||1R21CA119073-01A1 Interpret this number
||Investigatin of Measurement Error Correction Models for Physical Activity
DESCRIPTION (provided by applicant): Physical activity questionnaires (PAQ) used in epidemiologic studies are known to contain a substantial level of measurement error and it is generally believed that such errors result in bias towards the null, or an underestimation of the true association. However, little research has described systematic and random errors in physical activity (PA) reporting or their impact on epidemiologic measures of effect (e.g., relative risk). Existing measurement error correction (MEC) models provide a framework for quantifying the error structure of PAQ-based measurements, and for estimating correction factors that adjust for bias induced by these errors (e.g., attenuation). In this exploratory/developmental research (R21) we will utilize data from two cohort studies based in Shanghai, China (Shanghai Women's Health Study, N approximately 75,000 [CA70867], Zheng, PI; Shanghai Men's Health Study, N approximately 60,000 [RA82729], Shu, PI). Measurement studies conducted within the cohorts will obtain PAQ measures and repeated measurements by PA Logs and Actigraph (4 times/yr; 7-days/time) for 300 men and women. Using these data sources and the MEC models of Kipnis and colleagues (2003), we will complete three major objectives. First, we will quantify the error structure of reports derived from the PAQ and PA Log measures in terms of systematic (i.e., activity-related and person- specific biases) and random components, using the Actigraph as an unbiased measure to achieve identifiability in the MEC models. Second, we will calculate attenuation factors, de-attenuated validity coefficients for the PAQ, and estimate the magnitude of the correlation of person-specific biases between the PAQ and PA Log. Third, through simulation studies we will explore the robustness of the MEC modeling procedures relative to different assumptions required to achieve identifiability of the models using the Actigraph. This work represents a first step in applying and testing existing MEC methods in the context of PA measurement and will provide critical insight into the measurement properties of widely used reference measures of free-living PA. Collectively, this research will extend our understanding of the influence of measurement errors in PA studies and indicate the next steps necessary to minimize these errors in future studies.
Using repeated measures to correct correlated measurement errors through orthogonal decomposition.
, Zhang S.
, Friedenreich C.
, Matthews C.E.
Communications in statistics: theory and methods, 2017; 46(23), p. 11604-11611.