SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Dietary assessment instruments
  5. Study designs in nutritional epidemiologic research
  6. Nutrient databases
  7. Statistical issues in dietary analyses
  8. Conclusions
  9. References

Summary

Background

Nutritional epidemiology is the assessment of diet and its relationship to disease aetiology in populations. The choice of dietary assessment method depends on the disease pathology. Events such as cancer that are chronic and complicated by exposure time require methods that capture consumption patterns of populations over a period of years. Although several methods of dietary assessment exist for collecting information on groups of individuals, their application to epidemiologic studies requires an understanding of the effect of variability in nutrient intake, sources of measurement error, and statistical issues unique to the study of nutritional epidemiology.

Aim

This review provides an overview of commonly-used methods of dietary assessments in epidemiologic studies, and identifies their strengths and limitations and application to epidemiologic study designs. It concludes with a brief discussion of assumptions of nutrient databases and objectives of energy-adjustment and measurement error correction models.

Conclusions

Nutritional epidemiology has contributed significantly to our understanding of the relationships between diet and disease. Ongoing investigations that further characterize important exposure periods (early life, in utero) and clarify associations within the context of genetic susceptibility will continue to elucidate our understanding of the pathophysiology of complex diseases, and support future recommendations for disease prevention.

Key points

• The choice of dietary assessment depends on the research question and the pathophysiology of the disease.

• Long-term dietary patterns are most relevant to estimate chronic disease risk.

• Most risk models will need to adjust for total energy.

• Knowledge of potential sources of error in nutritional assessment is essential.

• Statistical methods exist to estimate and correct measurement error.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Dietary assessment instruments
  5. Study designs in nutritional epidemiologic research
  6. Nutrient databases
  7. Statistical issues in dietary analyses
  8. Conclusions
  9. References

One of the earliest applications of nutritional epidemiology was in the study of gastrointestinal diseases. In the late 1960s, Burkitt [1] observed differences in fecal bulk between individuals in rural Africa compared with industrialized Western countries and hypothesized that this was the result of the high fiber intake of the former. Subsequently, he hypothesized that dietary fiber protects against the development of colorectal cancer. To date, there are more than 400 published accounts on this topic.

Nutritional epidemiology is the assessment of diet and its relationship to the causes of diseases in populations. This includes the intake of essential nutrients (e.g., vitamins, minerals and amino acids), energy sources (protein, carbohydrate, fat and alcohol), naturally occurring food compounds (e.g., plant fiber, cholesterol and caffeine) or, for specific hypotheses, the intake of chemicals formed in cooking, such as heterocyclic aromatic amines formed in well-done or charred meats, or from food processing, such as trans fatty acids. An observed association between a nutrient and disease is complemented by statistical analysis of the nutrient’s food source (e.g., food, food groups) with the disease, which strengthens the hypothesis under study.

The investigator’s choice of dietary assessment method will depend on his or her knowledge of the disease pathology. Events that are acute and occur over a relatively short period, such as maternal dietary folate intake and risk of fetal neural tube defects, require methods that accurately and precisely assess an individual’s intake over the course of days or weeks. In contrast, events such as cancer that are chronic and are complicated by exposure time, require methods that capture patterns of consumption among populations over a period of years, because measurement of diet several years prior to disease manifestation probably represents the more relevant exposure period for understanding these diseases. Diet–disease associations may be confounded or modified by several factors, including body size, physical activity, other dietary factors and genetic susceptibility. Understanding the interplay among these factors is crucial to derive unbiased estimates of disease risk.

Dietary assessment instruments

  1. Top of page
  2. Abstract
  3. Introduction
  4. Dietary assessment instruments
  5. Study designs in nutritional epidemiologic research
  6. Nutrient databases
  7. Statistical issues in dietary analyses
  8. Conclusions
  9. References

Two methods of dietary assessment typically used in clinical settings have been modified for use in epidemiologic studies. Both the 24-hour recall and the diet record assess short-term dietary intake, but when used as repeated measurements, can inform of usual patterns of intake over a longer period.

