Design of the Brazilian Individual Dietary Survey
Study population
The 2008–2009 Brazilian HBS sample will include 60,000 households, assuming that non-response is around 20%. Thirty-percent of the households will be randomly selected to be surveyed on individual food intake. It is estimated that each household has a mean of 3.5 persons aged above 10 years; therefore, individual dietary intake data is expected to be gathered from about 70,000 individuals.
Data collection
In the selected households, all family members will be listed, and those aged ten years or more will be included in the individual dietary intake survey. For all children younger than 10 years of age, a separate set of questions will ascertain the place where their meals and snacks are usually eaten; an adult member, usually the mother, will provide this information. Pregnant and breastfeeding women will fill in the individual records; nevertheless their data will not be taken into account in the estimation of individual intake, because their potentially modified food consumption might introduce biases in the estimates. On the other hand, the number of pregnant and breastfeeding mothers is too small for us to gather enough data or to make specific estimations for this group; however, the information on their food intake will be used to compare individual data with the overall family food budget. Proxy information will be recorded for the elderly, children, and for those who are not able to write.
Data collection procedures will comprise the following steps:
(1) Completion of an open diary of food purchases for the household during a seven day period.
(2) Each household member 10 years-old or older will fill out a small notebook registering all items consumed during two non-consecutive days. Water intake will not be considered in the individual data collection. There will be a question regarding usual sugar intake (sugar or sweetener or both). The records must include:
(a) food items;
(b) amount consumed (referred in standard units of volume measurement – usually grams or milliliters – or in universally used cooking ware and tableware – such as cups, plates, tablespoons, ladles, skimmers, etc, or in commonly used serving portions, such as bunch, slices, etc.);
(c) place of the meal (at home or out of home);
(d) time of intake;
(e) preparation of food for specific items (raw, cooked, baked, grilled, fried, steamed, with tomato sauce, sauté, stewed, etc).
(3) Review of the records: an interviewer will visit the household at least twice during the week of data collection to get the filled notebooks on general household acquisitions and individual food intake data. The interviewers will be trained to record dietary data on a computer database, reviewing the individual notes and probing for items usually not mentioned in this kind of records, such as candies, snacks and beverages. Also, when the record shows more than three hours during the daytime without any food intake, the interviewer will make sure that nothing was eaten during that period.
Computerized datasets and data entry
The computer software designed to store individual data on food intake provides datasets on food items, prepared dishes, cooking methods, and measurements. The building of the food items and prepared dishes list was based on the records of 2002–2003 HBS, where 5,686 food items were reported. A great part of these items were synonyms, however, or had alternative spellings, or yet contained detailed variations that do not present significant differences in nutritional composition (such as different kinds of banana). The review of these items respected regional nomenclature and the diversity of food intake in Brazil. Hence, the computer-assisted interview for the individual dietary intake survey database includes around 1,200 food items and preparations. Interviewers will be able to include more items, but they will be trained to look for the food item that better describes the food reported by the subjects in the dietary records before adding new items.
The food list also includes the names of popular dishes, such as lasagna, pizza, hamburgers, etc. The intake of these food items will be recorded according to the way by which they are commonly referred (i.e., the name of the dish), instead of registering each ingredient (for example, a lasagna will be listed as such, instead of pasta, tomato sauce, cheese, etc).
Each item entered in the individual database will be linked to portions registered in grams, milliliters, plates, tablespoons, cups, etc. The interviewers will not be able to introduce other measurements in the dataset and they will be trained to find the better option in each situation. To provide the standardization of portions, utensils, and tableware, the interviewers' training will include the identification of the measurements listed in the database, and the field manual will include a set of tableware and cooking-ware photographs.
Data analysis
The sample will be large enough to capture the variations between individuals (S
b
), and the two records will assure the estimation of the variation within individuals (S
w
) for food groups, energy and nutrients intake, allowing for the deattenuation of individual data for random error.
Data on individual dietary intake (IDI) and food family budget (FFB) will be stratified by the five regions of the country (North, Northeast, Southeast, South, and Central West), and by rural or urban (the latter being subdivided into metropolitan and non-metropolitan). We are anticipating large differences among the five regions of the country as well as among rural, urban metropolitan and urban non-metropolitan areas within each region. Hence, analysis will be accordingly stratified, and for these 15 stratums, statistical models will be constructed assuming the following premises:
1- FFB and IDI are linearly associated (although food intake and household availability may not be linearly associated if there is a significant prevalence of eating out. This assumption will be tested).
2- Households without children younger than 10 years of age yield a good estimation of weekly total intake.
Estimation for the weekly individual food intake (WIFI) as follows:
Regression of the sum of all WIFI on FFB for a specific food or nutrient will give the correction factors for FFB data for those items purchased on a weekly basis, such as bread. Therefore, firstly the agreement between WIFI and FFB will be tested and for those foods with correlation greater than 0.80, weekly data will be analyzed. Foods with low correlations indicate that FFB data collected on a weekly basis will not correspond to the household availability, meaning that those foods such as sugar, oil, rice and beans are acquired on a monthly basis. Therefore, for these items showing lower correlations, WIFI will be regressed on the mean of all FFB of those primary sampling unit (about 15 households) of similar income per capita surveyed on a monthly basis. For all food groups, the mean acquisition will be adjusted by family size. These assumptions will be tested comparing household acquisitions and the sum of all WIFI. Models will also take into account all variables that are highly associated with intake, such as household composition, schooling of the head/mother, and income.
Daily energy and nutrients intake will be estimated by transforming the WIFI into calories and nutrients using food composition tables, preferably the Brazilian Food Composition Table [7].
WIFI and FFB data are complementary, and since they are collected through independent tools, they have independent source of errors and can be used for addressing limitations inherent in all data sources.