| Sumario: | Computerized adaptive testing (CAT), which administers only items relevant to respondents' ability, has the advantage of measuring persons' ability precisely with considerably fewer items than traditional tests. CAT has been proposed for use in healthcare to reduce the respondents', administrators' or researchers' burden in clinics/clinical trials. However, there have been few studies in healthcare that have investigated the optimal characteristics of a CAT. The objective of my study was to investigate: (1) the psychometrics of an item bank measuring upper extremity (UE) disorders for CAT use, (2) how different testing procedures affect the ability estimates from CAT, and (3) whether CAT produce better ability estimates than a traditional static short assessment. The psychometrics of the item bank developed by combining items from the Disabilities of the Arm, Shoulder and Hand and the Upper Extremity Functional Index were examined by confirmatory factor analysis, item response theory analysis, and differential item functioning (DIF) analysis across body part impairments (neck, shoulder, elbow, and wrist/hand). Repeated-measures MANOVA was used to investigate the standard error (SE) and bias of ability estimates from CAT with different testing procedures. Structural equation modeling was implemented to examine the correlations between ability estimates from the full test and different CAT structures and the full test with a short form. Further, paired-sample t test was performed to investigate the SE and bias of ability estimates from CAT and a static short form. In general, the item bank was found essentially unidimensional, the generalized partial credit model fit to the data better than partial credit model and there was no significant DIF. The ability estimates from CAT with expected a posteriori ability estimation method was found to be more precise and more comparable to those from full test than ability estimates derived from the maximum likelihood estimation. Further, CAT had better precision, comparability to full test and sensitivity to detect change than a static short form. The findings from this study suggest that UE CAT differ across estimations methods and are significantly better than a short form version.
|
|---|