A More Efficient Outcome for Alzheimer Disease Research: the Item Response Theory Based Score for the Clinical Dementia Rating (CDR®). (20th December 2022)

• Record Type: Journal Article
• Authors: Li, Yan; Wang, Guoqiao; Xiong, Chengjie; Moulder, Krista L.; McDade, Eric; Hassenstab, Jason J.; Aschenbrenner, Andrew J.; Benzinger, Tammie L.S.; Gordon, Brian A.; Fagan, Anne M.; Chatterjee, Saptarshi; Cheng, Yun‐Ju; Bateman, Randall J.; Morris, John C.
• Is Part Of: Alzheimer's & dementia. Volume 18(2022)Supplement 10
• Page Start: n/a
• Page End: n/a
• Publication Date: 2022-12-20
• Subjects: Alzheimer's disease -- Periodicals; Alzheimer Disease -- Periodicals; Dementia -- Periodicals; Démence; Maladie d'Alzheimer; Périodique électronique (Descripteur de forme); Ressource Internet (Descripteur de forme); 616.83
• DOI: 10.1002/alz.062591
• Languages: English
• ISSNs: 1552-5260

Abstract

Abstract: Background: CDR sum of boxes (CDR‐SB) is widely used as a primary outcome in Alzheimer disease (AD) research due to its reliability and accuracy. However, it is an ordinal rank measure instead of a continuous measure and therefore may not be able to capture subtle changes in cognition. Additionally, it weights each domain equally, without accounting for the patterns of the responses scores across CDR items. To generate a more optimal score, we developed a bi‐factor model based on the item response theory (IRT) which can account for different response pattern, item difficulty and discrimination level to provide an improved estimate of the dementia severity. We demonstrated the superiority of the IRT score in two independent cohorts. Method: Baseline item level CDR data from 2949 participants enrolled in the WashU Knight ADRC study were used to develop the IRT model. The best fitting IRT model was then applied to the longitudinal item level CDR data in both the Knight ADRC and DIAN‐TU‐001 (142 participants) studies to generate the IRT scores (a continuous measure) for each participant at each visit. Result: Correlations between the CDR IRT score and SB were consistent at different visits (0.89 ∼ 0.92 for DIAN‐TU cohort, 0.92 ∼ 0.94 for Knight ADRC cohort). The IRT score had a much larger effect size (> 60% for the DIAN‐TU cohort and > 40% for the Knight ADRC cohort) than that of CDR‐SB when comparing converter to stable cognitively normal participants. For CDR‐SB of 0 group, the IRT score was significantly associated with some of the clinical/cognitive outcomes and biomarkers. For CDR‐SB > 0 group, the IRT score had slightly higher correlations with other outcomes than CDR‐SB. The IRT score correlated better with the estimated age from dementia onset in the DIAN cohort. Using IRT score as the primary outcome in clinical trials could save 10% ∼ 30% of the sample size compared to that of using CDR‐SB. Conclusion: The CDR IRT score could potentially be used as a more efficient outcome in AD research to gain precision in detecting cognitive decline and improve power (or reduce sample size).