The flaws in the Lucy Letby case The convictions should be referred back to the Court of Appeal

In the case of the nurse Lucy Letby, a theory that she was a murderer was alighted upon by one of the ward’s consultant paediatricians very early on — after only a very few of the unexplained deaths or collapses that ended up on her indictment — and the theory was based almost entirely on her having been on shift at the relevant times. There can be little doubt that there was not, at that stage, enough evidence for murder to be anything other than a remote possibility. So either it was a very lucky guess, or this theory was later taken up by others in a way that led to inaccurate expert medical opinion, the misleading presentation of evidence, and a terrible miscarriage of justice.

Following her conviction last August, for murdering seven babies and trying to kill six others, a 13,000-word piece in The New Yorker told a compelling story of her likely innocence. More recently, and once the reporting restrictions were lifted after her conviction three weeks ago in a retrial for the attempted murder of one of the babies on which the jury could not agree, long pro-defence pieces have appeared in The Guardian, Telegraph, Independent and Private Eye. Also recently published — withheld pending the outcome of the retrial — is the written judgement of the Court of Appeal, who refused her application for leave to appeal.

An appeal first goes to a single judge, who makes a decision on the papers whether to let you argue it at a hearing. In Letby’s case, that judge was Sir Robin Spencer, who turned her application down. Coincidentally — or perhaps not — Spencer was the barrister who led the prosecution of Sally Clark in 1999, who was wrongly convicted of killing two of her young children. That notorious miscarriage of justice, as well as others from the same era, such as Angela Cannings and Donna Anthony, was based on flawed statistical evidence about cot death from the since-discredited and disgraced paediatrician Sir Roy Meadow.

“Unknown unknowns” back then included the possibility of a genetic factor that could explain multiple cot deaths in the same family. It was nearly 20 years before a good candidate was discovered. A growing collection of lawyers, doctors and statisticians believe that something similar — flawed statistics and paediatric misdiagnosis without due respect for unknown unknowns — may have occurred in Letby’s case.

If the first Court of Appeal judge turns you down, you can try again at an oral hearing in front of three others. Letby was again unsuccessful. Their written judgement is long, detailed and illuminating. But it will not, I suspect, relieve much anxiety about a possible miscarriage of justice.

The main issue in the original trial, and indeed the retrial, was whether the deaths and other “events” were deliberately caused. Letby’s biggest problem here was the two allegations of attempted murder by insulin — Babies F and L, committed some eight months apart. Blood test results showed high insulin accompanied by low levels of a chemical called c-peptide. When the pancreas produces insulin, it produces similar levels of c-peptide. Therefore, the Crown said, the insulin was not produced in the body. It was, in other words, artificial. Therefore, it was an attempt to kill.

The inference to foul play with these two babies is arguably much stronger than it is with the more numerous diagnoses of air embolism, which was a more subjective opinion, more dependent on the statistically questionable “shift-pattern” evidence. The two guilty verdicts on the insulin counts were the first the jury returned — the next not coming for three days — and were two of only three unanimous convictions in the trial.

The Court of Appeal records that “the integrity of the blood samples and reliability of the biochemical testing was challenged by [defence KC] Mr Myers. However, in her evidence at trial, the applicant admitted that both babies had been poisoned by insulin, but denied that she was the poisoner.” The Crown made much of this “admission”, but the point is seriously unfair.

Letby was being pushed quite hard to swallow an inference which she was in no position to accept or deny. Her evidence was that she did not poison them — simple as that. So her opinion of the validity of the blood samples and the inferences to be drawn from them is neither here nor there.

This is a common practice by prosecutors, easy to lapse into in the heat of the moment. I’ve done it myself. “So it’s all just a big coincidence is it? Or have these two complainants put their heads together?” — that sort of thing. You’re trying to make a little speech and put the defendant on the back foot, in the course of questioning them about matters within their knowledge — which is what you should confine yourself to. But a defendant’s answers to such questions are not a sound basis for treating a prosecution inference as agreed by the defence. If the reliability of a blood sample is not agreed, or an inference not accepted, there the matter should rest. Her barrister explicitly did not concede the accuracy of the results in his speech to the jury, much less the inferences to be drawn.

And it seems there may be good reason not to accept their validity. The Guardian has reported serious concerns about the results from senior experts in the field. The Telegraph quotes guidance from the relevant lab saying that its tests are “not suitable” where the introduction of artificial insulin is suspected. And The New Yorker piece quotes the Crown’s expert as saying that there was a third insulin case, with similar blood results, not attributed to Letby. The samples cannot be tested again as they were destroyed soon after they were taken — because the babies’ blood sugar levels recovered and there was no suspicion of wrongdoing at the time.

