Study population
An ongoing prospective cohort study, the Korean Genome Epidemiology Study (KoGES), was initiated in 2001. Detailed information on the study design and procedure has been previously reported [11]. At enrollment, the initial cohort of 10,030 participants, 40 to 69 years of age, was randomly recruited from two study sites, with 5012 participants from the urban community of Ansan (2518 men and 2494 women) and 5018 participants from the rural community of Ansung (2240 men and 2778 women). The participants were randomly selected from the general population by telephone, mail, and door-to-door visits. These participants have been biennially followed up. The participants underwent comprehensive tests, when visiting a research site, including physical examinations, biochemical and clinical examinations, and interviewer-administrated questionnaires. All participants provided written informed consent, and the study protocol was approved by the Human Subjects Review Committee at Korea University Ansan Hospital and Ajou University School of Medicine.
The SC type for each participant was identified from 2009 to 2012. Of 6878 eligible participants, 5840 were successfully classified by the SC type. Considering the SC type as a unique individual trait, we estimated the risk of developing a newly diagnosed metabolic syndrome according to the SC types for approximately 14 years from 2001 to 2014. At baseline, of 5840 participants who were classified into an SC type, we excluded 1860 adults who already had metabolic syndrome. Fifty-nine individuals were further excluded when diagnosed with an established cardiovascular disease. Additionally, fifty-six individuals were excluded due to missing data from any of covariates at the baseline examination (age; n = 1, sex; n = 1, body mass index (BMI); n = 1, education; n = 27, and glucose n = 26). Additionally, 336 participants were excluded for missing one or more follow-up examinations over the 14 years of the study. Therefore, 3529 participants (1769 men and 1760 women) were involved in the final analyses.
Classification of Sasang constitutional types
A new diagnostic model was used to identify the subjects’ SC types. Detailed information on this new diagnostic model has been previously reported [12]. Briefly, this is based on a probability model using multivariable logistic regression with individual data on facial image, body shape, voice features, and questionnaire results. Specifically, facial images obtained with a digital camera were processed to extract the variables of facial characteristics. Eight circumference measurements, at the forehead, neck, axilla, chest, ribs, waist, pelvis, and hips, were measured to process the variables of body shape. The variables of voice features were processed with the Hidden Markov Model Toolkit (Cambridge University Engineering Department, Cambridge, United Kingdom) and the Praat voice analysis program (University of Amsterdam, Amsterdam, Netherlands). The questionnaire consisted of 67 multiple-choice questions about general temperament, eating habits, and physiological symptoms, and was processed for the variables of personality characteristics and physiological symptoms. Since no participants were classified into the TY type, the SC types in this study involved just three types, TE, SE, and SY.
Definition of metabolic syndrome
Metabolic syndrome was defined according to the criteria of the National Cholesterol Education Program Adult Treatment Panel III [13]. Participants were identified with newly diagnosed metabolic syndrome if they had at least three out of the five following components: abdominal obesity (waist circumference for Asia-Pacific adults ≥90 cm for men and ≥80 cm for women) [14], elevated triglycerides (triglyceride ≥150 mg/dl), low HDL cholesterol (<40 mg/dl for men and <50 mg/dl for women), high blood pressure (systolic/diastolic pressure ≥ 130/85 mmHg or treatment with anti-hypertensive drugs), and elevated fasting blood glucose (≥110 mg/dl or treatment with anti-diabetic drug).
Other measurements
Following a standardized protocol, professionally trained interviewers and health professionals help all participants undergo anthropometric examinations. Height and weight of the participants were measured in light clothes with no shoes. Each participant’s BMI was calculated as weight divided by height (kg/m2). After resting five minutes in a seated position, participants had blood pressure measurements taken with a mercury sphygmomanometer (Baumanometer®, W.A. Baum Co., Inc., Copiague, NY, USA). After fasting at least eight hours overnight, participants underwent an early morning blood sample collection. The blood samples were delivered for assays at the Seoul Clinical Laboratory (Seoul, Korea). In this lab, the levels of triglycerides (TG), HDL cholesterol, and fasting glucose were assessed using a chemistry analyzer (ADVIA 1650, Siemens, Tarrytown, NY).
Statistical analysis
Continuous and categorical variables were examined using generalized linear models and chi-square tests. Multiple comparisons were conducted with Scheffe’s post hoc tests. Cox proportional hazard regression models were established to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for the risk of developing metabolic syndrome associated with the SC types. To compare the risk prediction models for metabolic syndrome, four models were established. Model 1 was adjusted for age, sex, BMI, education, income, smoking, drinking, and physical activity. Model 2 was further adjusted for total cholesterol and C-reactive protein (C-RP). Model 3 was additionally adjusted for fasting glucose, HDL cholesterol, triglyceride levels, and systolic and diastolic blood pressures. Model 4 was further adjusted for SC type. Estimated risks for the development of metabolic syndrome according to the SC types were calculated and displayed. To identify significant predictors according to each SC type, the forward selection method was used. The areas under the curve (AUC) from the receiver operating characteristic (ROC) according to the SC types were estimated to compare the accuracy of the risk prediction for incident cases of metabolic syndrome. Statistical analysis was performed with SAS version 9.4 (SAS Institute, Cary, North Carolina, USA). All p-values <0.05 were considered statistically significant.