The 24-hour recall interview is administered in person or by telephone. Subjects report their exact intake in the preceding 24 hours guided by the interviewer’s standard questions, which may also use visual aids to assist with recall of portion size (these may be mailed to subjects in advance of unannounced 24-hour recalls undertaken by telephone). In its favor, memory of recent intake may be more precise and quantities may be estimated with greater accuracy with minimal participant burden. Well-trained interviewers are required, however, and the nutrient analysis of food intake can be laborious. Because individual diets vary greatly from day to day, a single day’s dietary recall does not represent usual dietary intake.

The diet record is similar to the 24-hour recall, except that the subject records actual food and beverage intake prospectively over several days. Subjects are asked to provide detailed descriptions of preparation methods and food quantities, which are assessed by weighing, volume/dimension measurements or estimation assisted by the use of photographs. The prospective nature of diet recording reduces errors associated with recall and minimizes omission of foods consumed. However, the method requires a high level of subject literacy, motivation and training, and can be costly to analyze. Furthermore, consecutive days of dietary recording may result in food intake that is highly correlated from day to day (due to consumption of leftover meals or alteration of usual diet to include foods that are easy to record), possibly introducing bias. A trade-off is to collect fewer records per subject on a greater number of individuals. In the absence of objective assessments of long-term dietary intake, the diet record is considered the “alloyed” gold standard. Like the 24-hour recall, multiple days of records over several months or one year can reduce day-to-day correlation of intake, improve accuracy and precision of individual intake and capture seasonal variation in food intake.

For investigations of several hundreds or thousands of individuals, food frequency questionnaires (FFQs) are a viable option to assess long-term diet. These questionnaires consist of a list of foods and beverages that represent the major contributors to the macronutrient and micronutrient content of the diet of the population under study. Thus, they are population- or ethnic-specific [2, 3]. For each food or beverage item, the subject selects one of several options that best defines their frequency of intake over the past year with or without a selection for a portion size option (Fig. 12.1). Photographs of different serving sizes assist with portion recall. FFQs are easily administered in person or by mail, provide information on the intake of a large number of foods, food groups and individual nutrients, and are substantially less expensive to analyze particularly if in scannable form. Repeated FFQ administrations over several years can capture dietary changes over time. Table 12.1 summarizes characteristics of some large, prospective studies employing FFQs.

image

Figure 12.1.  Example of the format of the Block Food Frequency Questionnaire 2005. (Reproduced from http://www.nutritionquest.com, with permission from Block Dietary Data Systems.)

Download figure to PowerPoint

Table 12.1.   Characteristics of prospective cohort studies utilizing validated Food Frequency Questionnaires
StudyDate initiatedStudy characteristicsFFQ administrationOutcome ascertainmentContact
  1. ACS, American Cancer Society; IARC, International Agency for Research on Cancer.

Nurses’ Health Study I1976∼121 700 registered nurses aged 30–55 years from 11 US statesEvery 4 yearsMedical and pathology record review, linkage with national death indexhttp://www.channing.harvard.edu/nhs/
Nurses’ Health Study II1989∼117 000 registered nurses aged 25–42 years from 14 US statesEvery 4 yearsAs aboveAs above
Health Professionals Follow-up Study1986∼52 500 US male health professionals aged 40–75 yearsEvery 4 yearsAs abovehttp://www.hsph.harvard.edu/hpfs/
Iowa Women’s Health Study1986∼100 000 Iowa women aged 55–69 yearsBaselineLinkage with Iowa cancer registry, national death indexhttp://www.cancer.umn.edu/page/research/prevent6.html
European Prospective Investigation of Cancer1992∼521 000 men and women aged 35–70 years from ten European countries BaselineCancers reported by each country’s cancer registries to a central IARC databasehttp://www.iarc.fr/epic/Sup-default.html
ACS Cancer Prevention Study II1992∼184 000 men and women aged 50–74 years from 21 US statesEvery 2 yearsLinkage with state cancer registries, national death indexhttp://www.cancer.org/docroot/RES/content/RES_6_2_Study_Overviews.asp
Multiethnic Cohort Study1993∼215 000 men and women aged 45–75 from Hawaii and Los Angeles of Caucasian, Latino, African-American, Native Hawaiian and Japanese-American ethnicitiesBaselineLinkages to cancer registries and death certificate files in Hawaii and California and to the national death indexCancer Research Center of Hawaii, University of Southern California/ Norris Comprehensive Cancer Center

