A leopard goes to the doctor.
“Doc,” he says, “You gotta help me. Whenever I look at my wife, I see spots.”
The doctor looks at him for a minute and says, “Well, what do you expect? You’re a leopard.”
“I know that, Doc. But I’m married to a zebra.”
Healthcare serial murder enquiries are unusual in that they generally start without specific evidence of wrongdoing as such. Suspicion generally arises from an unusually large cluster of unexpected deaths. From there, prosecutors work backwards towards a perpetrator.1
Their job is made harder by two inconvenient facts: firstly, hospitals generally, and intensive care units specifically, are places where “unexpected deaths” are a regrettable part of a normal operating environment. There is a “going rate” of unexpected collapse, and a “suspect cluster” is simply one that exceeds that going rate, usually not by much: a standard deviation or two above the hospital’s average. The elevated rate might be what you’d expect once in ten years, say, or fifty — but not one in a million. According to Professor Sir David Spiegelhalter’s evidence before the the Thirlwall Inquiry, the increase in deaths at the Countess of Chester Hospital’s neonatal unit in 2015 would be within the expected range for hospitals across the UK in any year:
“If we assume an underlying rate of 3 neonatal deaths per year, then the probability of getting 8 or more deaths in 2015 is 0.02. This would generally be considered as constituting an “alert” signal. Again, to put this in perspective, we would expect around 3 such signals each year in the UK, just by chance alone.”
Secondly, “malicious nurses furtively murdering patients” is a one-in-a-million sort of thing. You would not expect to see it “every so often” in a given ICU.
All other things being equal, you would take the “one in fifty” option over the “one in a million” option. From a starting point of there being “an unusual cluster of deaths”, there is a quite likely explanation — it’s just one of those things — and a highly unlikely one — there’s a serial murderer in the ward — and no evidence, as yet, to tell us which way to lean. But as the saying has it: when you hear hoofbeats, think horses, not zebras. At least, until you see some stripes.
A prosecution intent on proving murder will need “posterior evidence” to support its hunch and rebut the far more likely explanation that the cluster is just “one of those things”. Investigators must therefore scratch around looking for secondary indicators of foul play that were not picked up at the time of the deaths and, if they find any, a plausible suspect having both the opportunity and disposition to murder.
The first question — “was it murder?” — can be addressed by technical medical evidence. That is not the topic of this post.
This post is about the second question: “given that it was murder, whodunnit?” Here, generally, you need solid “identification evidence” incriminating a specific person. Eyewitnesses. Fingerprints. Confessions. That kind of thing.
The problem with healthcare serial murder cases is that they start and often end without compelling identification evidence. Prosecutors are often obliged to resort to unusual things the suspect did, or said, that do not specifically indicate that she committed a crime, but are more broadly “consistent with” it. This is a bit of a recipe for confirmation bias.
In the 1970s, the FBI’s legendary Behavioural Science Unit learned how subtle behavioural patterns could lead them to notorious serial killers. Would their techniques apply, or be any use, in a classic healthcare serial murder case?
When the lambs stopped screaming
Lecter: Oh, Agent Starling. You think you can dissect me with this blunt little tool?
Starling: No! I thought with your knowledge —
Lecter: You’re so ambitious, aren’t you.
—Silence of the Lambs (1991)
In 1972, Howard Teten and Patrick Mullany established the FBI’s Behavioural Science Unit in Quantico, Virginia. You may remember it form the opening scenes of Silence of the Lambs. The FBI had long recognised that certain types of violent crime — “serial” ones, particularly — pose unique challenges to traditional policing methods. Serial killers murder in the shadows. They tend not to leave living witnesses, nor much in the way of physical evidence behind them.
By contrast, most normal violent crime happens between people who know each other and are embedded in stable, stationary, long-established social networks. When violence breaks out there tend to be many witnesses, lots of evidence, and only a few plausible suspects. Second-order psychological inference to work out who did what is rarely needed.
Serial offenders are different: they tend to be antisocial, itinerant, they target strangers, and those who are any good at it soon develop ways of avoiding detection. But they may still give themselves away by their habits and idiosyncrasies, predictors for which may be inferred from crime scenes and improbable commonalities between victims. The FBI developed behavioural profiling techniques designed to point to patterns not immediately obvious to your regular cop on the beat.
