Efficiency aspects of design and analysis of prospective cohort studies on diet, nutrition and cancer

This thesis presents and analyzes methodological approaches to improve the design and analysis of prospective cohort studies on the relations between diet, nutritional status and cancer. The first chapters discuss methods to optimize the measurement of the individuals' habitual dietary intakes, focussing on the use and design of sub-studies for the "validation" or "calibration" of baseline dietary questionnaire assessments. The power of prospective studies can be improved by maximizing the variation in true dietary intake levels actually distinguished - or "predicted" - by questionnaire assessments. This can be achieved by designing an optimal questionnaire method, using a preliminary validity study to evaluate its performance. An additional possibility is to broaden the range of dietary exposures by conducting multiple cohort studies in populations with different dietary habits. A main objective is to precisely estimate the magnitude of the predicted variation of intake levels, to account for the effect of measurement error as well as of the real variation in exposure, in the evaluation of the power or sample size requirements of a cohort study, and in the estimation of relative risks describing diet-disease relations. The predicted variation is estimated most efficiently by means of a "calibration" sub-study, which differs from validity studies in that it requires only a single (unbiased) reference measurement per person (e.g., based on a 24-hour recall), in a representative sub-sample of cohort members. In multi-cohort projects, calibration studies are essential to improve the between-cohort comparability of relative risk estimates, and to increase the power of a statistical test for the presence of a diet-disease association based on a pooled summary estimate. A simplified method is proposed for the estimation of sample size requirements of dietary calibration studies. When the exposure assessments are based on a biochemical marker, a most efficient design is to store biological specimens in a biobank, and to postpone laboratory analyses until cases with disease have been identified. Nevertheless, the number of scientific hypotheses potentially of interest is usually much larger than can be tested with limited amounts of biological specimens available. The last chapter of this thesis discusses the use of a sequential study design, to allow the evaluation of a maximum number of different hypotheses at the expense of as little biological material as possible.