The data of the participants with impaired kidney function were retrieved from the database of the Mei Jau (MJ) private health screening centers in Taiwan from 2008 to 2010. The MJ Group has four health screening centers located in Taipei, Taoyuan, Taichung and Kaohsiung, and provides health examination periodically to its members. Participants completed a questionnaire about sociodemographic data, lifestyle and dietary habits prior to anthropometric and biochemical measurements. All participants signed the informed consent authorized by the MJ health screening centers, and the data without personal identification were used for research only. Eligible participants (n = 112,140) were aged ≥ 40 years and had impaired kidney function with estimated glomerular filtration rate (eGFR) < 90 mL/min/1.73 m2 and positive urinary protein. We excluded those who had any types of cancer or virus infection (n = 48,169), history of any transplantation (n = 1,765), error values in blood analysis and anthropometric measurements (n = 1,266), missing data in dietary assessment and other covariates (n = 26,605), not complete the questionnaire (n = 212) and multiple entries (n = 8,554). Finally, 25,569 participants were included in the analysis. Taipei Medical University-Joint Institutional Review Board approved this study (TMU-JIRB N201802006).
Assessment of anthropometric and biochemical variables
Body weight and height were observed by an auto-anthropometer (Nakamura KN-5000A, Tokyo, Japan), and body mass index (BMI) was calculated as the ratio of weight (kg) to the square of height (m2). Waist or hip circumference was measured by a flexible tape. Blood pressure was recorded twice at a 10-minute interval after resting for 5 minutes in the sitting position using a standardized sphygmomanometer. Participants were overnight fasting at least for 8 hours before a blood test. Uncompensated Jaffe method with alkaline picrate kinetic test was used to measure creatinine levels and eGFR was calculated using the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation . Meanwhile, urinary protein was measured by Roche Miditron M semi-automated computer-assisted urinalysis system (Combur-10 test M dipstick, Basel, Switzerland). Fasting blood glucose (FBG) and blood lipids such as triglycerides (TG), high density lipoprotein cholesterol (HDL-C), low density lipoprotein cholesterol (LDL-C) and total cholesterol (TC) were analyzed (Toshiba C8000 auto-analyzer, Tokyo, Japan) at the MJ health screening central laboratory. The coefficient of variation for all variables ranged from 1% to 3%. Hypertension and type 2 diabetes were defined as described in the previous study . The definition of hypertension included at least one of the followings: systolic blood pressure ≥ 140 mmHg, diastolic blood pressure ≥ 90 mmHg, use of antihypertensive medication or self-reported hypertension. The definition of diabetes included at least one of the following: (1) FBG ≥ 7.0 mmol/L (≥ 126 mg/dL), (2) use of hypoglycemic medication or (3) self-reported diabetes. The definition of MetS for Asians was to have at least three or more of the followings: (1) waist circumference ≥ 90 cm in men or ≥ 80 cm in women, (2) systolic blood pressure (BP) ≥ 130 mmHg, diastolic BP ≥ 85 mmHg or on anti-hypertensive drug treatment, (3) TG ≥ 1.70 mmol/L (150 mg/dL) or on treatment for lipid abnormality, (4) HDL-C < 1.03 mmol/L (40 mg/dL) in men, < 1.30 mmol/L (50 mg/dL) in women or on treatment for lipid abnormality, (5) FBG ≥ 5.6 mmol/L (100 mg/dL) or on anti-diabetic drug treatment .
Assessment of dietary habits and other covariates
Dietary habits were obtained using standardized and validated self-administered semi quantitative food frequency questionnaire (SQ-FFQ) [17, 18]. Initially, the questionnaire had 85 closed-ended questions on individual food items, twenty-two non-overlapping food groups were classified after standardization and validation as mentioned previously . Participants reported the consumption frequency of each food group on a daily or weekly basis in the past month . The consumption frequency described by the portion size of a bowl, a glass or a serving for one-time intake was categorized into five response options from the lowest to the highest frequency as mentioned previously . The detailed information about the food groups are provided in additional file 1: Table S1.
