May 2, 2017
Anarina Murillo, PhD, Postdoctoral Scholar in the Nutrition Obesity Research Center (NORC) and Office of Energetics, has won the Third Place Award for “Predicting Physical Activity from 2D Body Composition Data Using Regression Calibration Methods for Measurement Error Correction” in the Postdoctoral/Junior Faculty Research Poster Presentation at the Twelfth Annual Arizona WAESO (Western Alliance to Expand Student Opportunities) Student Research Conference, held on March 15, 2017.
In the study, Murillo and her UAB colleagues applied regression calibration (RC) methods, which correct for measurement errors in health research, to predict physical activity levels as a function of body fat using a logistic regression model. Percentage of body fat estimated by dual-energy X-ray absorptiometry (DXA) scans, which are costly and time-consuming, were used as a reference to adjust model parameters of body fat assessed by 2D photographic images, which was assumed to involve measurement error and therefore may lead to biased estimates.
A cross-sectional sample of 723 European American/White and African American/Black adults between ages 19 and 79 participated in the study (57 percent male and 43 percent female). Logistic RC was applied to predict levels of physical activity based on body fat levels, utilizing error-corrected percent body fat parameter estimates obtained from the 2D photographic data (BFPhoto). The researchers then compared this data against the assumed error-free equivalent equation, utilizing body fat percentage measured by DXA (BFDXA). BFDXA was associated with BFPhoto (r=0.88; P<0.0001).
They found that “Bland-Altman analyses showed small differences between BFDXA and BFPhoto (mean difference=0.02; 95% CI:-9.87 to 9.87); However, biases in body fat estimates were observed based on physical activity levels, race, and sex (P<0.05). RC corrected BFPhoto parameter estimates were more similar to BFDXA coefficient estimates, in comparison to uncorrected BFPhoto parameter estimates, and therefore presumably less biased.”
Murillo and her team determined that the RC methods’ ability to adjust for measurement error may lead to less-biased estimates of model parameters for variables with small measurement error.
Co-investigators in the study are Olivia Affuso, PhD, associate professor in the Department of Epidemiology and associate scientist in the NORC; Courtney M. Peterson, PhD, MSc, MA, MS, assistant professor in the Department of Nutrition Sciences and scientist in the NORC; and David B. Allison, PhD, distinguished professor and director of the NORC and Office of Energetics.
Open the original version of this page.