Discharge Is a Clinical Decision Too — and the One Medicine Barely Teaches

Nobody has ever held a morbidity meeting about a cautious admission. The patient who got the extra night on a monitor, the bloods that came back boringly normal, the bed that was arguably wasted — none of that ends up on a slide with the lights dimmed and a room full of colleagues working out what went wrong. Over-caution is invisible. It costs money and beds and someone's evening, and it is never, ever the thing that gets reviewed.

Discharge is different. Every emergency clinician carries a small private gallery of the ones they sent home. The one who came back worse. The one who didn't come back at all, and you only found out months later, sideways. Ask any experienced doctor about a case that still sits with them and the odds are good it walked out of the department on its own two feet, with a leaflet and a vague instruction, into a trajectory nobody was watching. That asymmetry — admit-and-be-reviewed versus discharge-and-find-out — should tell us something about where the real risk lives. Mostly, it doesn't. We treat the decision to send someone home as the absence of a decision: the default that happens when nothing else does. It is, in fact, one of the highest-stakes clinical acts in the building.

Why discharge is the high-wire decision

Admission buys one thing above all others, and it isn't treatment. It's observation. The patient enters a system that will, however imperfectly, keep looking at them — repeat the observations, notice the trend, escalate if the picture turns. Discharge ends that. The instant someone leaves, the monitoring doesn't transfer to another professional; it transfers to the patient and whoever happens to be at home, who are, almost by definition, the people least equipped to do it. They don't know which change matters. They don't know how fast is too fast. They have no baseline to compare against except how they felt yesterday, and yesterday they came to hospital.

So the decision to discharge is really a prediction. Not a statement about how the patient is now — that part is usually the easy bit — but a bet on the trajectory: that whatever is happening will get better, or at least not get dangerously worse, in the window before anyone looks again. And predictions about early-course illness are genuinely hard, because early is when the signal is weakest. The serious thing that presents non-specifically often presents early, before it has organised itself into something a clinician would recognise on sight. You are forecasting the weather from the first cloud.

That is the high-wire part. The clinician is asked to act on a probability — reasonably confident, never certain — and then remove the safety net of continued observation at precisely the moment the illness is least legible. The scrutiny is asymmetric on top of that. An admission is reviewed within hours by the next team, the ward round, the consultant post-take. A discharge is reviewed by nobody, unless it goes wrong, in which case it is reviewed by everybody, in retrospect, with the answer already known. We audit the cautious decision in real time and the bold one only after the fact. No wonder the gallery fills up.

What real safety-netting contains

The thing that converts that bet into a managed risk has an unglamorous name: safety-netting. And most of what passes for it isn't. "Come back if you're worried" is not safety-netting. It's a verbal shrug — reassurance shaped like instruction, an offload of responsibility onto a worried person's judgement about a situation they have no framework to judge. Real safety-netting is a different object, and it has parts.

It names specific triggers, in the concrete. Not "if you get worse" but the particular changes that would matter for this particular presentation — described in terms a frightened person at home can actually recognise without a medical education. The whole point is to convert "worse", which is uninterpretable, into a short list of observable events. Vague deterioration is invisible to the person living inside it; named deterioration is something they can spot and act on.

It carries a time horizon. What should have happened by when — the expected course, so that a departure from it becomes legible as a departure. "This should start settling within a couple of days" does real work, because it hands the patient a yardstick. Without a timeframe, there is no such thing as "not improving as expected", because nothing was ever said about what to expect. The clock is half the information.

And it confirms the person can actually act on it. This is the part most often skipped, because it requires the clinician to think past the cubicle door. Is there transport, if returning becomes necessary at two in the morning? Is there someone at home, or does this person live alone? Did any of the words land, through the pain, the fear, the language barrier, the eight hours already spent in a waiting room? A perfect plan delivered to someone who cannot enact it is not a plan. It is a document.

None of this is exotic. It is just specific — and specificity, here as everywhere in medicine, is where the expertise hides. Anyone can say "come back if worse". It takes clinical judgement to know, for this presentation in this person, exactly which "worse" matters and how long the clock should run.

Why it degrades to ritual

If proper safety-netting is so clearly the thing that makes discharge safe, why is so much of it hollow? Not because clinicians don't know better. Because the leaving conversation is the most squeezable minute in the entire encounter, and everything in the environment squeezes it.

Throughput pressure does the bulk of the damage. The department is full, the corridor is filling, and the patient who is medically ready to leave has, in the brutal economy of the shift, already had their share. The diagnostic work felt like the real medicine; the discharge conversation feels like admin standing between you and the next sick person. So it compresses. The triggers go generic, the timeframe drops out, the check on whether any of it was understood never happens. The clinician hasn't decided safety-netting doesn't matter. They've been placed in a system that prices their attention by the minute and have spent it where the pressure was loudest.

Documentation then quietly finishes the job. "Safety-netting advice given" is a phrase that satisfies an audit and informs absolutely no one. It records that a conversation happened without recording anything that was in it — perfectly defensible, completely empty. Tick the box and the system is content, whether the patient left with a precise list of triggers or with nothing but the word "worried". The phrase that proves the work was done becomes a substitute for doing it.

And underneath both sits a comfortable fiction: that handing over a leaflet equals comprehension. It doesn't. A leaflet is a fine aide-mémoire for a conversation that happened; it is no replacement for one that didn't. The patient who couldn't take in the verbal version under stress will not absorb the printed version in the car park. Information transferred is not information received, and the gap between the two is exactly where the discharged patient gets lost.

Discharge as a system property

Pin the whole failure on the individual clinician's final five minutes and you'll fix nothing, because the quality of a discharge is mostly determined before that conversation ever starts. It is a system property.

A good handover home is written and verbal, pitched at the person's actual language and health literacy rather than the clinician's — which means the system has to make the better version the easy one to produce, not a feat of individual virtue performed against the clock. And departments already hold the feedback they need to know whether they're getting this right, if they choose to read it: re-attendance data. The patient who returns within days is not simply a statistic to be minimised or, worse, an irritation. They are a signal about the quality of the discharge that sent them out — sometimes about the illness, often about whether the leaving conversation actually equipped anyone. A department that treats its bounce-backs as failures of the patient learns nothing. One that treats them as feedback on its own disposition decisions is the one that improves.

This is also where the limits of the digital answer come into focus. Software can carry pieces of this well — a structured discharge summary that won't let you leave the timeframe blank, plain-language materials, a prompt that the trigger list is missing. Those are real gains, and worth having. But the parts that matter most are the parts a tool can't carry: whether the words landed, whether this specific person can act on the plan, whether the worried look across the cubicle means the message didn't get through. A discharge tool that produces an immaculate document while the human conversation withers has optimised the auditable half and abandoned the half that keeps people alive. Which is, depressingly, the easiest kind of tool to build and the easiest kind to sell.

What this means

The discharge conversation is the last clinical act of the encounter, and we have somehow agreed to treat it as the bit after the clinical acts are over. The diagnostic reasoning gets the glory; the disposition gets the leftover minute. But sending someone home ends observation, hands monitoring to the people least able to do it, and stakes everything on a prediction made at the moment the illness is hardest to read. That is not the end of the clinical decision. On the days it goes wrong, it was the clinical decision — the only one that mattered.

It deserves the same rigour as the first ten minutes, and the same intellectual honesty. Name the triggers. Set the clock. Check the person can actually act. Write it so the next clinician — and the patient — can use it, and read the re-attendance data as the verdict it is. The cautious admission will never trouble a morbidity meeting. The hollow discharge will, eventually, sit in someone's private gallery, where the cases that teach us the most are always the ones we'd most like to forget.