Objectives:

Bayesian statistical methods are increasingly popular as a tool for meta-analysis of clinical trial data involving both direct and indirect treatment comparisons. However, appropriate selection of prior distributions for unknown model parameters and checking of consistency assumptions required for feasible modeling remain particularly challenging. We compared Bayesian and traditional frequentist statistical methods for multiple treatment comparisons in the context of pharmacological treatments for female urinary incontinence (UI).

Data sources:

We searched major electronic bibliographic databases, U.S. Food and Drug Administration reviews, trial registries, and research grant databases up to November 2011 to find randomized studies published in English that examined drugs for urgency UI on continence, improvements in UI, and treatment discontinuation due to harms.

Review methods:

We fitted fixed and random effects models in frequentist and Bayesian frameworks. In a hierarchical model of eight treatments, we separately analyzed one safety and two efficacy outcomes. We produced Bayesian and frequentist treatment ranks and odds ratios (and associated measures of uncertainty) across all bivariate treatment comparisons. We also calculated the number needed to treat (NNT) to achieve continence or avoid harms from pooled absolute risk differences.

Results:

While frequentist and Bayesian analyses produced broadly comparable odds ratios of safety and efficacy, the Bayesian method's ability to deliver the probability that any treatment is best, or among the top two such treatments, offered a more meaningful clinical interpretation. In our study, two drugs emerged as attractive because while neither had any significant chance of being among the least safe drugs, both had greater than 50 percent chances of being among the top three drugs in terms of Best12 probability for one of the efficacy endpoints.

Conclusions:

Bayesian methods are more flexible and their results more clinically interpretable but require more careful development and specialized software.

Key Messages

• Both Bayesian and frequentist hierarchical models can be effective in multiple treatment comparisons.
• Bayesian models sensibly shrink estimates towards each other, encouraging more borrowing of statistical strength from the entire collection of studies. Bayesian methods also lead to more clinically interpretable results (through their ability to assign probabilities to events), as well as more sensible rankings of the pharmacological treatments as compared to traditional NNT-based methods.
• Further development of hierarchical Bayesian multiple treatment comparison methods is warranted, especially for nonbinary data models, simultaneous decisionmaking across multiple endpoints, assessing consistency, and incorporating data sources of varying quality (e.g., clinical vs. observational data).