| Food Composition Data: A User's Perspective (1987) |
|The uses of food composition data|
|Need for a standardized nutrient data base in epidemiologic studies|
Factors influencing diet-related epidemiologic studies, using diet and colon cancer studies as an illustration
Searching on the key words "diet or dietary" and "colon cancer," "colonic neoplasms," or "sigmoid neoplasms," a MEDLINE literature search yielded 166 citations dating back to 1980. Twenty-six or 16 per cent of these studies were population-based or epidemiologic in nature. Thirty-three population-based studies reported after 1977 were identified by cross-referencing colon cancer with dietary risk factors. These studies have been summarized in table 1. The studies have been grouped according to the most commonly cited dietary risk/protective factors: dietary fibre, fat/meat, beer/alcohol, and cruciferous vegetables. The headings in table 1 list major components of epidemiological studies, each of which can effect the outcome of the study. The major types of study design as seen in the table are: ecological and food disappearance studies, retrospective (case-control) studies, cross-sectional surveys, and prospective (cohort) studies. In addition to choosing the appropriate study design, the investigator must also decide how to collect dietary information.
Though there are many variations of each, there are four basic dietary data collection tools: diet diaries, diet recalls, diet histories, and food frequencies. If data on specific food or foodgroup intake or availability is obtained for individuals or groups. the information can be transformed into nutrient intake by interfacing the food intake data of study respondents with a food composition data base.
Each technique has inherent strengths and weaknesses. Retrospective data collection methods are subject to respondent memory bias while diary methods tend to distort usual intake patterns. In addition these standard methods measure different aspects of dietary intake. Therefore there will be differences in study outcome depending on the food-intake datacollection instrument chosen. (Notice that all four intake tools were employed in the studies reported in table!.)
The type of food or nutrient data base selected is dependent on the study design, the data collection method, the study objectives, and the endpoints to be measured. However, a lack of standardized definitions of dietary study variables has been a major weakness in interpreting study outcomes. Definition has presented problems for developing standardized food names as well as for food composition tables. For example, dietary fibre, the first risk factor listed in table 1, is a complex of a number of physically and chemically different entities found in foods. They include cellulose, hemicellulose, lignins, pectins, and gums, and the ratio of these materials varies in fibre-containing foods. Until recently, data bases reported only crude fibre values, in which food samples were subjected to strong acid and then alkali solutions. These values are not equivalent to dietary fibre, which is the residue of undigested food.
The last column in the table describes the outcome or risk-factor association found in the studies. Drawing correct conclusions from the data concerning the strength of association of study variables and the attributable risk for diseases is dependent on choosing appropriate statistical tests. In addition one must control for confounding variables and adjust for covariables. Unlike other clinical or laboratory studies, epidemiological studies are based mainly on relative rather than absolute differences of risk factors between exposed and unexposed groups. However, these studies lose power if real differences exist in the nutrient content of foods consumed by different population groups. This problem is analogous to regressing to the mean by not utilizing significant differences in food composition consumed by study populations. Increasing the power of a study is important, since the influence of diet is often obscured by stronger overriding etiological factors encountered in multi-etiological chronic disease studies. Also, epidemiologic methods and techniques are sometimes inadequate or inappropriate for the evaluation of diet and disease relationships, especially if one assumes that nutrient variables are independent of other dietary or environmental factors. Furthermore, much of the confusion in outcomes of diet-related epidemiologic research may stem from inappropriately comparing studies that differ in design, analytical techniques, or food composition data bases.