The Endpoint Is Not the Patient

Elena Sinclair
May 24
10 min read

A clinical trial sponsor post regulatory approval is scratching his head wondering why his new drug is being ignored by the patients

Summary:

This article argues that clinical development programs which succeed operationally routinely fail strategically — because they optimize each functional workstream (regulatory, clinical operations, biomarkers, HEOR, medical affairs) independently, without integrating the evidence requirements of the full stakeholder ecosystem. Regulators, payers, physicians, patients, and guideline committees each inhabit a distinct evidentiary world; designing for one produces predictable failures in the others. Three documented case studies — Alzheimer's anti-amyloid therapy, oncology accelerated approvals, and cell and gene therapy — illustrate the systemic cost of functional siloing. Part 2 of 3 in The Evidence Ecosystem series.

Picture this. A drug gains FDA approval. The regulatory team did exactly what it was supposed to do — designed a clean pivotal trial, hit the primary endpoint, navigated the submission flawlessly. The clinical operations team enrolled on time. The biostatistics team delivered a statistically significant p-value. Every function performed.

Every metric was green.

Then the drug launched.

ICER rated it 'low value.' CMS restricted Medicare coverage to clinical trial settings. Physicians, uncertain about which patients to treat and lacking head-to-head data against standard of care, hesitated. Patients who qualified on paper couldn't access it — the infusion infrastructure didn't exist at community centers. The HEOR team, now scrambling to build an economic model from trial data never designed for that purpose, found itself without the comparative effectiveness endpoints payers actually needed.

The drug reached a fraction of the patients the team had planned for. At lower value than the system could sustain.

Every function did its job. And yet.

What does it mean to succeed functionally and fail systemically? And why has this pattern become so common that it barely registers as a failure anymore — just the expected friction of launch?

The answer, it turns out, is structural. And it begins well before the first patient is enrolled.

The Org Chart Is the Problem

The standard clinical development organization looks like this: regulatory affairs optimizes for approval, clinical operations optimizes for enrollment speed and data quality, the biomarker team optimizes for assay validity, HEOR builds economic models, medical affairs plans physician education, market access works the payer landscape. Each function has its own leadership, its own metrics, its own definition of success.

This structure has a logic. Specialization produces depth. Parallel workstreams accelerate delivery. Departmental accountability creates clear ownership. When you're trying to move a complex program efficiently, decomposing it into manageable parts makes sense.

The problem is that it's the same logic that produces local optimization and global failure in biological systems.

In Part 1 of this series, we argued that the history of biology is largely a history of this mistake — discovering that mechanisms which work beautifully in isolation fail unexpectedly in the system that contains them. Encainide and flecainide suppressed cardiac arrhythmias perfectly. The patients died at twice the rate of placebo. The antiarrhythmic drugs weren't wrong about the arrhythmia. They were wrong about the patient.

A clinical development program organized into functional silos is the organizational equivalent of a biological system whose components have been separated for individual study. Each component performs as designed. The system-level behavior is not what the components predicted.

The relevant unit of design for a clinical development program is not the functional workstream. It is the evidence need of the full stakeholder ecosystem the program must ultimately satisfy. Design the workstreams in service of that — not in service of functional metrics.

Right now, that inversion is rare. What is common is something else entirely: a drug that convinces the FDA and almost no one else.

Five Stakeholders, Five Evidentiary Worlds

Here is a useful way to think about what goes wrong. Every major stakeholder in the healthcare system lives in a different evidentiary world — constituted by their specific decision context, accountability structure, and success criteria. The same trial data means fundamentally different things to each of them.

The FDA and EMA need substantial evidence of safety and efficacy on a pre-specified primary endpoint in an adequate, well-controlled study. The pivotal RCT is their world. They are not structurally designed to evaluate real-world effectiveness, comparative value, or patient-level burden. A clean p-value satisfies them. It does not satisfy anyone else.

Payers and HTA bodies — CMS, NICE, ICER — need evidence that the drug provides sufficient incremental clinical benefit, relative to the current standard of care, to justify its cost. Their world is comparative effectiveness, cost per QALY, and real-world feasibility. A pivotal trial that didn't include a comparator arm, didn't collect health utility data, and didn't define a cost-effectiveness threshold is nearly invisible to them — regardless of statistical significance. In 2025, 70% of all commercial insurance new-to-brand prescriptions received an initial denial from payers. Regulatory approval, in this environment, is necessary but radically insufficient.

Physicians need to know where the drug fits in their treatment algorithm, for which specific patients, with which safety profile, and with what practical monitoring requirements. A drug approved in a highly selected trial population — without comparative data against the most commonly used alternatives, without clear patient selection criteria — forces them to extrapolate in ways they are reluctant to do. The result isn't opposition. It's inertia.

