The participants were patients who were admitted in the Department of Metabolic Medicine, Osaka University Hospital, between June 2021 and August 2022, and had undergone abdominal computed tomography (CT) scans for underlying diseases during their hospital stay or within 3 months before admission. The sample size was set according to our previous research (8). We excluded patients with type 1 diabetes mellitus, pancreatic diseases (e.g., pancreatic tumors or pancreatitis), endocrine diseases (e.g., Cushing syndrome or pheochromocytoma), or taking oral corticosteroid (prednisolone 10 mg or more). Diagnosis of pancreatic cancer and pancreatitis was made by experienced radiologists. The study was approved by the institutional ethics review board of Osaka University Hospital (approval number: 21062). The participants were informed about the study and gave written consent to participate.

We collected information on lifestyle factors through self-administered questionnaires distributed to the patients during the hospitalization period. For dietary habits, a validated food frequency questionnaire developed by the Japan Public Health Center (JPHC) (9, 10), including 128 food items and 13 beverage items, was used. For exercise habits, the International Physical Activity Questionnaire short form (11) was used to yield a score in MET-hours/week. The participants were also asked about the number of meals consumed per day, snacking habits, timing of each meal, and eating speed. Information about eating speed and sleeping habits (sleep duration, frequency of difficulty falling asleep, mid-awakening, early morning awakening and deep sleep disorder) were collected using a questionnaire based on the JPHC-NEXT interview form (12). The questionnaire on these lifestyle factors is shown in Supporting Information Table S1 . Smoking were assessed by the Brinkman index (number of cigarettes consumed per day multiplied by years of smoking) (13). Late-night eating was defined as eating after 9:00 p.m., and fast eating was defined as those who responded that their eating speed was fast/very fast according to previous study (14). Short sleep duration was defined as sleeping 5 hours or less per day (15).

In order to measure the extent of IPFD and intra-hepatic lipid accumulation (IHLA), unenhanced CT values were utilized, which have been demonstrated to exhibit strong correlation with histologically determined organ fat content (16). As previously reported (8, 17), we determined the CT values of the pancreas (P) by calculating the mean value of three 1cm² regions of pancreas (head, body, and tail). The measurements were carefully conducted by excluding the pancreatic duct and margins from the selected areas. Likewise, the CT values of the liver (L) was determined by calculating the mean CT value of three 1cm² regions of liver (anterior, posterior, and lateral). The CT values of the spleen (S) was determined by calculating the mean CT value of three 1cm² regions of spleen (upper, middle, and lower). Subsequently, by following previous studies, we quantified IPFD by calculating the difference between the mean pancreatic values and the mean splenic values (P-S) (8, 17, 18). Similarly, we quantified IHLA by calculating the difference between the mean liver values and the mean splenic values (L-S) (17, 19). Lower CT values indicate higher levels of fat accumulation. Measurements were performed by an experienced physician blinded to each patient’s information. To determine inter-observer variability in IPFD and IHLA measurements, another physician randomly selected 27 patients and independently measured their IPFD and IHLA. The intraclass correlation coefficients (ICC) and Bland-Altman plot between these two physicians were evaluated for these measurements. All unenhanced CT scanning was performed with a slice thickness of 5 mm. The images were analyzed using Aquarius Net Viewer Version 4.4 (TeraRecon, Inc., Tokyo, Japan).

The following data was obtained from the medical records at the time of participants’ admission: age; sex; height; body weight; Hemoglobin A1c (HbA1c); fasting plasma glucose; triglycerides (TG); high-density lipoprotein-cholesterol (HDL-C); and low-density lipoprotein-cholesterol (LDL-C) concentrations. Blood samples were obtained in the morning, prior to breakfast, on the day following admission. Diabetes mellitus were diagnosed according to Japanese diagnostic criteria (20).

