Lucy Letby And The Evidence That Isn’t There

On August 5th the Crown Prosecution Service confirmed something which ought to have detonated across the front pages and instead surfaced in the trade press. A prosecutor independent of the original decision had re-examined six allegations of attempted murder against Lucy Letby, relating to six further babies, and concluded there was no case to bring.

The families had requested the review under the Victims' Right to Review scheme.

They lost. It was the second refusal.

In January 2026 the CPS had already declined to charge eleven proposed offences involving nine babies where Letby trained. The evidential test was not met.

Cheshire Constabulary had worked through thousands of neonatal records across a window running from 2012 to 2016, and produced a file which the prosecuting authority examined twice and put down twice.

None of which disturbs a single existing verdict. The convicted cases may simply have been evidentially stronger, perhaps.

What the refusals supply is something more useful than a rebuttal: a comparison group. A second collection of clinical events associated with the same nurse was assembled under conditions of maximum suspicion, by officers who already believed they were investigating a serial killer. Handed to prosecutors who already had fifteen convictions in the bank.

It failed. Twice.

Which invites a question nobody in authority has answered: what evidential feature separates the collapses which produced whole-life orders from the collapses which cannot clear the threshold for a charge?

If the answer is materially different evidence, the state should identify it. If the true distinction amounts to nothing more than which babies were examined first, something has gone badly wrong.

What The Crown Actually Alleged

The convictions cover a thirteen-month period from June 2015. Seven murders; seven attempted murders; one baby attacked twice; fifteen whole-life orders in total. The alleged methods were extremely different:

1. Air injected into the bloodstream.
2. Air or excessive milk delivered through feeding tubes.
3. Insulin poisoning.
4. Physical assault, including inflicted trauma.

A single offender, on a single small unit, allegedly rotating between techniques over the course of a year.

Nobody witnessed an attack. There is no recording, no forensic residue, and no moment at which anybody observed Letby doing anything to a baby which she should not have been doing.

Every alleged act was reconstructed backwards, months or years later, from the clinical record of a collapse. A retired paediatrician reviewed clinical notes and formed opinions about which deteriorations bore the signature of deliberate harm. Those opinions became the medical case.

Once the collapses were characterised as attacks, presence acquired its meaning. Consultants had drawn up a chart of suspicious events and the nurses on duty for each. Letby appeared beside every one. The chart was shown to the jury on the first day of trial and again on the last, and prosecuting counsel described her as the common denominator.

Around that skeleton the Crown assembled corroboration.

• Handwritten notes recovered from her home.
• Facebook searches for the parents of babies who had died.
• Handover sheets kept rather than shredded.
• The two insulin cases, treated as proof somebody on the unit was deliberately harming infants, which in turn made deliberate harm elsewhere on the unit more credible.

If this seems weak and circumstantial, well, yes. It is as it appears.

The structure is cumulative by design, and English law permits it. The trial judge allowed the jury to consider the counts together, and the Court of Appeal upheld his directions in its judgment of July 2nd 2024.

The doctrine is neither novel nor improper. Its reliability rests on a condition which receives far less attention than it deserves: the incidents lending weight to each other must have been correctly identified as crimes in the first place.

Where the classification is contested, mutual support becomes mutual dependency. Incident A acquires its sinister character from a pattern established by B, C, and D. B acquires its character from A, C, and D.

The aggregate carries a force which no individual component possesses.

Possible, Probable, and Proved

The Crown's proposition was not that a nurse harmed a child. Individual acts of cruelty by healthcare staff are rare but documented, and juries convict on them without much difficulty.

The proposition here was considerably more... ambitious.

One woman with no prior history:

1. Became a covert infant killer of exceptional prolificacy;
2. Crossed the same enormous moral threshold repeatedly over thirteen months;
3. Varied her methods;
4. Produced no direct evidence of any act;
5. Generated no established motive, and;
6. Continued while colleagues, doctors, and management circled ever closer.

Such a person can exist. Beverley Allitt existed.

Rarity is not impossibility. What rarity does affect is how plausible ambiguous evidence is.

