P3‐254: A COMPARISON OF NOVEL APPROACHES FOR EXAMINING DIFFERENTIAL ITEM FUNCTIONING IN THE MONTREAL COGNITIVE ASSESSMENT: A MULTIPLE COVARIATE APPROACH

Accurate clinical diagnosis of individuals into the correct disease status depends on the validity of the assessment instrument. Differential item functioning (DIF) in these tests due to group membership differences may weaken the validity of the cognitive assessment.Traditional item response theory (IRT) methods to study DIF allow the consideration of only few subpopulations, like racial group, when estimating if the probability of responding to an item depends upon the membership to a subpopulation. We evaluated items from the MoCA scale for DIF relative to multiple observed characteristics concomitantly and compared the results to traditional IRT methods. Data from participants in the multi-site Alzheimer's Disease Neuroimaging Initiative, with item-level data in MoCA (N=1,108; age=72.77, SD=7.14 years; 54.6% male; education=16.11, SD=2.73 years; 89% White), were used to assessed DIF due to age, education, and gender using the following four traditional IRT-based approaches: 1) Lord's chi-square test (Lord, 1980); 2) Raju's area test (Raju, 1990); 3) Likelihood-ratio test (Thissen, Steinberg and Wainer,1988), and 4) Rasch modeling with accurate item-parameter error variance-covariance matrices and expectation-maximization (Cai, 2008). Novel IRT-based methods included: 1) Regularization based on penalty using penalized maximum likelihood (Tutz, 2013), 2) Recursive-partitioning and tree-method, and 3) Random-item mixture modeling (Frederickx, 2011). The analysis included individuals across the full spectrum of disease status. In general, there was moderate to almost perfect agreement across methods. Seven items were found to exhibit DIF and one item (Serial 7; measuring Attention) consistently across all methods showing a higher probability to be answered correctly by males and individuals with higher education level. There was an interaction between gender and education; females with lower education (<14 years) showed a lower probability of answering Serial 7 items correctly compared to females with 17 or more years of education. The same pattern was not found in males. The extent that commonly-used screening tests are useful in making diagnostic decisions depends largely on their validity. Issues concerning measurement invariance are important in research and clinical assessments. Methods, as those presented, can be routinely used to enhance measurement and decisions about what instrument to use for diagnosis.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.