The P-S and L-S values were square root transformed to normalize their distributions and then multiplicated to the range of 0–100 (with 0 representing lowest fat deposition and 100 as highest fat deposition). The numbers were defined as “IPFD score” and “IHLA score”, respectively. Multiple regression analyses were conducted to assess the associations between each lifestyle factor and the scores of IPFD and IHLA, adjusting for covariates including age (continuous), sex (male or female), diabetes status (presence of diabetes mellitus), and body mass index (BMI, continuous). Least squares geometric mean concentrations were calculated for lifestyle factors. The results were showed as Model 1, not adjusted; Model 2, adjusted for age and sex; Model 3, adjusted for age, sex, and diabetes status; and Model 4, adjusted for age, sex, diabetes status, and body mass index. In addition, to evaluate generalizability, sensitivity analysis was performed by stratifying by the presence or absence of diabetes. Each macro-nutrient and micro-nutrient intake provided by FFQ is adjusted for total energy intake by residual method. Statistical analyses were performed using R version 4.2.2. (https://www.r-project.org/) and two-sided p-value < 0.05 was considered to represent statistical significance. Any missing data were excluded from the statistical analyses.

Ultimately, 99 patients were enrolled in the study ( Figure 1 ). Table 1 presents the clinical characteristics of the participants upon hospitalization. The median age was 64 years and median BMI was 25.3 kg/m². The median CT values for the pancreas, liver and spleen were 40.0 HU, 56.9 HU and 49.0 HU, respectively. ICC (2,1) for IPFD (P-S) was 0.86 (95%CI: 0.71-0.93) and ICC (2,1) for IHLA (L-S) was 0.98 (95%CI: 0.96-0.99). Bland-Altman plots for IPFD and IHLA were shown in Supporting Information Figures S1 , 2 . For both measurements, more than 95% values were included within the limits of agreement. No significant fixed errors (p = 0.08 and 0.41, respectively) and proportional errors (p = 0.59 and 0.39, respectively) were observed. IPFD score 18.0 and IHLA score 18.7 correspond to 1SD.

We examined the associations of IPFD score with lifestyle factors ( Table 2 ) and nutrient intakes ( Supporting Information Table S2 ). Fast eating speed, late-night eating, low alcohol intake, early morning awakening and deep sleep disorder were significantly associated with high IPFD score after adjusting for age and sex (p=0.02, 0.01, 0.03, 0.02 and 0.03, respectively). After further adjustment for diabetes status and BMI, fast eating speed, late-night eating, and early morning awakening were significantly associated with high IPFD score (p=0.04, 0.01 and 0.01, respectively). For nutrient intakes, vitamin B12 intake tended to be negatively associated with IPFD in the age, sex, diabetes, and BMI adjusted model (p=0.05).

In addition to IPFD, we examined the associations of IHLA score with lifestyle factors ( Table 3 ) and nutrient intakes ( Supporting Information Table S3 ). After adjusting for age, sex, diabetes status, and BMI, Brinkman index, difficulty falling asleep, deep sleep disorder were significantly associated with high IHLA score (p=0.04, 0.03, 0.03, respectively). Fast eating speed, late-night eating, and early morning awakening, which were significantly associated with IPFD score, were not associated with IHLA score. For nutrient intakes, while negative associations were observed between intakes of potassium, β-carotene, vitamin B6, vitamin C, and folate and IHLA in the univariate model (p=0.02, 0.03, 0.01, 0.04, 0.02, respectively), all these associations were attenuated after adjusted for age and sex.

Considering the possibility that IPFD changes depending on the presence or absence of diabetes, analysis was performed separately in the patients with and without diabetes ( Table 4 ). In the patients with diabetes, fast eating speed were significantly associated with IPFD score (p<0.001), after adjusting for age, sex, and BMI. In the patients without diabetes, number of snacks, late-night eating and short sleep duration were significantly associated with IPFD score (p=0.02, 0.002 and 0.003, respectively), after adjusting for age, sex, and BMI.