For decades, oncology trials have been anchored to a familiar set of endpoints. Overall survival (OS) and progression-free survival (PFS) remain the benchmarks against which new treatments are judged. These measures are indispensable. They provide clarity, comparability and regulatory consistency. Yet, they tell only part of the story.
Patients undergoing cancer treatment experience outcomes that extend far beyond tumour shrinkage or time-to-progression. Fatigue, neuropathy, gastrointestinal toxicity and long-term impairments in daily functioning are often as meaningful as months of added survival. Nevertheless, such dimensions are typically relegated to secondary or exploratory endpoints, rarely determining whether a therapy is deemed successful. The result is a gap between what trials measure and what patients actually live through.
There is a growing consensus that oncology trials require a more inclusive framework that integrates efficacy and safety outcomes into a single evaluation. Generalised pairwise comparisons (GPC) offers a rigorous and interpretable way to achieve this shift.
From isolated endpoints to integrated outcomes
Traditional statistical analyses assess outcomes one at a time. Survival curves are compared using hazard ratios, adverse events are summarised separately and quality-of-life scores are often reported descriptively. Each endpoint provides information, but none captures the totality of patient experience. This uni-dimensional approach risks underestimating clinically relevant improvements or, conversely, overlooking harms that offset gains in efficacy.
GPC addresses this limitation by allowing multiple outcomes, of any type, to be evaluated jointly. First described more than a decade ago, the method generalises the Wilcoxon-Mann-Whitney test to accommodate a hierarchy of prioritised outcomes. Rather than treating all endpoints equally or relying on a single variable, GPC can mirror the way clinicians and patients weigh decisions in practice.
The analysis proceeds by forming all possible pairs between patients in the experimental and control groups of a randomised clinical trial. Each pair is compared according to the ranked list of outcomes. If one patient does better on the most important outcome (eg, survival), the comparison is recorded accordingly. If there is no clear difference, the analysis moves down to the next outcome in the hierarchy, and so forth.
The result of these comparisons is summarised in a single, interpretable metric known as the net treatment benefit (NTB). The NTB is defined as the net probability that a randomly selected patient from the treatment group fares better than a randomly selected patient from the control group, considering the prioritised outcomes. An NTB of 15%, for example, means that any random patient in the experimental arm has a 15% higher probability of experiencing a better overall outcome than a patient in the control arm. Importantly, because NTB is expressed as a probability difference, it can be translated into the number needed to treat (NNT), a familiar measure for clinicians.
By integrating efficacy, safety and patient-reported outcomes into one quantitative framework, GPC provides a more holistic and clinically relevant assessment of treatment benefit.
Existing use of generalised pairwise comparisons
While still gaining traction in oncology, GPC has already had a meaningful impact in other therapeutic areas. In cardiovascular disease, it played a pivotal role in the approval of tafamidis and, more recently, acoramidis for transthyretin amyloid cardiomyopathy. Both trials pre-specified NTB-based analyses as their primary endpoints, prioritising survival and hospitalisation events. This approach allowed regulators and clinicians to interpret the therapies' benefits in a way that reflected real-world patient priorities.
In oncology, GPC has been applied to both prospective and retrospective studies. The ongoing SHAPERS trial for elderly patients with rectal cancer exemplifies its utility. Rather than relying on a standard non-inferiority design centered on a single endpoint, SHAPERS uses GPC to integrate survival, tumour progression, neuropathy and high-grade toxicity. This design allows the trial to test for superiority, while simultaneously reflecting trade-offs that matter to patients who must balance treatment intensity with tolerability.
Retrospective re-analyses have also demonstrated the method's value. In a JAMA Oncology study, Péron and colleagues applied GPC to existing oncology trials and showed that NTB provided clearer, more intuitive insights than hazard ratios, particularly in settings where survival curves crossed or delayed treatment effects emerged. By reframing results as probabilities of net benefit, the analysis offered a more patient-centered interpretation of data that was already available.