Patients need to believe the treatment will improve how they feel or function, that the side effects are manageable, and that the logistics fit their lives. Patient-reported outcomes data isn't a regulatory nicety — it's the primary evidence relevant to the patient's decision. Trials that don't collect robust PRO data leave patients — and the clinicians serving them — without the information they actually need to decide.

Clinical guideline committees — ASCO, ESMO, NCCN — need evidence that is reproducible, generalizable, and interpretable in the context of current clinical practice. A program that wins approval but fails confirmatory trials, or whose results are restricted to a biomarker-selected subpopulation without community diagnostic infrastructure, struggles to earn guideline endorsement. And without guideline endorsement, physician adoption stalls regardless of what the label says.

A program that designs its evidence strategy for one stakeholder world — typically the regulatory one — will produce evidence that satisfies the regulator and fails the others. This is not bad luck. It is the predictable consequence of optimizing for the wrong system level.

Where Siloing Shows Up — and What It Costs

The structural mismatch between how development programs are organized and what the stakeholder ecosystem actually requires shows up in specific, documentable ways. For each functional dimension, there is a siloed default, a systemic consequence, and an integrated alternative that development teams almost never build in time.

Endpoint selection. Biostatistics and regulatory affairs select an endpoint that maximizes statistical power and regulatory acceptability — PFS in oncology, CDR-SB in Alzheimer's. The systemic consequence: approved on a surrogate that ICER rates as insufficient evidence of clinical benefit; CMS restricts coverage to clinical trial settings. A 2025 ASCO analysis of nearly 800 oncology Phase 3 trials found that fewer than 30% of positive surrogate-endpoint trials showed overall survival improvement. Only 6% improved both OS and quality of life. The integrated alternative designs endpoint strategy alongside payer evidence requirements from Phase 2 onward — not from the day the submission is filed.

Biomarker strategy. The biomarker team selects the biomarker that most reliably predicts mechanism-of-action engagement and regulatory acceptance of enrichment strategy. The systemic consequence: the companion diagnostic is analytically valid but requires infrastructure that doesn't exist in community practice. Drug uptake restricts to academic centers. The population that benefited in the trial bears little resemblance to the population seeking treatment in the real world.

Inclusion/exclusion criteria. The clinical science team tightens eligibility to maximize internal validity and signal clarity. The systemic consequence: the trial population is so selective that physicians cannot identify patients matching the label. Real-world effectiveness diverges from trial efficacy. Payers restrict coverage to trial-eligible patients. A highly powered trial produces results that generalize to almost no one actually seeking treatment.

Regulatory strategy. Regulatory affairs pursues accelerated approval on a surrogate endpoint to reduce time to market. The systemic consequence: approval granted, CMS coverage restricted pending confirmatory evidence, ICER evaluates the drug as low value, commercial launch stalls. Fifteen percent of oncology drugs granted accelerated approval have been withdrawn because of failed or incomplete confirmatory trials — yet many remained in clinical guidelines for years, creating a landscape where physician guidance no longer reflects the evidence.

Payer evidence. HEOR builds an economic model post-approval using trial data not designed for this purpose. The systemic consequence: the model relies on surrogate-to-OS correlation assumptions that ICER and NICE challenge; cost-effectiveness is uncertain; formulary coverage is delayed or restricted. The payer conversation that should have been designed into Phase 2 becomes a post-approval negotiation conducted from a position of evidentiary weakness.

These are not unrelated failures. They share a common cause: each was produced by a team optimizing within its own evidentiary world without integrating the requirements of the other worlds the program must ultimately satisfy.

Three Case Studies in Predictable Failure

The pattern above is not hypothetical. It plays out in the public record, repeatedly, with the same structural signature.

Case Study A: Alzheimer's Anti-Amyloid Therapy — The Full-System Failure

Aducanumab and lecanemab represent the most complete available illustration of stakeholder misalignment in a major therapeutic program. What was optimized locally: amyloid clearance as a biomarker of target engagement; CDR-SB change as evidence of clinical effect; accelerated approval pathway for speed to market.

What was missed: payer evidence for the clinical meaningfulness of CDR-SB change; physician infrastructure for ARIA monitoring and infusion delivery; healthcare equity implications of access barriers; comparative value against supportive care at list price.

The chain of systemic consequences is now extensively documented. FDA granted accelerated approval of aducanumab over its advisory committee's 10-0 vote against approval. CMS explicitly restricted Medicare coverage to patients in clinical trials, citing absence of evidence of clinical benefit — a direct contradiction of FDA's decision. ICER rated lecanemab at its $26,500/year list price as 'low value' unanimously (15-0). Physician adoption has been limited by monitoring requirements, equity concerns, and uncertainty about clinical meaningfulness. Real-world uptake of lecanemab is six times higher among white versus Black patients, and 24 times higher among patients with higher versus lower socioeconomic status.

