Reporting of missing data and methods used to accommodate them in recent analgesic clinical trials: ACTTION systematic review and recommendations

Abstract

Missing data in clinical trials can bias estimates of treatment effects. Statisticians and government agencies recommend making every effort to minimize missing data. Although statistical methods are available to accommodate missing data, their validity depends on often untestable assumptions about why the data are missing. The objective of this study was to assess the frequency with which randomized clinical trials published in 3 major pain journals (ie, European Journal of Pain, Journal of Pain, and Pain) reported strategies to prevent missing data, the number of participants who completed the study (ie, completers), and statistical methods to accommodate missing data. A total of 161 randomized clinical trials investigating treatments for pain, published between 2006 and 2012, were included. Approximately two-thirds of the trials reported at least 1 method that could potentially minimize missing data, the most common being allowance of concomitant medications. Only 61% of the articles explicitly reported the number of patients who were randomized and completed the trial. Although only 14 articles reported that all randomized participants completed the study, fewer than 50% of the articles reported a statistical method to accommodate missing data. Last observation carried forward imputation was used most commonly (42%). Thirteen articles reported more than 1 method to accommodate missing data; however, the majority of methods, including last observation carried forward, were not methods currently recommended by statisticians. Authors, reviewers, and editors should prioritize proper reporting of missing data and appropriate use of methods to accommodate them so as to improve the deficiencies identified in this systematic review.

Percutaneous image-guided cryoablation for the treatment of phantom limb pain in amputees: a pilot study

“…Multilevel linear growth modeling was uniquely suitable for the present study, as it permits (a) the responses to treatment to vary heterogeneously both over time and between individual patients, (b) the number of assessments and time between assessments to be distinct for each individual, and (c) accurate estimation in the presence of missing data (13)(14)(15). Changes in efficacy endpoints in response to cryoablation were analyzed using a 2-level model (level 1: visit; level 2: patient) with the variable for visit (baseline/day 0, day 7, day 45, long-term follow-up) included both as a fixed effect and as a random effect at the patient level.…”

Reporting of adverse events and statistical details of efficacy estimates in randomized clinical trials of pain in temporomandibular disorders

“…The most frequent method reported was a complete case analysis (that is, including only participants who provided all necessary outcome data), which assumes that the data are missing completely at random (that is, missingness is unrelated to observed or unobserved outcomes), an assumption that is unlikely to be accurate because participant withdrawals in pain clinical trials are often due to AEs or lack of effectiveness. 34 Other than 1 article in which the authors used a 2-way mixed analysis of variance model to accommodate missing data, no authors of published articles reported methods currently recommended by statisticians, such as multiple imputation, linear mixed models (for example, mixed-effect model repeated measure), or weighted generalized estimating equations. [9][10][11] In addition, none of the authors in any of the articles reported multiple methods to accommodate missing data to evaluate the sensitivity of the results to the assumptions made about the pattern of missing data (that is, sensitivity analysis).…”

“…We adapted a coding manual (Appendix 2; available online at the end of this article) from previous methods of Gewandter and colleagues 33,34 and Smith and colleagues 39 to evaluate reporting of details that can affect accuracy of treatment effect estimates (for example, identification of primary analyses, methods to adjust for multiple primary analyses, and methods to accommodate missing data), as well as fulfillment of the 10 CONSORT harms reporting recommendations. 25 We coded the method to accommodate missing data as unsure if the reported statistical test accommodates missing data only when certain statistical packages are used in specific ways, but insufficient information was provided about what was implemented.…”

“…[33][34][35][36][37][38][39][40][41] The goal of this systematic review was to evaluate reporting practices in clinical trials of treatments for TMD from 1969 to 2013 and highlight practices that are in need of improvement.…”

“Unsettling circularity”: Clinical trial enrichment and the evidentiary politics of chronic pain

“…IMMPACT was conceptualized as a series of meetings involving industry, academia and the FDA to "develop consensus reviews and recommendations for improving the design, execution and interpretation of clinical trials for pain" (IMMPACT, 2002). IMMPACT has published recommendations on patient-reported outcomes (Turk et al, 2006), assay sensitivity (Dworkin et al, 2012), abuse-deterrence (Dworkin et al, 2012), multiple endpoints and missing data analysis (Turk et al, 2008a;Gewandter et al, 2014b), disclosure of authorship contributions (Hunsinger et al, 2014) and interpretive "spin" (Gewandter et al, 2015). Together, these articles argue for the necessity of widely agreedupon standard forms of recording and reporting data from clinical trials, as well as on alternative design strategies such as enrichment.…”