An unusual event observed against a very improbable hypothesis (where ordinary alternatives remain available) moves the needle less than the same evidence would against a commonplace one.

And in a neonatal unit the ordinary alternatives are abundant: extreme prematurity, sepsis, respiratory collapse, line infections, feeding complications, medication errors, misplaced tubes, delayed escalation, inadequate senior cover, and the plain unpredictability of babies born months too early.

The serial-killer hypothesis has one considerable advantage over those alternatives. It explains everything at once.

A single cause accounting for a dozen heterogeneous events is intellectually satisfying in a way an untidy list of unrelated clinical failures is not.

Elegance is not evidence, and a hypothesis does not earn a discount on proof by being tidier than the truth.

Twenty-Five Events Out of How Many?

The chart which anchored the prosecution has attracted sustained criticism from people with the relevant expertise. Peter Green, a professor of statistics and former president of the Royal Statistical Society, pointed out how the chart depicts twenty-five events and omits the ones occurring when Letby was not on duty.

Richard Gill, the statistician whose work helped exonerate Lucia de Berk in the Netherlands, puts the objection more bluntly: the events are called suspicious because she was there.

The technical name for the error is the Texas sharpshooter fallacy, and academic commentary has applied it to this chart repeatedly. Fire at a barn door, paint a target around the tightest cluster of holes, and marvel at the accuracy.

Letby's column is full of crosses because a collapse at which she was absent could never become a charge, and only charges appear on the chart.

Gill's published analysis records two further details worth more attention than they have received.

1. One charged event had to be dropped early in the trial because she had not been present after all, meaning the police spreadsheet contained at least one event classified as a deliberate attack on the strength of a presence which did not exist.
2. And the prosecution's lead expert identified roughly twenty additional suspicious events for which no charges followed. Defence counsel asked for a specification of how suspicious events had been defined and identified. The judge refused, on the entirely orthodox basis that each count falls to be assessed on its own medical evidence.

The Royal Statistical Society had anticipated all of this.

Its 2022 report Healthcare serial killer or coincidence? opens by identifying what makes these prosecutions structurally different from ordinary murder: investigators face uncertainty not merely about who did it, but about whether anybody did anything.

The report warns specifically about which deaths get counted, since investigators naturally focus on events already flagged as unexpected, and thereby build the very cluster they later treat as evidence. It recommends experts assessing such cases be shielded from the suspect's identity until they have reached their conclusions.

The document was sent to both prosecution and defence before the trial.

A companion paper by Gill, Fenton, and Lagnado had already traced identical failures through the Lucia de Berk and Ben Geen prosecutions.

Establishing A Crime Before ID-ing A Criminal

Every count required the Crown to prove two separate things: a) a baby was deliberately harmed, and b) Letby was the person who harmed her. The second is unreachable without the first, and uncertainty about whether a murder occurred cannot be cured by confidence about who was standing nearby.

Clinicians at the time encountered deteriorations they found distressing and unexplained. Reviews were commissioned.

In July 2016 the trust asked the Royal College of Paediatrics and Child Health to examine the mortality rate, and the resulting review found no definitive explanation for the increase.

Only afterwards, once police were involved, did a prosecution expert working through the records begin identifying particular collapses as deliberate attacks. Later still, other specialists disputed those characterisations.

In February 2025 a panel of fourteen international specialists convened by Dr Shoo Lee concluded there had been no murders at all, attributing the deaths and injuries to natural causes or poor medical care.

Lee's own 1989 paper on air embolism had been relied on by the prosecution, and he says his findings were misread in court.

Lawyers for the bereaved families dismissed the panel as a rehash of a defence case two juries had already rejected. The senior investigating officer described the criticism as ill-informed and based on partial knowledge of the evidence.

Adopting the panel's conclusions is unnecessary and would be presumptuous. Their significance lies in a different direction.

Specialists of that standing publicly disputing whether any crime occurred establishes the actus reus remains scientifically arguable across a substantial portion of the indictment.

And the honest answer to the question of when each event became a crime is: at the point somebody decided to look for one.

