I have been thinking about adjuvant therapy, and about why it produces the hardest conversations I have in clinic. The reason is structural. In the metastatic setting everyone can see the disease, we want to treat it, and the next scan tells us whether we were right. In the adjuvant setting the cancer is already gone. We are treating something that may not even be there, weighing a harm that is certain and immediate against a benefit that is probabilistic and years away, and we will never find out whether the decision we made was the right one for that person, regardless of what happens.
Let me make that concrete.
A patient with nothing to show on a scan
A 58-year-old woman comes to see me after a nephrectomy for a clear cell kidney cancer (the usual kind). The pathology reads pT3, clear margins, no involved nodes. Her scans are clean. She feels well. By the staging math she sits in the intermediate-high risk group, which means adjuvant pembrolizumab is on the table.
I can tell her what KEYNOTE-564 showed. A year of pembrolizumab after surgery improved overall survival, with a hazard ratio for death of 0.62. At four years, about 91 percent of treated patients were alive, against 86 percent on placebo. That is a real benefit, and it is the first time an adjuvant therapy has moved survival in this disease, which is a real advance.
Then she asks the only question that matters to her: How much does this add for me.
And I do not have that number at my fingertips. What I have is a single trial and a population inside it. Most of the women who look like her on paper are already cured by the surgery, and for them the year of immunotherapy offers no benefit at all, only the chance of a thyroid that never recovers, a colitis, or in rare cases something worse and permanent. A smaller group will recur no matter what we do, and the drug may delay that without changing how it ends. Somewhere inside the average sits the woman who is genuinely saved by it.
Here is the part I want to be honest about. It is not that her number is unknowable. There are ways to get closer to it than I usually let on. Prognostic calculators turn her stage and grade into a recurrence risk. Subgroup analyses, from KEYNOTE-564 and from the trials around it, speak to patients more like her. More involved calculations combine all of it into something approaching a personal estimate. A great deal of that work already exists, and more of it is published every month.
What I cannot do is hold all of it in my head. Not for her, and not for the next twenty patients, each sitting in a different corner of a different disease, each with its own calculators and its own updating evidence. The numbers are out there, scattered across papers and tools that are revised faster than any one clinician can keep pace with. In the time I have with her, I cannot reliably gather the right ones, run the calculation, and hand her a figure I trust to reflect what is currently known. So the precision stays in the literature, and she gets my judgment instead.
The grey zone is wider here than almost anywhere
This is the part I think people outside clinic underestimate. The guideline is not ambiguous. It says offer adjuvant pembrolizumab to intermediate-high and high risk disease. I know the guideline. I know the trial. That was never the hard part.
The hard part is that the risk groups are built from anatomy, from tumour size and grade and stage, and anatomy is a coarse instrument for what is really a biological question. PD-L1, which helps stratify in other settings, did not separate benefit here. The pieces of her number exist, scattered across calculators, subgroup tables, and a literature that updates faster than any of us can read it. What does not exist is a way to pull them together at the point of decision, into a single estimate of her recurrence risk and her likely benefit that I can trust and defend in the room. The population result is solid. Her number is computable, and uncomputed. And her number is the entire conversation.
So I do what every honest oncologist does. I lay out what we know, I give my judgment, and we decide together under real uncertainty. That is good medicine. It is also a long way from the precision the trial was actually capable of supporting, and it leaves a great deal of the drug’s value stranded somewhere between the evidence and the bedside.
Why this should matter to anyone launching an adjuvant therapy
Adjuvant indications are uniquely exposed to this gap. In the metastatic setting the benefit is visible, the patient is symptomatic, and the decision largely makes itself. In the adjuvant setting the benefit is invisible by definition and the harm arrives first. That is precisely why real-world adjuvant uptake tends to be lower, slower, and far more variable between oncologists than the trial would lead you to expect. The decision is probabilistic, and probabilistic decisions are the ones clinicians defer, soften, or quietly talk a patient out of on a busy afternoon.
For a diagnostics company, that is the opening, with a catch worth naming. A test that could tell my 58-year-old whether she still harbours minimal residual disease, with circulating tumour DNA the obvious candidate, would not merely confirm a decision I had already made. It would change it. It would pull her three possible selves apart and begin to separate the woman who is already cured from the one who actually needs treating.
But a test result is only ever one more input. On its own it tells me no more than her stage or her grade does in isolation. Its value appears only when it is folded into the same calculation as everything else, when knowing she is positive or negative moves her recurrence risk and her expected benefit by an amount I can actually see. Without a formula to receive it, a new marker is just another number on the pile I already cannot hold in my head, and the honest answer to how much the test is worth stays unclear, to me and to the company that built it. Give the test a defined place inside that calculation, and its value becomes legible. Leave the calculation implicit, and even a good test struggles to prove what it is worth.
For a pharma company, the lesson is about the last mile, and about timing. The hardest part, generating the survival benefit, is already done and already paid for. What remains is giving the community oncologist a way to turn that population result into an individualized estimate she can defend in the room. Without it, the drug quietly underperforms its own evidence, not because anyone doubts the data, but because the data never reaches the patient in a form she can use. And the moment to build that is at launch, while the field is still forming its habits around a new indication, not three years later once the undertreatment has already set in.
I find this a genuinely hard and interesting problem to sit with, the decisions where the evidence is strong and the individual answer is still missing. If this resonates with something you are working on, I’d like to hear about it. Contact.
Dr. Henry Conter is a Medical Oncologist and Hematologist at William Osler Health System and the founder of Kesis & Sisters. He trained in Medical Oncology at MD Anderson Cancer Center and spent six years at Hoffmann-La Roche in progressively senior roles spanning oncology clinical development, portfolio strategy, and medical and regulatory affairs, across both national and global functions.