Hence, the premise of a “behavioural checklist”: a range of traits which, when found in unique combination, operate as a strong predictor that an individual is predisposed to behaving in a highly unusual, uniquely antisocial way.
Focus on individuals like this and you have far a better chance of finding your man.
So was born the legend of the “mindhunters”: the FBI profiler brilliantly decoding the machinations of an malevolent evil genius, tracking him to his hidden lair using just his inadvertent habits and idiosyncratic behaviours. It is a beguiling, but somewhat romantic, image. Real-life “hits” from behavioural profiling are not always as sophisticated as Buffalo Bill’s dressmaking connection in Silence of the Lambs. They tend to involve relatively straightforward patterns that require some expertise to see, but are still not especially esoteric.
Regular serial killers versus healthcare serial killers
Being more prosaic, “regular” serial killer investigations have particular features that lend themselves to behavioural profiling:
Serial murder conundrums
Likelihood it was murder: certain.
Method of murder: certain.
Common perpetrator: certain.
Direct evidence: none.
Perpetrator identity: unknown.
But mainly: you have no clue who the murderer is. What you do know, for sure, is that there has been a series of murders. You’re pretty sure they are linked: there is a specific modus operandi, and freaky features that make a coincidence highly improbable.
This part of your case — were they all murders by the same individual — is not really in doubt: the corpses all have bullet-holed playing cards in their breast pocket and an ice pick in the ear. They were definitely murdered by the same guy.
The question is, rather, who is that guy? A behavioural profiler asks, “What are the characteristics we would expect in a person who would commit this kind of crime?”
Profiling, thus, works forward from probable patterns in certain behaviour to predict what sort of person the perpetrator will be. It narrows down who you are looking for. Once you find a suspect with the predicted characteristics, it then becomes a question of finding hard evidence to tie him to the crimes.
Importantly, you develop your theory before you settle on a suspect. You use these characteristics not to verify suspects, but to weed them out. The risk of confirmation bias is therefore low.
If you find a suspect who fits the profile, you must then look for hard evidence that definitively connects that suspect to the crimes — ballistics, trophies, incriminating possessions like ice picks and decks of bullet-holed playing cards — things that an innocent person is highly unlikely to have. You don’t rely on the profiling exercise itself to prove anything in court. You probably don’t even mention it. It was a preliminary filtering tool in the investigation, not a conviction technique itself.
Interesting side note here, quickly: much is made in healthcare serial killer cases of the defendant’s sophistication in avoiding detection, covering her tracks and disposing of evidence: in this regard, it is assumed, healthcare serial murderers are like real serial murderers. But real serial murderers are not like this. When they are finally identified — with or without the help of behavioural profiling — they tend to be found with lots of incriminating evidence of their crimes: trophies, weapons, body parts and so on. The challenge is finding the murderer, rather than proving what he did when you do find him.
Anyway, all of the above applies to regular serial murderers. Healthcare serial murderers — like killer nurses — are a different ball of wax. They don’t lend themselves to behavioural profiling:
Healthcare serial murder conundrums
Likelihood it was murder: not certain.
Method of murder: not certain.
Common perpetrator: not certain.
Direct evidence: none.
Perpetrator identity: Known and already central to the investigation.
Here the main thing behavioural profiling is useful for — finding a likely suspect — is already done. There is no need to scour a large, widely distributed, poorly documented and inchoate group itinerant individuals: suspicion starts and finishes with statistical clustering of deaths in a single facility staffed by a small number of people. If it is murder, the potential culprits are few and already known. Usually2 there is only one: the nurse with the highest correlation of shifts to collapses naturally fills the frame. Who else could it realistically be?
But note: the hard question to be answered here is not “whodunnit” but was it even a murder? The suspect’s behavioural traits won’t help with that.
Or will they?
Behavioural profiling to backfill identification evidence
Seeing as there is no direct evidence of her doing anything, behavioural traits seem to make up a large part of the identification evidence against Lucy Letby. Seemingly innocuous behaviour like retaining handover sheets or making Facebook searches, while plainly not direct evidence of wrongdoing, are presented as the sorts of things a murderer would do. I wondered: are there others? Could we look at similar nurse serial murder cases and identify other behaviours that might mark Ms. Letby out as an unusually likely candidate? There must be something that disposes a small handful of healthcare professionals to murder vulnerable patients, when millions of common or garden nurses plainly don’t?
