Efficiency and sensitivity of multidimensional computerized adaptive testing of pediatric physical functioning

Purpose. Computerized adaptive tests (CATs) have efficiency advantages over fixed-length tests of physical functioning but may lose sensitivity when administering extremely low numbers of items. Multidimensional CATs may efficiently improve sensitivity by capitalizing on correlations between functional domains. Using a series of empirical simulations, we assessed the efficiency and sensitivity of multidimensional CATs compared to a longer fixed-length test. Method. Parent responses to the Pediatric Evaluation of Disability Inventory before and after intervention for 239 children at a pediatric rehabilitation hospital provided the data for this retrospective study. Reliability, effect size, and standardized response mean were compared between full-length self-care and mobility subscales and simulated multidimensional CATs with stopping rules at 40, 30, 20, and 10 items. Results. Reliability was lowest in the 10-item CAT condition for the self-care (r=0.85) and mobility (r=0.79) subscales; all other conditions had high reliabilities (r0.94). All multidimensional CAT conditions had equivalent levels of sensitivity compared to the full set condition for both domains. Conclusions. Multidimensional CATs efficiently retain the sensitivity of longer fixed-length measures even with 5 items per dimension (10-item CAT condition). Measuring physical functioning with multidimensional CATs could enhance sensitivity following intervention while minimizing response burden.