Air, Milk, and Insulin

Air embolism supplied the explanation for several collapses, supported by descriptions of unusual skin discolouration. The Court of Appeal considered Lee's evidence on this and rejected the suggestion prosecution experts had made the specific error attributed to them.

The judgment acknowledges some findings as consistent with air embolism without being diagnostic of it, and claims they were properly assessed alongside the surrounding circumstantial evidence. Findings which do not independently establish an attack derive their force from other allegations, whose own status depends on the same reasoning.

The insulin counts were treated as the hardest evidence available, resting on laboratory readings showing high insulin with low C-peptide.

Five experts subsequently told the Telegraph the immunoassays could not bear that weight, since antibody interference produces false readings and confirmatory testing is routine before conclusions are drawn. They also raised the possibility of expert witnesses and the judge conflating results which were unquantifiable with results which were absent.

The consultant who received the Child F result in 2015 told the Thirlwall Inquiry she had assumed a laboratory error, on the grounds deliberate poisoning struck her as absurd.

The Court of Appeal treated deliberate insulin administration as compelling proof somebody on the unit was harming babies, and the finding then informed its assessment of other charges.

Loosen the criminal insulin anchor and the chain of guilty evidence drifts.

Even taken at its highest, an abnormal insulin result establishes a contaminated bag. It does not establish who contaminated it, or when, or whether the contamination was deliberate rather than a dispensing failure.

And this is where the Letby case finds its enormous canyon: the lack of hard evidence which is overcome via contestable inference. Not what was found, but what probably happened.

Every alleged mechanism in this case has an innocent twin. Air enters lines by accident. Feeding errors happen. Insulin ends up in the wrong bag through ordinary pharmacy mistakes. The physical trace is identical whichever occurred.

A Unit Which Could Not Staff Itself

Comparing the prosecution hypothesis against a blank space is how these cases go wrong. The competing explanation has nothing to do with the competence of prosecution doctors. These babies were being treated in an environment perfectly capable of producing, unaided, the abnormal events later reconstructed as crimes.

The RCPCH review found significant gaps in medical and nursing rotas, poor decision-making, and insufficient senior cover.

Medical staffing was adequate for a special care unit and inadequate for the level 2 local neonatal unit the Countess was operating, taking babies from 27 weeks and providing short-term intensive care.

Investigation reports on the deaths themselves showed a pattern of insufficient senior cover and a reluctance to seek advice. Transport services across Cheshire and Merseyside were under-resourced.

In its opening statement to the Thirlwall Inquiry, the College recorded a 21 per cent staffing deficit on the unit at the time the deaths occurred.

Letby herself told the jury the unit had been at overcapacity, caring for eighteen babies in sixteen places, and she had described staffing to a colleague at the time as completely unsafe.

Coming from the defendant, that is self-serving.

It is also consistent with everything the College found independently.

The mortality figures are worth seeing whole.

The unit handled roughly 400 babies a year. A rise from three to eight is striking in a graph and thin as a dataset. Small-number series produce clusters routinely, which is the entire subject of the RSS report.

The prosecution's answer to the whole picture was that deaths returned to normal once Letby was removed.

Two things happened in July 2016:

1. Letby was moved to clerical duties, and
2. The unit stopped accepting babies born before 32 weeks, suspending intensive care and reverting to level 1.

Attributing the subsequent improvement to the first change alone requires ignoring the second, which involved sending the sickest babies elsewhere.

There is a further detail from the College's evidence which sits awkwardly with the settled account.

At the time of the deaths, the two consultants who later took their concerns to the police did not consider Letby's presence at every incident significant, and one wondered whether something was being missed in the case reviews.

Independent Anchors in Comparable Cases

The standard reply to any complaint about thin evidence is straightforward: covert medical killing leaves almost nothing behind. Harold Shipman never explained himself either. Both true.

Shipman's psychology stayed opaque to the end and it never mattered, because toxicology and paperwork established the killings without reference to his inner life.

Take away his motive and the case survives intact.

The British comparators which have held up possessed evidential anchors not requiring anyone first to accept the suspect as the common cause. The cases which collapsed did not.

