This was a cross-sectional analysis of the Saku cohort, which was launched from 2009 in Saku Central Hospital Human Dock Center in Saku city, Nagano Prefecture, Japan. Participants who visited for a health checkup between May 5, 2009 to September 30, 2010, and who agreed to participate in the cohort were included in the study. From the study population at baseline (n = 2,565), we excluded subjects with missing data (n = 30), age < 50 years old (n = 350), and age ≥ 80 (n = 16). Of the remaining 2,169 participants, 301 participants were defined to have diabetes. According to the WHO criteria, diabetes was defined by either fasting plasma glucose levels ≥126 mg/dL, 2-h post-load glucose levels ≥200 mg/dL after a 75 g oral glucose tolerance test, or diabetes diagnosed by physicians. Of the remaining 1,868 participants, 542 participants who were defined to have impaired glucose tolerance or impaired fasting glucose according to the WHO criteria were excluded. Control participants were randomly selected from the remaining 1,326 participants and individually matched to cases on age and sex (n = 301). Of these 602 participants, 1 case–control pair was excluded from the analysis because of no remaining serum sample for 1 male case, leaving 300 diabetes cases and 300 matched controls in the analysis. This study was reviewed and approved by the Ethical Committee of the National Institute of Health and Nutrition and Saku Central Hospital. Participants received a precise explanation of the study and provided their written informed consent.
The height (cm) and weight (kg) of the subjects were measured with an automatic scale (Tanita, BF-220, Tokyo, Japan), in light clothing. The body mass index (BMI) was calculated as the weight (kg) divided by the squared height (m2). Waist circumference was measured twice at the umbilicus level while the subject was in a standing position using a fiberglass measuring tape; the average measurement was used for the analysis. Blood pressure was measured while the subject was in a sitting position using a validated automated blood pressure monitor (ES-H55; Terumo, Tokyo, Japan). The physical activity levels were obtained by asking the participants about their average frequency of physical activity: rarely/never, 1 to 3 times per month, 1 to 2 times per week, and more than 3 times per week.
Following an overnight fast, blood samples were collected at the time of each health checkup at the Saku Health Dock Center. Blood samples were collected in tubes containing EDTA and heparin for the measurement of the fasting plasma glucose, insulin, and HbA1c levels, and the remaining frozen serum samples were sent to the laboratory at the National Institute for Health and Nutrition and were stored in deep freezers. Serum gel separator tubes were used for the measurement of the total cholesterol, HDL cholesterol, and triglyceride (TG) levels. Routine laboratory blood analyses were performed at the Saku Central Hospital. HbA1c levels were measured using a high-performance liquid chromatography method (TOSOH HLC-723 G8; Tosoh Corporation, Tokyo, Japan), with intra- and inter-assay coefficients of variation (CVs) of 0.5–1.4% and 0.6–1.3%, respectively. The plasma glucose levels were analyzed using an enzymatic method (ECO glucose buffer; A&T Corporation, Kanagawa, Japan), with intra- and inter-assay CVs of 0.3–0.5% and 0.6–0.8%, respectively. The plasma insulin levels were analyzed using an electrochemiluminescence immunoassay (Modular E170; Roche Diagnostics, Mannheim, Germany), with intra- and inter-assay CVs of 0.5–2.0% and 3.2–3.6%, respectively. The serum albumin levels were measured with a modified bromocresol green method (Aqua-auto Kainos ALB Test Kit; KAINOS Laboratories Inc., Tokyo, Japan), intra- and inter-assay CVs of 1.2–2.2% and 1.7–2.4%, respectively. The serum γ-glutamyl-transferase (GGT) levels were analyzed with the Japan Society of Clinical Chemistry reference method (Cica Liquid γ-GT J; Kanto Chemical Co, Tokyo, Japan) and an autoanalyzer BM-2250 (Nihon Denshi, Tokyo, Japan), with intra- and inter-assay CVs of 1.0–3.7% and 0.96–3.65%, respectively. The serum total cholesterol, HDL cholesterol, and TG concentrations were determined using enzymatic methods (serum total cholesterol: Detaminar L TC II, Kyowa Medex, Tokyo, Japan; HDL cholesterol: Cholestest N HDL,Sekisui Medical Co. Ltd., Tokyo, Japan; and TG concentrations: Mizuho TG-FR Type II, Mizuho Medi, Saga, Japan) and an autoanalyzer BM-2250 (Nihon Denshi, Tokyo, Japan), with intra- and inter-assay CVs of ≤ 1.7% and ≤ 2.3%, respectively.
