When Tumors Stop Listening: Why the Right End Point Can Be the Wrong Story

May 22, 2026

10 min read

3 views

84%

The seduction of a shrinking tumor

What if the most reassuring number in cancer medicine is also one of the easiest to misread?

A tumor that gets smaller feels like proof. Proof that the treatment is working. Proof that the biology is being controlled. Proof that the next dose, the next cycle, the next approval is justified. But this confidence hides a deeper problem: not every cancer responds in the same language, and not every endpoint tells you whether the disease has truly been confronted or merely silenced for a moment.

That is the tension at the heart of modern oncology. One side of the tension is practical and regulatory: if a therapy produces a high objective response rate, especially in a single arm trial, that seems like clean evidence of activity. The other side is biological: some tumors no longer rely on the pathway the drug was designed to target. In those cases, response is not just a matter of magnitude, but of mismatch. The tumor has changed the terms of the conversation.

The deeper question is not simply, “Did the tumor shrink?” It is this: what kind of tumor are we testing, what kind of evidence do we need, and what does success really mean when the disease has already evolved beyond its original identity?

The hidden assumption behind response rates

Objective response rate has enormous appeal because it offers something rare in medicine: apparent clarity. A shrinkage on imaging can be measured, compared, and counted. In settings where spontaneous regression is extraordinarily rare, a response can be directly attributed to the treatment. That makes ORR especially attractive in early drug development, where time is short and the need to identify active agents is urgent.

But ORR also carries a quiet assumption: that the treatment’s target is present, reachable, and still governing the tumor’s behavior. That assumption is often invisible in trial design, yet it determines whether the endpoint means what we think it means.

Imagine testing a key in a lock without asking whether the lock is still there. A response rate tells you the key turned. It does not always tell you whether the door was the one you needed to open. In oncology, the door may be a temporary shrinkage, while the real disease trajectory continues elsewhere, shaped by cellular escape routes, alternative signaling, or lineage switching.

This is why a high ORR can be both impressive and incomplete. It is strong evidence of biological activity, but not necessarily of durable relevance for every tumor subtype. The challenge is not that ORR is useless. The challenge is that ORR is a local truth, while cancer progression is an ecological one.

When the target disappears, the endpoint changes meaning

A striking example comes from metastatic castrate resistant prostate cancer, where some tumors become AR-null. In plain terms, these cancers no longer express the androgen receptor at meaningful levels. That matters because androgen receptor signaling inhibitors are designed to suppress a pathway that the tumor may no longer depend on.

This creates a profound clinical puzzle. If a tumor no longer expresses the receptor, then continuing to judge its behavior by the response to receptor-targeted therapy can become misleading. The treatment may be biologically elegant and clinically irrelevant at the same time. The disease is no longer asking the question the drug is built to answer.

The AR-null phenotype is especially important because it can be non-neuroendocrine, which means it may not fit older categories of aggressive prostate cancer behavior. It may also be associated with TP53 and RB1 alterations, suggesting a more profound genomic remodeling, not just simple resistance. In other words, the tumor is not merely dodging the bullet. It is rebuilding the battlefield.

This is the key insight: the value of a clinical endpoint depends on whether the underlying biology still honors the premise of the endpoint. ORR is compelling when the target exists, the drug engages it, and shrinkage reflects meaningful pathway inhibition. But if the tumor has become AR-null, then the same response logic may obscure the need for a different therapeutic frame entirely.

A response rate is only as informative as the biology it assumes.

The real divide is not response versus survival, but alignment versus mismatch

Most discussions of endpoints frame the choice as a tradeoff between speed and rigor. ORR is fast. PFS and OS are slower but more consequential. That framing is useful, but incomplete.

A better framing is alignment versus mismatch.

A high ORR is persuasive when three things line up:

The cancer still expresses the target.
The treatment mechanism directly disrupts that target.
Tumor shrinkage is a reliable proxy for meaningful disease control.

When those conditions hold, ORR can function like a clean signal in a noisy system. It can reveal a therapy with true breakthrough activity, especially in single arm studies where randomized comparisons are not yet feasible.

But when alignment fails, even a dramatic response can become biologically ambiguous. The tumor may shrink temporarily while a resistant subclone expands. The measurable lesion may improve while the disease elsewhere accelerates. The treated pathway may be intact in one metastatic site and absent in another. In such settings, a single endpoint can accidentally flatten a heterogeneous reality.

This is especially important in metastatic disease, where different lesions within the same patient can behave like different species in the same ecosystem. One site may remain androgen-driven. Another may have already abandoned that dependence. The clinician sees a single patient. The biology is often a consortium of divergent populations.

This is why pathology and biomarkers matter not as ornamental diagnostics, but as endpoint calibration tools. They tell us whether the measure we are using is still attached to the thing we care about.

A useful mental model: the map, the terrain, and the weather

Here is a simple framework that helps connect endpoint design to tumor biology.