In evaluating dietary assessment methods, it is worth commenting on two important dimensions that affect nutritional epidemiologic research. The first is the distinction between group and individual, and the second is between quantitative precision and classification or ranking of individuals [4]. A clinician might value a dietary assessment method that gives accurate results for each individual, in micrograms of folate or kilocalories of total energy; that is, patient A’s dietary assessment accurately reflects patient A’s intake. However, such accuracy at an individual level and in such precise quantities is not essential to produce valid and useful research on diet and disease at the group level, provided the dietary assessment instrument is valid for the population under study [4]. Indeed, John Snow did not need to know the exact dose of the organism necessary to cause cholera in order to produce a tremendous advance in public health. Valid but less precise methods that locate individuals on the distribution in broad categories of low, medium and high intake still permit the examination of nutritional hypotheses and the assessment of dose–response relationships [4], whereas invalid instruments will bias associations.

Study designs in nutritional epidemiologic research

  1. Top of page
  2. Abstract
  3. Introduction
  4. Dietary assessment instruments
  5. Study designs in nutritional epidemiologic research
  6. Nutrient databases
  7. Statistical issues in dietary analyses
  8. Conclusions
  9. References

The most common application of nutritional assessments using FFQs is for epidemiologic investigations utilizing case-control, prospective cohort and cross-sectional study designs.

The reference period of dietary intake differs between study designs. For case-control studies, the disease process may alter the subject’s dietary intake during the period leading up to diagnosis, through food intolerances or changes in appetite; therefore, the FFQ typically asks about usual eating habits “before one year ago” and not including any recent dietary changes. On the other hand, dietary assessment among disease-free subjects in a prospective cohort study is based on recollection of usual eating habits in the past year.

A major limitation of case-control studies is recall bias of exposure among the cases, who may over-report foods that they believe may have contributed to their diagnosis and under-report healthier foods that they believe may have prevented their disease. This biases relative risks further from the null value than would be observed in a prospective study of the same association. Moreover, selection bias, driven by the eagerness of cases to find the “cause” of their disease, probably contributes to their higher participation than controls in epidemiologic studies [5]. Controls who participate may be more health conscious, for example, consume more fruits and vegetables and less fat. The effect of recall and selection bias is not trivial and could lead to apparent inverse associations with fruits and vegetables and positive associations with dietary fat [6].

Differences in findings even among cohort studies may be due to various reasons, including differences in the populations or in the endpoints studied (e.g., colon adenomas vs carcinomas), follow-up duration, the choice of nutrient database (discussed below), and the range of intake captured by the FFQ. For example, two national US cohorts examined the ratio of red to white meat intake with risk of colon cancer among women (Fig. 12.2). Higher intakes increased risk in both cohorts, which were similar in cohort size and duration of follow-up (6–9 years). The overlap in distributions, however, suggests the full extent of increased risk is observed only at very high intakes.

image

Figure 12.2. Ratio of red to white meat intake and risk of colon cancer among women in two national US cohorts. CPSII, Cancer Prevention Study II; NHS, Nurses’ Health Study. Vertical bars represent 95% confidence intervals. (Data from Willett et al., 1990, and Chao et al., 2005)

Download figure to PowerPoint

Identifying the role and extent of dietary stimuli in the development of disease is usually easier and more efficient when comparing and contrasting culturally heterogeneous populations (e.g., ethnic groups) who have differences in lifestyle practices. For example, the Multiethnic Cohort was established to study diet and cancer among 215 251 adult men and women living in Hawaii and Los Angeles, who showed baseline differences in incidence for common cancers according to ethnicity [2] (Fig. 12.3). The inclusion of multiple ethnic groups within a single study permits interethnic comparisons of diet–disease associations by using common data collection methodology in all groups and, where no heterogeneity exists among ethnic groups in the estimates of disease risk, allows the pooling of data for a wide range of dietary intakes to estimate the overall effect with disease. As the numbers of cancer cases accrue, this study promises to evaluate the extent to which dietary and other environmental exposures explain interethnic differences in disease incidence.

