What a Shrinking Tumor and a Hidden Risk Factor Have in Common

May 26, 2026

10 min read

4 views

58%

The Strange Problem with Medical Proof

What counts as evidence when the stakes are life and death?

That question sounds simple until medicine forces a harder choice: do we trust a result because it happens quickly and clearly, or because it predicts what we ultimately care about? In cancer care, this tension appears everywhere. A tumor can shrink dramatically, but does that mean a patient lives longer? A risk factor can quietly shape where a disease shows up, but is it strong enough to change what clinicians do today?

These are not separate problems. They are two faces of the same deeper challenge: medicine often has to act before it has complete truth. It must decide under uncertainty, using imperfect signals that are sometimes vivid, sometimes delayed, and sometimes misleading.

That is why the relationship between a visible response and an invisible risk is so revealing. One asks how we know a therapy is working. The other asks how we know a patient is in danger. Together they expose a central fact about modern medicine: the most measurable signal is not always the most important one, and the most important outcome is often the hardest to measure.

The Seduction of the Visible Signal

A shrinking tumor is emotionally powerful. It is immediate, tangible, and easy to photograph. In a trial setting, a response rate can seem almost elegant in its simplicity: either the tumor shrank or it did not. That makes objective response rate, or ORR, attractive, especially in single-arm trials where randomization is absent and a rapid readout is essential.

There is a reason such endpoints are used to support accelerated approvals. If a drug produces a substantial response in a setting where spontaneous regression is rare, that response appears to be directly tied to the therapy. In practical terms, ORR becomes a kind of biological receipt. It says, in effect, this intervention clearly did something.

But clear does not mean complete. A response rate can tell you that a drug hits the target. It cannot fully tell you whether the hit matters over months or years. A therapy may shrink tumors without altering the course of the disease in a durable way. Another may not create dramatic early changes but still extend life. The seduction of ORR is that it gives us an answer fast, at a moment when waiting for survival data feels intolerable.

This is the first half of the dilemma: the more direct the signal, the less comprehensive it may be.

A measurable effect is not the same thing as a meaningful outcome. In medicine, that distinction can separate progress from illusion.

The Quiet Signal We Ignore at Our Peril

Now consider a different kind of evidence: chronic glucocorticoid use and the risk of advanced prostate cancer at presentation. This is not a dramatic signal. It does not flare up on a scan. It does not produce a crisp before-and-after image. Instead, it works in the background, shaping disease in ways that may only become obvious when the cancer is already advanced.

This kind of factor matters because it changes the distribution of risk before any treatment begins. It affects who shows up late, who has a more aggressive presentation, and who may be misclassified if we look only at the disease state in front of us rather than the biological and clinical forces that produced it.

That is the opposite of ORR, but it is no less important. ORR asks, “Did the tumor visibly respond?” Risk-factor research asks, “What conditions made this cancer behave the way it did by the time we noticed?” One is a reading of treatment effect, the other a reading of disease context. Yet both are trying to solve the same problem: how to infer the truth of a disease process from partial, time-bound evidence.

The temptation in medicine is to privilege what is easiest to count. We like clean categories, neat thresholds, and endpoints that can be plugged into a regulatory framework. But cancers do not always respect those preferences. Some signals are loud but shallow. Others are quiet but structurally decisive.

A chronic medication like glucocorticoids may operate less like a switch and more like a climate. It does not always cause a single event that can be traced in the way a tumor shrinkage can. Instead, it may influence immune function, symptom masking, and the timing of detection. The result is not a flashy biomarker. The result is a changed reality.

Why Cancer Teaches Us to Distrust Simple Endpoints

The deeper question behind both topics is not merely technical. It is philosophical: what kind of evidence should we trust when the thing we care about is longer, larger, and less visible than the thing we can observe today?

This is where many debates about cancer endpoints become distorted. People often talk as if the choice is between bureaucracy and speed, or between rigor and access. In truth, the real issue is mismatch. We use one measurement when we really need another. We confuse an intermediate signal with the final outcome.

That does not mean ORR is unimportant. In the right setting, it can be exactly the right tool. If a single agent produces striking tumor shrinkage in a refractory cancer, that is valuable evidence. It can justify further development, especially when the disease is severe and the clock is short. But ORR should be understood as a proxy with boundaries, not a universal verdict.

Likewise, a chronic exposure or medication history is not a fate sentence. It is a clue about underlying vulnerability, not an oracle. Yet clues matter precisely because they are incomplete. Clinicians are rarely handed a perfect dataset. They work with probabilities, not certainties. The art lies in knowing which imperfect signal to trust for which decision.

A useful way to think about this is through three layers of evidence:

Immediate signal: a tumor shrinks, a lab changes, a symptom improves.
Trajectory signal: the disease stays controlled over time, or keeps returning.
Context signal: the patient's background, medications, physiology, and exposures change the meaning of both the immediate and trajectory signals.