Intention Without An Observed Act

Ordinary murder rarely demands this kind of reconstruction. A knife driven into a chest carries its own meaning, and no jury needs a reason for the act to be sure of the intention behind it. The reasoning runs in one step: observed act, therefore intent.

Here it runs in four.

1. Physiological outcome.
2. Inferred act.
3. Inferred deliberate character of that act.
4. Inferred intention to kill.

Each step is an inference resting on the one before, and the second is precisely where the medical dispute sits.

For eight of the fifteen orders the fourth step is heavier still, because those relate to attempts.

Attempted murder demands more than murder does. An intention to cause really serious harm will not suffice. Recklessness will not suffice. The Crown had to prove, on each of those counts, a settled intention the child should die.

A clinically trained nurse intending specifically to kill a ventilated premature infant possesses knowledge unavailable to an ordinary assailant.

Most of the alleged attempts nevertheless failed. Babies were resuscitated, sometimes with Letby among those raising the alarm.

Making sense of a killer who repeatedly forms a genuine intention to kill and repeatedly does not accomplish it requires an account of what she was actually trying to achieve.

• She wanted the emergency.
• She wanted to participate in the resuscitation.
• She wanted proximity to grief.
• She was escalating.

Every available answer is a claim about motive.

Motive Matters, Even If Not Technically

Prosecutors were candid about this. On the day Letby received her fifteenth whole-life order the CPS stated plainly it had no idea why she had done it and was under no obligation to prove a motive.

Correct as a matter of law.

Motive is not an ingredient of murder, and the reason for forming a murderous intention has never needed establishing.

Which makes it interesting so much of the trial was devoted to supplying one.

Jurors were shown handwritten scraps found at her home and were reminded of them repeatedly, the most famous reading "I am evil I did this". Prosecuting counsel told the jury she had left a confession, and described voyeuristic tendencies driving searches for the parents of her patients.

In the strict sense none of it is motive. All of it performs motive's function, which is to make an otherwise unintelligible person intelligible.

The same sheets of paper carried her insistence she had done nothing wrong, demands to know what allegations had been made and by whom, and the words slander, discrimination, and victimisation.

It has since been reported she may have been advised by occupational health staff to write her intrusive thoughts down, and the jury heard nothing of it. If that stands up, the point stops being one of interpretation and becomes a disclosure question of extreme seriousness.

Journalling, or writing down disturbing thoughts, is encouraged in general psychotherapy, according to Richard Curen, the chair of the Forensic Psychotherapy Society, who has worked as an expert witness and with victims and offenders for 25 years. “Doodling, journalling is a way of taking control of your thoughts. I don’t think it relates to a confession of any kind,” he said.

He added that Letby’s response on the notes in court was “robust, and seems right – she wrote down how she was being made to feel”.

Is it conceivable Letby was doubting the sanity of her own mind and writing down her cycling thoughts under extreme stress? Absolutely.

The handover sheets and the searches are genuinely ambiguous rather than exculpatory, and pretending otherwise would be as dishonest as the prosecution's use of them.

Nurses hoard paper. Staff involved in a death look up the family afterwards. Neither behaviour is normal in the sense of admirable, and neither is rare.

The state said it need not explain why.

The evidence nevertheless required twelve people to imagine a reason vivid enough to convert ambiguous clinical events into a campaign of murder.

Finality Set Against Accuracy

Two juries convicted. The Court of Appeal refused leave twice. Under section 23 of the Criminal Appeal Act 1968, fresh evidence is received only where the court considers it necessary or expedient in the interests of justice, and one statutory factor is whether there is a reasonable explanation for the failure to adduce it at trial. Defendants may not lose with one set of experts and return with another.

The policy behind that is sound. Litigation has to end. Trials would become rehearsals if every convicted person could reopen the science afterwards, and the Court of Appeal has construed the provision narrowly for good reason.

The sequence Letby exposes is nastier than the rule was built for.

1. The defendant is convicted.
2. Specialists afterwards challenge the mechanism.
3. The appellate system asks why they were not called the first time.
4. Finality therefore hardens in proportion to how badly the original defence failed to anticipate a scientific challenge which had not yet been formulated.