The mechanism was scientifically coherent. The evidence ecosystem was not designed to convince anyone beyond the regulator. The result: a therapy with genuine biological plausibility reaching a fraction of intended patients, at a price the system declined to sustain.

Case Study B: Oncology Accelerated Approvals — The Surrogate Chain

The oncology accelerated approval pathway has produced a well-documented pattern: drugs approved on progression-free survival or objective response rate that fail or are inconclusive in confirmatory trials, while guideline bodies maintain endorsement, patients receive therapies of uncertain benefit, and payers face coverage decisions without adequate evidence.

What was optimized locally: regulatory pathway speed; PFS as a statistically powerful primary endpoint; signal in an enriched patient population.

What was missed: confirmatory trial design that could verify clinical benefit; clinical guideline alignment with post-approval trial results; physician guidance on patient selection after confirmatory failure.

Eighteen cancer drug indications received accelerated approval but failed to improve primary endpoints in post-approval trials; 11 (61%) were voluntarily withdrawn — yet NCCN guidelines provided high-level endorsement for most, sometimes even after withdrawal. Patients received drugs that had been effectively pulled in over one-quarter of eligible patient-years between approval and withdrawal across five studied indications. Only 16% of accelerated approval drugs had their guideline status noted after withdrawal. The signal-to-noise ratio in clinical guidance has been degraded across the field.

Case Study C: Cell and Gene Therapy — Approval to Access Failure

Cell and gene therapies present a structurally distinct version of the same problem. Development teams designed programs for regulatory approval: clinical efficacy demonstration, regulatory pathway navigation, list pricing consistent with a clinical value argument.

What was systematically missed: the payment architecture. The US healthcare system is built around annual recurring costs. CGTs are durable, one-time interventions with uncertain long-term follow-up data. Outcomes-based contracting frameworks were not operationally available at scale. Treatment center capacity was not planned for commercial demand. The list price spend on CGTs in the US over the next decade has been estimated at $35–40 billion — a number the current payment infrastructure is not designed to absorb.

Despite regulatory approvals, multiple CGTs have faced commercial struggles driven by access barriers: payers restricting coverage to narrow populations, outcomes-based contracts not yet operational at scale, and treatment centers limited in number and geography. For rare disease CGTs, where patient populations are already small, the gap between regulatory approval and actual patient access has been particularly acute.

The lesson is the same as in Alzheimer's and oncology. Approval is not access. Access is not adoption. And none of those outcomes are addressable after launch with data the program was never designed to generate.

Why Siloing Persists — And What Would Have to Change

If the consequences of functional siloing are this well-documented, why does the structure persist? The answer is incentives — and incentives are stubborn.

Development organizations measure functional performance locally. Regulatory affairs is measured on approval. Clinical operations is measured on enrollment and data quality. Biomarker teams are measured on assay validation. HEOR is measured on published models. No single function is measured on systemic evidence quality — the degree to which the evidence generated will satisfy the full stakeholder ecosystem.

This creates a principal-agent problem at the program level. Each agent (functional team) is rational in optimizing its own metric. The collective outcome is suboptimal for the principal — the program's ultimate strategic goal, which is not approval but access, adoption, and durable patient benefit.

The same dynamic appears in biology. Each cell, each organ, each physiological subsystem pursues its local optimization. It is the regulatory architecture — homeostasis — that produces integrated, adaptive, whole-organism behavior. In biology, homeostasis prevents local optimization from producing global failure. In clinical development, integrated evidence planning is the equivalent architecture. And in most development organizations, it does not exist.

The problem is not that clinical development teams lack expertise. The problem is that each team's expertise is applied within its own evidentiary world, without a shared architecture for designing evidence that satisfies all of those worlds simultaneously. And no one is accountable for the gap.

The Pattern Is Avoidable

Return to the opening scenario. Every function did its job. The regulatory team got approval. The operations team enrolled on time. The biostatistics team hit the endpoint. And the drug reached fewer patients than the team had planned, at lower value than the system could sustain.

The lesson is not that the functional teams failed. The lesson is that each team's success was defined too narrowly, measured against a local metric, and disconnected from the systemic outcome the program was designed to achieve.

This is not a talent problem. It is not a science problem. It is a design problem — the organizational equivalent of treating the clinical trial as a workflow to be executed rather than an evidence ecosystem to be architected.

In biology, this error is called reductionism taken too far. In clinical development, it is called functional siloing. The name is different. The structure of the mistake is identical.

The good news: it is also a design problem with design solutions. In the final article in this series, I describe what an integrated evidence ecosystem looks like in practice — and the specific tools that development teams can adopt without dismantling the functional organization they depend on.