Surgical volume-outcome relationship

Why does the same operation carry less risk in a hospital that does little else?

Take a pancreaticoduodenectomy — the Whipple operation. Done at a hospital that performs two or three a year, and done at one that performs a hundred, it is nominally the same procedure, by surgeons with the same qualification, using the same instruments. Yet the mortality differs by several times over. Birkmeyer and colleagues, working through Medicare data in 2002, found roughly sixteen per cent at the lowest-volume hospitals against roughly four at the highest.

The instinctive explanation is that the high-volume surgeon is simply better with their hands — practice makes perfect. Plausible. But before accepting it, notice that it makes a prediction we can go and check, and that the checking pulls the answer somewhere unexpected.

If manual skill were the mechanism, the difference should show up during the operation. High-volume centres should have fewer complications: fewer leaking anastomoses, less bleeding, cleaner margins. That is exactly what the "practice makes perfect" story asserts.

Now look at what was found. Ghaferi, Birkmeyer and Dimick reported in 2009 that complication rates across American hospitals were remarkably similar. Patients at low-volume and high-volume centres suffered adverse events at close to the same rate. What differed, and differed sharply, was what happened next. At the better hospitals, a patient who developed a complication was far more likely to survive it. They called the gap failure to rescue.

Sit with how much that rearranges. The scarce commodity is not a pair of hands during four hours in theatre. It is what the institution does across the following week, when a patient is slightly tachycardic on the third day and looks otherwise unremarkable.

So ask what a specialised institution has that a general one does not, at two in the morning on day three. The ward nurse has seen that particular presentation before and knows it is not ordinary post-operative malaise. The on-call registrar does not have to reason from first principles about an unfamiliar anatomy. There is a CT protocol for it, and a radiologist who reads three of these a week rather than three a year. Interventional radiology can drain a collection tonight. The anaesthetist knows this operation's physiology. There is a route back to theatre that does not require anyone to be persuaded.

Every one of those is a property of the system, not of a person — and each is a form of specialisation, in the ordinary economic sense: repetition permits division of labour, and division of labour builds capabilities that a generalist cannot hold. Volume is not the cause. Volume is what makes the specialisation affordable, because a hospital cannot justify a dedicated pathway, a trained team and a protocol for an operation it does three times a year.

That reframing makes better predictions. It says the volume effect should be large for complex operations with dangerous, subtle complications, and small for common ones where any competent unit already has the recognition and the rescue capacity. And that is the pattern: the relationship is strong for pancreatic and oesophageal resection and aortic surgery, weak or undetectable for cholecystectomy and hernia repair. It also predicts that surgeon volume and hospital volume should both matter, but not identically — and in the 2003 follow-up work, hospital volume's effect was substantially explained by surgeon volume for some procedures and not others.

One caution that has been live since the very first paper on this, by Luft, Bunker and Enthoven in 1979. The correlation runs both ways. Hospitals with good outcomes attract referrals, so high volume may be partly a consequence of quality rather than a cause — selective referral. Both effects appear to be real, and separating them requires designs that exploit something like distance to the nearest centre as a natural experiment. The causal component holds up in that work, but the honest statement is "substantially causal", not "purely causal".

Three refinements are worth carrying away.

The first is that "practice makes perfect" is not so much wrong as aimed at the wrong practice. What repetition builds is not primarily dexterity but recognition — of the deviation from normal recovery, in a person who has seen a hundred normal recoveries. That is a perceptual skill distributed across a whole team, and it is why the effect survives even when the individual surgeon is held constant.

The second is a caution about using volume as policy. Volume is a proxy, and a crude one: a low-volume hospital may have excellent outcomes and a high-volume one poor ones. Minimum-volume standards, like those the Leapfrog Group publishes and the 2015 Take the Volume Pledge adopted, are defensible mainly because volume is easy to measure reliably while risk-adjusted outcomes at small numbers are statistically noisy. Where outcomes can be measured well, measuring them directly is better.

The third is that centralisation has costs the mortality figures do not show. Concentrating complex surgery means longer journeys, families further from the ward, and rural units losing the case mix and the staff that keep their emergency capability alive. There is also a real question of where the boundary sits between complex work that should be concentrated and routine work that should not, and reasonable people place it differently.

• How instrumental-variable studies use travel distance to separate selective referral from a genuine causal effect.
• Whether failure-to-rescue rates could replace volume as the standard a regulator actually measures.
• The equity arithmetic of centralisation — how many extra deaths from delayed access offset how many prevented in theatre.

Every term the collection defines is gathered in the glossary.