Introduction

Each year new health-related quality of life (HRQOL) questionnaires are developed or revised from previous measures in the hope of obtaining instruments that are more reliable, valid within the study population, and sensitive to a patient's change in health status. Also, it is important that the measures provide interpretable scores that accurately characterize a patient's HRQOL. While several quality instruments have emerged in cancer outcomes research, we presently lack the ability to crosswalk scores from one instrument to another so that researchers are able to combine or compare results from multiple studies when different instruments are used. Developing these psychometrically strong measures requires analytical methods that will allow researchers to choose the best set of informative questions to match study objectives and to crosswalk scores from one assessment to another, despite use of different sets of questions.

There has been growing interest in learning how applications of item response theory (IRT) modeling can be used to respond to these analytical needs of the cancer outcomes measurement field. This interest is generated by the ability of IRT models to analyze item and scale performance within a study population; to detect biased items that may occur when translating an instrument from one language to another, or when respondents from two different groups hold culturally different meanings for the item content; to link two or more instruments on a common metric and thus facilitate crosswalking of scores; and to create item banks that serve as a foundation for computerized adaptive assessment.