Foods that we eat and their relationship to health

  • Food history questionnaires

    A lot of nutrition research is done through the use of the food history questionnaire. This is different from food recalls in that it is shorter, easier to administer and does not require special interviewing skills.

    Sometimes studies need to identify patterns of food consumption in a large population. This is done, for instance, as part of a general population health survey. These are done on a fairly regular basis, usually funded by governments, to be proactive in the formulation of heath related polices. This saves our tax money in the long run.

    Food history questionnaires are related to food recalls in that they collect information on commonly eaten foods. These questionnaires tend to be simpler in that they do not seek to know each individual food that was eaten. They ask questions like “When you eat pasta how much do you eat and do you eat it once a week, or 3-5 times a month”. You get the idea. Typically a food history questionnaire will have fewer than 100 such questions.

    Before a questionnaire like this can be formulated it is wise to know what kinds of foods the population usually eats. A smaller number of individuals are asked what they ate recently, using the recall format. Using the example above, from the results of these recalls, a pattern of “pasta” consumption can be determined. A composite “pasta” food can then be created and used as the nutrient profile for the “pasta” question in the food history questionnaire.

    The downfall is that it is a fairly “blunt” instruments. With a bit of care in its construction some of the bluntness can be removed.

    The following techniques can be used to make them better research instruments:

    • Make up your own;
    • Make them specific to your target population;
    • Make them as detailed as possible while keeping them still realistically long;
    • Base the questions on foods your population eats. Collect as many food recalls as you can and base the questions on the foods in those recalls;
    • Create composite foods from similar foods in the recalls and use those to calculate nutrient contributions for corresponding questions in the food history questionnaire;
    • Validate your questionnaire… does it over/under-estimate specific nutrients in comparison to your food recall data?

    There is enough variability in food intake and nutrient concentrations of individual foods to try to minimize the effect of further variability introduced by blunt food history questionnaires.

    Given the amount of variability in this data it is a wonder that any conclusion can be reached about the predicted effect of a nutrient intervention or of specific nutrient consumption.

    By all means find and use good software that will make your task bearable. Variability can only be reduced by eliminating mistakes, using the best sources of data possible and collecting large amounts of data. A significant task indeed.


  • Questionnaire concepts

    A subject’s nutrient profile, whether obtained from food recall information or from food history (questionnaire) information, is obtained by a simple arithmetic calculation. Each nutrient is the product of the food quantity consumed and the nutrient’s concentration in that food. The subject’s consumption is the sum of these products over the foods consumed in one day. Simple.

    Problems:
    In a recall… using the right food code to represent the food consumed and estimating the quantity consumed as accurately as possible. For a recall there are thousands of food codes to choose from, one of those is likely to represent fairly accurately the actual food consumed. The quantity of that food consumed can also be fairly accurately recorded as reported.
    An example of a recall record would be “I ate a banana for breakfast”. The corresponding coding would have the food code for bananas and quantity being typically “one medium banana”. Hence, nutrient profile and quantity fairly accurately recorded and yield fairly good nutrient information.

    In a questionnaire very few questions (usually less than 200, sometimes as few as 100) are used to represent historical intake. Every question is matched to a food with a nutrient profile representing the consumption for that question. A single food taken from a database used in recalls is not likely to be indicative of  the group of foods represented by any one question in the questionnaire.

    An example question could be “Do you consume soup?”. There are many soups with different nutrient profiles. Which one to use for the question as a nutrient profile? The soup code used  should be a “composite” of all the possible soups. Which composite to use? One formed of the relative use all the soups consumed in the population in question. This information can be obtained from a food recall study of the population. This “composite” food could then be taken as the food representing the question’s nutrient profile… a good estimate on a population basis, probably not so good on an individual basis.

    Composite food calculations from food recalls:

    • Determine the unique food codes from all the recalls. Typically should be 400-600 food codes
    • Divide those codes into food groups corresponding to questions you would want in the questionnaire
    • Run the nutrient calculations on the recalls looking at food details by food group, sorted in descending order of food quantity in grams
    • Create a recipe from each food group using the main foods in that group and their corresponding quantities as recipe ingredients
    • Run the nutrient analysis on the recipe file and export each recipe and its nutrients per 100G to a food file
    • Use this food file in your nutrient calculations of the relevant questions

    Of course, this assumes you have the software to do all these calculations and conversions automatically. Doing the calculations manually or using a spreadsheet would be very onerous indeed.

    The question of quantity to record is a bit more difficult. Usually such a question asks “How often do you consume this soup? Per day? Per week? Per month?”. No problem here, just a mathematical calculation.
    The problem is in the next part of the estimate, the portion size. If the portion size is indicated precisely as in .5 cup, 1 cup, 2 cups… again, no problem. The composite soup can have a weighted density based on the density of the soups making up the composite. Cup weights can then be precisely calculated.  250 ml x 1.06 G/ml would give us a cup weight of 265 G.

    Technique A:
    How does one estimate portion sizes when the portion is not so precisely indicated? As in, .5 of a cup or less, .5 cup to 2 cups, 2 cups or more? One logical estimate would be to take the mid-point of the ranges.
    For minimal consumption to .5 cups, use .25 cup;
    for .5 to 2 cups, use 1.25 cups; for 2 cups or more use 4 cups (maximal consumption assumed to be 6 cups).

    Technique B:
    Much more intensely computational… not using the composite weighted densities…
    An alternative to the above would be to use population based estimates. For each of the range of consumption, .5 cup or less, .5 cup to 2 cups and 2 cups and more, establish the distribution of consumption and calculate the median or average value. The median value would probably be best as it would negate the effects of outlier consumption.

    In the population there is no consumption of the composite soup. The distributions have to be calculated for each and every soup making up the composite. One median per soup! For example for the lower range, less than .5 cup, how does one obtain a composite median from the individual soup medians? A weighted average of medians? Based on what weighting factor? The relative weight of the total weight of the soup consumed in the population (used to get the weighted density of the composite) or the relative weight of the total weight consumed in the range less than .5 cup? I would guess the latter to be the better estimate.

    How does one establish the cut-off weight for .5 cup. The best value would be obtained by using the density for the soup whose median is being calculated. If the soup has a density of 1.06, one would look at all consumption of that soup of .5 cup or less or of 265G/2 = 132 G or less. The range .5 and above would start at 133G…

    Should the basis of the distribution of consumption be each consumption of soup or the total soup consumption for the day? This question may not seem relevant here (each consumption would be the best information for the typical portion size) but what about other foods, such as milk in all its possible consumption portion sizes (see below)?

    Assumptions:
    Technique B assumes that all consumptions recorded in the population recalls are based on portions that are cups. In soups this is probably reasonable. What about questions that ask questions about foods such as milk. “Do you consume milk?” If yes, how many times per day/week/month and how many glasses? Recalls will record all kinds of consumptions of milk. In cereal, in coffee or tea, in glasses or cups. Each one of these will be converted to Grams. The total of those consumptions, on a daily basis, or on a per consumption basis, may not reflect typical population median gram values for typical glass or cup portion sizes.
    Estimates of portion sizes for questionnaire data should be based on recall data collected using those same portion sizes.