image

Figure 12.3. Incidence (1988–1992) of colorectal cancer in the Multiethnic Cohort Study by ethnicity, age-adjusted to the 1970 US standard population. (Data from Kolonel et al. [2])

Download figure to PowerPoint

Secular trends can also identify changes in incidence rates caused by environmental factors. Figure 12.4a shows the age-specific incidence for esophageal cancer per 100 000 of the US population among white males and females over three time periods. For all age groups over 40, the incidence increases sharply during the most recent period, 1998–2002. Similar trends are observed for African-American males and females. During this period, the number of overweight and obese individuals also increased (Fig. 12.4b), and vegetable intake decreased, suggesting possible links that require further investigation. Indeed, investigators recently explored the hypothesis that increased carbonated soft drink consumption is associated with this trend [7].

image

Figure 12.4. (a) Age-specific incidence of esophageal cancer per 100 000 in the US for three time periods among white males (M) and females (F). (Data from Surveillance, Epidemiology and End Results registries; http://seer.cancer.gov/) (b) Percent of adults aged 20–74 years who were at a healthy weight, overweight or obese: 1971–74 to 2000–02. Data are age-adjusted to the 2000 population standard. (Reproduced from National Cancer Institute Cancer Trends Progress Report 2005 Update; http://progressreport.cancer.gov/)

Download figure to PowerPoint

The hypothesis that an individual’s genetic susceptibility to developing disease may be modified by diet is emerging as an active area of investigation. A well-known example of a diet–genetic association study is that of folate and the 677C→T polymorphism in the gene encoding the folate-metabolizing enzyme, methylenetetrahydrofolate reductase (MTHFR), which impairs the conversion of 5,10-methylenetetrahydrofolate (5,10-mTHF) to 5-methyltetrahydrofolate (5-mTHF). 5,10-mTHF is involved in essential one-carbon transfer reactions that are important in DNA synthesis and replication whereas 5-mTHF functions in the methylation of many compounds including DNA, RNA, proteins and phospholipids [8] (Fig. 12.5). Folate deficiency is implicated in cancer development by either pathway [8]. On the basis of the functional effects of the polymorphism, and the inverse association between folate status and disease, it might have been expected that the variant would be associated with increased risk of colorectal cancer [9]. On the contrary, most studies showed that the variant (TT) genotype is associated with moderately reduced colorectal cancer risk, which may result from accumulation of 5,10-mTHF that serves as a cofactor for DNA synthesis and repair reactions. Alternatively, higher dietary folate can lower risk by stabilizing the MTHFR enzyme among individuals with the variant T allele. This favours 5-mTHF production, critical for methylation reactions [9].

image

Figure 12.5.  The role of folate in pathways of DNA synthesis and methylation reactions. MS, methionine synthase; MSR, methionine synthase reductase; MTHFR, methylenetetrahydrofolate reductase; TS, thymidylate synthase. (Reproduced from Sharp et al. [9], with permission from Oxford University Press.)

Download figure to PowerPoint

Nutrient databases

  1. Top of page
  2. Abstract
  3. Introduction
  4. Dietary assessment instruments
  5. Study designs in nutritional epidemiologic research
  6. Nutrient databases
  7. Statistical issues in dietary analyses
  8. Conclusions
  9. References

Most software for analyzing the nutrient content of foods combines various sources, such as the US Department of Agriculture (USDA) database, with data provided by food manufacturers. These databases can contain upwards of 25 000 different foods and provide information on over 100 different nutrients. The nutrient value of foods varies by plant variety, soil mineral content, storage, processing and cooking conditions, and country-specific food fortification practices. Furthermore, the values of items such as dietary fiber can vary according to different definitions (e.g., plant lignin, cellulose, non-starch polysaccharides) and analytic techniques used for quantification, and could partly explain differences in risk estimates across studies. During the design of their studies, investigators interested in nutritional hypotheses are encouraged to involve personnel who are knowledgeable with these databases.

