Mr. X is fifty-eight, and until a month ago he thought the cough was nothing. Now there is a shadow on the upper part of his right lung, a biopsy that confirms it is cancer, and a scan showing the disease has reached the lymph nodes in the centre of his chest — the drainage stations between the two lungs, which is as far as it has gone. It is not the earliest stage, and not the last; it sits in the difficult middle, where the decisions are hardest. Had he taken his reports to three different hospitals that week, he could have come away with three different plans, each one defensible.
At the tumour board, his file opens with questions, in a sequence every oncologist would recognise. Do we actually know how far it has spread — have we checked the brain, sampled those central nodes with a needle, or are we trusting the scan to tell the truth? Is the cancer sitting in one node station or several, because that single detail changes everything that follows. Could his lungs survive an operation that removes a section of one? And where is the report on the tumour's genetics, because one particular fault in the cancer's wiring quietly rewrites the whole plan.
The surgeon sees disease in one spot in a fit man, and wants to shrink it with treatment, then remove it. The radiation specialist sees cancer in several node stations, too scattered to cut, and wants to treat it with radiation and chemotherapy together, without an operation. And I am waiting on the genetic report, because the drug I would add afterwards for one version of this cancer is precisely the wrong drug for another.
That is what a tumour board is. It is structured disagreement — a room where a surgeon, a radiation specialist, and I are each expected to see the same patient differently, and to defend what we see. The plan that survives all three of us is stronger than any plan one of us would have written alone.
Now consider what happens when that room does not exist.
Cancer care in our country has spread out faster than cancer expertise has. Hospitals have opened in towns that had none, which is genuinely good news, until you look at how the difficult cases move through them. A surgeon finishes a morning list in the city and drives to a hospital two hours away for an afternoon one. A specialist covers four centres across a week. Everyone is stretching to keep up with everyone else, and so it has quietly become normal for the operating surgeon to meet the patient for the first time on the morning of the operation, trusting a hospital he does not work in to have judged, correctly, that this patient is fit for this surgery and that this surgery is the right one.
For a gallbladder stone, that chain usually holds. For a cancer, it can break in ways nobody sees until it is too late, because in cancer two things decide a great deal, and neither survives being handled in pieces. The first is sequence — whether surgery should come first, or whether chemotherapy or radiation should come before it. Operate in the wrong order and you can close a door that was open that morning. The second is the operation itself — whether enough was removed, with a wide enough margin, the right nodes sampled, the tissue handled so the pathologist can actually read it. A cancer operation done well the first time is a real chance at cure. The same operation done in a hurry, by someone who met the patient an hour earlier, is often the last good chance being spent.
So here is what I would say to any family across my desk, and will say here plainly. When it comes to cancer, the instinct to be treated at the nearest and most convenient place does not serve you well. It is the one illness where the inconvenience of travelling to a centre built to handle it is not really an inconvenience; it is part of the treatment. This is your one life, and the disease is unforgiving of a first attempt done poorly.
I say this knowing it is easier said than done. For many families the nearer hospital is the only one they can afford to reach, the only one they can miss work for, the only one that speaks their language and sits within a bus ride. These are real constraints, not failures of judgement, and I do not pretend a paragraph dissolves them. But a family that understands what is at stake can at least weigh it, and can ask, before the surgery date is fixed, one question: has anyone looked at this whole case together, or only at their piece of it?
We have always known cases need that room, and we have always improvised the infrastructure for it. Twenty years ago the infrastructure was a train ticket. We sent the patient to another city with a file of reports to sit across from the expert. What came back was rarely a different plan, but it came back with conviction, for the patient and, honestly, for us.
WhatsApp has replaced the train. Most oncologists I know sit in half a dozen case-discussion groups, where a difficult case gets argued at midnight and a quick poll settles what a conference lecture couldn't. It is faster, wider, and free, and the instinct behind it is sound; it is the tumour board, extended. But it is also, too often, a fragmented case dressed up as a discussed one — a single photo, a single question, without the room behind it.
Meanwhile, colleagues have begun publishing studies that measure how often the board's decision matches what an AI would have suggested. High agreement gets read as reassurance: the machine agreed with us, so we were right. I read it differently. The board's value was never in the agreement; it was in the argument. A machine that nods along with the consensus has only added a vote. The more useful question is what happens when it does not agree, and whether it can tell you why.
That is where I think clinical AI has to go: built the way the board is built. Not one model handing down one confident answer, but several, each reasoning from a different starting point — the staging rules, the current guidelines, the patient's own physiology — with the disagreement shown to you rather than smoothed away. For Mr. X, a tool like that would not deliver a verdict. It would lay the same three directions on the table, each with its evidence and its objection attached. When the models disagree, that is worth paying attention to: it usually marks the case that most needs the humans in the room. What no tool can do is give a fragmented case the depth of a discussed one; it can only make sure that when the case is discussed, nothing was missed.
One more honesty, because the group chats deserve it. Mr. X, as you may have guessed, is a composite, invented from many patients, the way every case we discuss in public should be. The country's new data-protection law asks us to strip out anything that identifies a patient before we share their story, and a photograph of a real report in a group of two hundred people, with the name and hospital number in plain view, is not that. The train-era consult carried one confidentiality risk; the group chat carries hundreds. Whatever replaces the train next has to make that stripping-out automatic, rather than a discipline we remember, or forget, at midnight.
The tumour board never needed a machine's vindication. It needs its rigour carried into the tools that are replacing the train, and its lesson carried to every family choosing where to be treated.
If your centre has run one of these comparison studies, formal or just a conversation over coffee, I would like the other half of the data: where did the machine disagree, and what did you do?