Invited commentary: meta-physical activity and the search for the truth

Measurement error in self-reported data from questionnaires is a well-recognized challenge in studies of physical activity and health. In this issue of the Journal, Lim et al. (Am J Epidemiol. 2015;181(9):648-655) used data from accelerometers in a small measurement study to correct self-reported physical activity data from a larger study of adults from New York City and to develop an error correction model. They showed that correction of measurement error in self-reported physical activity levels strengthened the associations of physical activity with both obesity and diabetes by 30%-50% compared with using the self-reported questionnaire data alone. Thus, Lim et al. demonstrated a method to improve potentially biased estimates of the association between self-reported physical activity and disease. However, as this field develops, we feel it is important to call attention to a sometimes overlooked problem that occurs when comparing these instruments: Questionnaires and accelerometers are often calibrated (i.e., designed) to measure different types of physical activity, and accelerometers are still subject to measurement error. Thus, physical activity estimates corrected with an imperfect accelerometer measurement might over- or undercorrect the strength of the associations. We take this opportunity to further comment on physical activity measurement in epidemiologic studies and the implications for research.