15specific research needs, either by manually selecting items that are believed to
closely match the levels of the measured trait of interest in the target population or
by computerized adaptive testing (CAT) algorithms. CAT algorithms sequentially
statistically optimize estimated trait levels by presenting only items that match the
trait level of the respondent, as estimated from previously responded to items or
other prior information [ 51 ]. By tailoring the assessment to the level of the mea-
sured trait of individual patients, trait levels can be estimated with an optimal, pre-
specified level of precision, while using a lower number of items compared with
traditional instruments. This is achieved by capitalizing on the fact that reliability is
locally defined in IRT. That is, item response models can be used to construct the
so-called information functions, which describe the measurement precision of indi-
vidual items at each of the different levels of the measured trait. Information func-
tions also provide researchers with more detailed information regarding the
influence of measurement error on the information that is provided by the PRO
instrument and its items, compared with the CTT-based reliability indices.
Consequently IRT-based analysis is also ideally suited for scale construction and
refinement. This information can be utilized to develop measures that are optimally
relevant to the disability levels of the studied population, while response burden can
be managed. Another application of IRT-based item banking is the establishment of
a common IRT-based metric between existing outcome measures that are used for
the same purpose in a setting. Once established, such metrics facilitate the compa-
rability and interpretability of outcomes by the development of IRT-based scoring
procedures that allow study results to be expressed on 1 standardized metric, irre-
spective of the questionnaire that was used in a specific study. This would allow
results obtained in different settings to be compared despite that different individual
measures were used in individual studies.
Future Directions
Over the last few decades, a variety of PROs have become available for use in vari-
ous settings in rheumatology. While traditionally PROs have been used primarily to
assess beneficial effects of treatment in clinical trials in this field, the increasing
attention for the patient as the center of healthcare has led to the proliferation of
PROs in other settings, such a clinical practice, comparative effectiveness research
and in the assessment of quality of care. Different settings require a different focus
and different measurement attributes of a PRO instrument. For instance, feasibility
issues are often considered key in clinical practice, whereas reliability is more
important in clinical trials. To accommodate the needs of specific settings, a host of
new PRO measures have been developed over the last decade. Unfortunately, indi-
vidual instruments are frequently only used within a single disease population and/
or specific setting. For example, the HAQ-DI (Health Assessment Questionnaire
Disability Index) and WOMAC (Western Ontario and McMaster Universities
Arthritis Index) are both measures of disability primarily used in respectively
1 PROMs and Quality of Care