An Improved Method for Simple, Assumption-Free Ordinal Analysis of the Modified Rankin Scale Using Generalized Odds Ratios. Issue 8 (December 2014)

• Record Type: Journal Article
• Authors: Churilov, Leonid; Arnup, Sarah; Johns, Hayden; Leung, Tiffany; Roberts, Stuart; Campbell, Bruce C. V.; Davis, Stephen M.; Donnan, Geoffrey A.
• Is Part Of: International journal of stroke. Volume 9:Issue 8(2014:Dec.)
• Page Start: 999
• Page End: 1005
• Publication Date: 2014-12
• Subjects: clinical trial -- epidemiology -- outcome and process assessment -- randomized controlled trials -- statistical analysis -- stroke; 616.8005
• DOI: 10.1111/ijs.12364
• Languages: English
• ISSNs: 1747-4930

Abstract

Background and Purpose: : Developing improved methods for analysis of the modified Rankin Scale (mRS) remains a critical issue for the stroke research community. A recently proposed permutation-based approach is assumption-free and easily interpretable but computationally intensive and does not provide confidence intervals to quantify the precision of the effect size estimate. We aimed to develop a method to overcome these limitations. Methods: : We propose a procedure using generalized odds ratios to estimate the odds that a patient who received the investigational treatment will have a better outcome than a patient receiving standard treatment. This approach was validated against the permutation method using hypothetical clinical trial scenarios of neuroprotective effect, early recanalization effect, late recanalization effect, and random benefit. Results: : The generalized odds ratio approach had strong agreement with the permutation approach provided sample size was >15 patients per treatment arm. Simulation established that the confidence intervals generated were accurate. Ignoring patient pairs with tied mRS scores overestimates the treatment effect compared with splitting tied mRS scores. Conclusions: : In addition to all the advantages of the recently proposed permutation-based approach, our method generates confidence intervals without the need for intensive computational power. The resulting generalized odds ratios are particularly suitable for inclusion in meta-analyses and have a simple and intuitive connection with the number-needed-to-treat measure.