Study design and participants
The study design is a retrospective analysis of 24-hour recalls from participants of the BALANCE program study (https://www.clinicaltrials.gov/; NCT01620398). A description of the original study has been published elsewhere(20).
Briefly, between March 2013 and January 2015, 2534 individuals who experienced one or more indicators of established CVD in the preceding 10 years were enrolled in the trial and followed until December 2017 in one of the 35 research sites in Brazil. They were then randomised to either the control group or the intervention group (BALANCE Program). The study was approved by the Research Ethics Committee of the Hospital do Coração de São Paulo and all participants signed an informed consent form.
The participants in the control group received generic nutritional advice on low-fat, low-energy, low-sodium, and low-cholesterol diet. The participants in the intervention group had frequent contact with registered dietitians and received diet prescriptions based on the Brazilian dietary guidelines for treatment of CVD(22), including specific educational intervention for improvement in dietary patterns, especially regarding the consumption of locally-available foods with cardioprotective role (i.e., 50% to 60% of energy from carbohydrate, 10% to 15% from protein, 25% to 35% from total fat, 7% from saturated fatty acids, 10% polyunsaturated fatty acids, 20% monounsaturated fatty acids, 1% trans fats, 200 mg/day cholesterol, 20 to 30 g/d fiber, and 2400 mg/d sodium). Trained registered dietitians provided individualized dietary advice (face-to-face or telephone sessions) for each participant.
For the present study, per-protocol analysis was carried out with adherence to protocol being defined as 80% presence in the individual sessions and telephone calls during the 36 months of follow-up. The adherence cut-off was defined considering prior studies that demonstrated higher treatment compliance due to closer contact with health professionals(23,24). The final sample consisted of 1,161 individuals (576 in the intervention group and 585 in the control group, 40.2% of the original sample).
Dietary intake data from 24-hour recalls were used to estimate diet costs and quality. Trained researchers collected five 24-hour recall during the 36 follow-up months: two at the beginning of the study (before the intervention with a 15-day interval between them), and three in the following years, once a year.
The economic assessment for comparison of the two strategies (control and BALANCE Program) was performed based on cost-effectiveness ratios (CER) and incremental cost-effectiveness ratios (ICER). The analyses were conducted using the perspective of the patients. This perspective was adopted because patients are directly affected by the intervention, and decide to follow or not the prescribed diet. The time horizon was established in 36 months, based on health outcomes presented in the trial paper previously published. (21) Costs were valued in the same time period so the discount rate was not applied
Effectiveness was assessed using diet quality from the Brazilian Health Eating Index Revised (BHEI-R)(25), and direct costs were calculated from food items reported during the 24-hour recall interviews. The direct costs to perform the intervention (e.g., researchers wages, transport of subjects to study centers, biochemical analyses, etc.) were not included in the analysis, considering that the program was designed to comprise a strategy for secondary prevention at primary health care level; therefore, should be implemented in local settings near individuals’ residence.
Estimation of diet costs
The cost of the diet was assessed at individual level, based on a dataset of food items reported by the participants. The dataset was constructed by compiling a full list of food items from each of the 24-hour recalls, resulting in 1,103 standardized food items and recipes.
Prices of food items were collected between October and December 2018 in three local supermarkets located in the Northeast, Midwest and South of Brazil, and two nationwide supermarket chains with online stores. Prices were registered for usual retail purchase of food items, promotions referring to sales or bulk acquisition were discarded during consistency analysis. In the case of more than one product available of the same food, data were collected from up to three items, and the average price of the items was used as the final price.
The price of each raw food was converted into price per gram of food ready to eat by applying correction and cooking factors. The recipes were obtained from the BALANCE Program recipes book or a Brazilian standard book for recipes(21), using the amount of main ingredients in the recipes, and including standardized amounts of seasonings (e.g., 5% of sugar, 1% of salt and 2% of soy oil for cooked or 10% for fried foods), in order to allow its inclusion in costs. Prices of food items were deflated to the period of 24-hour recall interviews using official data on specific inflation rates for each item at local level, in order to properly represent relative prices of purchase at the time of the interview, as published by the Brazilian Institute for Geography and Statistics (IBGE)(26).