But even without these post-conviction queries, if this were the only evidence, and these the only allegations, I would hope no jury would convict a nurse for attempted murder on that basis. The timing and detail of the insulin results meant that the Crown had to say that not only must Letby have poisoned the feeding bags she herself put up, but also must have poisoned one or more other bags in the store cupboard so that they would unwittingly be administered by other nurses after she had left. (The bags themselves, of course, are long gone.) Would you feel sure a random nurse tried to kill two babies eight months apart, or would you think this might be an instance of why the lab says its tests are “not suitable” for proving the presence of artificial insulin? But of course, this was not the only evidence against her.

There are also Letby’s online searches for bereaved families, and the collection of medical notes found at her home relating to some of the events she was charged with. Neither of these, I feel, are anything like as bad as they look: they both stand against a background of a much larger number of non-suspicious searches and medical notes.

But for many observers, and perhaps for the jury, the “confession” note is persuasive. For criminal lawyers, less so. That’s because false confessions are surprisingly common. So much so, that a law was enacted to prohibit a confession being given in evidence unless the prosecution can prove beyond reasonable doubt that it was not obtained by oppression or as a result of anything “said or done” that was likely to render a confession unreliable.

How could any nurse subjected to a long investigation into whether they have been murdering babies find the process anything but oppressive? If the case against her can be proved beyond reasonable doubt, one might think it could — and even should — be proved without the mixed inculpatory-exculpatory scrawls of a woman on the brink.

Almost all of the rest of the convictions — that is, all seven murders and three of the alleged attempts — involved the deliberate injection of air, leading to air embolism. One ground of Letby’s appeal impugned the prosecution experts’ diagnosis-by-exclusion — i.e. “we don’t know what else it could be, so it is likely to be air embolism” — as well as their diagnosis on the basis of various kinds of unusual skin discolouration that were observed. The 1989 research paper relied on by the Crown’s experts, Pulmonary Vascular Air Embolism in the Newborn, is very short, and says relatively little about skin discolouration other than that, in about 10% of the 50 or so cases they reviewed, “migrating pallor” in the skin was observed — which, it was not disputed, may be caused by other things.

The Court of Appeal was not impressed by this complaint, or by fresh evidence from the surviving author of the 1989 paper in which he criticised the use to which his research had been put in the Letby case. The experts were entitled to use skin discolouration as supportive of the diagnosis, the court said, as being “consistent with” embolism; the fresh evidence did not undermine that; and the lack of any other plausible explanation is significant.

“Consistent with” does a lot of heavy lifting for the Crown in criminal trials. A piece of evidence only helps to prove something if — and to the extent that — there is a difference between the probability of finding the evidence if the hypothesis (e.g. guilt, or air embolism) is true, and the probability of finding it if it is false. More precisely, it is the ratio between these two that determines the strength of the evidence in either direction. That, in a nutshell, is Bayes’s theorem. So, if you can say no more than that a piece of evidence is “consistent with” the hypothesis, but it is also “consistent with” its negation, and you can’t say how much more “consistent” it is with one rather than the other, then the evidence does not assist you either way. For example, if a prosecutor says “the fact you were wearing gloves makes you more likely to be guilty of this robbery”, that evidence is in fact worthless if it’s winter time and everyone is wearing gloves: wearing gloves is “consistent with” both guilt and innocence, to the same degree.

It seems, though, that the experts used skin discolouration as more than merely “consistent with” air embolism. They used it to prove, to some extent, that diagnosis. Which they are entitled to do, if it is a justifiable professional opinion, but it does point to a fundamental danger with expert evidence: the difficulty a jury has in getting behind the curtain and judging what weight to give it.

One way of achieving this is to look at whether the experts have been wrong before. In the case of the lead expert in Letby’s case, Dr Evans, the defence discovered halfway through the trial that he had been utterly excoriated in a ruling by a Court of Appeal judge in a different case. “No effort to provide a balanced opinion… No attempt made to engage with the powerful contradictory indicators… The report has the hallmarks of an exercise ‘working out an explanation’ that exculpates the applicants… tendentious and partisan expressions of opinion that are outside [his] professional competence.” That level of criticism from a senior judge is not made lightly: they know it can be career-ending. And if the Crown had known of it in advance, I suspect they would have looked for a different expert.

