The failure mode
NHS teams already have dashboards, lists and alerts. The harder problem is making sure that every intended next step is visible, owned and acted upon.
A follow-up that has not been booked is a particular kind of problem. The missing link may not present as an ordinary waiting list entry or visible breach, and an absence does not raise its hand. Finding it means comparing what was intended against what was booked, patient by patient.
Leva Health is researching whether software that can only take actions from a fixed, pre-authorised list can find those cases, prepare the permitted operational response, and route every clinical judgement to a clinician.
One case, worked through
A synthetic worked example. Northstar University Hospitals NHS Trust is fictional — no such Trust exists. P10042 is an invented reference, not a patient and not an NHS number, and there is deliberately no name: what should make this recognisable is that the workflow is precise, not that the record impersonates somebody. No real patient, staff or NHS data appears anywhere on this site.
Case P10042 starts from seven recorded conditions:
- The pathway is open and awaiting follow-up
- The priority band recorded on it is routine — recorded, never decided here
- The specialty is an ophthalmology follow-up
- A first outpatient attendance is recorded
- A clinician has recorded that a review is intended
- An interval for that review is recorded
- Responsibility for booking it sits with a named team
Nothing above is a judgement. Every line is something a person or a system wrote down, and the exception is a comparison of two of them:
A review was due by 10 July 2026, and as at 15 July 2026 no appointment could be matched to it.
That is the whole finding. It is not a statement that the patient came to harm, that anything was saved, or that anyone did something wrong — it is a deterministic comparison of an intended review against the bookings recorded against it, and the demonstrator shows every field it rests on beside it.
A second synthetic case in the same run has an appointment on the pathway that nothing links to the intended review. The system does not report that one as unbooked. It reports that an appointment exists and the record does not say whether it serves this review, declines to conclude, and routes it to a person. A system that cannot tell no appointment from an appointment I cannot match will eventually tell an operational team the first when it means the second.
Why a follow-up list is harder to keep than it sounds
- 01The intention and the booking live in different placesA clinician records that someone should be seen again in four months. Booking it is a separate task, in a separate system, done by a separate team. The intended review and the booking may be held in different records, and reliable reconciliation depends on local systems and processes.
- 02A list that is never wrong is also never looked atWhere the intended date is held as free text, or in a letter, or in a field that is optional, it can be right for most patients and quietly missing for some — and the ones it is missing for are invisible rather than flagged.
- 03Nobody owns the gapThe clinic has moved on and the booking team may have no reason to look again. Where reconciliation is incomplete, the patient may not appear on the worklist used by the responsible team.
- 04It may surface late, or when someone asksDepending on how a service is configured, a follow-up that has not happened may generate no alert, no breach and no queue position — so the first signal can be a patient asking when they will be seen.
What the published evidence says
Three sources. Two are historic ophthalmology research and one is a current national safety investigation. Each is shown with its date, the population it covers, what it supports, and what it does not.
This evidence establishes that the problem exists. It does not establish that Leva Health works in an NHS environment or improves patient outcomes.
External evidence Reviews that a clinician intended and that did not happen on time are a documented source of harm in NHS ophthalmology, and difficulty tracking patients through a pathway is a current national patient-safety concern.
The demonstration works through three situations, and each one is harder to see than the one before it.
Patients on planned follow-up are the group with no equivalent national tracking requirement: a national patient safety investigation found that the mandated 18-week referral to treatment target has resulted in newly referred patients being prioritised over follow-up patients, and reporting in July 2026 stated there is no requirement to record and publish data on follow-up waiting times.
A national patient safety investigation examined a discharge document that instructed both the hospital and the general practice to arrange the same follow-up appointment, and found the practice could not access the hospital's appointment booking systems and reported it would not have taken any action to book one.
A national patient safety investigation found no feedback mechanism in the trust's process to confirm an intended outpatient appointment had been booked and no audit process to capture missed bookings, with staff keeping personal notebooks as workarounds.
The first is a problem. The second is that problem where no report can see it, because every record involved is individually correct. The third is what happens after the finding — and a report has no actions, so it has nothing to follow up.
- Historic researchSurveillance 2015–16, published 2017
Royal College of Ophthalmologists / British Ophthalmic Surveillance Unit
Population and period: Every UK consultant ophthalmologist, reporting over twelve months from March 2015
Historic surveillance estimated that between 16 and 22 patients per month may have experienced preventable sight loss from health-service-initiated ophthalmic delays. The published study reported that 80% of cases involved delayed follow-up and the median delay was 22 weeks.
What it does not say. This is not a current figure. It describes a twelve-month surveillance period that ended a decade ago, and it says nothing about how often this happens today.