Think of cancer assessment as three layers:

The map: the biomarker or receptor status that tells you what kind of disease you are dealing with.
The terrain: the actual tumor architecture, including heterogeneity, lineage changes, and resistance mechanisms.
The weather: the measured clinical endpoint, such as ORR, PFS, or OS, which captures short term conditions but not always the deeper landscape.

If the map is wrong, the weather becomes hard to interpret. If the terrain has changed, the map can be outdated. If the weather is used as the whole story, you miss the difference between a passing storm and a long climate shift.

AR-null prostate cancer is a perfect example of map drift. The tumor no longer matches the receptor based map that justified the therapy. A high ORR in a receptor positive population might be meaningful. The same endpoint in an AR-null subgroup could be structurally misleading. The biology has changed, so the signal from the endpoint must be reinterpreted.

This model also explains why pathology is not a backward looking ritual. It is a way of keeping the clinical map synchronized with the terrain. Without that synchronization, the trial may generate numbers without insight, or worse, insight that applies only to a disappearing slice of disease.

Why single arm trials are both powerful and fragile

Single arm trials occupy a special place in oncology because they are often the first place a therapy can prove itself. They are efficient, ethically useful in some contexts, and well suited to finding signals in diseases with few options. If a tumor response is rare without treatment, then a substantial ORR can be an important clue that a drug deserves further attention.

But single arm trials are also fragile in a subtle way: they can overstate certainty when the patient population is biologically mixed. If some tumors remain target driven and others do not, the aggregate response rate may blur two different truths. The trial might show promise without revealing that the benefit is concentrated in a molecular subset.

That means the question is not whether single arm trials should use ORR. In the right context, they absolutely should. The deeper issue is whether the study is designed with enough biological resolution to distinguish response because of mechanism from response despite heterogeneity.

This is where AR immunohistochemistry becomes more than a lab test. It becomes a gatekeeper for interpretation. If AR immunohistochemistry can distinguish AR-null from AR-expressing cases in the metastatic setting, then it helps decide whether a response endpoint is tracking the therapy’s true mechanism or merely averaging across incompatible disease states.

In practical terms, endpoint selection is not just statistical. It is ontological. It asks, what is this disease, really, at the moment we measure it?

The clinical lesson: stop asking only whether the drug works, and ask for whom it works

The obsession with a single global number often masks the more useful question: for which biology does the drug work?

This is especially relevant when a therapy class has a clear mechanism but a shifting target population. A pathway inhibitor may remain powerful for receptor expressing disease and nearly irrelevant for receptor absent disease. If the clinical conversation stays focused only on aggregated outcomes, the field risks two errors at once: it may underuse an effective therapy in the right subgroup, and overuse it in the wrong one.

That is the paradox of precision medicine. The more precisely we understand tumor biology, the less satisfied we should be with average outcomes. A mean response rate can hide the fact that one subgroup is experiencing real benefit while another is effectively being treated with a beautifully designed placebo.

This is not an argument against ORR. It is an argument for endpoint humility. ORR is a tool, not a verdict. It can illuminate activity, but only if we know the biological context in which that activity matters. In the presence of AR-null disease, the endpoint must be read alongside receptor status, lineage features, and resistance patterns. Otherwise we risk rewarding a therapy for doing exactly what the tumor has stopped needing.

Key Takeaways

A response is not the same as relevance. A tumor can shrink even when the disease has already escaped the pathway the treatment was meant to target.
Endpoints depend on biology. ORR is most informative when the target is present and the tumor still depends on that pathway.
Biomarkers are endpoint calibrators. Tests such as AR immunohistochemistry help determine whether a response measure still reflects the mechanism of action.
Heterogeneity changes interpretation. Mixed metastatic disease can produce aggregate results that obscure who is actually benefiting.
Ask the subgroup question early. Before trusting a global response rate, ask which tumors remain target driven and which have become biologically independent.

The bigger lesson: medicine should measure compatibility, not just change

The deepest connection between these ideas is not about prostate cancer alone, and not even about oncology alone. It is about a general problem in measurement: we often mistake visible change for meaningful alignment.

A tumor can become smaller. A number can improve. A graph can turn upward. But if the underlying system has already shifted its dependencies, the apparent success may be incomplete, temporary, or irrelevant to long term control. Good medicine therefore requires more than endpoints that are easy to count. It requires endpoints that are attached to the real structure of the disease.

That is why the most mature question in cancer care is not, “Did it work?” It is, “Did it work on the disease that exists now?”

Once you see that distinction, the meaning of response rates changes. They are no longer final judgments. They are clues. And the best clinical decisions come from reading those clues in the light of biology, not in spite of it.

The future of treatment evaluation will belong to the therapies, trials, and clinicians that can hold both truths at once: measure the change, and verify the match. When a cancer stops listening to the signal you are sending, the right next step is not louder repetition. It is better diagnosis of what the tumor has become.