Endpoint Agnosticism in Late-Phase Trials: Shifting from Fixed Hierarchies to Patient-Driven Value Functions

Late-phase trials often lock endpoint hierarchies years before data collection begins. A primary endpoint is chosen, secondary endpoints are ranked, and the entire statistical analysis plan is built around that fixed order. Yet by the time results are read out, patient priorities may have shifted, new evidence may have emerged, and the rigid hierarchy can obscure what actually matters to the people who will use the therapy. Endpoint agnosticism offers a way out—not by abandoning structure, but by making the structure responsive to patient-driven value functions. In this guide, we walk through the rationale, the mechanics, and the practical steps to shift from fixed hierarchies to value-driven endpoint selection.

Why Fixed Hierarchies Fall Short in Patient-Centric Trials

Traditional endpoint hierarchies were designed for regulatory clarity. Regulators needed a single primary endpoint to power a study, and secondary endpoints were ranked to control multiplicity. This approach works well when the disease is well understood and the patient experience is homogeneous. But in many chronic, rare, or heterogeneous conditions, a single endpoint cannot capture the multidimensional benefit that patients value.

For example, in a trial for a neurodegenerative disease, a fixed hierarchy might prioritize a motor function scale as primary, with cognition and quality of life as secondary. Yet patients often rank quality of life or cognition as equally or more important. When the primary endpoint fails to show significance, the hierarchy prevents the secondary endpoints from being interpreted as confirmatory, even if they show strong patient-relevant improvements. This leads to missed opportunities for demonstrating meaningful benefit.

The Gap Between Statistical Significance and Patient Value

A statistically significant result on a primary endpoint may not translate to a clinically meaningful change for patients. Conversely, a non-significant primary endpoint may mask real improvements on patient-important outcomes. Fixed hierarchies assume that the primary endpoint is the most important, but that assumption is often made by regulators or sponsors, not by patients. Endpoint agnosticism acknowledges that value is context-dependent and that the relative importance of endpoints can vary across patient subgroups, disease stages, and treatment goals.

Regulatory Evolution and the Push for Flexibility

Regulatory agencies have begun to recognize the limitations of rigid hierarchies. Guidance documents from the FDA and EMA increasingly encourage the use of patient-reported outcomes and composite endpoints that reflect patient priorities. The concept of endpoint agnosticism aligns with this trend by formalizing a flexible framework where endpoints are not pre-ranked but are instead evaluated through a value function that weights them according to patient preferences. This shift is not about abandoning rigor but about redefining what rigor means in a patient-centric context.

One composite scenario: a team developing a therapy for a rare metabolic disorder initially planned a fixed hierarchy with a biomarker as primary and physical function as secondary. After conducting patient preference surveys, they found that patients valued reduced fatigue and improved social participation more than biomarker changes. By adopting an endpoint agnostic approach, they reweighted the endpoints in their analysis, allowing the patient-relevant outcomes to drive the interpretation of efficacy. The result was a more compelling dossier that addressed both regulatory and patient needs.

Core Frameworks: Value Functions vs. Fixed Hierarchies

At the heart of endpoint agnosticism is the concept of a value function—a mathematical representation of how different endpoints contribute to an overall assessment of treatment benefit. Unlike a fixed hierarchy, where endpoints are ranked ordinally, a value function assigns continuous weights to each endpoint, reflecting their relative importance as determined by patient input.

How Value Functions Work

A value function takes the form V = w1 * E1 + w2 * E2 + ... + wn * En, where w are weights and E are endpoint scores (often standardized). The weights can be derived from patient preference studies, such as discrete choice experiments or best-worst scaling. The function can be linear or non-linear, depending on the nature of the endpoints. For example, if a patient group values a 10% improvement in pain relief three times more than a 10% improvement in physical function, the pain endpoint would receive a weight three times larger.

The key difference from a hierarchy is that endpoints are not treated as binary pass/fail gates. In a hierarchy, if the primary endpoint fails, the secondary endpoints are often considered exploratory. In a value function, all endpoints contribute to the overall value score, and the trial can be evaluated on the composite value rather than on individual endpoint success. This reduces the risk of missing a truly beneficial treatment because one endpoint did not meet its threshold.

Comparison of Three Approaches

When to Choose Each Approach

Fixed hierarchies remain appropriate when there is a well-established single endpoint that is universally accepted as the gold standard. Value functions are better suited when the disease affects multiple domains that patients value differently. Hybrid approaches can serve as a transitional strategy for sponsors who want to incorporate patient preferences without fully abandoning traditional frameworks. The choice should be guided by the specific trial context, the availability of patient preference data, and the regulatory environment.

