High-Dimensional Symptom Data in Cancer Trials: Exploring Item Redundancy in the Patient-Reported Outcomes Version of the Common Terminology Criteria for Adverse Events (PRO-CTCAE)

• Activity Number: 663
• Type: Contributed
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Sandra A. Mitchell*+ and Amylou C Dueck and Tito R Mendoza and Steven B Clauser and Bryce B Reeve and Thomas M Atkinson and Antonia V Bennett and Yuelin Li and Jeff A Sloan and Ethan Basch
• Companies: National Cancer Institute and Mayo Clinic and MD Anderson Cancer Center and National Cancer Institute and The University of North Carolina at Chapel Hill and Memorial Sloan-Kettering Cancer Center and Memorial Sloan-Kettering Cancer Center and Memorial Sloan-Kettering Cancer Center and Mayo Clinic and Memorial Sloan-Kettering Cancer Center
• Keywords: Patient-reported outcomes ; symptoms ; dimensional reduction ; psychometrics ; clinical trials ; cancer

Abstract

NCI PRO-CTCAE is a new patient-reported outcome measure designed to enhance adverse event (AE) reporting in clinical trials by integrating the patient experience. PRO-CTCAE includes items representing the frequency (F), severity (S) and interference (I) with usual activities of 78 symptomatic AEs. Data from our validation study for PRO-CTCAE (N=869 patients receiving treatment for a range of cancer types) have posed analytic challenges typical of high-dimensional symptom self-report data including collinearity among symptom dimensions of F, S and I; multi-dimensionality; and clusters of co-occuring symptoms. In this paper we briefly summarize these challenges in analyzing symptom data, and highlight the approaches employed (e.g. weighted Kappa and Bowker's test for symmetry) and the insights derived through efforts to explore item redundancies. This work sets the stage for future efforts to create composite scores, derive unidimensional subscales, and model patterns of symptom co-occurrence. Analytic challenges include the need to accommodate data features that are common in symptom research including large numbers of symptoms, numerous zero values, and mixture distributions.