The patient will see you now

For a long time a patient’s opinion of a hospital was a courtesy, collected politely and filed. That is changing. England’s ten-year health plan proposed “patient power payments”. Patients would decide whether part of what a hospital is paid for their care is handed over or diverted to a regional improvement fund. A pilot in some gynaecology services is due to run in 2026/27, with recommendations to follow in the NHS payment rules for 2028/29.

Elsewhere the stick is publicity rather than money. In Switzerland, most of this year’s audits under the hospital quality contract go to hospitals whose national results stood out for the wrong reasons. Those that do not improve can end up publicly marked as non-compliant. In Spain, from 28 December, large companies selling to consumers must measure satisfaction and publish an audited evaluation of their service. Private hospital groups are among them. Germany has written patient surveys into its national quality assurance, starting with heart-catheter procedures. Italy’s regions are spreading continuous surveys of inpatients, one region at a time.

The direction is right, because patients see things nobody else records. Tom Reader and Alex Gillespie of the London School of Economics studied 59 British hospitals. They set 23,287 patient surveys and 2,017 complaints, some 2.5m words, against 49,302 staff surveys and 242,859 incident reports filed by staff. The more serious the error and neglect described in patients’ complaints, the higher a hospital’s death rate tended to be, over and above everything the staff data showed. Patients, the authors concluded, are a valid source of safety data in their own right.

The trouble starts when the measure carries money. Picker, a charity that researches patient experience, has set out the risks of the English plan. Take the Friends and Family Test, the NHS’s standard feedback question. It is anonymous and does not limit anyone to one answer. Were payments to hang on it, a coordinated campaign could push a hospital’s money down. A patient angry with a family doctor may take it out on the gynaecology ward. Results depend on how and when a survey is run, so providers could be rewarded for their methods rather than their care. And evidence that patient-reported measures work in pay-for-performance schemes is thin. Low response rates make matters worse: Switzerland’s national survey drew 28% last time.

Words resist these distortions better than scores. A tick-box rating can be moved by a campaign, or by sending the survey on a better day. A complaint that describes a missed dose on a named ward at a named hour is harder to fake and harder to dismiss. It also says what to fix, which a score cannot.

Until recently hospitals had an excuse for leaving the words unread. Reading thousands of comments takes a team, and complaints departments were built to close cases, not to find patterns. Language models have removed most of that cost. A hospital can now read every comment and complaint as it arrives and sort it by ward, theme and severity. The alarming ones can be in front of a clinician the same day. The judgement still has to be human. The reading no longer does.

Regulators will keep counting scores, because scores are easy to compare and to pay against. Hospitals should report them, and then read what their patients actually wrote. The scores say how a hospital is judged. The words say what to fix, and, if the British evidence holds, where the danger lies.