A better understanding of the limitations of current physical assessment measures is needed. Both random and systematic errors in an assessment method can have a large impact on estimating the relationship between physical activity and disease. By using a calibration method with an unbiased measure of physical activity, a prediction of truth (which is unobservable) may be obtained for an observed value of a biased instrument. A correction factor can be estimated to correct the relative risk of developing disease when a biased measure of physical activity is used to predict disease status. The utility of this method is that the equation predicting true physical activity may be developed on a subset of individuals, rather than an entire cohort. Clearly, assessing whether a measure is unbiased is important, as unbiased measures may be used in both validation and calibration studies, as well as in surveillance studies. The Observing Protein and Energy Nutrition (OPEN) Study explored the measurement error structure of self-report dietary instruments using an unbiased biomarker for total energy expenditure, doubly labeled water (DLW), for 451 men and women ages 40-69 y. A physical activity questionnaire (PAQ) was also administered to the OPEN participants, the same PAQ that was used in National Health and Nutrition Examination Survey (NHANES) 2003-06. We anticipate data from approximately 3,670 participants ages 40-69 y who completed the PAQ and wore Actigraph accelerometers will be available from 2003-06 NHANES. Although the Actigraph is an objective measure, it is not clear if it is unbiased, or if it measures activity equally well for different gender, age, race/ethnicity, or weight status groups. This study will utilize data from the OPEN study and NHANES to develop a measurement error framework for physical activity assessment, and to apply this framework to OPEN and NHANES data for middle-aged adults.
The specific aims of this study are: (1) To validate and calibrate the PAQ in the OPEN Study using DLW as an unbiased measure of energy expenditure;(2) To determine whether the Actigraph measure of physical activity in NHANES is unbiased;and (3) to compare estimates from the measurement error models by personal characteristics. This study will provide statistical methods for correcting physical activity data for measurement error in an epidemiologic study. It will provide insight as to whether the Actigraph may be used as an unbiased measure, and provide a statistical modeling approach for assessing other accelerometers in other studies.
Surveys and other methods of measuring physical activity are prone to measurement error. Due to this error, it is difficult for researchers to make the correct conclusions about the relationship between true physical activity and the development or progression of disease, or to make recommendations about how much physical activity people need to do to stay healthy. Statistical methods are needed to describe the types of errors that exist, to examine their impact on conclusions made in research studies, and to develop better ways of analyzing data from these studies.
|Tooze, Janet A; Troiano, Richard P; Carroll, Raymond J et al. (2013) A measurement error model for physical activity level as measured by a questionnaire with application to the 1999-2006 NHANES questionnaire. Am J Epidemiol 177:1199-208|