The care the model delayed

A clinician examines a patient, weighs the options, and orders a treatment. This is the moment the whole apparatus of medicine exists to reach — a trained person, looking at a particular body, deciding what that body needs now. And then, before the treatment can begin, the order stops. Somewhere between the clinic and the pharmacy, or between the referral and the procedure, an automated review opens the request, checks it against a coverage rule and a necessity model, and returns a verdict: not yet, or not this, or not without more. The clinician files an appeal. The patient goes home to wait. And while the paperwork moves at the speed of paperwork, the illness moves at the speed of the illness, which was never going to consult the calendar.

This is prior authorization, and it is worth being precise about what it is. It is the point at which an insurer's automated decision is permitted to override or postpone a clinical judgment — to sit between the person who examined the patient and the treatment that person ordered, and to decide whether, and when, care may proceed. The party running that review bears none of the clinical consequence of getting it wrong. It will not be in the room if the patient deteriorates. And the timescale on which its decision can be contested — the days or weeks of an appeal — is mismatched, sometimes catastrophically, to the timescale on which the harm accrues, which is set by the disease. Hold those three facts together and a fourth follows. When care is time-sensitive, the delay is not a preface to the decision. The delay is the decision.

The delay is the decision

We are trained to think of a denial as the harmful act and a delay as something gentler — a hold, a pause, a request for more information. In ordinary commerce that intuition is fine. A delayed refund is an annoyance; the money arrives eventually and nothing is lost but patience. But medicine does not run on that clock. A treatment that would have worked on Monday can be useless by the following week, not because anyone reversed the decision but because the window closed while the decision was pending. When the thing being rationed is time, withholding it is not a neutral state between yes and no. It is a substantive act with its own consequences, and those consequences fall on the patient whether or not a denial is ever formally issued.

Notice what this does to the usual defense of automated review, which is that the patient can always appeal. An appeal is a remedy calibrated to a world where the harm waits for the ruling. Prior authorization operates in a world where it does not. The appeal may well succeed — the treatment may be authorized in the end — and the patient may still have been harmed by the interval, because the interval was itself part of the injury. A process that can be right on the merits and still injurious in its timing is a process whose timing has to be counted as part of what it decides. The pause is not free. Someone is paying for it in the only currency the disease accepts.

A denial you can appeal in three weeks is a decision made in three seconds about an illness that will not wait for either.

Deciding without the consequence

The deeper problem is a separation between who decides and who bears the outcome. The clinician who ordered the treatment carries the consequence directly: they know the patient, they will answer for the choice, and they are accountable in every sense the profession recognizes. The automated review that overrides them carries none of that. It applies a rule and scores a model, and then it is done; it does not follow the patient home, does not revise its judgment as the illness declares itself, and cannot be asked, at the bedside and at speed, why it decided as it did. The authority to override a clinical judgment has been placed with a party structurally insulated from the clinical consequence of overriding it.

This is the accountability gap that runs through everything the series argues, and prior authorization is one of its sharpest instances. A decision that grants something is one kind of act. A decision that withholds something a professional has judged to be owed is a different and heavier kind, and it demands a correspondingly stronger account — stronger, and faster. The graver the thing being denied, and the shorter the fuse on the harm, the higher the bar for the party doing the denying. It is exactly backward for the most consequential and most time-critical decisions to be the ones delivered with the least immediate, least contestable justification. If a reviewer will not be present for the consequence, the very least it owes is a full account of the decision, delivered at the tempo the consequence requires.

What the account owes the patient and the clinician

So what would accountability actually require here, for this patient, in this case? Not more accuracy in the abstract, and not a better model. It would require that the decision arrive with its account attached, in a form the people affected can use in the time they actually have.

Concretely, four things. The clinician is owed the actual basis of the determination — the specific coverage rule and the specific necessity criterion that were applied, as they stood at the moment of decision, in a form precise enough to rebut immediately rather than a form letter that gestures at guidelines. There must be a fast human path — a route to a person with the authority to authorize, reachable on the clock the illness sets, not the leisurely clock of ordinary appeals. Time itself must be recognized as part of the stakes, so that a decision touching a life or a limb cannot be allowed to run out its harm in a queue. And there must be a preserved record — a Decision Receipt carrying the rule in force, the inputs relied on, and enough state to replay how the verdict was reached — so that the determination can be contested on its merits and, if it was wrong, seen to have been wrong. These are the public properties the rest of this series argues for: standing, contestability, provenance, and the plain requirement that an adverse determination account for itself to the party it lands on. A treatment that a clinician ordered and a patient needs cannot be adjudicated in the dark, on a timescale set for the convenience of the adjudicator, by a party who will not be there when the clock runs out.

The scenario above is illustrative — a composite drawn to show a pattern, not an account of any real person, company, or event.

— Dispatches · Summit Cognitive