Where the disputed expert evidence goes to whether any crime occurred, the doctrine does more than protect a verdict. It insulates a potential scientific error from correction, and works most effectively in exactly the cases where correction matters most.

Finality and factual accuracy are both constitutional goods. Whole-life imprisonment is the point at which they stop being reconcilable by procedural tidiness.

The Criminal Cases Review Commission quango is the only remaining route. Its chair published a timeline in February after criticism of the pace, recording a preliminary application received on February 3rd 2025, with material continuing to land throughout the year, with further submissions and expert reports on January 21st 2026.

The Commission has no power to decide guilt. Its test is whether new evidence or argument creates a real possibility a conviction would not be upheld.

An Inquiry Which Begins With the Answer

The Thirlwall Inquiry has now slipped four times and offers no date. Its terms of reference direct it to examine the conduct of staff and management in relation to concerns raised about Letby, and whether governance structures contributed to a failure to protect babies from her.

The premise is the conclusion.

Ten million pounds of public money has gone into establishing the governance failures surrounding a conviction currently before the body which exists to overturn convictions.

In March, David Davis used an adjournment debate to set out failings by Cheshire Constabulary, drawing on two senior former officers. One wrote the national guidelines on investigating deaths in healthcare settings, and the other led the Beverley Allitt investigation.

Both had believed Letby guilty.

Both changed their minds.

Davis also reminded the House of an earlier Cheshire prosecution: Sally Clark, convicted of murdering her two infant sons, freed by the Court of Appeal in 2003 as a gross miscarriage of justice. She was dead of alcohol poisoning four years later.

The Home Office minister replied with the formula such occasions require. Due process had been followed. There had been a jury trial. The convictions were upheld on appeal. Cheshire Constabulary issued a statement rejecting every point made about it.

The Speaker had to waive the sub judice rule for the debate to happen, because inquests into the deaths of six babies remain live and adjourned.

Who Carries The Risk Of Uncertainty?

The trial ran for ten months. Most of the evidence was never reported in detail, and the CCRC has observed how much of the commentary comes from people with a partial view. Nobody outside the courtroom is positioned to return a verdict, and campaigners announcing Letby's innocence with total confidence are committing the identical error they attribute to the prosecution.

The real issue goes right to the core of how our judicial process operates and the principles on which it rests.

Reasonable doubt operates as a rule rather than a technicality. It decides who bears the risk when reality stays uncertain, and it exists because the state's power to end a life without ending it is otherwise limited only by the confidence of the officials exercising it.

• An unexplained death does not require the defence to explain it.
• An anomalous laboratory reading does not require the defence to name an alternative culprit.
• An unresolved collapse does not become a murder because nobody can identify the natural mechanism.
• A defendant facing a murder charge is entitled to say she does not know what happened; where the Crown cannot establish its own answer, it is a complete defence rather than an evasion.

The state asserted four separate things: a) these babies were attacked, b) Letby attacked them, c) she intended each of them to die, and d) she deserved imprisonment until her own death. Having asserted them, it owns the burden for every one.

Where the mechanism is disputed by qualified specialists, the difficulty belongs to the prosecution. Where the clustering has no denominator, the deficiency is the prosecution's. Where a laboratory result establishes an abnormality without identifying a hand, the gap is the prosecution's.

Constitutional protections are never tested by sympathetic defendants and tidy evidence. They are tested when the accusation is monstrous enough that doubt itself starts to feel like complicity, and every institution which has touched the case acquires an interest in the answer already given.

Surely all of this cannot be coincidence.

Surely the doctors cannot all be wrong.

Surely someone must have done this.

Each surely marks the precise spot where the burden requires policing.

Fifteen whole-life orders now rest on an inference whose supposedly independent components may not be independent of one another at all. When the same investigation produced eleven further proposed offences, the state could not satisfy its own test to prosecute a single one, and could not satisfy it again on review.

The state's obligation ran a great deal further than telling the best available story about Lucy Letby. It had to prove an extraordinary one beyond reasonable doubt.

It didn't. The burden was never anyone else's.

If Letby is innocent, the British judicial system is about to pay out its largest compensation penalty in legal history.