Problems in the development and validation of questionnaire-based screening instruments for ascertaining cases with...

To determine if screening for symptomatic knee osteoarthritis (OA) for clinical trials and epidemiologic studies could be satisfactorily done without performing knee radiographs and to develop efficient screening instruments for symptomatic knee OA based on self-reported symptoms and functional limitations.

We administered a mailed questionnaire containing many different questions on knee symptoms and functional limitations to 1,921 participants of the Framingham Study who had previously been screened for symptomatic OA with a history and knee radiographs. Recursive partitioning methods (using the Classification and Regression Trees [CART] program) were used to create a set of screening instruments for symptomatic knee OA, which was defined as knee symptoms on most days and radiographic evidence of OA. Three screening instruments were developed to maximize the sensitivity, specificity, and efficiency.

The sensitive instrument had 84% sensitivity and 73% specificity. The specific instrument had 46% sensitivity and 94% specificity. The efficient instrument had 56% sensitivity and 85% specificity. Sensitivity was lower and specificity was higher when these instruments were used to screen for radiographic OA. All instruments had higher sensitivity but lower specificity when used for older subjects (age >60) with greater disease prevalence. However, using any of these instruments as a single-step screening mechanism resulted in considerable misclassification.

We conclude that none of these instruments has adequate diagnostic test performance to serve as a single-step evaluation of the presence or absence of symptomatic knee OA.