Skip to Main Content

An official website of the United States government

Principal Investigator
Name
Zezhao Chen
Degrees
Ph.D
Institution
University of Illinois at Urbana Champaign
Position Title
Ph.D Candidate
Email
About this CDAS Project
Study
IDATA (Learn more about this study)
Project ID
IDATA-28
Initial CDAS Request Approval
Nov 13, 2018
Title
Establishing Equivalence of Commonly Used Physical Activity Measures Using Testing Equating
Summary
Physical activity (PA) is characterized as body movement resulting in energy expenditure. Participating in PA has significant benefits for maintaining health and preventing and managing chronic diseases. Accurate measurement of PA EE are essential in epidemiological studies and in determining the effectiveness of PA intervention.

The gold standard for assessing TEE is the DLW method. TEE is composed of the energy costs of the processes essential for life in a free-living environment, which is comprised of resting energy expenditure, physical activity energy expenditure , and diet-induced thermogenesis. Therefore, PAEE can be calculated, while REE is estimated from Mifflin-Joe equation.

Due to the cost and impracticality of the DLW method, simple field measures are often used in practice. The most commonly used field measures of PA are self-report questionnaires; however, the subjectivity of these recall measures has often been criticized. Recently, there has been a proliferation of low-cost, user-friendly apps of consumer-wearable devices with potential in research settings, but the discrepancy among these measures still exists: a recent review stated no consumer wearable devices fell within the acceptable accuracy limits for EE.

Multiple field tests measuring the same construct might have different outputs, which could not be compared. Even when a new field test is proposed, their validity and reliability validation, as well as their cut-off score classification take a long time to establish. Fortunately, a set of statistical procedures known as “testing equating,” in which two or more measures were equated onto the same scale, can be utilized to address the discrepancy limitation.

Test equating methods can generally be classified as traditional equating and item response theory (IRT) equating. Linear and equipercentile equating are the two most commonly used traditional equating methods. In traditional equating methods, score-correspondence of tests is established by setting characteristics of the score distributions equal for a specified group of examinees. In the linear equating method, the means and standard deviations of the two tests for a particular group of examinees are set equal. Linear equating, therefore, can be considered as the establishment of equivalent z scores for two different tests. In equipercentile equating, score distributions are set to be equal so that the same percentile ranks from different tests are considered to indicate the same level of performance.

Given the available data set of the IDATA, establishing the equivalence of the commonly used PA measures becomes possible. Using measured PAEE as an example, the criterion measure is DLW-determined PAEE, the primary field measure is ActiGraph (AG), and the alternative field measures is ActivPAL(AP). After equating AP to the scale of AG, a new variable will be generated, ActivPAL equated, which were transfer to AG scales to estimate PAEE. Therefore, it is possible to evaluate and compare. Finally, field test scores will be correlated with the criterion measure for further accuracy. The purpose of this study is therefore to establish the equivalence of PAEE measures in IDATA, including physical activity intensity classifications.
Aims

Study1. Equating calibration study

(a) Examine the EE equivalent relationship between AG and AP and set AP on the scale on AG, a new variable will be generated, ActivPAL equated (AP-EQ)

To predict METs from AG: a) from vertical counts, equation METs-min = 1.439008 + (0.000795× counts/min) . b) from triaxial counts, equation METs-min = 0.000863× (VM3) + 0.668876, where VM3 is calculated as the square root of the sum of the three axis data squared, over the 60 second epoch .

(b) A variety of equating methods (linear, mean, equipercentile equating, and kernel equating) will be employed so that the best equivalence can be achieved. The accuracy of equating will be examined using a set of error-index, such as standard error of equating (SEE), root mean squared difference (RMSD), and mean signed difference (MSD).

Study 2. Cut-off points agreement
(a) Based on the equivalent relationship established, new classifications for PAEE in classifying sedentary, light, moderate, and vigorous PA will be reset and compared with the ones previously set.

Statistics like the ROC will be employed to assist in the classification setting. The classification agreements in sedentary , light , moderate, and vigorous among AG, AP, and AP-EQ will be examined using Kappa statistics and the Bland-Altman method.

(b) Further examining the PAEE measurement accuracy with criterion measures. Correlation coefficients and pair t-test will be used to determine the relative and absolute EE differences between AG, AP, AP-EQ, and criterion measure by gender. Sum all METs-min (valid wear time) from each AG, AP, and AP-EQ, and divided by 60 to get METs-h per day, compared with DLW-determined PAEE

Study 3. Cross-validation study
Procedure:
(a) Based on the equivalent relationship established, new classifications for PAEE in classifying sedentary, light, moderate, and vigorous PA will be reset and compared with the ones previously set. For example, for data based on vertical axis counts only, thresholds described previously as primarily reflecting locomotor activity: Light-intensity physical activity (<3.00METs)=100-2020; moderate intensity (3 – 5.99 METs) = 2020-5999 counts/min; and vigorous-intensity (≥ 6 METs) = ≥ 6000 counts/min (Troiano et al.,2008).
Statistics like the operating characteristic curve (ROC) will be employed to assist in the classification setting. The classification agreements in sedentary (<1.5 METs), light (1.5-2.99 METs), moderate (3-5.99 METs), and vigorous (≥6 METs) among AG, AP, and AP-EQ will be examined using Kappa statistics and the Bland-Altman method.

(b) Further examining the PAEE measurement accuracy with criterion measures. Correlation coefficients and pair t-test will be used to determine the relative and absolute EE differences between AG, AP, AP-EQ, and criterion measure (DLW-determined PAEE) by gender. Sum all METs-min (valid wear time) from each AG, AP, and AP-EQ, and divided by 60 to get METs-h per day, compared with DLW-determined PAEE: (TEE×0.9) - REE (PAEE in MET-hr/day is numerically equivalent to PAEE in kcal/kg/day based on the fact that 1.0 MET = 1.0 kcal/kg/hr).

Collaborators

Xiaofei Wang, Tsinghua University
Weimo Zhu, Ph.D, University of Illinois at Urbana-Champaign