Aims, study design, and participants
The aims of this study were to classify the dietary behaviour of overweight or obese individuals into subgroups using LCA and to explore the relationship between each of these subgroups and cardiometabolic risk factors. This cross-sectional study was based on a retrospective review of the medical records of 259 patients who visited an outpatient weight management clinic at a tertiary hospital and underwent a dietary behaviour assessment between January 2014 and February 2019. Patients who were younger than 18 years, those for whom height or weight data were missing, and those with no dietary behaviour assessment records were excluded. The study protocol was approved in advance of enrolling patients by the Institutional Review Board at Seoul National University Bundang Hospital (IRB no. B-1904–532-114). The need for informed consent was waived in view of the retrospective nature of the study and the anonymity of the data.
Sociodemographic and clinical variables
Our outpatient weight management clinic is attended by obese patients who want to lose weight or are referred by other departments. Patients who attend this clinic initially complete a self-administered questionnaire designed to collect information on sociodemographic and lifestyle factors, a weight history, and eating-related behaviours, and then attend a consultation with a family physician specialised in obesity management. At the next session, a dietician performs a dietary assessment, provides the patient with detailed education on nutrition, and prescribes a low or very low calorie diet. Next, the patient attends an appointment with a doctor who checks for adherence or barriers to the prescribed diet or the exercise recommendations and, if needed, prescribes appetite-suppressing medication. These sessions are repeated at intervals of 2 weeks for 1 month, 4 weeks for 2 months, and 8 weeks for 4 months until the patient has achieved a weight decrease of at least 10% of the initial body weight. Patients are then encouraged to attend a weight loss maintenance program, where they learn strategies for maintaining the weight lost, and regular follow-up sessions for at least 12 months.
In this study, we collected sociodemographic information on sex, age, income, and level of education from the self-completed questionnaires. Lifestyle factors, such as smoking, hazardous drinking, and frequency of exercise, were included. Anthropometric measurements, including height, weight, systolic and diastolic blood pressure, and body fat were obtained by bioelectrical impedance analysis using the InBody 720 device (BioSpace Inc., Urbandale, IA, USA). Clinical variables included total cholesterol, high-density and low-density lipoprotein cholesterol (HDL and LDL cholesterol), triglycerides (TG), fasting blood glucose (FBS), glycated haemoglobin (HbA1C), estimated glomerular filtration rate and liver function tests including aspartate transaminase (AST), alanine transaminase (ALT) and gamma-glutamyl transferase (GGT).
Since mood disorders such as depression might affect eating behaviours of study subjects, we extracted the patient’s diagnosis of depression by using the tenth revision of the International Statistical Classification of Diseases and Related Health Problems (ICD-10). An ICD-10 diagnosis of F20.4, F31.3-F31.5, F32.x, F33.x, F34.1, F41.2, or F43.2 and two depression prescriptions given within a 1-year period were considered proxies for having depression [18]. Diagnosis was extracted using electronic medical records for a period of approximately 2 years from the day the questionnaire was conducted.
Using the National Cholesterol Education Program—Adult Treatment Panel III criteria for Asian populations [19], a diagnosis of metabolic syndrome was made when at least three of the following five conditions were met:
- High blood pressure (systolic ≥130 mmHg or diastolic ≥85 mmHg) or on antihypertensive medication
- High FBS level (≥100 mg/dL) or current use of oral antidiabetic medication
- High TG level (≥150 mg/dL) or current use of lipid-lowering medication
- HDL cholesterol (<40 mg/dL in men, <50 mg/dL in women)
- Abdominal obesity (≥90 cm in men, ≥85 cm in women)
Dietary behaviour assessment
The self-administered dietary behaviour questionnaire included the following items, each of which was rated on a 5-point Likert scale: frequency of meals, skipping meals, frequent snacking or night eating, overeating or binge eating, preference for fatty foods or carbohydrates, consumption of instant food such as ramen, fast foods such as pizza or hamburgers, greasy food such as fried chicken, intake of sugar-sweetened beverages, and frequency of intake of carbohydrates, vegetables and fruit, protein, fat and dairy products; Table 1). Based on previous studies [20-23], we organised our dietary behaviour questions into three main domains with the following nine categories: food choice (frequently eating out, consumption of fast food, instant food, and takeaway food), eating behaviour (irregular meals, frequent snacking including at night, emotional eating, and overeating/binge eating), and nutritional intake (high-fat/high-calorie foods, salty food, and poorly balanced food intake).
We reclassified the food choice and eating behaviour domains as dichotomous variables and recorded ‘yes’ if items in those categories were responded to as ‘frequently’ or ‘very frequently’. For nutritional balance, we referenced the nutrition quotient calculation for Korean adults [24]. Briefly, the Korean nutrition quotient consists of nutrition balance, food diversity, moderation of amount of food intake, and dietary behaviour. Nutritional balance assesses the intake of vegetable/fruit, protein (fish, eggs, and beans), nuts and milk, and dairy products. If the calculated nutrition balance score was in the lower quartile, food intake was classified as unbalanced.
High-fat/high-calorie intake included high consumption of fatty foods, such as fried food, meat, ham, or sausage, and high consumption of sugar-sweetened beverages or foods containing sugar. Each score was summed to yield a mean score. If the mean high-fat/high-calorie score was >4, it was classified as ‘yes’, as for a dichotomous variable. Other latent variables were classified as ‘yes’ or ‘no’ according to the corresponding mean scores.
Statistical methods
Continuous variables are presented as mean and standard deviation and categorical variables as the frequency count and percentage. LCA was performed using PROC LCA (version 1.3.2) [25-27] and was used to classify the patients into nine subtypes of dietary behaviour. The Akaike information criterion and Bayesian information criterion were used to find the best representative number of subgroups in these data. The number of classes was selected by a combination of parsimony and interpretability, and the class number can be possibly used as a meaningful label for each class [27]. Each class was described in terms of sociodemographic factors, anthropometric measurements, and laboratory results.
We examined the association between LCA-derived classes and cardiometabolic risk factors using the LCA distal outcome Macro program (http://methodology.psu.edu/downloads). Briefly, LCA with a distal outcome was constituted with a ‘three-step’ method (Bolck–Croon–Hagenaars method) [28], whereby the parameters of the LCA model are first estimated without the distal outcome, the posterior probabilities of class membership are then used to compute a weighting variable, and finally, the weighting variable is used to calculate a weighted average of outcomes for each class. The LCA-driven classes were subsequently entered into a logistic regression model to test for associations between each class (with the relatively healthy class chosen as the referent) and the risk of metabolic syndrome with adjustment for age, sex, healthy behaviours, BMI, and other clinical variables. All statistical analyses were performed using SAS version 9.3 (SAS Institute Inc., Cary, NC, USA).