Ever the have-a-go hero, I thought I would try it out. I might not be Silence of the Lambs’ Agent Clarice Starling, but I could at least be space-cadet Olive Budgie.
I set about compiling some data from public sources — Wikipedia, Fandom’s Serial Killer Database, the FBI Vault and so on — about healthcare serial murderers to see what they are generally like. Do they have a type? Are they likely to have a history of criminal offending? Psychiatric illness? A motive? What age, sex, and marital status do they tend to be? What are their popular murder methods? Do they confess, before or after conviction?
Across North America and Western Europe, I found 42 nurse serial murder convictions since 1970. That is not many. There are a few more women than men — though given the 85% female skew in the profession that still suggests a significant predominance of males. Most murder elderly or adult patients though, of the 5 convicted of murdering infants, all were female.
Beyond that, information about them is thin.
They are, however, unusually likely to be wrongfully convicted: more than a fifth of the 42 convictions have been overturned or are subject to outstanding formal challenge.
So that is the first problem: the sample is small and, apparently, prone to error. It is likely to include known unknowns — people who aren’t healthcare serial murderers, but have been convicted of it anyway — and to omit unknown unknowns — people who are healthcare serial murderers, but haven’t been caught. There is significant “epistemic uncertainty”, in other words: we don’t know what we don’t know about healthcare serial murderers.
So the behavioural traits we might derive from this list are somewhat qualified.
The more I learned about base rate neglect, confirmation bias and prosecutor’s tunnel vision the more this amateur criminal profiling seemed a waste of time, so I retired space cadet Budgie. But kept my spreadsheet and resolved to see if I could find studies by better qualified people who know that they are doing.
Soon enough, I did.
Dr. David Wilson, emeritus professor of criminology at Birmingham City University is well-known in the serial killer field. He has written several books on the subject and intervened in a number of high profile cases, notably Lucy Letby’s. With his bright blue eyes and close-cropped silver beard he has quite the face for TV, in the same way I have one for radio and a voice for semaphore.
His frequent Letby soundbites have been somewhat ambivalent — but in the main, he’s in the guilty camp. This is interesting: he is, apparently, a rare expert with no personal interest, who is prepared to support the prosecution.
He’s also written about healthcare serial killer profiling. Notably, with fellow BCU Professor Elizabeth Yardley, in 2014 he published In Search of the “Angels of Death”: Conceptualising the Contemporary Nurse Healthcare Serial Killer — I didn’t realise they had changed since the olden days — a paper aiming to:
“establish insights into this particular subcategory of healthcare serial killer ... [and] ... test the usefulness of an existing checklist of behaviours among this group of serial murderers.”
The paper draws heavily on American criminologist Dr. Katherine Ramsland’s 2007 book Inside the Mind of Healthcare Serial Killers: Why they Kill. In her book, Professor Ramsland sets out 22 common healthcare serial murderer “behaviours”:
1. Moves from one hospital to another
2. Secretive/difficult personal relationships
3. History of mental instability/depression
4. Predicts when someone will die
5. Makes odd comments/claims to be “jinxed”
6. Likes to talk about death/odd behaviours when someone dies
7. Higher incidences of death on his/her shift
8. Seems inordinately enthused about his/her skills
9. Makes inconsistent statements when challenged about deaths
10. Prefers nightshifts—fewer colleagues about
11. Associated with incidents at other hospitals
12. Been involved with other criminal activities
13. Makes colleagues anxious/suspicious
14. Craves attention
15. Tries to prevent others checking on his/her patients
16. Hangs around during investigations of deaths
17. In possession of drugs etc. at home/in locker
18. Lied about personal information
19. In possession of books about poison/serial murder
20. Has had disciplinary problems
21. Appears to have a personality disorder
22. Has a substance abuse problem
Hmmm.
Professor Ramsland is an engaging writer and, by all accounts, an experienced criminologist. But while, intuitively, many of these traits seem plausible, few bear close scrutiny. She offers little by way of explanation or amplification about how she arrived at them, or why these, and not other traits, are germane. Nor does she say how they should be interpreted or applied.
