Data Pooling and Analysis to Build a Preliminary Item Bank
An Example Using Bowel Function in Prostate Cancer
Published in: Evaluation and the Health Professions, v. 28, no. 2, June 2005, p. 142-159
Posted on RAND.org on January 01, 2005
Assessing bowel function (BF) in prostate cancer can help determine therapeutic trade-offs. The authors determined the components of BF commonly assessed in prostate cancer studies as an initial step in creating an item bank for clinical and research application. They analyzed six archived data sets representing 4,246 men with prostate cancer. Thirty-one items from validated instruments were available for analysis. Items were classified into domains (diarrhea, rectal urgency, pain, bleeding, bother/distress, and other) then subjected to conventional psychometric and item response theory (IRT) analyses. Items fit the IRT model if the ratio between observed and expected item variance was between 0.60 and 1.40. Four of 31 items had inadequate fit in at least one analysis. Poorly fitting items included bleeding (2), rectal urgency (1), and bother/distress (1). A fifth item assessing hemorrhoids was poorly correlated with other items. Our analyses supported four related components of BF: diarrhea, rectal urgency, pain, and bother/distress.