Methods for Imputing Missing Efficacy Data in Clinical Trials of Biologic Psoriasis Therapies: Implications for Interpretations of Trial Results

August 2017 | Volume 16 | Issue 8 | Original Article | 734 | Copyright © 2017

Richard G.B. Langley MD FRCPC,a Kristian Reich MD PhD,b Charis Papavassilis MD PhD,c Todd Fox PharmD ACPR,c Yankun Gong PhD,d and Achim Güttner PhDc

aDalhousie University, Halifax, Canada bDermatologikum Hamburg and Georg-August-University Göttingen, Germany cNovartis Pharma AG, Basel, Switzerland dBeijing Novartis Pharma Co. Ltd., Shanghai, China

Abstract

BACKGROUND: An issue in long-term clinical trials of biologics in psoriasis is how to handle missing efficacy data. This methodological challenge may not be understood by clinicians, yet can have a significant effect on the interpretation of clinical trials.

OBJECTIVE Evaluate the effects of different data imputation methods on apparent secukinumab response rates.

METHODS: Post hoc analyses were conducted on efficacy data from 2 phase III, multicenter, randomized, double-blind trials (FIXTURE and ERASURE) of secukinumab in moderate to severe plaque psoriasis. Per study protocols, missing data were imputed using strict non-response imputation (NRI), a highly conservative method that assumes non-response for all missing data. Alternative imputation methods (observed data, last observation carried forward [LOCF], modified NRI, and multiple imputation [MI]) were applied in this analysis and the resultant response rates compared.

RESULTS: Response rates obtained with each imputation method diverged increasingly over 52-weeks of follow-up. Strict NRI response estimates were consistently lower than those using the other methods. At week 52, Psoriasis Area and Severity Index (PASI) 90 rates for secukinumab 300 mg based on strict NRI were 9.2% (FIXTURE) and 8.7% (ERASURE) lower than estimates obtained using the least conservative method (observed data). Estimates obtained through LOCF and modified NRI were closest to those produced by MI, currently regarded as the most methodologically sophisticated approach available.

CONCLUSION: Awareness of differences in assumptions and limitations among imputation methods is necessary for well-informed interpretation of trial data.

J Drugs Dermatol. 2017;16(8):734-742.

Purchase Original Article

Purchase a single fully formatted PDF of the original manuscript as it was published in the JDD.

Download the original manuscript as it was published in the JDD.

Contact a member of the JDD Sales Team to request a quote or purchase bulk reprints, e-prints or international translation requests.

To get access to JDD's full-text articles and archives, upgrade here.

Save an unformatted copy of this article for on-screen viewing.

Print the full-text of article as it appears on the JDD site.

→ proceed | ↑ close

INTRODUCTION

Thorough evaluation of biologic treatments for chronic diseases such as psoriasis requires data from long-term clinical trials. In lengthy trials, a non-negligible portion of data may be missing at the study end, as a substantial portion of patients may discontinue, or data may be not available for other reasons. The reasons for missing data are varied and may be highly relevant to the investigation, such as discontinuation due to lack of efficacy. Regardless of the cause, missing data always represents a loss of information, reducing the power and precision of estimates, and complicating analyses. For example, the proportion and pattern of missing data may differ between treatment groups (eg, differing discontinuation rates), affecting the validity of conclusions; or, differences between observed and missing data may produce a study population that no longer represents the intended target population. Although the best approach to missing data is to prevent it, the acquisition of a complete data set is unlikely for clinical trials. The method by which missing data is handled, whether replaced with substituted values (imputation) or ignored, affects the interpretation of the trial results and may itself introduce bias. Effects of bias may include an inaccurate estimation of treatment effect and statistical significance, and an undermining of the comparability between treatment groups.1 The intention-to-treat (ITT) principle mandates that a study maintain the balance between treatment groups obtained through randomization and that all randomized patients be included in analyses regardless of discontinuation, completion, or adherence to the protocol. The European Medicines Agency (EMA) and US Food and Drug Administration (FDA) support the ITT principle, the pre-specification of the methods for handling missing data in the protocol or statistical analysis plan, and a conservative approach for imputation of missing data using analyses that exclude bias in favor of the therapy under study.2–4 Performing several imputation methods has been suggested for analysis of clinical trial endpoints in order to assess the sensitivity of the outcome.5 In long-term clinical trials with biologics, the following strategies for managing missing data have been used alone or

↑ back to top


Related Articles