Firstly, they are notably vague. If you’re hunting serial murderers by reference only to their behaviour, you would want criteria a little more hard-edged and categorical than this. What counts as “secretive”? Does “mental instability” include moodiness? What does being “associated with incidents at other hospitals” involve?
Secondly, some flags contradict others. “Craving attention” is a red flag, but so is “being secretive”. “Avoiding colleagues and working night-shifts” counts, but so does being “inordinately enthusiastic”, “hanging around the investigation” and “getting in early and staying after knock-off”.
Thirdly, some flags duplicate others: “moving around a lot” and “being involved in incidents at multiple hospitals” cross over, as do “predicting when someone will die”, “joking about killing” and “talking about death”.
Fourthly, the flags are not “risk weighted”: “owning a book about serial murder” — I’ve got tons of those — carries the same weight as “being found in possession of unauthorised poisons”.
Fifthly, one trait — “higher incidences of death on shift” — isn’t a behavioural trait at all. It is simply being at work when you are required to be. It has nothing to do with a suspect’s behaviour as such. And failing to have this trait is not just an absent check-mark from a long list of optional traits: it is a stone cold alibi. Surprise, surprise, this is, by a distance, the most prevalent trait.
Sixthly, not all flags unambiguously indicate misbehaviour: you would expect your very best nurses to be “enthusiastic about their skills”, to arrive early, stay late and be prepared to work night shifts.
Lastly, many of the flags fit a wide range of behaviours, including ones you would expect from any nurse, and which would be far more probable, alternative explanations than “probable serial murderer”.
I could go on.3
And just as troubling as the dubious flags on the checklist are the obvious ones that are missing. For example, there is no “suspect was witnessed murdering patients,” nor “suspect was found with a murder weapon,” nor “suspect has a cogent motive,” nor “suspect made an unambiguous confession”.
This may seem trite, but bear in mind over half of the 42 healthcare serial murderers admitted their crimes.
Behavioural profiling was invented to identify unknown criminals. It is a detection tool: it helps you hunt through a haystack of innocent passers by to find a villainous needle. Once you have your needle, its utility quickly dwindles. There tends to be plenty of corroborating evidence — DNA matches, fingerprints, ballistics and so on — linking an actual culprit to his crimes. If there isn’t, he is not very likely to be the culprit. It is that corroborating hard evidence that secures the conviction, not behavioural typing. This ought to be equally true of healthcare serial murderers.
Matching “observed traits” to already convicted healthcare serial murderers, therefore, is a bit of a rum business. But this is what Professor Wilson proposes. He applies the Ramsland checklist to 16 “confirmed” serial killer nurses convicted of at least two murders between 1977 and 2009 in hospitals in North America and Europe:4
Charles Cullen, Kimberley Saenz, Kristen Gilbert, Robert Rubane Diaz, Sonia Caleffi, Beverley Allitt, Cecile Bombeek, Vickie Dawn Jackson, Aino Nykopp-Koski, Orville Lynn Majors, Benjamin Geen, Petr Zalenka, Christine Malevre, Irene Becker, Stephan Letter, and Colin Norris.
Dr. Wilson’s cohort scored a median of just 6 out of 22 of Ramsland’s red flags — five, if you exclude the non-behavioural “higher incidences of death on shift”.
This is disappointingly low. Given how vague and innocuous many of the flags are, you would expect a stronger signal to emerge from the noise than that. It prompts a question: how would a “control group” of randomly chosen “normal” healthcare professionals score?
Oddly, there isn’t one. This seems quite an oversight, especially where the traits are so vague and the targeted behaviour so unlikely. I have friends who work at hospitals who would score higher than six on the Ramsland scale. I don’t think any of them are serial murderers. I scored six myself, and I don’t even work in healthcare.5
It is hard to believe that six of these “red flags” would mark anyone out as especially unusual, let alone a one-in-a-million serial killer.
Speaking of which, there is one other candidate we could apply the “Ramsland checklist” to: Lucy Letby. After two trials, multiple convictions and a decade in the public eye, there is a lot of her recorded behaviour available for convenient retrospective pattern matching. How would she get on?