Statistical issues in dietary analyses

  1. Top of page
  2. Abstract
  3. Introduction
  4. Dietary assessment instruments
  5. Study designs in nutritional epidemiologic research
  6. Nutrient databases
  7. Statistical issues in dietary analyses
  8. Conclusions
  9. References

Energy adjustment

Statistical adjustment for energy intake in models of diet and disease is important for several reasons. Because intakes of nutrients, particularly macronutrients, are correlated with total energy intake, these nutrients may be noncausally associated with disease from confounding by total energy intake [10]. Residual confounding from factors difficult to measure or measured with error that are associated with energy intake (including body size, physical activity and metabolism) can attenuate associations with disease risk. Failure to account for total energy intake can obscure associations between nutrient intakes and disease risk or possibly reverse the direction of the association. Several disease-risk models are described to control for energy intake in epidemiologic studies [10], although recent studies show the superiority of one or two statistical models over the others [11].

Measurement error correction

The recall of diet is associated with both random and systematic error. The former attenuates diet–disease risk estimates by introducing noise and reducing precision, whereas the latter biases risk estimates from overestimating or underestimating a person’s “true” intake. Although neither error can be completely removed, the best safeguard against biased data is to evaluate the validity of a FFQ, described in detail elsewhere [10]. To reduce the effects of random measurement error, several statistical approaches exist to correct estimates of correlation coefficients and regression coefficients to produce approximate unbiased point and interval estimates from linear, Cox and logistic regression models [10]. These correction methods, however, rely on data from a validation study.

Conclusions

  1. Top of page
  2. Abstract
  3. Introduction
  4. Dietary assessment instruments
  5. Study designs in nutritional epidemiologic research
  6. Nutrient databases
  7. Statistical issues in dietary analyses
  8. Conclusions
  9. References

Nutritional epidemiology has contributed significantly to our understanding of the relationships between diet and disease over the past three decades. Ongoing investigations that further characterize important exposure periods (early life, in utero) and clarify associations within the context of genetic susceptibility will continue to elucidate our understanding of the pathophysiology of complex diseases, and support future recommendations for disease prevention.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Dietary assessment instruments
  5. Study designs in nutritional epidemiologic research
  6. Nutrient databases
  7. Statistical issues in dietary analyses
  8. Conclusions
  9. References
  • 1
    Burkitt DP. Epidemiology of cancer of the colon and rectum. Cancer 1971; 28: 3.
  • 2
    Kolonel LN et al. A multiethnic cohort in Hawaii and Los Angeles: baseline characteristics. Am J Epidemiol 2000; 151: 346.
  • 3
    Kelemen LE et al. Development and evaluation of cultural food frequency questionnaires for South Asians, Chinese, and Europeans in North America. J Am Diet Assoc 2003; 103: 1178.
  • 4
    Block G. A review of validations of dietary assessment methods. Am J Epidemiol 1982; 115: 492.
  • 5
    Morton LM et al. Reporting participation in epidemiologic studies: a survey of practice. Am J Epidemiol 2006; 163: 197.
  • 6
    Willett WC. Diet and cancer: an evolving picture. JAMA 2005; 293: 233.
  • 7
    Mayne ST et al. Carbonated soft drink consumption and risk of esophageal adenocarcinoma. J Natl Cancer Inst 2006; 98: 72.
  • 8
    Choi SW et al. Folate and carcinogenesis: an integrated scheme. J Nutr 2000; 130: 129.
  • 9
    Sharp L et al. Polymorphisms in genes involved in folate metabolism and colorectal neoplasia: a HuGE review. Am J Epidemiol 2004; 159: 423.
  • 10
    Willett WC. Nutritional Epidemiology, 2nd edn. New York: Oxford University Press, 1998.
  • 11
    Michels KB et al. The effect of correlated measurement error in multivariate models of diet. Am J Epidemiol 2004; 160: 59.