Description
Aims: Computerized adaptive tests (CATs) are instruments that areadapted to the patient on the fly, and are increasingly used in health
measurement. Several studies conducted in the field of educational
testing have shown that empirical information about a test-taker can be
used to make a CAT more efficient and more precise. However, it is
currently not sufficiently clear to what degree these findings can be
generalized to health measurement. The aim of this simulation study is to
evaluate the risks and return associated with using prior information in a
healthcare setting. Methods: An empirical item bank calibrated with the
graded response model was used to simulate unidimensional CATs. The
bank was based on an instrument measuring personality dysfunction,
which consisted of 12 polytomous items. The prior was based on a global
score set by a clinician prior to administering the instrument, which was
scored on an ordinal scale (0-4). The correlation between the prior and
the estimated trait scores (personality dysfunction) equaled .76. Fivehundred simulees were generated, and two CATs were administered to
them: one with a standard normal prior (default) and one with an
empirical prior. For each global score, the theta value with the highest
probability was taken as the mean for the empirical prior. The lower this
highest probability, the larger the variance of the empirical prior:N(-3,
0.685) if global score is 0 N(-1, 0.625) if global score is 1 N(0, 0.612) if
global score is 2 N(1, 0.812) if global score is 3 N(2, 0.500) if global
score is 4 Results: For 63.6% of the simulees, using the empirical prior
resulted in a reduction in test length; for the remaining 31.7% test length
was equal across the two conditions. On average, the empirical prior was
associated with a reduction in test length of 20%. Bias in the final
estimates was comparable across the two conditions, and was generally
low (well below |.5|). More detailed results will be presented at the
conference. Conclusion: Based on these findings, using empirical prior
information can be expected to be ad
| Period | Oct-2020 |
|---|---|
| Event title | 27th Annual Conference of the International Society for Quality of Life Research |
| Event type | Conference |
| Degree of Recognition | International |