Over 500,000 cancer deaths are expected to occur in the US in 2005. Investigations of the impact of diet on cancer have been mixed. Errors in dietary assessment methods lead to biased estimates of the effect of diet on cancer outcomes, a possible explanation for the lack of consensus of diet-cancer studies. The goal of this proposal is to quantify bias, reliability and validity of dietary assessment methods using existing data from a longitudinal dietary intervention study. The validity coefficients estimated in this proposal will provide statistical adjustment factors that can be used to obtain corrected estimates of diet-cancer associations. The implications of bias in dietary assessment methods for the design of cancer trials will also be examined. The analysis will focus on the ongoing Women's Healthy Eating and Living (WHEL) study, a randomized trial examining the effect of a high fruit, vegetable, fiber and low-fat diet on breast cancer recurrence. The study randomized 3088 women between 1995 and 2000, and will close in 2007. The dietary exposure of WHEL subjects is assessed at study entry, 1 and 4 yrs by 3 methods: repeat 24-hour recall interviews; the Arizona Food Frequency Questionnaire; blood samples analyzed for plasma carotenoid concentrations (a marker, well correlated with fruit/vegetable intake).
The specific aims of this proposal are (1) To assess the reliability of the plasma marker over time (2a) To develop models for estimating bias and measurement error in self-report dietary instruments (2b) To assess whether participating in a diet intervention reduces bias (2c) To examine what factors (eg age, body-size) contribute to bias/error (3) To develop (and make publicly available) software implementing statistical methods derived in this proposal.
Aim 1 will estimate reliability measures (intraclass correlation, within-subject coefficient of variation) for the plasma marker.
For Aim 2, mixed-effects models will be fit with flexible distributions for """"""""true"""""""" (unobservable) dietary intake (eg mixture of normals). Simulations will be used to investigate the robustness of our findings. Relevance: Lifestyle factors such as diet and physical activity are modifiable behaviors that could protect against cancer incidence and improve survivorship after a cancer diagnosis. Statistical methods that quantify biases in dietary assessment tools and correct these biases could lead to more valid estimates of diet-cancer associations. ? ? ?
| Natarajan, Loki; Pu, Minya; Fan, Juanjuan et al. (2010) Measurement error of dietary self-report in intervention trials. Am J Epidemiol 172:819-27 |
| Natarajan, Loki (2009) Regression calibration for dichotomized mismeasured predictors. Int J Biostat 5:Article 12 |
| Messer, Karen; Natarajan, Loki (2008) Maximum likelihood, multiple imputation and regression calibration for measurement error adjustment. Stat Med 27:6332-50 |
