Eight benefit-risk methods, one withdrawn drug: what rimonabant teaches regulators

Every new medicine walks a tightrope. On one side hangs the benefit: the weight lost, the pain eased, the tumor shrunk. On the other hangs the risk: the depression triggered, the heart attack provoked, the immune system turned against its owner. Deciding whether a drug’s benefits outweigh its harms is the fundamental judgment that underlies every approval, every label, and every withdrawal in modern pharmacology — and yet there is no universal agreement on how that judgment should be made.

A team of benefit-risk epidemiologists at Sanofi, led by Jian-Yu E, has now put eight competing assessment methodologies through their paces on a single, notoriously difficult case: rimonabant, the cannabinoid type 1 receptor antagonist once marketed as Acomplia and Zimulti for weight loss. Approved in Europe in 2006, the drug was pulled from the market in January 2009 after post-marketing data revealed serious psychological side effects, including suicide risk. Crucially, the researchers note, no formal quantitative benefit-risk analysis integrating benefits, risks, and their relative importance to physicians or patients had ever been conducted at the time of withdrawal. Their study, published in Frontiers in Drug Safety and Regulation, uses the drug not to relitigate the regulatory decision but as a well-documented testing ground for the tools of structured decision-making.

The work grew out of the Innovative Medicines Initiative’s PROTECT consortium, a public-private partnership coordinated by the European Medicines Agency whose benefit-risk group aimed to identify, characterize, and test methods for benefit-risk assessment. The team evaluated eight methodologies spanning four main categories: two descriptive frameworks, two quantitative frameworks, and four metrics and estimation techniques. All eight were applied to the same rimonabant case study, drawing on data from European Public Assessment Reports, six randomized controlled trials, and published literature, so that their strengths and limitations could be compared directly under identical conditions.

The two descriptive frameworks — PrOACT-URL, an eight-step decision guide covering Problem, Objective, Alternatives, Consequences, Trade-offs, Uncertainty, Risk tolerance, and Linked decisions, and the Benefit-Risk Action Team framework developed under the pharmaceutical industry’s PhRMA umbrella — proved transparent and systematic at structuring the assessment. Both broke the decision down into manageable criteria and summarized benefit-risk outcomes into simplified value trees or effects tables. Building the final value tree was an iterative process requiring stakeholder input on which benefits and risks carried the most substantial impact, while guarding against double-counting correlated endpoints. But the frameworks remained purely descriptive: they lacked formal quantitative analysis, required no stakeholder weighting of criteria, and were not decision analysis tools in themselves.

The quantitative frameworks — Multi-Criteria Decision Analysis (MCDA) and Stochastic Multi-criteria Acceptability Analysis (SMAA) — extended the descriptive models by integrating objective treatment-effect data with subjective preference data. Both built on the value trees and effects tables from the descriptive stage, but transformed them into numerical outputs. For MCDA, the team gathered preference weights through a direct elicitation questionnaire in which physicians, regulators, and statisticians scored each criterion from 0 (not important) to 10 (very important), with statisticians serving as layman perspectives. Using meta-analyzed data from the randomized trials with layman weighting, the total weighted score favored rimonabant over placebo. Yet the results proved sensitive to the type of weighting used: small adjustments in the weights assigned to benefits or risks could alter the preferred treatment, and the layman-weighted model, while relatively stable under moderate weight changes, differed from the regulatory-weighted model.

SMAA, regarded as an extension of MCDA, was better equipped to handle uncertainty in the data through simulation. Rather than demanding precise stakeholder weightings upfront, SMAA could assume a uniform weight distribution across criteria, or incorporate ranking preferences if available. Where MCDA produced a single overall utility score, SMAA generated an acceptability index for a particular ranking, transforming data into utility scores using linear criteria functions and building common scales from 95 percent confidence intervals. The final results were interpretable in both graphical and numerical form using free software, making the approach flexible and replicable even when statistical knowledge was limited.

The four metrics and estimation techniques — Number Needed to Treat and Number Needed to Harm, Impact Numbers, the Benefit-Risk Ratio, and the Probabilistic Simulation Method — offered straightforward numerical evaluations but proved narrower in scope. Impact Numbers, conceptually similar to NNT but framed from a public health perspective, could extrapolate rimonabant’s potential impact across the United Kingdom’s obese population, yet the researchers found these metrics not particularly useful for benefit-risk decision-making, since real-life conditions involve multiple interventions and outcomes that single metrics struggle to integrate. The Benefit-Risk Ratio, useful for directly comparing one benefit to one risk with optional weighting, could not incorporate uncertainties, suffered scaling issues, and required a separate calculation for each benefit-risk criterion. The Probabilistic Simulation Method addressed the uncertainty limitation by repeatedly sampling each parameter of interest to capture natural variation and calculate 95 percent confidence intervals, but even with uncertainty handled, these techniques remained best suited to specific public health questions such as resource allocation rather than complex regulatory assessments.

The findings align with a growing consensus in the literature that no single benefit-risk methodology suits all decision settings. Tervonen and colleagues reported expert consensus on good practices for quantitative benefit-risk assessment, emphasizing the research question, model development, preference elicitation, data analysis, and communication of results. Sullivan and colleagues introduced an industry operating model classifying approaches as descriptive, semi-quantitative, or quantitative, highlighting the value of a systematic framework across drug development stages from first-time-in-human studies to regulatory submission. More recently, a scoping review by Suzumura and colleagues identified 129 approaches for conducting benefit-risk assessment and 37 visual tools for reporting results — a heterogeneity that limits consistency in real-world application. Against this backdrop, the present study adds value by applying eight methodologies to a single case study, bridging the gap between methodological development and practical implementation.

The researchers acknowledge limitations. The choice of rimonabant, a drug withdrawn due to psychiatric adverse events such as depression and anxiety, reflected the need for a well-documented example with a complex benefit-risk profile, but may not capture the diversity of decision-making for other anti-obesity medications. Preference weights were collected directly from team member surveys rather than through formal elicitation methods, and the study was limited to eight main methodologies. The authors emphasize that the study does not advocate any specific methodology, nor does it replace or comment on any regulatory decisions made by national agencies or the European Medicines Agency. Their conclusion is pragmatic: no single approach fully captures all aspects of a benefit-risk assessment, but a combination of a descriptive framework supplemented by quantitative models accommodates multiple criteria, reflects varying stakeholder preferences, and addresses uncertainty — a context-driven pathway for one of the most consequential judgments in medicine.

Subject of Research: Benefit-risk assessment methodologies compared using the rimonabant case study

Article Title: Methodological comparisons for benefit-risk assessment in complex decision-making: applications and insights from a case study

Article References: E, J.-Y., Zhu, K., Tcherny-Lessenot, S., & Juhaeri, J. (2026). Methodological comparisons for benefit-risk assessment in complex decision-making: applications and insights from a case study. Frontiers in Drug Safety and Regulation, 6, Article 1792441. https://doi.org/10.3389/fdsfr.2026.1792441

Image Credits: AI Generated

DOI: 10.3389/fdsfr.2026.1792441

Keywords: Methodological, comparisons, benefit-risk, assessment, complex, decision-making, applications, insights, case, scientific research