Study design and population
This was a cross-sectional study, which utilized data collected between October 2015 and November 2018 during screening of MHD patients for recruitment into the Palm Tocotrienols in Chronic Haemodialysis study, as described elsewhere14,18. Patients were recruited from 14 dialysis centres within the urban area of Klang Valley, Malaysia. Inclusion criteria included MHD patients aged at least 18 years old and dialyzed for at least 3 months. Exclusion criteria were patients with poor adherence toward haemodialysis treatment, unfit for assessment due to physical or mental disability, or with terminal illness such as HIV/AIDS or malignancy. Ethical approval was obtained from the Medical Research and Ethics Committee, Ministry of Health, Malaysia (reference number: NMRR-15-865-25260). All eligible patients gave written informed consent and all research procedures were conducted in accordance with relevant guidelines and regulations.
Variables and data collection
Patients’ sociodemographic data, medical history, and the most recent drug prescription were retrieved from medical records. Patients’ self-reported compliance to phosphate binder prescriptions was assessed via face-to-face interviews. Biochemical results were extracted from in-centre patient laboratory reports relevant to within 2 weeks of collection of dietary data information. All analyses were performed by accredited laboratories19 in accordance with the operating procedures mandated by the Ministry of Health, Malaysia.
Dietary assessment was performed by research dieticians using the 3-day dietary recall (3-DDR) method, which included a dialysis day, a non-dialysis day, and a weekend day20. Common household measurement tools (bowls, spoons, and glasses) were used to optimize the portion size estimation. In addition, patients were asked about their weekly frequency of eating out. Food and beverages consumed in household units were transformed into absolute weight (g) and volume (ml) before analysis for nutrient composition using the Nutritionist Pro Software (First DataBank Inc., USA), which references the Malaysian Food Composition21 and Singapore Food Composition22 databases. The Goldberg’s index was used to identify 3DDRs of acceptable reporters to ensure the quality of dietary data. In brief, patients’ basal metabolic rate (BMR) was estimated using the Harris–Benedict equation23. Based on the reported energy intake (EI), EI:BMR ratios of < 1.2, 1.2–2.4, and > 2.4 were considered as under-, acceptable-, and over-reporting of 3DDRs respectively24.
Food items from the 3DDRs were classified into either animal or plant protein categories25. The animal protein group consisted of the following food items: fish, shellfish, eggs, poultry, red meat, milk, dairy products, processed or preserved meat, seafood, and eggs. The food sources of plant protein group were rice, cereals, beans, legumes, fruits, leafy vegetables, and starchy vegetables. An individual food item was directly assigned to the respective food group. For cooked dishes with a mixture of ingredients consisting of animal and plant proteins, recipes from the Malaysian Food Composition21 were referenced to determine the protein content of each ingredient, which was then assigned to the corresponding protein group.
The total dietary phosphate intake derived from the nutrient composition analysis of 3DDRs was categorized into organic phosphates (plant and animal foods), or inorganic phosphates (processed food or beverages). This categorization was based on the assumption that organic phosphate is naturally found in food, while inorganic phosphate is an additive in processed foods26. Accordingly, all food items from 3DDRs were divided into two food groups according to the source of phosphate. Food groups as the source of organic phosphate included cooked rice, soda crackers, fresh or frozen vegetables and fruits, eggs, beans, legumes, nuts, milk, fresh or frozen poultry, seafood, meats, and plain tea or coffee. The phosphate content of these food groups was further subdivided into either organic phosphate from plant or animal, based on the protein category as mentioned previously. Food groups as sources of inorganic phosphate were processed cheese, frozen meals, ready-to-eat cereals, cookies, canned, processed, and luncheon meat, poultry or seafood, canned soups, fast food, cola beverages, and beverages added with sweetened condensed milk. As these food groups might contain a combination of organic and inorganic phosphates, an assumption model was used to derive the added inorganic phosphate based on the difference of total phosphate content and protein content of foods in the unprocessed form. For composite dishes, standard recipes were referred to determine the content and source of phosphate of each constituent ingredient, which was then assigned to the corresponding group.
As the 3DDR method was chosen to assess dietary intake instead of a food frequency questionnaire, aggregation of food items into food groups was first carried out as previously described14 prior to performing the dietary pattern analysis. All food items from 3DDRs were extracted and sorted by alphabetical order. Then, duplicates were removed, and the food items were grouped based on similarity, food preparation method, and nutrient content. Initially, 47 food groups were developed and then, based on the consumption of each food group, food groups with consumption by less than 5% of patients were either excluded or collapsed with similar food groups. This narrowed the final food listing to 27 food groups.
The Diet Monotony Index (DMI) was calculated as described by Zimmerer et al.27 to assess food variety. From the 27-food group listing developed, the total quantity of each food group consumed was converted into servings according to the Malaysian Dietary Guideline 201028. A diet consumed with a wide variety of food groups resulted in smaller proportions of total servings for each food group, and a smaller index value was derived, or vice versa.
Continuous variables were presented as mean ± SD or median [interquartile range (IQR)], while categorical variables were presented as frequency (percentages). Non-normally distributed variables were log-transformed before statistical analyses. Chi-square test was used to determine associations between two categorical variables. Simple linear regression was used to determine the correlation between serum phosphorus (dependent variable) and clinical and dietary parameters (independent variables). Variables with p-value < 0.1 on univariate analysis were included in the subsequently multiple linear regression analysis. Tolerance and variance inflation factor were used to check for multicollinearity.
Dietary patterns were derived using à posteriori approach as previously described14. The input variable for factor analysis was the weight of each food group. Principal component analysis (PCA) was used to derive dietary patterns and the derived patterns were orthogonally rotated (varimax rotation) to enhance the difference between loadings to improve the interpretability of factors. The number of dietary patterns retained was determined based on eigenvalue > 2.029, scree plot examination, and interpretability of the derived patterns30. The eigenvalue indicates the total variance explained by a given factor30. Dietary patterns were named in accordance to the food group with the highest factor loading. Patients were assigned to factor scores computed for each pattern identified, which indicated adherence to that pattern. Based on the factor score, patients were categorized into tertiles (T1–T3) for each dietary pattern, where tertile 1 (T1) represented the lowest adherence while tertile 3 (T3) was the highest adherence to that pattern.
One-way ANOVA was used to compare nutrient intakes by tertiles of identified dietary patterns, with Bonferroni test for post hoc analyses. Kruskal–Wallis with Dunn’s post hoc test was used for comparison of non-normally distributed variables. One-way analysis of covariance adjusted for covariates was used to compare serum phosphorus by tertiles of each dietary pattern, and Bonferroni test was used for multiple comparisons. Multiple logistic regressions were used to estimate the odds ratios (ORs) of hyperphosphatemia (serum phosphorus above 1.78 mmol/l and 2.00 mmol/l) associated with T3 of each dietary pattern. Missing covariate data (less than 2%) were imputed using the mean of the existing value of all patients. All analyses were computed using the SPSS version 25 (IBM, Chicago, IL, USA). Statistical significance was set at p-value < 0.05 for all evaluated parameters.