ORR sits in the first layer. Chronic glucocorticoid use influences the third. But good medicine requires all three. If you focus only on immediate signal, you risk confusing activity with benefit. If you focus only on context, you may miss a real therapeutic breakthrough. The goal is not to choose one layer. It is to interpret each layer in relation to the others.

The Regulatory Mindset and the Clinical Mindset Are Not the Same

One of the most important tensions in oncology is the gap between what regulators can approve and what patients need. Regulators must make decisions with finite time and incomplete data. That is why an endpoint like ORR can be useful. It creates a disciplined, reproducible way to recognize promising activity before long-term data mature.

Clinicians, however, live in a different time scale. They do not merely ask whether a drug is active in a cohort. They ask whether this specific patient will live longer, feel better, avoid toxicity, or regain function. A therapy can clear the bar for approval and still disappoint in practice if the initial response does not translate into durable benefit or tolerability.

This is where the hidden-risk perspective becomes instructive. A patient on chronic glucocorticoids may arrive with advanced disease, and that background can affect the interpretation of everything that follows. If a response is less durable, is that because the drug is weak, the disease biology is worse, or the host environment is altered? If diagnosis is delayed, was the cancer more aggressive, or was it simply harder to detect? These are not academic distinctions. They shape treatment decisions and trial design.

The broader lesson is that endpoints are not neutral. They create a lens, and every lens distorts as well as reveals. The regulatory lens favors speed and measurability. The clinical lens favors lived consequence. The best system does not pretend these lenses are identical. It asks how to move between them without fooling itself.

One practical implication is that early response endpoints should be paired with contextual risk assessment. A response rate is more convincing when it appears in a population whose baseline risks are well understood. Likewise, a risk factor becomes more actionable when we know how it alters presentation, detection, and likely treatment response. Evidence becomes stronger when it is not just observed, but placed.

A Better Mental Model: From Proof to Prediction

The real breakthrough comes when we stop treating endpoints as verdicts and start treating them as parts of a prediction system.

Think of oncology as navigating with incomplete instruments. ORR is like checking whether the engine starts. It tells you the machine is alive and doing something. But it does not tell you whether the car will make it across the country. Chronic glucocorticoid use is like knowing the terrain and weather before you begin. It tells you the road may be steeper, the visibility worse, and the chance of delay higher. Neither fact alone is enough.

This suggests a more mature framework for evidence:

Activity: Is the intervention biologically doing something?
Durability: Does that effect persist in a clinically meaningful way?
Context: What baseline conditions alter the interpretation of activity and durability?

Seen this way, the apparent opposition between ORR and survival endpoints becomes less absolute. ORR is not inferior simply because it is early. It is incomplete because it answers a different question. Similarly, risk factors are not secondary because they are upstream. They are essential because they shape what later data will mean.

This framework also guards against a common error: overvaluing a single elegant number. A high response rate can be genuinely exciting, but if the disease context is poorly understood, that excitement may be fragile. Conversely, a modest response rate in a high-risk population may be more meaningful than it first appears. Meaning depends on comparison, not just magnitude.

In medicine, the right question is rarely “Did it work?” The better question is “What did it work on, for whom, and toward what end?”

Key Takeaways

Do not confuse a fast signal with a final outcome. A shrinking tumor is important, but it is not the same as living longer or living better.
Context changes meaning. Chronic exposures, medications, and baseline risk factors can reshape how we interpret disease presentation and treatment response.
Use endpoints as tools, not truths. ORR can be valuable for early evidence, especially when spontaneous regression is rare, but it should be paired with longer-term measures when possible.
Think in layers of evidence. Immediate activity, durability, and patient context each answer different questions and should be interpreted together.
Ask what the signal predicts, not just what it shows. The most useful evidence helps forecast the course of disease, not merely describe a moment in time.

The Real Lesson: Medicine Is a Discipline of Interpreted Incompleteness

The deeper connection between a response endpoint and a hidden risk factor is not statistical. It is epistemic. Both reveal that medicine is built on interpreted incompleteness. We rarely get the full story at once. We get snapshots, proxies, and clues. We then build decisions from them, hoping to balance urgency with humility.

That is why the best medical thinking resists both cynicism and enthusiasm. Cynicism says proxies are never good enough. Enthusiasm says any visible change is enough. Both are wrong. Progress comes from knowing which signals are strong enough to guide action and which must be held provisionally until a larger truth appears.

Seen this way, a response rate and a risk factor are not opposites. They are complementary forms of uncertainty management. One tries to verify that a therapy has leverage. The other tries to identify where the disease has leverage over the patient. Together they remind us that the real task is not simply to detect change, but to understand what kind of change matters.

And that may be the most important shift of all: medicine does not advance by counting what is easiest to see. It advances by learning how to read what is easy to miss.