Brief reports: Impact of informed starting value on longitudinal computer adaptive tests in PROMIS® assessments

Objective This study examined whether respondent burden in PROMIS assessments can be reduced by incorporating an informed prior. The convergence of computer adaptive test (CAT) algorithms may improve when the initial θ i is close to the true θ t . Prior research in educational testing has demonstrated these benefits using the dichotomous model. This study investigates whether these findings extend to the Graded Response Model (GRM) utilized by PROMIS. Methods A test–retest simulation was conducted. The initial CAT assessment uses the prior mean θ i = 0. For the retest assessment, the prior mean θ i is set to final θ estimate of the previous CAT. Simulations were performed across each of the domains in the adult and pediatric PROMIS profiles. Efficiency and precision were calculated as the count of items administered and the root mean squared (RMSE) respectively. Results Analyses were stratified by the degree of alignment between examinees' true theta values (θ t ) and the information provided by the PROMIS domain banks (e.g., well-targeted vs. poorly targeted). Conclusions When prior information about trait levels was available, CATs could incorporate this information into subsequent assessments. However, simulations indicated that using an informed prior produced only marginal improvements in efficiency and precision with PROMIS bank items.