If that doesn't look like 'reasonable doubt' then I'm, erm, I don't know, the pope, maybe. Certainly not someone who did actually get a B grade in A-level maths which involved a fair amount of stats & prob, and can follow your reasoning.
One larger probability comparison I would love to see is with other comparable NICUs. Statistically speaking, I mean, given that every baby in an NICU has an elevated probability of dying, what is the average for a level 2 NICU mortality cluster. Likewise, what would be the probability of any randomly selected NICU nurse being associated with a cluster of deaths. I would imagine we're talking about 'standard deviations' here. Given the number of NICUs, and over a given number of years, I would imagine the percentage chance is 99.9% or something. And the more years which pass the tendency towards 100% increases, until 'it happens'. Thus, what really determines whether a nurse gets accused (let alone prosecuted) is the personality of the people around her, in particular the specialists. I would not be surprised if Lucy wasn't the only nurse statistically associated with such a cluster, just that she happened to be the only one surrounded by sociopaths, including a 'TV Doctor' with a sunk cost fallacy. If you get my meaning.
Anyhow, I think the point here is that serious monitoring, psychologically speaking, of the personality types in high risk environments like an NICU needs to be a matter of course. That kind of monitoring could've prevented this entire miscarriage of justice.
You have done amazing work here, by the way. I'm not surprised you haven't had much sleep! And I do hope that all of this can get to whichever uncorrupted future defence team LL might have, and thus lead to a total acquittal. And sooner, rather than later.
I see one issue with your idea. If one compares outcomes of NICU's, these all have to meet certain quality standards. They have to be comparable. CoCH had a serious problem with the plumbing, before the "Lucy" events started to happen.
I agree with you on the psychological monitoring. With all the stories that came out I kept wondering: did these nurses get counseling when one of "their" babies died ? Apparently Lucy tried to deal with it in her own way, and it has cost her.
While every hospital will claim they have someone in a basement office (probably with no windows and a squeaky door) who is available to support the front-line clinical staff with these distressing and emotional issues - my knowledge and experience of these things would suggest that most staff know to stay well away from it. When you attend a medical or psychological service that is run or funded by your employer, your employer has the ability to access the records of that service use. Further, the employer has a vested interest in the outcome (perhaps in claiming that they provided the service to you and hence, are no longer responsible if you still feel you aren't 'right with the world').
I would find it really helpful if you could note on past discussion of particular babies what the relevant trial verdict was.
Thank you for this article. I was aware of the general pattern and principles but very helpful to see the risks quantified. Given the significant increase in high risk babies on the change to level 2, the consultants should have expected the number of deaths to increase. Are they bad at maths/lacking numeracy? Another possibility is that they tried to discredit a nurse whom they knew would be a whistleblower when there were problems? Probably hoping she would move hospitals rather than prison for life.
Is there anything that NICU nurses, or any nurses for that matter, can do to try to prevent this sort of thing happening? Beyond understanding the statistical arguments and making sure NOT to hire a KC as a defence lawyer, should it go to court. Do they keep log books as engineers do?
The way Prof Fenton teaches us to develop these Bayesian network models requires that we embed ourselves with experts (so they are expert-driven models). As a result, this model was developed with direct input from an obstetrician and midwives and reams of academic research. Further, this Bayesian network is not trained on estimates or made up numbers like a Neil Ferguson model - every node is trained or learned from UK national statistics primarily from the ONS for a single calendar year. Each node represents a sign/symptom/test/demographic of the patient. The arcs between them represent the relationships between nodes that my experts felt were the most important or influential (this doesnt mean that the influence of one node isn't also reflected in other child nodes, just that the arcs are identified on the nodes they felt were more directly influenced). There is no statistical mumbo jumbo here - no P values, no strange T tests, Poisson or other statistical approaches that would require explanation or investigation. Each node is simply the learned values straight from the ONS data. For example - the learned number of white, black and asian mothers that gave birth, the number of mothers who delivered in each gestation window etc. No tom foolery involved.
The data collection and ingestion to the model took about 6 weeks. The current version of the model that we just published is based on 2021 data. If we were going to use this as actual evidence for a trial or paper directly on the Letby case I would spend the same amount of time sourcing and collecting the 2015 or 2016 datasets and put the model through the same learning process.
My own cursory look at some of the research and numbers from the time suggest that neonatal deaths across the country might actually have been slightly higher in the period as a percentage of overall pregnancies... but that is just observational anecdote until we bring it all together and give to the computer
I would expect a cynical judge (a part of the Establishment) to be looking for Establishment input/sign off, much like MHRA for drugs. Has Stats Regulator, Royal Stats Society or others a useful part to play?
Keep up the good work. The only other area of new evidence i can see is a successful GMC complaint, and I'm not hopeful of that.
I cannot understand how it is that her defense lawyers could have failed to produce enough doubt such that the standard of proof in a criminal court was not reached. I seem to remember reading something that certain sorts of evidence were not allowed to be followed by her defense KC. The other well known instances of nurses convicted in similar circumstances, in the Netherlands and Italy for example, all spent several years in jail before being exonerated. To reverse miscarriages of justice in the UK all seem to need sustained campaigns and pressure form influential voices within the establishment. Is there any such movement beginning to reverse what clearly must be an unsound conviction?
If that doesn't look like 'reasonable doubt' then I'm, erm, I don't know, the pope, maybe. Certainly not someone who did actually get a B grade in A-level maths which involved a fair amount of stats & prob, and can follow your reasoning.