Well, apart from “being on shift a lot,” which is not a behavioural trait at all, you could argue none of these traits squarely describe her behaviour. The one unusual thing is her failure to fit these any of these traits. Given how vague they are, it should be easy to confabulate some red flags. But have a go: it is not.
So what does Professor Wilson make of all this, then? Has he ever brought his “blunt little tool” to bear on the Letby case?
After a fashion he has. In a recent piece for the Observer he agreed: she does not score high on the Ramsland checklist. Does this give him grounds for doubt?
Apparently not.
She simply did not display the “red flags” we have come to associate with nurses who kill within hospital settings [...].
Yes, there were spikes of deaths that corresponded with her shifts, but hospitals are places where people with severe health issues are located and sadly some of those patients will die. A spike in a number of deaths on a unit or a ward can simply be a random occurrence and does not have to be an indication that a serial killer is active.
We now know that what was happening in the neonatal unit in Chester was not random but was indeed the work of a murderer.
So why did she do it?
We might never be able to answer this question definitively, but like many of those nurses who have been convicted in the past 50 years there was probably a “hero complex” at work – a desire to be the centre of everyone’s attention; a toxic narcissism that demanded that they should have power and control, which they might have felt had been denied to them in other aspects of their lives. And what better way of demonstrating that power than deciding who should live and who should die?
“Toxic narcissism”? “Hero complex”?
Where on Earth did these come from? What evidence is there that Ms. Letby suffered anything like this?
Rather than concluding that, in failing to display any of the behavioural traits he has identified for healthcare serial murderers, Lucy Letby may well be a false positive, Professor Wilson doubles down. Remarkably citing a refusal to explain her motive — also not on the Ramsland checklist! — Professor Wilson compares Ms. Letby to Harold Shipman:
“Like most of the serial killers I have worked with and studied, Letby has been silent about what might have driven her to kill. In that respect she is like our worst British serial killer, Dr. Harold Shipman, who also maintained his innocence and refused to accept any responsibility for the deaths of at least 215 of his often elderly female patients.”
Now: besides people like Harold Shipman, one other group tends to remain silent about their motivation for serial murder, of course: people who have not committed it.
Confirmation bias ahoy
As the “serial offender” theory develops around a suspect, using “behavioural profiling” to work backward from the suspect to a crime carries an unusually high risk of circular reasoning: “this cluster of deaths look suspicious because of the constant presence of the suspect, and the suspect looks suspicious because of the cluster of deaths”. With 22 criteria as woolly as this, you could convict almost anyone unlucky enough to fall victim to statistical correlation.
The uncomfortable reality about behavioural profiling in a healthcare environment is that if the prosecution needs to rely on behavioural profiling to get home, the risk of miscarriage of justice is already elevated. These cases are necessarily the hard ones: all the obvious markers of criminality and guilt are missing. There is already good reason to doubt a serial murderer is at work. If you are this far down the barrel, it might be time to put down the scraper and return to back to baseline probabilities. A statistical cluster is far more probable than serial murder.
Remember, the baseline chance, all other things being equal, that a given nurse is a serial murderer is something like one in a million.6 Cases with no hard evidence to improve those odds are exactly the ones most likely to yield false positives. “False positive,” in this case, is a statistical term for “miscarriage of justice”.
When you hear hoofbeats think horses, not zebras.
This is a common pattern. See Jane Bolding, Daniela Poggiali, Lucia de Berk, Rebecca Leighton, Victorino Chua, Beverley Allitt, Colin Norris and Lucy Letby, for example.
Not always, as Rebecca Leighton might tell you: she was at first suspected of the murders subsequently attributed to Victorino Chua.
For those with an unusually high boredom threshold, I do: there is a detailed critique of the “Ramsland checklist” in a table in the appendix.
All of these were on Olive Budgie’s list. Also on Olive’s list but not included were Marianne Nölle (Germany), Niels Högel (Germany), Richard Angelo (USA) and Roger Andermatt (Switzerland).
I will leave readers to guess which ones I got.
This is a rough estimate. There has been roughly one conviction for healthcare serial murder a year since 1970. The number of registered nurses in North America and Europe in 2025 something like 15 million, so one in five million is a conservative estimate.
Share this post