Michael J Palmer1,2,3, Harriet Richardson4,5,6, Dongsheng Tu4,6, Michael Brundage4,5,6. 1. Department of Public Health Sciences, Queen's University, Kingston, ON, Canada. 81mjp6@queensu.ca. 2. Cancer Care and Epidemiology, Cancer Research Institute, Queen's University, Kingston, ON, Canada. 81mjp6@queensu.ca. 3. Division of Cancer Care and Epidemiology, Cancer Research Institute, Queen's University, 10 Stuart Street, Level 2, Kingston, ON, K7L 3N6, Canada. 81mjp6@queensu.ca. 4. Department of Public Health Sciences, Queen's University, Kingston, ON, Canada. 5. Cancer Care and Epidemiology, Cancer Research Institute, Queen's University, Kingston, ON, Canada. 6. Canadian Cancer Trials Group, Cancer Research Institute, Queen's University, Kingston, ON, Canada.
Abstract
PURPOSE: Missing patient-reported outcome (PRO) data can seriously threaten the validity of randomized clinical trials (RCTs). Identifying which factors predict missing instruments may help researchers develop strategies to prevent it from happening. This study examined the association of factors with time to the first missing instrument after randomization in three cooperative group RCTs. METHODS: We performed descriptive analyses and Cox proportional hazards regressions for three RCTs selected from the Canadian Cancer Trials Group: MA17 (breast cancer), PR7 (prostate cancer), and LY12 (non-Hodgkin's lymphoma). The outcome was the time from randomization to the first missing instrument. Variables for 15 factors were used as covariates based on availability and previously-reported putative associations with missing PRO data. RESULTS: Nine percent of 1352 subjects on MA17, 37% of 923 subjects on PR7, and 59% of 477 subjects on LY12 had a missing instrument. Twenty-five percent of subjects on MA17 had first missing instrument within 4.6 years. The median time to first missing instrument was: not observed for MA17, 7.3 years for PR7, 0.12 years for LY12. Cox regression revealed statistically significant independent associations with outcome for only five factors: baseline age (PR7) and level of well-being (LY12), and centre level of activity (LY12), presence of post-graduate residency training program (MA17, PR7), and centre geographic location (PR7, LY12). CONCLUSION: Many factors reported to have association with missing instruments do not seem to predict time to the first missing instrument after randomization in RCTs. Context is important in understanding the few that may.
PURPOSE: Missing patient-reported outcome (PRO) data can seriously threaten the validity of randomized clinical trials (RCTs). Identifying which factors predict missing instruments may help researchers develop strategies to prevent it from happening. This study examined the association of factors with time to the first missing instrument after randomization in three cooperative group RCTs. METHODS: We performed descriptive analyses and Cox proportional hazards regressions for three RCTs selected from the Canadian Cancer Trials Group: MA17 (breast cancer), PR7 (prostate cancer), and LY12 (non-Hodgkin's lymphoma). The outcome was the time from randomization to the first missing instrument. Variables for 15 factors were used as covariates based on availability and previously-reported putative associations with missing PRO data. RESULTS: Nine percent of 1352 subjects on MA17, 37% of 923 subjects on PR7, and 59% of 477 subjects on LY12 had a missing instrument. Twenty-five percent of subjects on MA17 had first missing instrument within 4.6 years. The median time to first missing instrument was: not observed for MA17, 7.3 years for PR7, 0.12 years for LY12. Cox regression revealed statistically significant independent associations with outcome for only five factors: baseline age (PR7) and level of well-being (LY12), and centre level of activity (LY12), presence of post-graduate residency training program (MA17, PR7), and centre geographic location (PR7, LY12). CONCLUSION: Many factors reported to have association with missing instruments do not seem to predict time to the first missing instrument after randomization in RCTs. Context is important in understanding the few that may.
Authors: Andrew Bottomley; Madeline Pe; Jeff Sloan; Ethan Basch; Franck Bonnetain; Melanie Calvert; Alicyn Campbell; Charles Cleeland; Kim Cocks; Laurence Collette; Amylou C Dueck; Nancy Devlin; Hans-Henning Flechtner; Carolyn Gotay; Eva Greimel; Ingolf Griebsch; Mogens Groenvold; Jean-Francois Hamel; Madeleine King; Paul G Kluetz; Michael Koller; Daniel C Malone; Francesca Martinelli; Sandra A Mitchell; Carol M Moinpour; Jammbe Z Musoro; Daniel O'Connor; Kathy Oliver; Elisabeth Piault-Louis; Martine Piccart; Francisco L Pimentel; Chantal Quinten; Jaap C Reijneveld; Christoph Schürmann; Ashley Wilder Smith; Katherine M Soltys; Rajeshwari Sridhara; Martin J B Taphoorn; Galina Velikova; Corneel Coens Journal: Clin Trials Date: 2018-08-24 Impact factor: 2.486