Demographic (age, gender, education level, income and marital status) and lifestyle variables (smoking, drinking, sleep quality and physical activity) were recorded using a self-administered questionnaire. Smoking status was classified as ‘yes’ if the participant smoked a cigarette occasionally or daily and as ‘no’ if otherwise. Drinking alcohol was also categorized as ‘no’ (< 1 time/week) and ‘yes’ (≥ 1-2 times/week). Physical activity was assessed by self-reporting intensity (light, moderate and heavy or intense), duration (hours) and frequency (per week) in the last two weeks. For sleep quality, participants filled the questions regarding sleep quality and average daily sleep duration in the last month. Sleep quality had five response options (difficulty to fall asleep, difficulty maintaining sleep, feeling of non-restorative sleep, use of sedatives or sleeping pills and no problem to sleep well), and sleep duration had six response options (≤ 4 hours, 4-< 6 hours, 6-< 7 hours, 7-< 8 hours, 8-< 9 hours and > 9 hours). We defined sleep quality as ‘well’ if the participants had ≥ 7 hours of sleep duration with sleep quality of “no problem to sleep well” and as ‘not well’ if otherwise. For physical activity, the detailed examples of different intensities were described in the self-administered questionnaire. The metabolic equivalent task (MET) for different intensities of physical activity was determined according to previous study . The MET expressed as hours per week was calculated by multiplying the corresponding MET coefficient by duration and frequency of physical activity.
The statistical analysis was performed by SAS 9.4 (SAS Institute Inc., USA) and STATA version 13 (StataCorp LP, College Station, TX, USA). Continuous (non-normal distributed) and categorical variables were presented as median (interquartile range, IQR) and number (percentage), respectively. The characteristics of study subjects with or without MetS were compared using Mann-Whitney or chi-square test for continuous or categorical data, respectively. The multivariable linear regression [β and 95% confidence interval (CI)] and logistic regression (odds ratio (OR) and 95% CI) were used to examine the association of dietary pattern scores with the risk of MetS, components of MetS and their related biomarkers. The P-value for trend was analyzed using post-estimation contrast and linear hypothesis test. Moreover, a subgroup analysis based on impaired kidney function categories was used for sensitivity analysis.
Dietary patterns analysis
Dietary patterns were identified by PCA and RRR methods using PROC FACTOR and PROC PLS, respectively. For PCA method, the orthogonal varimax rotation was used and we decided to retain only one from two factors for the comparison. For RRR method, six response variables (waist circumference, TG, HDL-C, systolic BP, diastolic BP and FBG) associated with MetS were used to generate the MetS-specific dietary pattern (Fig. 1). As six response variables were included in the MetS-specific dietary pattern, six factors were generated by RRR method. However, we only retained the first factor that explained the largest percentage (2.4%) of variation in the response variables. The absolute factor loading (Pearson’s correlation coefficient) values ≥ 0.20 for each food group were the cutoff point to derive the dietary patterns in both PCA and RRR methods. Dietary pattern scores for an individual were calculated by summing intake frequency scores of food groups weighed by their respective factor loading values. However, six food groups had a factor loading ≥ 0.20 in both PCA-derived dietary patterns. For characterizing the dietary pattern, these food groups could only belong to one factor with a greater factor loading value. Hence, the dietary scores of four food groups (beans/legumes, fried vegetables/salad dressing, rice/flour products and seafood) were neglected in the calculation of the first extracted dietary pattern (Additional file 2: Table S2). For further analysis, dietary pattern scores were divided into quartiles and two adjustment models were performed: model 1 adjusted for age and gender and model 2 adjusted for model 1 variables and education level, income, marital status, smoking, drinking, sleep quality, physical activity and cardiovascular disease status. A P-value < 0.05 was considered statistically significant.