Shift handover

Why do the worst mistakes in a hospital or plant cluster around the moment one shift hands over to the next?

Handover looks like the safest moment in the day. Two competent people are present instead of one, the notes are open, and the entire point of the exercise is to check that nothing is dropped. Yet reviews of serious incidents in hospitals, refineries and control rooms keep finding communication failures at or around shift change among the most common contributing factors, and both aviation and medicine eventually wrote formal protocols for it.

So the puzzle is not "people are careless at 7am". It is stranger: why does doubling the competent people present, at a moment of full documentation, make things more dangerous rather than less?

Ask first what the outgoing worker actually holds. Some of it is plainly factual — readings, doses, what was done at what time — and all of that is in the record. But most of what makes them safe is not factual in that sense. They hold a direction: this patient looked worse than the numbers at four o'clock and has been drifting since. They hold a hypothesis under test, and the three explanations they have already quietly eliminated. They hold anticipations — if this alarm sounds, worry, but if that one does, ignore it, it has been spurious all night.

Now notice what a written record is shaped to store. It stores states, at instants, and what was done. What it does not naturally hold is a trajectory, an unfinished line of reasoning, or the negative knowledge of what has already been ruled out and why — often the most expensive knowledge in the room, because it was bought with hours the incoming worker does not have.

So handover is not the transfer of a document. It is the attempt to reconstruct a mental model inside a second person's head, in fifteen minutes, using speech.

Stack the conditions on that attempt. The outgoing worker is at the end of a long shift and wants to go home. Both people are being interrupted. And the deepest asymmetry: the receiver can only ask about what they know is missing. If the outgoing nurse never mentions that the family is distressed and asking about a transfer, the incoming nurse cannot ask, because nothing marks the absence. Absent knowledge leaves no hole to point at.

Which is why the response in both aviation and medicine was to standardise the form rather than trust the participants. If the order of the conversation is fixed, an omission becomes visible: a slot went unfilled. SBAR — Situation, Background, Assessment, Recommendation — came into healthcare from naval practice and does exactly this, and note the third slot: Assessment is a compulsory demand for the speaker's interpretation, not merely the readings. Aviation attacked the other flank too, protecting the moments when a model must be maintained: the sterile cockpit rule bars non-essential conversation below ten thousand feet.

Does formalising it work? Partly, and the evidence is real. When a group of paediatric hospitals adopted the structured I-PASS handoff bundle, medical errors fell by roughly a quarter and preventable adverse events by around a third.

But hold the honest line. Such interventions reduce omissions. They do not transfer situational awareness, because the thing being transferred is partly unavailable to the speaker too — the outgoing worker's unease about a patient is often a summary of cues they cannot list. A protocol can guarantee every field is spoken. It cannot guarantee the receiver ends up worried about the right person.

And there is a cost that the protocols create rather than remove. A very good handover installs the outgoing worker's model faithfully — including its errors. If they had settled on the wrong explanation, a fluent briefing hands the incoming worker that explanation pre-formed, and the fresh pair of eyes, which is the one genuine safety benefit of shift change, is spent before it is used.

The tempting misreading is that handover fails through laziness or poor documentation, and that better records would fix it. Better records help, but with the part that was never the problem. The bottleneck is that awareness is constructed, not stored: the outgoing worker built theirs over eight hours of continuous observation, and no compression of that into speech survives intact.

That reframes what a good protocol is for. It is not a higher-bandwidth channel; it is a device for making omissions detectable and forcing interpretation into the open. The read-back — the receiver stating the model in their own words — earns its place for the same reason: it is the only cheap way to discover that the model arrived wrong.

Two qualifications. Handover risk is hard to measure, because shift change is also when a deteriorating situation has had the most time to develop, so some of what looks like handover-caused harm is harm that merely surfaced then. And the fix is not fewer handovers: rosters designed to minimise them push shifts longer, and fatigue is a well-evidenced hazard in its own right.

• Why read-back, borrowed from radio procedure, catches more than any amount of extra documentation.
• Whether overlapping shifts, where both workers are present and working, buy more than a longer briefing does.

Every term the collection defines is gathered in the glossary.