Secondary Logo

Institutional members access full text with Ovid®

Share this article on:

Stratified Probabilistic Bias Analysis for Body Mass Index–related Exposure Misclassification in Postmenopausal Women

Banack, Hailey R.a; Stokes, Andrewb; Fox, Matthew P.c; Hovey, Kathleen M.a; Cespedes Feliciano, Elizabeth M.d; LeBlanc, Erin S.e; Bird, Chloef; Caan, Bette J.d; Kroenke, Candyce H.d; Allison, Matthew A.g; Going, Scott B.h; Snetselaar, Lindai; Cheng, Ting-Yuan Davidj; Chlebowski, Rowan T.k; Stefanick, Marcia L.l; LaMonte, Michael J.a; Wactawski-Wende, Jeana

doi: 10.1097/EDE.0000000000000863

Background: There is widespread concern about the use of body mass index (BMI) to define obesity status in postmenopausal women because it may not accurately represent an individual’s true obesity status. The objective of the present study is to examine and adjust for exposure misclassification bias from using an indirect measure of obesity (BMI) compared with a direct measure of obesity (percent body fat).

Methods: We used data from postmenopausal non-Hispanic black and non-Hispanic white women in the Women’s Health Initiative (n=126,459). Within the Women’s Health Initiative, a sample of 11,018 women were invited to participate in a sub-study involving dual-energy x-ray absorptiometry scans. We examined indices of validity comparing BMI-defined obesity (≥30 kg/m2), with obesity defined by percent body fat. We then used probabilistic bias analysis models stratified by age and race to explore the effect of exposure misclassification on the obesity–mortality relationship.

Results: Validation analyses highlight that using a BMI cutpoint of 30 kg/m2 to define obesity in postmenopausal women is associated with poor validity. There were notable differences in sensitivity by age and race. Results from the stratified bias analysis demonstrated that failing to adjust for exposure misclassification bias results in attenuated estimates of the obesity–mortality relationship. For example, in non-Hispanic white women 50–59 years of age, the conventional risk difference was 0.017 (95% confidence interval = 0.01, 0.023) and the bias-adjusted risk difference was 0.035 (95% simulation interval = 0.028, 0.043).

Conclusions: These results demonstrate the importance of using quantitative bias analysis techniques to account for nondifferential exposure misclassification of BMI-defined obesity. See video abstract at,

From the aDepartment of Epidemiology and Environmental Health, School of Public Health and Health Professions, University at Buffalo, The State University of New York, NY

bDepartment of Global Health and Center for Global Health and Development, Boston University School of Public Health, MA

cDepartment of Epidemiology, Boston University School of Public Health, MA and Department of Global Health, Boston University School of Public Health, MA

dDivision of Research, Kaiser Permanente Northern California, Oakland, CA

eKaiser Permanente Center for Health Research NW, Portland, OR

fRAND Corporation, Santa Monica, CA

gDepartment of Family Medicine and Public Health, University of California, San Diego, CA

hThe Department of Nutritional Sciences, College of Agriculture and Life Sciences, The University of Arizona

iCollege of Public Health, University of Iowa, IA

jDepartment of Epidemiology, University of Florida, Gainesville, FL

kDepartment of Medical Oncology and Therapeutics Research, City of Hope National Medical Center, Duarte, CA

lDivision of Cardiovascular Medicine, Department of Medicine, Stanford University School of Medicine, Stanford, CA.

Submitted November 29, 2017; accepted May 28, 2018.

Data are available from the Women’s Health Initiative coordinating center as well as through the NHLBI BioLINCC program. Software code for replication is available in the supplemental digital content.

This work was supported by grants HHSN268201600018C, HHSN-268201600001C, HHSN268201600002C, HHSN268201600003C, and HHSN268201600004C from the National Heart, Lung, and Blood Institute, National Institutes of Health, US Department of Health and Human Services and the Banting Postdoctoral Fellowship Program from the Canadian Institute of Health Research.

The authors report no conflicts of interest.

Supplemental digital content is available through direct URL citations in the HTML and PDF versions of this article (

Correspondence: Hailey Banack, Department of Epidemiology and Environmental Health, School of Public Health and Health Professions, University at Buffalo, The State University of New York, 270 Farber Hall, Buffalo, NY 14214. Email:

Copyright © 2018 Wolters Kluwer Health, Inc. All rights reserved.