes are provided in Subsequently, strati ed analyses by subgroup de ned by age, sex, smoking status, and regular exerciser were carried out, andthe positive association between lean NAFLD and future risk of diabetes persisted in all subgroups. Interestingly, this relationship appeared to be more substantial in females (p for interaction = 0.005, Figure 2 and   CI, con dence interval; NAFLD, non-alcoholic fatty liver disease.
The reference group was lean without NAFLD; a 95% CI that does not include 1.00 were considered statistically signi cant.
*Adjusted for sex, smoking status, grade of alcohol consumption, regular exercise, presence or absence visceral fat obesity, blood pressure levels, total cholesterol, HDL-c, and HbA1c at baseline.

Discussion
In this longitudinal study of 14,482 euglycemic adults without IFG and diabetes at baseline, subjects with a lean status and NAFLD diagnosed using ultrasound had a signi cantly higher risk of new-onset type 2 diabetes. Notably, among our study participants, those with NAFLD who were lean had approximately the same risk of eventually developing diabetes type 2 as individuals with NAFLDwho were overweight/obese, even though the former exhibited lower levels of metabolic syndrome risk factors. This positive effect was evident in all subgroups considered and more pronounced in female participants. Moreover, lean subjects with a low probability of hepatic brosis indicated by a BAAT score also had a higher risk of developing diabetes after adjusting for sociodemographic and key confounding factors.
Convincing evidence has shown that NAFLD is involved in the early pathogenesis of type 2 diabetes, as it contributes substantially to the development of insulin resistance. [13,28,29] Our data also con rm that the incidence rate of T2DM in NAFLD patients compared to those without NAFLD was more than sevenfold. Regardless, there are limited studies investigating the impact of lean NAFLD on diabetes incidence. For example, a cohort study of 4629 Japanese adults conducted by Takuya et al. rst indicated that the diabetes risk in NAFLD individuals who were not overweight (BMI < 23.0 kg/m 2 ) was higher than that in subjects without NAFLD who were overweight;however, the sample size was relatively small, andthere was a lack of adjustment for metabolic factors. [30] Recently, another large cohort study from Korea (n = 51,463) reported that the presence and severity of NAFLD in adults of normal weight (BMI of 23 -24.9 kg/m 2 ) had a close association with increased diabetes risk. [31] Unfortunately, the in uence of waist circumference on the association of lean NAFLD with diabetes risk was not fully considered in these studies, and in fact, waist circumference is closely linked to diabetes, regardless of BMI. [32] In our study, after adjustment for waist circumference and other known confounders, we further con rmed that individuals with NAFLD who were lean had a signi cant risk of incident T2DM.
Moreover, our study found that lean NAFLD carries a risk of T2DM that rivals the risk in overweight/obese NAFLD,even though the overall changes in metabolic risk factors were more slight in the lean NAFLD group, which might indicate that subjects with NAFLD who are lean have a severe liver disease similar to that of subjects with NAFLD who are overweight/obese. Such speculation is supported by a recent study with 664 biopsy-proven NAFLD Asian patients showing that the severity of hepatic histology in nonobese NAFLD is comparable to that in obese NAFLD. [33] Consistently, a study of 466 Caucasian subjects with biopsy-proven NAFLD demonstrated that lean NAFLD is characterized by a severe histological picture similar to that of obese NAFLD and the former cases are more progressed compared to overweight cases. [34] In our study, we attempted to evaluate the association between lean NAFLD severity and future diabetes risk. Numerous noninvasive scoring systems, including the NAFLD brosis score, BAAT score, and BARD score, have been proposed to stage NAFLD, though there is no consensus to date. [35]To eliminate potential confounding, the BAAT score was selected to de ne NAFLD with advanced liver brosis, as it does not involve information about glycemic status. [26] It has been reported that the sensitivity of BAAT scores for the identi cation of advanced brosis is 94.9% with a cutoff value of 2. [35] Unfortunately, as very few subjects who were lean had evidence of advanced brosis basedon BAAT score in our study (n = 37), there was limited power to evaluate associations between liver brosis and future diabetes risk in lean NAFLD. Nevertheless, we also noted that NAFLD subjects who were lean, even those with a low BAAT score, had a signi cantly elevated incidence of diabetes.
In addition, the positive association between NAFLD and incident diabetes was more prominent in female participants. Similarly, a recent study conducted by Alina M.Allen et al. showed that females with NAFLD might lose the cardiovascular disease protection conferred by sex. [36]Although the exact reason for this sexual difference remains unclear, there are several possible explanations.First, NAFLD refers to a spectrum of lesions ranging from pure steatosis to nonalcoholic steatohepatitis (NASH) and advanced brosis. Most studies note a sex difference in this process, though con icting data exist. A study conducted by Bambha et al. pointed out that womenare already twice as likely as men to develop from NASH after adjusting for demographic and metabolic factors. [37]In agreement, a recent systematic review involving 54 studies showed that once NAFLD is established, femalesare at a higher risk for advanced brosis than are men,especially after the age of 50 years. [38]It has been proven that patients with NASHhave more severe adipose tissue insulin resistance and hyperinsulinemia thando patients with simple steatosis, which might promote the pathogenesis of diabetes. [39]Additionally, differences in skeletal muscle mass according to sex is another potential mechanism. [40] Several limitationsshould be considered in interpreting the results of our study. First, fatty liver was de ned by abdominal ultrasound, which might be inaccurate when fat in ltration upon liver biopsy is < 30%. Liver biopsy is the gold standard for diagnosing and staging NAFLD; however, it is not applicable and realistic for large-sample surveys due to its invasive nature and high cost. Regardless, a metaanalysis containing 46 studies revealed that ultrasound had high sensitivity and speci city for the evaluation of hepatic steatosis when compared to histological data (73.3% and 69.6%, respectively).
[41]Second, the diagnosis of diabetes in our study was mainly based on self-reporting or single measurements of HbA1c or FPG, without repeated con rmation on at least two separatedays; nevertheless, this is an intrinsic limitation of all large observational surveys. Third, selection bias existed, in that our study participants were recruited from a health promotion center, and thus they might be more concerned about their health than the general population;hence, the generalizability of our study to other populations is uncertain. Finally, we adjusted as many important variables as possible (e.g., age, waist circumference, lifestyle factors). Given the nature of observational studies, residual confounding by unmeasured factors (e.g., insulin resistance and genetic variability) is unavoidable. Despite these limitations, the large sample size, long follow-up time, and use of a comprehensive and standardized databaseshould provide reliable support for the relationship between lean NAFLD and future diabetes risk and pave the way for future prospective and histologically based studies.

Conclusions
In summary, this analysis of a large community cohort supports the existence of a signi cant association between lean NAFLD and an increased risk of incident diabetes. More importantly, our study adds novel evidence that lean NAFLD has a similar diabetes risk as overweight/obese NAFLD. Considering that patients with lean NAFLD are typically asymptomatic and always fail to seek medical advice, individuals at high risk for lean NAFLD should be followed up and assessed regularly to prevent associated complications through appropriate intervention in clinical settings. Currently, the optimal management for subjects with lean NAFLD is unclear, and lifestyle interventions, including diet and physical activity, might be effective in this population[42] and should be further explored. This study conformed to the Declaration of Helsinki. Given that our date were obtained from the public database, no prior ethical approval was required. The requirement for informed consent was also waived as the data were anonymous.

Consent for publication:
Not applicable.
Availability of data and materials: The datasets generated during and/or analysed during the current study are available in the DRYAD repository (https://datadryad.org). [16] Competing interests: Flow diagram for cohort recruitment.