Received 12/13/95; revised 1/13/97; accepted 2/17/97. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked ads’ertisernent in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. I This project was supported by NIH Grant P01-CA41 108. The views expressed here represent the views of the authors and not those of NIH. 2 To whom requests for reprints should be addressed, at Arizona Prevention Center, University of Arizona, Tucson, AZ 85724-5126. Phone: (520) 626-7863; Fax: (520) 321-7754. 3 The abbreviations used are: FFQ, food frequency questionnaire; AFFQ, Arizona FFQ; PPV, positive predictive value; NPV, negative predictive value; USDA, United States Department of Agriculture; NFCS, Nationwide Food Consumption Survey; WBFT, Wheat Bran Fiber Trial; DR. diet record; 4DFR, 4-day food record. Vol. 6, 347-354, May 1997 Cancer Epidemiology, Biomarkers & Prevention 347 Use of a Food Frequency Questionnaire to Screen for Dietary Eligibility in a Randomized Cancer Prevention Phase III ri1 Cheryl Ritenbaugh,2 Mikel Aickin, Douglas Taren, Nicolette Teufel, Ellen Graver, Kathleen Woolf, and David S. Alberts Department of Family and Community Medicine [C. R., M. A., D. T., N. T., E. G., K. W.] and Department of Medicine [D. S. A.], The Arizona Cancer Center, The University of Arizona, Tucson, Arizona 85724 Abstract Cancer prevention clinical trials use food frequency questionnaires (FFQs) to assist in eligibility screening. FFQ reliability and validity studies are available, but these studies do not evaluate FFQs as screening tools. The Wheat Bran Fiber Trial of the University of Arizona used a FFQ as an eligibility screen with the goal of screening out subjects whose true daily calcium intake was less than 500 mg per day (for safety) and whose true dietary fiber intake was greater than 30 g per day (for safety and trial efficiency). Subjects ineligible by FFQ were interviewed for final dietary eligibility determinations. A study was undertaken within the Wheat Bran Fiber Trial to evaluate the sensitivity, specificity, positive predictive value, and negative predictive value (NPV) of the FFQ used in this context. Four-day food records were collected on 183 potential participants before entry into the study. Using the 4-day averages as the “true” value, sensitivity, specificity, positive predictive value, and NPV were calculated for men and women separately under two screening conditions: using the target calcium and dietary fiber values and using “revised” values identified in interim analysis within the study. NPV was acceptable in all analyses. Sensitivity for low calcium intake was inadequate under the original criteria (0.33 for men and 0.09 for women) but acceptable under the revised criteria (0.80 for men and 0.81 for women). With the revised criteria, specificity declined, resulting in heavy screening burdens deemed worthwhile for the safety considerations. Dietary fiber eligibility screening worked well at target values. These differences were not predicted by reliability/validity studies. Introduction Randomized controlled trials are needed to determine the long- term effects of diet in the etiology and prevention of cancer. Ideally, such trials recruit panicipants who are not already on the intervention and randomize them to experimental or control conditions. FFQs3 are currently in use as screening tools to identify suitable participants in the Colon Cancer Prevention Program Project at The University of Arizona, and in the Women’s Health Initiative, a NIH-funded nationwide trial (1). The FFQ method has been applied in epidemiological research to the etiology of cancer and cardiovascular diseases, in which one or more dietary components is either the factor of interest or a confounding variable. FFQs have been subjected to reli- ability and validity analysis for use in these epidemiological studies. However, use in eligibility screening requires informa- tion on the sensitivity and specificity characteristics of the instrument, analyses that have not been widespread to date. Multiple DRs have generally been considered to yield the most valid estimates of individuals’ usual intakes but they are far from perfect (a flawed “gold standard”). In the United States, recent studies comparing energy estimates from DRs with energy estimates from doubly labeled water find an aver- age underestimate of about 200 kcal/day, although this varies with the population studied (2-4). However, even the use of DRs is in many study settings limited by the time and expense they require. The quantitative FFQs are designed to measure average long-term diet or usual consumption of specified food items rather than to provide a precise estimate of short-term intake. The low cost and ease of administration of FFQs facil- itates their use for screening in large clinical trials; they impose the least respondent burden of all dietary assessment methods in common use and, if accompanied by clear instructions, can be self-administered successfully in many populations. Assessment of the quality of a measurement entails con- sideration of its reliability and of the validity of the technique used to derive the measurement. Reliability may be defined as the level of agreement between replicate measurements under the same conditions (repeatability), and validity may be defined as the extent to which a measurement technique measures that which it aims to measure (accuracy). These two aspects of quality are related in that a valid measurement technique gives measurements that are repeatable. A measure that is not repeat- able may lead to misclassification of individuals upon catego- rization with respect to that measurement. Dietary assessment methods usually contain a high degree of random misclassifi- cation due to random measurement error. Random measure- ment error often attenuates risk estimates of the diet-disease on January 10, 2022. © 1997 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from