Finally, the cost of the diet was estimated by multiplying the mean price per gram of the food item by the amount reported in the 24-hour recall for each participant. The cost of the diet was updated and converted into U.S. dollars using the official exchange rate published by the Brazilian Central Bank at the reference date of December, 2018 (1 US dollar = 3.88 Brazilian Reais).
Assessment of diet quality
The BHEI-R(25) is a validated adaptation of the Healthy Eating Index 2005 for the Brazilian population(27), based on nutrition recommendations from the Brazilian Ministry of Health Food Guide, the World Health Organization, the Institute of Medicine, and the guidelines of the Brazilian Society of Cardiology(27).
The BHEI-R score for each 24-hour recall was obtained by the sum of scores referring to twelve components: nine based on the consumption of food groups (total cereals; whole grains; total fruits; whole fruits; vegetables; dark green and orange vegetables and legumes; milk and dairy products; meats, eggs and legumes and oils), two based on the intake of nutrients (saturated fat and sodium), and one resulting from the energy intake from solid fat, alcohol, and added sugar (SoFAAS). Each component can contribute from 0 to 20 points to the total score, depending on component type (food group or nutrient intake). Minimum to maximum values were determined according to the nutritional recommendations of each component based on national and international guidelines; e.g., in the case of the group “total fruit”, dietary intake equal to or greater than the recommended per 1,000 kcal was given the maximum score of the item (five points).
Nutritional values of 24-hour recalls were calculated using the Nutriquanti software, and BHEI-R score, ranging from 0 to 100 points, was considered the outcome variable in the economic assessment and in the statistical analysis. Higher BHEI-R values indicate better dietary quality, whereas low scores indicate less adherence to recommendations.
Between-group differences in baseline characteristics were analyzed with the Wilcoxon rank-sum test for continuous variables and chi-square for categorical variables. Regarding costs and diet quality, mean differences at baseline and post-intervention (1, 2 and 3 years later) were compared between groups using generalized estimating equation (GEE) models with unstructured and exchangeable correlation matrix, adopting time, treatment, and interaction between time and treatment as predictors.
Differences in scores of BHEI-R components, total energy, total amount of food, amount of food group, costs, energy density, and macronutrients between the intervention and control groups at end point were analysed using ANCOVA, with baseline variables as covariates. Mean difference was presented as difference between the BALANCE and the control group.
Analyses considered a 2-tailed statistical significance level of a=0.05, and were performed using R software, version 3.6.0 (R Foundation for Statistical Computing).
Economic analyses were conduct with data from baseline and 36 month 24-hour recall only. The cost-effectiveness ratios (CER) were calculated with the following equations:
Where CERZi = cost-effectiveness ratio of patient i in group Z; CZi = direct cost of the patient’s i diet; EZi = diet quality index of the patient’s i diet (BHEI-R). Direct cost was defined as:
Where pRij = means price of food item j in region of residence R of the patient i; qRij = amount consumed of food item j of the patient i living in region R. Finally:
Where ICER = incremental cost-effectiveness ratio of group intervention in comparison to group control; CIm = mean direct diet cost in intervention group; CCm = mean direct diet cost in control group; EIm = mean diet quality in intervention group; ECm = mean diet quality in control group.
In addition, Monte Carlo simulations were performed to confirm trends in the comparison between control and intervention groups, resulting in 10,000 cases of patients in the intervention group and 10,000 cases of patients in the control group. Estimations on costs and effectiveness were based on mean and variance of direct costs and diet quality within each group using g distribution. The average of expected costs and effects were used to estimate different ICER for diverse scenarios, plotted onto incremental cost-effectiveness diagrams, with costs plotted in the horizontal axis and effects in the vertical axis.