- Historic researchEpisodes from 2007–2012, published 2016
Moorfields Eye Hospital NHS Foundation Trust
Population and period: 145,234 ophthalmic patient episodes at a single centre
Historic evidence of the scale and complexity of reconciling ophthalmology follow-up records. The study reported that 75% of the seriously harmed patients examined had no appointment booked.
What it does not say. The 145,234 figure is the number of episodes that study reviewed. It is not a current backlog, it is not a Moorfields waiting list today, and it cannot be scaled up to a national number.
- CurrentInterim report, 18 August 2026
Health Services Safety Investigations Body
Population and period: NHS advice and guidance services and the NHS electronic referral service in England
Current evidence that interoperability, communication and pathway tracking problems can contribute to delay, missed diagnosis and patient harm.
What it does not say. This investigation studied neither ophthalmology follow-up appointments nor Leva Health. It is interim, so its findings may change, and it puts no number on anything.
What it does, and where it stops
The distinction the demonstrator enforces, in the terms a service would use. The system finds an administrative fact: a review expected by a recorded date, with no matching booking. Whether that patient should be seen sooner is a clinical judgement, and it is never the system's to make.
Synthetic demonstration In a synthetic scenario the demonstrator identifies a review that was expected by a recorded date and has no matching booking, prepares the permitted administrative response, and routes the cases it cannot resolve from the record to a named person rather than deciding them.
What it can observe
Facts a person recorded, compared with each other. Nothing inferred about the patient.
- The date a clinician recorded that the patient should next be reviewed.
- Today’s date, and how far past the recorded one it is.
- Whether an appointment is recorded against that review.
- Whether the field the date came from is flagged as out of date.
- Who recorded it, and when.
What it can prepare without asking
One thing, and it is deliberately unexciting.
- A coordination task for a named person: check whether an appointment has been booked against this recorded review, and record what is found.
What needs a named person to approve
Anything with an effect outside the organisation, or on a patient.
- Escalating a review that is beyond twice its recorded interval to a named manager.
- Offering an appointment slot, which needs two people and a human confirmation of clinical suitability — the system never infers that.
- Anything sent to a patient, which needs two people.
What it refuses outright
Not "asks first". Refuses, records the refusal, and routes it to a clinician.
- Deciding that a patient should be seen sooner.
- Setting or changing clinical priority, urgency or suitability.
- Anything touching diagnosis or treatment.
- Any action not in the agreed list, whatever it is and whoever asks.
There is one more case, and it is the one that matters most. Where the record does not say enough to be sure — an appointment exists but nothing links it to the review, or no interval was ever recorded — the system does not guess. It says it is not sure, and sends it to a person. A confident wrong alert about a patient who already has an appointment costs more than the alert is worth.
What we do not know
Longer than the list of what we do. That is the honest shape of the work at this stage.
Whether the intended review date is reliably recorded at all
Everything here rests on that date existing and being right. In a real service it may be missing, stale, held somewhere the system cannot read, or recorded in a way that varies by clinician. We do not know, and it is the first thing discovery has to answer.
Whether an intended review can be matched to an actual booking
The demonstrator compares a recorded intention with a booking flag. A real reconciliation has to cope with appointments that exist but are for the wrong thing, were cancelled, or sit on another pathway — and it will get some of them wrong.
What rate of false alerts a service would tolerate
A list that is wrong often enough stops being read, and then it is worse than nothing because it looks like cover. Nobody has told us where that line sits.
Whether this generalises beyond ophthalmology
Ophthalmology is where the published evidence is. That is a reason to start there, not evidence that the same approach helps anywhere else.
Whether any of it changes what happens to patients
Not demonstrated, not modelled and not claimed. It would need a prospective evaluation in a real service with a declared comparator.
No NHS deployment, integration, partner, customer, pilot, certification, saving or patient outcome. None of these is claimed and none exists.
Tell us where this is wrong
We are looking for people who run or support follow-up in an NHS service, and for the specialists who would have to be satisfied before any of this went near a patient: clinical safety, information governance, evaluation and health economics.
The three questions we most want answered are the ones an outsider cannot guess. Which field records when the patient should next be reviewed, and how reliably is it filled in? If a system noticed on day 15 that a step due on day 14 had not happened, who should it tell and what should it be allowed to do without asking? And how do you find out today?
If your service already solves this, please tell us how. Which system holds the intended review date, what reconciles it against bookings, who owns the exception when the two disagree, and what it took to get there. An answer of “we do not have this problem” is a genuinely useful outcome for us: it would mean the thing we are building is not the thing worth building, and we would rather learn that from you now than after two years of engineering.
Reviewing the demonstrator takes about 25 minutes and is by individual invitation. It is not a sales conversation and no endorsement is implied by taking part.