Implementing Endpoint Agnosticism: A Step-by-Step Process

Shifting from a fixed hierarchy to a value function requires careful planning and execution. Below is a step-by-step process that teams can adapt to their specific trial.

Step 1: Identify Relevant Endpoints

Begin by listing all potential endpoints that capture the treatment's effects. Include clinical, biomarker, patient-reported, and functional outcomes. Engage patients and clinicians to ensure the list is comprehensive and meaningful. Avoid excluding endpoints prematurely based on past conventions.

Step 2: Elicit Patient Preferences

Conduct preference elicitation studies to determine the relative importance of each endpoint. Methods include discrete choice experiments, conjoint analysis, and best-worst scaling. Ensure the sample represents the target patient population, including subgroups by disease severity, demographics, and treatment experience. The output is a set of weights that reflect how patients trade off improvements across endpoints.

Step 3: Define the Value Function

Using the preference weights, construct a value function. Decide whether to use a linear additive model or a more complex non-linear function. Consider interactions between endpoints—for example, patients may value a combination of pain relief and physical function more than the sum of their individual improvements. Pre-specify the function in the statistical analysis plan.

Step 4: Power the Trial for the Value Score

Calculate the sample size needed to detect a meaningful change in the value score. This may require simulations to account for the correlation between endpoints and the distribution of weights. The value score should be the primary analysis, with individual endpoints analyzed as secondary to provide granularity.

Step 5: Plan for Multiplicity and Sensitivity Analyses

Because the value function aggregates multiple endpoints, multiplicity adjustment is simplified—there is one primary analysis. However, sensitivity analyses should assess the robustness of results to different weighting schemes, such as equal weights or weights from different subgroups. Also plan for analyses that examine each endpoint individually to aid interpretation.

Step 6: Communicate with Regulators Early

Engage regulatory agencies during the design phase. Present the rationale for the value function, the preference data, and the planned analyses. While regulatory acceptance varies, early dialogue can reduce uncertainty. Some agencies may require a hybrid approach as a bridge, with the value function as a supportive analysis.

Tools, Economics, and Maintenance Realities

Implementing endpoint agnosticism requires investment in new tools and processes. Below we discuss the practical considerations.

Software and Statistical Tools

Standard statistical software (R, SAS) can handle value functions with custom code. Specialized packages for conjoint analysis and preference elicitation are available (e.g., Sawtooth Software, R packages like 'support.CEs'). For simulation-based sample size calculations, tools like East or ADDPLAN may be used with custom macros. Teams should budget for statistical programming time and validation.

Economic Considerations

The main additional costs are for preference elicitation studies (survey design, recruitment, analysis) and increased statistical complexity. These costs can range from tens of thousands to a few hundred thousand dollars, depending on the scope. However, the investment may pay off by reducing the risk of failed trials—if a value function captures a real benefit that a fixed hierarchy would miss, the cost of a single successful trial can be saved many times over.

Maintenance and Updates

Patient preferences can change over time, especially as new treatments become available or disease understanding evolves. For long-term trials, consider re-assessing preferences at interim points or using adaptive weighting. However, changing weights mid-trial introduces complexity and must be pre-specified to avoid bias. A pragmatic approach is to use baseline preferences for the primary analysis and conduct sensitivity analyses with updated weights.

One composite scenario: a sponsor developing a device for chronic pain conducted preference surveys at the start of a three-year trial. Two years in, a new standard of care emerged, shifting patient priorities from pain intensity to functional independence. The team had pre-specified a sensitivity analysis using updated weights, which showed that the device provided greater value than the original analysis suggested. This allowed them to adjust their messaging and regulatory strategy.

Growth Mechanics: Scaling Endpoint Agnosticism Across Programs

Once a team has successfully implemented endpoint agnosticism in one trial, scaling the approach across a portfolio requires organizational change.

Building Internal Expertise

Create a cross-functional working group that includes biostatistics, clinical development, patient engagement, and regulatory affairs. This group can develop standard operating procedures for preference elicitation and value function design. Training sessions and case study reviews help disseminate knowledge.

Creating Reusable Assets

Develop a library of patient preference data for common indications. While preferences are disease-specific, certain patterns (e.g., patients valuing symptom relief over biomarkers) may generalize. Reusing validated survey instruments reduces costs and accelerates timelines.

