The Safest-Looking Hospitals Are Sometimes the Ones Learning the Least

Here is a trap that catches careful, well-run hospitals.

You decide, sensibly, to track patient safety. You start counting adverse events: the cases where something went wrong and a patient was harmed. The number is low. It stays low. Quarter after quarter, the report comes back clean.

You conclude that you are safe.

But look at what you actually built. You built a metric where silence looks like success, and then you gave your team every reason to stay silent. Nobody has to be dishonest for this to work. The measure does the work on its own.

The iceberg you aren't counting

Most of what happens in a busy veterinary hospital never becomes an adverse event. It becomes a near miss.

The wrong drug drawn up and caught before it's given. The decimal point spotted a second before the dose goes in. The allergy noticed on the record just in time. The surgical site marked, then re-checked, then found to be the wrong limb. The patient whose IV line was almost connected to the wrong bag, at 2 a.m., by someone on their eleventh hour.

In each of those cases the system that was supposed to protect the patient did partially fail, and then something caught it. Sometimes that something was a designed defense: a checklist, a second signature, a hard stop. Very often it was a person who happened to be paying attention at the right moment.

That distinction is the whole game. When a near miss is caught by design, you learned that your defenses work. When it is caught by a person who happened to glance down, you learned that you got lucky, and you have no idea how many times the same setup will occur before nobody glances.

Either way, the near miss just handed you a free, complete, no-consequence map of exactly where your system is fragile. No patient harmed. No difficult client conversation. No claim. No regulatory exposure. The most valuable safety information you will ever receive, and it arrives at zero cost.

Most hospitals throw it away within the hour.

Aviation figured this out; medicine is still catching up

The comparison to aviation is made so often it has gone slightly stale, but the specific mechanism is worth being precise about, because the popular version of the story is usually wrong.

Aviation's safety record was not built primarily on crash investigation. Crashes are catastrophic, rare, and, crucially, statistically useless for prediction precisely because they are rare. What changed the industry was the systematic capture of the events where nothing happened: the altitude deviation, the near-collision on approach, the fatigue-driven checklist skip. Confidential, non-punitive reporting systems collected these in volume, and volume is what makes patterns visible.

The key design choice was decoupling the report from consequences for the reporter. Not decoupling it from accountability, but from punishment. Pilots reported because reporting was safe, routine, and demonstrably led to changes they could see.

Human medicine adopted versions of this decades ago. Veterinary medicine largely has not, and the published record reflects it. Studies of clinical incident reporting across UK first-opinion practice show that structured, voluntary reporting captures substantial volumes of events once a system exists to receive them. Which is another way of saying the events were always there, and were simply going unrecorded. A 2025 analysis of first-opinion electronic health records identified more than a thousand medication errors in routine clinical records, roughly a third of which caused some degree of harm. Those errors were sitting inside data hospitals already had. Nobody was looking.

Why the near miss evaporates

The near miss doesn't get suppressed. It gets dissolved, and understanding the difference matters, because the fixes are different.

There is nowhere to put it. Most hospitals have an incident form designed around harm. It asks what happened to the patient. When the answer is "nothing, I caught it," the form doesn't fit, and the person correctly concludes it isn't meant for them.

There is no time. The catch happens at the busiest possible moment, which is usually why the error was set up in the first place. The window in which someone might log it is the same window in which four other things need doing. Sixty seconds later the moment has passed.

It doesn't feel like an event. This is the quiet one. To the person who caught it, nothing happened. The system worked. They fixed it. Reporting a non-event feels self-important, or like tattling on a colleague who made the original slip.

And underneath all of it: a worry about what raising it will look like. Even in a hospital where nobody would actually be punished, the reporter can't be certain of that in advance. Emerging veterinary work points at professional identity as the mechanism here: the sense that admitting involvement in error is admitting you aren't a good clinician. If your self-concept is on the line, a quiet "phew" is a much cheaper option than a report.

So the lesson evaporates. Not because people don't care. Because caring wasn't enough, and nothing in the environment made the alternative easy.

The two numbers, and what each one actually tells you

Set them side by side.

Your adverse event count tells you how often the holes lined up all the way through, how often you were unlucky. It is a real number and it matters. But it is small, it is lagging, and it moves for reasons that have as much to do with case mix and chance as with the quality of your systems. A hospital can have a clean year and be one staffing change away from a bad one.

Your captured near-miss count tells you something different: how much of the fragility in your system you can actually see. It is not a measure of how dangerous your hospital is. It is a measure of how much you are learning.

Which produces a counterintuitive rule that is worth stating plainly, because it reverses the instinct of every leader who has ever looked at a safety report:

A rising near-miss count is usually good news.

It almost never means your hospital became more dangerous last quarter. It means people started telling you things. The fragility was always there; the visibility is new.

This is also why near-miss counts make terrible performance targets and excellent diagnostic instruments. The moment a number becomes something a department is judged on, it stops describing reality. If you take one structural lesson from this piece, let it be that: never tie a near-miss count to anyone's evaluation.

What to do this week

You don't need a platform to start. You need a place and a habit.

Make one thirty-second channel. A card by the drug cabinet, a shared note, a standing line in the shift handover. If logging a catch takes longer than the catch itself took, it won't happen.

Ask for catches by name, in rounds. "Did anyone catch anything this week?" is a different question from "did anything go wrong," and it gets a different answer. It also frames the reporter as the person who worked, not the person who erred.

Separate the catch from the person who set it up. Log the conditions, such as look-alike vials, an interrupted count, or a handoff at shift change. Not the initials.

Feed one back, visibly, within two weeks. Pick a single near miss, change one concrete thing, and tell the team you did it because of their report. The first closed loop is worth more than the next fifty forms.

The count of adverse events tells you how often you got unlucky. The count of near misses you actually capture tells you how well you're learning. One of those is a measure of safety. The other is mostly a measure of luck.

A question for your team

When someone catches a problem just in time this week, where does that catch go? Into a system you'll learn from, or into thin air?