Selecting optimal screening items for delirium: an application of item response theory
1 Harvard Medical School, Beth Israel Deaconess Medical Center, Division of Gerontology, Institute for Aging Research, Hebrew SeniorLife, 1200 Centre Street, Boston, MA 02131, USA
2 Department of Medicine, Harborview Medical Center, University of Washington, Box 359780, 325 Ninth Avenue, Seattle, WA, 98104, USA
3 Harvard Medical School, Brigham and Women′s Hospital, Division of Aging, 75 Francis St, Boston, MA, 02151, USA
4 Harvard Medical School, Beth Israel Deaconess Medical Center, Division of General Medicine and Primary Care, 330 Brookline Ave; CO-230, Boston, MA, 02215, USA
5 Harvard Medical School, Beth Israel Deaconess Medical Center, Divisions of General Medicine and Primary Care and Gerontology, 330 Brookline Ave, CO-216, Boston, MA, 02215, USA
BMC Medical Research Methodology 2013, 13:8 doi:10.1186/1471-2288-13-8Published: 22 January 2013
Delirium (acute confusion), is a common, morbid, and costly complication of acute illness in older adults. Yet, researchers and clinicians lack short, efficient, and sensitive case identification tools for delirium. Though the Confusion Assessment Method (CAM) is the most widely used algorithm for delirium, the existing assessments that operationalize the CAM algorithm may be too long or complicated for routine clinical use. Item response theory (IRT) models help facilitate the development of short screening tools for use in clinical applications or research studies. This study utilizes IRT to identify a reduced set of optimally performing screening indicators for the four CAM features of delirium.
Older adults were screened for enrollment in a large scale delirium study conducted in Boston-area post-acute facilities (n = 4,598). Trained interviewers conducted a structured delirium assessment that culminated in rating the presence or absence of four features of delirium based on the CAM. A pool of 135 indicators from established cognitive testing and delirium assessment tools were assigned by an expert panel into two indicator sets per CAM feature representing (a) direct interview questions, including cognitive testing, and (b) interviewer observations. We used IRT models to identify the best items to screen for each feature of delirium.
We identified 10 dimensions and chose up to five indicators per dimension. Preference was given to items with peak psychometric information in the latent trait region relevant for screening for delirium. The final set of 48 indicators, derived from 39 items, maintains fidelity to clinical constructs of delirium and maximizes psychometric information relevant for screening.
We identified optimal indicators from a large item pool to screen for delirium. The selected indicators maintain fidelity to clinical constructs of delirium while maximizing psychometric information important for screening. This reduced item set facilitates development of short screening tools suitable for use in clinical applications or research studies. This study represents the first step in the establishment of an item bank for delirium screening with potential questions for clinical researchers to select from and tailor according to their research objectives.