Linda G. Marc, MPH, MS¹, Lisa F Berkman, PhD¹, Charles Lewis, PhD², and Marcia A. Testa, MPH, PhD³. (1) Department of Society, Human Development and Health, Harvard School of Public Health, 677 Huntington Avenue, 7th FLoor, Boston, MA 02115, 617-421-0164, lmarc@hsph.harvard.edu, (2) Department of Psychology, Fordham University, Rose Hill Campus, Dealy Hall 226G, 441 East Fordham Road, Bronx, NY 10458, (3) Department of Biostatistics, Harvard School of Public Health, 655 Huntington Avenue, Boston, MA 02115

Few quality-of-life (QoL) studies apply differential item functioning (DIF) methodology to evaluate differing item endorsement across groups. This study is the first to investigate measurement bias across educational subgroups on the HIV MOS Scale, which consists of 21 items assessing 9 dimensions of QoL. Twenty items are scored using an ordinal Likert scale, one-item is a visual analog scaled from 0 to 100. Data are from the baseline measures of the Adult AIDS Clinical Trial Group, Protocol-384 (ACTG-384), which enrolled randomized antiretroviral-naïve individuals (N=980) infected with HIV-1 to three- and four-drug regimens between 1998-99. Educational attainment (EA) was measured as an ordinal categorical variable. DIF across EA, conditioning for total scale score, is explored using ordinal logistic regression to predict item responses. Fit of model is assessed using likelihood ratio and chi-square difference tests; and test of significance (p-value). Measures of magnitude consist of the p-value, McKelvey-Zavoina R-square and Odds Ratio. Unidimensionality, uniform DIF (confounding) and non-uniform DIF (effect modification) are evaluated. Items in several domains show presence of uniform DIF. Questions asked ‘Has your health interfered with social activities’ OR=0.79 [95% CI 0.70-0.90]); ‘vigorous activities’ (OR=0.82 [0.68-0.98]); ‘walking uphill’ (OR=1.33 [1.09-1.62]); ‘being happy’ OR=0.83 [0.73-0.94]); and ‘general health perceptions’ OR=1.27 [1.10-1.46]). Non-uniform DIF is present for ‘social activities’ OR=0.98 [097-0.99]. These findings do not constitute item bias. DIF is meaningful with an odds ratio of >=2.0 or, conversely, <=0.50.

Learning Objectives:

• Assess differential item endorsement across educational subgroups using DIF techniques.

Keywords: Behavioral Research, Social Class

Presenting author's disclosure statement:
I have a significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.
Relationship: Recipient of a GlaxoSmithKilne and Novartis unrestricted educational grant.