Positioning for Regulatory Acceptance

As more sponsors adopt endpoint agnosticism, regulatory agencies will gain familiarity. Sponsors can contribute to this evolution by publishing their methods and results, even if the value function was used only as a supportive analysis. Collaborative initiatives with patient advocacy groups can also build evidence for patient-driven endpoints.

Measuring Impact

Track metrics such as the proportion of trials using value functions, the correlation between value scores and traditional endpoints, and the success rate of trials that use value functions versus fixed hierarchies. Over time, this data can demonstrate the value of the approach and justify further investment.

Risks, Pitfalls, and Mitigations

Endpoint agnosticism is not without risks. Below are common pitfalls and how to address them.

Overfitting to Preference Data

If preference studies are not robust, the value function may reflect noise rather than true patient values. Mitigation: use rigorous study designs with adequate sample sizes, test-retest reliability, and validation in independent samples. Pre-specify the weighting scheme and avoid post-hoc adjustments.

Stakeholder Misalignment

Regulators, payers, and clinicians may be skeptical of a composite value score they do not understand. Mitigation: include individual endpoint analyses alongside the value score. Educate stakeholders on how the weights were derived and why they are meaningful. Use visualization tools to illustrate the value score's components.

Regulatory Rejection

Some agencies may not accept a value function as the primary analysis. Mitigation: plan a hybrid approach as a backup. Engage regulators early and be prepared to present the value function as a key secondary or supportive analysis. Build a dossier that shows the value score aligns with traditional endpoints.

Complexity in Multiplicity Control

While the value function simplifies multiplicity for the primary analysis, secondary analyses of individual endpoints still require adjustment. Mitigation: use a hierarchical testing procedure for individual endpoints, with the value function as the gatekeeper. Alternatively, use a global test that accounts for correlations.

Data Quality and Missing Data

Missing data on one endpoint can affect the value score. Mitigation: pre-specify methods for handling missing data, such as multiple imputation or mixed models. Consider using a value function that can accommodate partial data by weighting available endpoints.

Decision Checklist and Mini-FAQ

Before committing to endpoint agnosticism, review the following checklist and common questions.

Decision Checklist

• Have we identified all relevant endpoints from the patient perspective?
• Do we have robust patient preference data to derive weights?
• Is the trial powered for the value score?
• Have we pre-specified the value function and sensitivity analyses?
• Have we engaged regulators early in the process?
• Do we have a hybrid backup plan if regulators are not receptive?
• Are we prepared to communicate the value score to diverse stakeholders?

Frequently Asked Questions

Q: Does endpoint agnosticism mean we can't have a primary endpoint? A: No. The value function itself becomes the primary endpoint. It is a composite that aggregates multiple measures, but it is still a single pre-specified analysis.

Q: How do we handle multiplicity if we also analyze individual endpoints? A: Use a hierarchical gatekeeping strategy where the value function is tested first. If significant, individual endpoints can be tested in a pre-specified order, with alpha controlled by the hierarchy.

Q: What if patient preferences change during the trial? A: Pre-specify whether you will use baseline preferences or re-assess. If re-assessing, plan for adaptive weighting with clear rules to avoid bias. Sensitivity analyses with different weights can show robustness.

Q: Is this approach only for patient-reported outcomes? A: No. It can include any type of endpoint—clinical, biomarker, functional, or PRO. The key is that all endpoints are weighted according to patient values.

Q: How do we convince regulators? A: Start with a hybrid approach where the value function is a key secondary analysis. Build evidence from preference studies and show that the value score aligns with traditional endpoints. Pilot in less critical indications first.

Synthesis and Next Actions

Endpoint agnosticism represents a paradigm shift from fixed hierarchies to flexible, patient-driven value functions. It acknowledges that the importance of endpoints varies across patients and contexts, and it provides a rigorous framework for incorporating that variability into trial design. While the approach requires additional upfront investment in preference elicitation and statistical planning, it can reduce the risk of failed trials and produce evidence that truly reflects what matters to patients.

For teams considering this shift, we recommend starting with a pilot in a single trial where patient preferences are well understood and regulatory risk is manageable. Use a hybrid approach to build confidence and gather data. Document your methods and outcomes to contribute to the growing body of evidence supporting patient-centric endpoint innovation. As regulatory acceptance grows and tools become more accessible, endpoint agnosticism may become the standard for late-phase trials in many therapeutic areas.

This article is for general informational purposes only and does not constitute professional advice. Readers should consult qualified clinical, regulatory, and statistical experts for their specific trial designs.