How Much Is Too Much: Impact of Missing Data Rates in Patient Reported Outcomes Research

• Activity Number: 303 - Patient-Centered Outcomes, Patient Preferences, and Patient Reports
• Type: Contributed
• Date/Time: Tuesday, July 31, 2018 : 8:30 AM to 10:20 AM
• Sponsor: Health Policy Statistics Section
• Author(s): Katie L Kunze* and Paul J. Novotny and Jeff A. Sloan and Blake T. Langlais and Amylou C. Dueck
• Companies: Mayo Clinic and Mayo Clinic and Mayo Clinic and Mayo Clinic and Mayo Clinic
• Keywords: missing data; patient reported outcomes; quality of life; mixed models; simulation

Abstract

Quality of life (QOL) research examines patients' symptoms and well-being via patient-recorded responses. Often used in clinical trials, patient reported outcomes (PRO) data may be missing due to the nature of the treatment (e.g., symptoms lead to missing or incomplete responses). Common practice recommends allowable PRO missingness rates ranging from 0% to 25%. Using the Wisconsin Ginseng Study (N07C2; Barton et al. 2013) as a motivating scenario and previous research by Bell et al. (2013) and Bell, King, and Fairclough (2014), we conducted a series of simulations to identify the rates of missingness at which mixed models and all-observations-available analysis (e.g., paired t-tests) results are impacted. Data for two groups (treatment and control) were generated to follow two response patterns (linear decline and plateau) based on observed values in the Wisconsin Ginseng Study with three time points of data collection (e.g., baseline, 4 weeks, and 8 weeks). Ten to 50% missingness was imposed using missing completely at random, missing at random, and missing not at random mechanisms. We will present results of our findings along with a discussion of implications for QOL research.