Comparison of statistical approaches for analyzing incomplete longitudinal patient-reported outcome data in randomized controlled trials
Received 2 August 2017
Accepted for publication 15 March 2018
Published 21 June 2018 Volume 2018:9 Pages 197—209
Checked for plagiarism Yes
Review by Single-blind
Peer reviewer comments 4
Editor who approved publication: Dr Liana Bruce (formerly Castel)
Ines Rombach,1,2 Crispin Jenkinson,3 Alastair M Gray,1 David W Murray,2 Oliver Rivero-Arias4
1Health Economics Research Centre, Nuffield Department of Population Health, University of Oxford, Oxford, UK; 2Nuffield Department of Orthopaedics, Rheumatology and Musculoskeletal Sciences, University of Oxford, Oxford, UK; 3Health Services Research Unit, Nuffield Department of Population Health, University of Oxford, Oxford, UK; 4National Perinatal Epidemiology Unit, Nuffield Department of Population Health, University of Oxford, Oxford, UK
Purpose: Missing data are a potential source of bias in the results of RCTs, but are often unavoidable in clinical research, particularly in patient-reported outcome measures (PROMs). Maximum likelihood (ML), multiple imputation (MI), and inverse probability weighting (IPW) can be used to handle incomplete longitudinal data. This paper compares their performance when analyzing PROMs, using a simulation study based on an RCT data set.
Methods: Realistic missing-at-random data were simulated based on patterns observed during the follow-up of the knee arthroscopy trial (ISRCTN45837371). Simulation scenarios covered different sample sizes, with missing PROM data in 10%–60% of participants. Monotone and nonmonotone missing data patterns were considered. Missing data were addressed by using ML, MI, and IPW and analyzed via multilevel mixed-effects linear regression models. Root mean square errors in the treatment effects were used as performance parameters across 1,000 simulations.
Results: Nonconvergence issues were observed for IPW at small sample sizes. The performance of all three approaches worsened with decreasing sample size and increasing proportions of missing data. MI and ML performed similarly when the MI model was restricted to baseline variables, but MI performed better when using postrandomization data in the imputation model and also in nonmonotone versus monotone missing data scenarios. IPW performed worse than ML and MI in all simulation scenarios.
Conclusion: When additional postrandomization information is available, MI can be beneficial over ML for handling incomplete longitudinal PROM data. IPW is not recommended for handling missing PROM data in the simulated scenarios.
Keywords: missing data, repeated measures, patient-reported outcome measures, PROMS, multilevel mixed-effects models, multiple imputation, inverse probability weighting
This work is published and licensed by Dove Medical Press Limited. The full terms of this license are available at https://www.dovepress.com/terms.php and incorporate the Creative Commons Attribution - Non Commercial (unported, v3.0) License. By accessing the work you hereby accept the Terms. Non-commercial uses of the work are permitted without any further permission from Dove Medical Press Limited, provided the work is properly attributed. For permission for commercial use of this work, please see paragraphs 4.2 and 5 of our Terms.Download Article [PDF] View Full Text [HTML][Machine readable]