One larger probability comparison I would love to see is with other comparable NICUs. Statistically speaking, I mean, given that every baby in an NICU has an elevated probability of dying, what is the average for a level 2 NICU mortality cluster. Likewise, what would be the probability of any randomly selected NICU nurse being associated with a cluster of deaths. I would imagine we're talking about 'standard deviations' here. Given the number of NICUs, and over a given number of years, I would imagine the percentage chance is 99.9% or something. And the more years which pass the tendency towards 100% increases, until 'it happens'. Thus, what really determines whether a nurse gets accused (let alone prosecuted) is the personality of the people around her, in particular the specialists. I would not be surprised if Lucy wasn't the only nurse statistically associated with such a cluster, just that she happened to be the only one surrounded by sociopaths, including a 'TV Doctor' with a sunk cost fallacy. If you get my meaning.
Anyhow, I think the point here is that serious monitoring, psychologically speaking, of the personality types in high risk environments like an NICU needs to be a matter of course. That kind of monitoring could've prevented this entire miscarriage of justice.
You have done amazing work here, by the way. I'm not surprised you haven't had much sleep! And I do hope that all of this can get to whichever uncorrupted future defence team LL might have, and thus lead to a total acquittal. And sooner, rather than later.
I see one issue with your idea. If one compares outcomes of NICU's, these all have to meet certain quality standards. They have to be comparable. CoCH had a serious problem with the plumbing, before the "Lucy" events started to happen.
I agree with you on the psychological monitoring. With all the stories that came out I kept wondering: did these nurses get counseling when one of "their" babies died ? Apparently Lucy tried to deal with it in her own way, and it has cost her.
While every hospital will claim they have someone in a basement office (probably with no windows and a squeaky door) who is available to support the front-line clinical staff with these distressing and emotional issues - my knowledge and experience of these things would suggest that most staff know to stay well away from it. When you attend a medical or psychological service that is run or funded by your employer, your employer has the ability to access the records of that service use. Further, the employer has a vested interest in the outcome (perhaps in claiming that they provided the service to you and hence, are no longer responsible if you still feel you aren't 'right with the world').
That would certainly mess things up a bit, for sure.
I would find it really helpful if you could note on past discussion of particular babies what the relevant trial verdict was.
Thank you for this article. I was aware of the general pattern and principles but very helpful to see the risks quantified. Given the significant increase in high risk babies on the change to level 2, the consultants should have expected the number of deaths to increase. Are they bad at maths/lacking numeracy? Another possibility is that they tried to discredit a nurse whom they knew would be a whistleblower when there were problems? Probably hoping she would move hospitals rather than prison for life.
All cases are discussed by an Irish doctor, now retired.
https://jameganx.notepin.co
Is there anything that NICU nurses, or any nurses for that matter, can do to try to prevent this sort of thing happening? Beyond understanding the statistical arguments and making sure NOT to hire a KC as a defence lawyer, should it go to court. Do they keep log books as engineers do?
Very interesting work, thank you.
New evidence is needed, and this could fit the bill. One question, why should the judge accept the model as accurate?
The way Prof Fenton teaches us to develop these Bayesian network models requires that we embed ourselves with experts (so they are expert-driven models). As a result, this model was developed with direct input from an obstetrician and midwives and reams of academic research. Further, this Bayesian network is not trained on estimates or made up numbers like a Neil Ferguson model - every node is trained or learned from UK national statistics primarily from the ONS for a single calendar year. Each node represents a sign/symptom/test/demographic of the patient. The arcs between them represent the relationships between nodes that my experts felt were the most important or influential (this doesnt mean that the influence of one node isn't also reflected in other child nodes, just that the arcs are identified on the nodes they felt were more directly influenced). There is no statistical mumbo jumbo here - no P values, no strange T tests, Poisson or other statistical approaches that would require explanation or investigation. Each node is simply the learned values straight from the ONS data. For example - the learned number of white, black and asian mothers that gave birth, the number of mothers who delivered in each gestation window etc. No tom foolery involved.
Thanks, just playing Devil's Advocate. Is the data you are using 2015/16 to fully replicate the situation at that time?
A great question!
The data collection and ingestion to the model took about 6 weeks. The current version of the model that we just published is based on 2021 data. If we were going to use this as actual evidence for a trial or paper directly on the Letby case I would spend the same amount of time sourcing and collecting the 2015 or 2016 datasets and put the model through the same learning process.
My own cursory look at some of the research and numbers from the time suggest that neonatal deaths across the country might actually have been slightly higher in the period as a percentage of overall pregnancies... but that is just observational anecdote until we bring it all together and give to the computer
I would expect a cynical judge (a part of the Establishment) to be looking for Establishment input/sign off, much like MHRA for drugs. Has Stats Regulator, Royal Stats Society or others a useful part to play?
Keep up the good work. The only other area of new evidence i can see is a successful GMC complaint, and I'm not hopeful of that.
Great work!
I cannot understand how it is that her defense lawyers could have failed to produce enough doubt such that the standard of proof in a criminal court was not reached. I seem to remember reading something that certain sorts of evidence were not allowed to be followed by her defense KC. The other well known instances of nurses convicted in similar circumstances, in the Netherlands and Italy for example, all spent several years in jail before being exonerated. To reverse miscarriages of justice in the UK all seem to need sustained campaigns and pressure form influential voices within the establishment. Is there any such movement beginning to reverse what clearly must be an unsound conviction?
Some comment on low birth weight and prematurity that effect mortality. I am sure you will have it already, but just in case from 28mins https://open.spotify.com/episode/14ul4Co892rqyYATSx7d3o?si=fb4580f9d337466a