Regulatory context and patient-focused guidance
The growing attention to GPC reflects a broader regulatory movement toward patient-focused drug development. The US Food and Drug Administration (FDA) has issued a series of guidance documents emphasising the importance of incorporating patient perspectives in trial design and benefit-risk assessments. The FDA's Project Patient Voice and Project Optimus further underscore the need for methodologies that evaluate both efficacy and tolerability in an integrated way.
The European Medicines Agency (EMA) has similarly advanced frameworks encouraging the involvement of patients in defining relevant outcomes and assessing benefit-risk trade-offs. The agency's engagement framework highlights the importance of systematically capturing and reflecting patient priorities in the evidence base.
These initiatives are not merely aspirational. They signal a regulatory environment that is increasingly open to robust methodologies capable of integrating multidimensional outcomes. GPC, with its validated statistical foundation and successful use in multiple therapeutic areas, aligns well with these expectations.
Challenges and practical considerations
Despite its promise, adopting GPC in oncology is not without challenges. Defining the hierarchy of outcomes requires careful deliberation and structured input from patients, investigators and clinicians. This prioritisation must be determined before trial initiation and justified in the protocol to ensure interpretability and regulatory acceptance.
A cultural barrier also remains. Clinicians and sponsors are accustomed to traditional endpoints such as OS, PFS and objective response rates. Expanding to multidimensional measures demands a shift in mindset, as well as education on how to interpret NTB and its clinical relevance.
Finally, operational considerations must be addressed. Collecting patient-reported outcomes consistently across trial sites and ensuring data quality requires planning and infrastructure. Yet these are surmountable challenges, particularly considering the potential for more meaningful, efficient and patient-aligned evidence generation.
Conclusion
Oncology research is at a turning point. Survival and tumor-based endpoints remain critical, but they are no longer sufficient to capture the full spectrum of treatment impact. Patients and regulators increasingly expect evidence that reflects not only how long people live, but how well they live during and after treatment.
Generalised pairwise comparisons provide a rigorous, interpretable and regulatorily acceptable way to meet this expectation whilst also allowing for better patient engagement when designing clinical trials. By integrating multiple prioritised outcomes into a single measure of net treatment benefit, GPC aligns oncology trial design with the principles of patient-focused drug development endorsed by both the FDA and the EMA. Its track record in cardiovascular diseases, and its emerging applications in oncology, demonstrate that this methodology is not theoretical but practical, actionable and compatible with modern regulatory frameworks.
As cancer therapies grow more complex, so too must our approaches to evaluating them. GPC offers a way forward: scientifically robust, clinically relevant and firmly grounded in the priorities of the patients trials are meant to serve.
References
1. Buyse M (2010), 'Generalized pairwise comparisons of prioritized outcomes in the two-sample problem', Stat Med 29(30): pp3245-3257.
2. Buyse M et al (2025), Handbook of Generalized Pairwise Comparisons: Methods for Patient-Centric Analysis, Chapman & Hall/CRC.
3. Maurer MS et al (2018), 'Tafamidis treatment for patients with transthyretin amyloid cardiomyopathy', N Engl J Med 379(11): pp1007-1016.
4. Gillmore JD et al (2024), 'Efficacy and safety of acoramidis in transthyretin amyloid cardiomyopathy', N Engl J Med 390(2): pp132-142.
5. Saúde-Conde R et al (2024), 'Efficacy and safety of short-course radiotherapy versus total neoadjuvant therapy in older rectal cancer patients: a randomised pragmatic trial (SHAPERS)', ESMO Gastrointest Oncol 4: p100067.
6. Péron J et al (2020), 'Assessment of net benefit of new oncology treatments', JAMA Oncol 6(9): pp1345-1352.
7. U.S. Food & Drug Administration (2020), 'Patient-focused drug development: Collecting comprehensive and representative input', FDA Guidance, June 2020.
8. European Medicines Agency (2022), 'Engagement framework: EMA and patients, consumers and their organisations', EMA, January 2022.