With the stored frozen samples, SHBG, testosterone, and estradiol were measured in the laboratory at SRL (Tokyo, Japan), blinded to cases and controls. Serum SHBG levels were analyzed with an immunoradiometric assay; (Siemens Medical Solutions Diagnostics, Los Angeles, CA, USA), and testosterone levels with an electro chemiluminescence immuno assay; (Roche Diagnostics GmbH, Mannheim, Germany). The intra- and inter-assay CVs for SHBG in the laboratory at SRL have been reported to be 1–3% and 7–8%, and the lower limit of detection for SHBG was 1.1 (nmol/L). The intra- and inter-assay coefficient of variations (CVs) were 1–3% and 2–3%, and the lower limit of detection was 0.04 ng/mL for testosterone. Of 600 total testosterone measurements, undetectable readings (1 control and 1 case among men, and 4 controls and 3 cases among women) were set to missing. Free testosterone levels were calculated using the methods described by Södergård et al. and Vermuelen et al..
The values for HbA1c were collected as Japan Diabetes Society (JDS) values, and then converted to National Glycohemoglobin Standardization Program (NGSP) values using the following conversion formula: HbA1c (NGSP) = 1.02×HbA1c (JDS)+0.25%.
We evaluated the fatty liver condition with the validated FLI derived from TG levels, BMI, waist circumference, and GGT levels as follows: exp[0.953×ln(TG) + 0.139×BMI + 0.718×ln(GGT) + 0.053×waist−15.745]/(1+exp[0.953×ln(TG) + 0.139×BMI + 0.718×ln(GGT) + 0.053×waist−15.745])×100. The FLI has a relatively high accuracy in detecting fatty liver (0.84 [95% CI, 0.81–0.87]) and studies have shown that the FLI is associated with higher hepatic-related and cardiovascular disease mortality, incidence of diabetes, and insulin resistance, risk of coronary heart disease, and early atherosclerosis.
We conducted all analyses by sex. Baseline characteristics were compared between case patients and control subjects using the paired t-test for continuous variables and the McNemar’s test for categorical variables. Pearson correlation coefficients (r) were calculated to evaluate associations between testosterone and SHBG levels, fasting insulin and glucose levels, HbA1c levels, BMI, and FLI among controls.
Odds ratios (ORs) and 95% confidence intervals were calculated according to quartiles based on the joint distribution of cases and controls by sex; the lowest quartile was used as the reference. We fitted conditional logistic regression models to estimate the association between SHBG and diabetes. In Model 1, we stratified on matched pairs using conditional logistic regression models. We further adjusted for smoking status (never, past, or current), physical activity (rarely/never, 1 to 3 times per month, 1 to 2 times per week, and more than 3 times per week), history of hypertension, family history of diabetes, alcohol use (almost none, occasional, or regular), menopausal status (women only), and BMI (continuous) (Model 2). To examine the impact of a fatty liver in the association, we additionally included the FLI (quartiles) in the model (Model 3). In Model 4, we further included levels of fasting insulin (quartiles), total testosterone (quartiles), and SHBG (quartiles) that were not examined as the primary independent variable. As sensitivity analyses, we additionally adjusted for total energy intake (quartiles) but this additional adjustment did not materially change the estimates. P- values for trend were computed based on median levels in categories. To assess nonmultiplicative interactions of testosterone and SHBG levels with gender, we fitted logistic models with product terms for each of these interactions, treating the biomarkers as continuous variables.
To further provide visual representation of the dose–response curve, we fitted restricted cubic spline models by including transformed variables of exposure variables to multiple logistic regression models (with 2 knots at the 33.3th and 67.7th percentiles) with adjustment for covariates included in Model 3. We conducted statistical analyses using SAS (version 9.3; SAS institute, Cary, NC) and STATA (version 12.0; StataCorp, College Station, TX).