Developing a brief, accurate and adaptive version of the Boston Naming Test based on high‐fidelity estimates of item properties
Thomas A. Giauque, Lynn Shaughnessy, George Lin, Maurice Smith, Daniel Z. Press- Psychiatry and Mental health
- Cellular and Molecular Neuroscience
- Geriatrics and Gerontology
- Neurology (clinical)
- Developmental Neuroscience
- Health Policy
- Epidemiology
Abstract
Background
Current naming tests such as the Boston Naming Test (BNT) and the Multilingual Naming Test have demonstrated significant clinical utility in detecting naming deficits in mild cognitive impairment (MCI), assisting with differential diagnosis among neurodegenerative disorders, and measuring disease progression in AD. However, the tests are time‐consuming due to their large number of items, and pen‐and‐paper administration. A computerized, adaptive naming test could increase the accuracy and speed of assessment, enhance usability, and minimize cross‐cultural testing disparities.
Method
Current naming tests that produce accurate results over a range of naming abilities are lengthy, because they must both include items that densely cover the large range of potential abilities. In contrast, an adaptive test can be more efficient, by quickly zeroing in on a narrow range and then concentrating test questions only in the vicinity of the participant’s underlying ability. This maximizes efficiency in terms of information obtained per item, by tailoring the test to avoid items that would, based on previous responses, be either too easy or too difficult for the current participant.
Result
We show via a series of probabilistic simulations that the precision of item property estimates would limit effectiveness of adaptive strategies previously attempted (Fergadiotis et al., 2015). In fact, we find that the mismatch between the true properties of items and existing estimates, due to the broad confidence regions of these estimates (Embretson & Reise, 2000) considerably reduces achievable accuracy. We are therefore in the process of delineating item properties with considerably higher fidelity than those currently available in order to achieve the efficiency improvement that adaptive testing can in principle deliver, analyzing the results of individual item responses from over a 1000 BNT administrations from participants distributed across a range of ability levels. We will test the ability of adaptively‐chosen 12‐item subsets of the BNT to estimate the full scores from our clinics via simulation and implement a new ipad‐based adaptive tool, prospectively testing its accuracy in predicting the BNT score and its ability to identify MCI and AD.
Conclusion
The development of a computerized, adaptive, naming test may allow considerable time savings while maintaining accuracy.