REVIEW  PROBLEMS WITH STATISTICS IN REHABILITATION FROM THE RCTRACK PROJECT Free access

European Journal of Physical and Rehabilitation Medicine 2020 December;56(6):799-816

DOI: 10.23736/S1973-9087.20.06428-X

Copyright © 2020 EDIZIONI MINERVA MEDICA

lingua: Inglese

Influence of attrition, missing data, compliance, and related biases and analyses strategies on treatment effects in randomized controlled trials in rehabilitation: a methodological review

Susan ARMIJO-OLIVO ^{1, 2, 3} ✉, Wendy MACHALICEK ⁴, Liz DENNETT ⁵, Nikolaus BALLENBERGER ¹

¹ Faculty of Business and Social Sciences, University of Applied Sciences, Osnabrück, Germany; ² Department of Physical Therapy, Faculty of Rehabilitation Medicine, Rehabilitation Research Center, Edmonton, AB, Canada; ³ Cochrane Rehabilitation, IRCCS Fondazione Don Carlo Gnocchi, Milan, Italy; ⁴ Faculty of Special Education and Clinical Sciences, College of Education, University of Oregon, Eugene, OR, USA; ⁵ Scott Health Sciences Library, University of Alberta, Edmonton, AB, Canada

INTRODUCTION: Attrition, missing data, compliance, and related biases are three interrelated concepts. Previous research has found that these biases can affect the treatment estimates of randomized trials (RCTs). The extent to which the effects of attrition, missing data, compliance and related biases influence effect size estimates in rehabilitation as well as the effect of analytic strategies to mitigate these biases is unknown.
EVIDENCE ACQUISITION: To compile and synthetize the empirical evidence regarding the effects of attrition and compliance related biases on treatment effect estimates in rehabilitation RCTs. Electronic searches were conducted. Studies were included if they investigated the effects of attrition, missing data, compliance and related biases on treatment estimates. The seven studies meeting inclusion criteria were coded for type of biases and summarized using a narrative and/or quantitative approach when appropriate.
EVIDENCE SYNTHESIS: Findings demonstrated that trials reporting higher levels of attrition (differences in ES: 0.18 [95%CI: 0.15, 0.22 ]), exclusion of participants from analyses (differences in ES: 0.13 [95% CI: -0.03, 0.29]), lack of good control of incomplete outcome data (differences in ES: 0.14 [95%CI: -0.02, 0.30]) and analysis by “as treated”(differences in ES:-0.39 [95%CI: -0.99, 0.2]) or “per protocol” (differences in ES:-0.46 [95%CI: -0.92, 0]) analyses were more likely to have higher effects than those that did not.
CONCLUSIONS: These findings suggest that attrition, missing data, compliance, and related biases have an influence in treatment effect estimates in rehabilitation trials. Therefore, these results should be taken into consideration when designing, conducting and reporting trials in the rehabilitation field.

KEY WORDS: Compliance; Bias; Rehabilitation; Randomized controlled trials; Systematic review