Now, this was on one view a gift to the defence: they could now undermine Dr Evans’s credibility by cross-examining him on it in front of the jury. True, his conclusions in the Letby trial were supported by other experts who had checked his work, but the defence argued that those experts were merely rubber-stamping his opinions. The jury may have had a difficult task in assessing what weight to give the criticisms of Evans’s previous work — one with somewhat binary options: disregard his evidence, or believe it. Ultimately, the Court of Appeal upheld the trial judge’s rejection of the defence’s application to exclude his evidence. It might not have been an easy decision for the trial judge, or the Court of Appeal, and although it may have been the correct decision, I cannot imagine any lawyer would claim the situation overall was entirely satisfactory.

Zooming out, the big question remains: without knowing of the police investigation and the suspicion of foul-play, and the fact that one nurse was on duty for many or all of the incidents under review, would any expert have diagnosed air embolism by the deliberate injection of air, in any of those cases? Certainly none of the pathologists did at the time.

In his recent Private Eye piece, Dr Phil Hammond gives a startling statistic: in a review of more than 1,000 cases of infant death in south-east London, “the cause of mortality was unexplained for about half of the newborns who had died unexpectedly, even after post-mortem examinations”.

So, the shift-pattern evidence — i.e. the fact that Letby was on duty for so many unexplained deaths and collapses — does seem to have been central to the case against her. And statisticians are questioning its probative power. Even leaving aside the issue of deaths and other negative events that occurred when Letby was not on duty, coincidences of that scale occur — just as someone will win the lottery each week, at odds of 14 million to one.

Also very unusual is the fact that Dr Evans “pitched” his services to the police, when he heard about the investigation in the media. This is not the way experts are normally selected. The defence argued it indicated pro-prosecution bias. The Court of Appeal disagreed.

And then there’s the fact that in the last few weeks Dr Evans has commented publicly on the defence not calling their own expert, who was advising them throughout the trial. Indeed, that expert, Dr Hall, has himself gone on record saying that he “does not know” why he was not called. And the same question was posed by the Court of Appeal, somewhat pointedly: without having called their own expert, can the defence really complain if the jury agree with the Crown’s?

It’s an entirely valid question. But I would be slow to assume the worst. One reason the defence might have chosen not to call Dr Hall (and I am speculating here) is if they felt they had so thoroughly undermined Dr Evans that they were better off leaving it at that. That would be a big tactical call, but covering your own back is not always consistent with protecting your client’s interests.

So, for example: if your own expert would contradict much, but not all, of the Crown’s expert’s conclusions, and you believe the credibility of the Crown’s expert has been totally destroyed, you might judge it prudent not to put your own expert in the witness box. Avoiding the potential for future criticism and hindsight regret by putting your client in what you believe will be a worse position — by calling your own expert — is not a boss move. This is all speculation, but I guarantee that careful thought went into the decision not to call Dr Hall. None of this, though, would mean the conviction is safe.

There is evidence against Letby. And the fact that it’s “circumstantial” is not a valid criticism: circumstantial evidence can be, and often is, utterly overwhelming. But this is where comparison with the Sally Clark case really comes in. The famous statistical error in that trial was that the expert Professor Roy Meadow wrongly treated two putative cot deaths as probabilistically independent. But arguably the more important point is that while the probability of “coincidence” might indeed have been very low, so too is the probability of a mother murdering her children. In Letby’s case, the chance that a randomly selected nurse is a baby killer is vanishingly small. As Carl Sagan put it, paraphrasing the French mathematician Laplace, extraordinary claims require extraordinary evidence.

Defence lawyers often say to juries that their clients start off innocent: the fact that they are in the dock is no evidence at all. But this is not quite right. If the probability of guilt is zero, no amount of evidence — that is, no Bayesian likelihood ratios, no matter how high or how numerous — will alter that chance. Rather, a defendant’s prior probability of guilt — that is, before any evidence has been called — is just very, very small. But not all crimes are equally common. A randomly selected member of the public is much more likely to have stolen a bike — given the number of bikes that get stolen — than a randomly selected nurse is to be a serial killer of her patients. But there isn’t much room for these fundamental concepts to be explored and explained to a jury.

There has been some commentary to the effect that those protesting Letby’s innocence should pipe down, that they are behaving irresponsibly. But miscarriages of justice do occur, and the Court of Appeal sometimes gets things wrong.

The only route for Letby now would be a referral back to the Court of Appeal by the Criminal Cases Review Commission, which would require substantial new evidence of one sort or another. The process is not quick, and the Commission do not refer many cases. But I would not be surprised if one day it happens. I hope it does. If Letby has been wrongly convicted, it would be one of the worst miscarriages of justice we have ever seen. If she is guilty, well, true verdicts, like true scientific theories, are only strengthened by rigorous testing and challenge.

---