In 2014, Italian nurse Daniela Poggiali was arrested and convicted of murdering two hospital patients. Richard Gill, a statistician in the Netherlands, followed the case and became suspicious of the analysis used to convict her. Gill enlisted the help of Italian forensic statistician and ASA member Julia Mortera and, together, they secured an acquittal for Poggiali after finding errors in the original statistical analysis.
We wanted to know more about what it takes to become a forensic statistician, so we asked Mortera to answer the following questions:
Julia Mortera is a statistics professor at the University Roma Tre and honorary professor at the University of Bristol. She is also a member of the Royal Statistical Society Statistics and Law Section and the ASA Forensic Science Committee. The editor of Law, Probability, and Risk, she has made contributions to forensic statistics, decision theory, and Bayesian statistics.
What motivated you to become a forensic statistician?
It was quite by chance. I was asked to discuss a paper on the “island problem” that Phil Dawid had written about. The island problem is a toy example that illustrates the uses and misuses of statistical logic in forensic identification. I then worked with Phil on some generalizations of the problem and the results were published in a 1996 Journal of the Royal Statistical Society B joint paper, “Coherent Analysis of Forensic Identification Evidence.”
I so enjoyed the statistical problems that arose in the area of forensic statistics that I have never left the field in more than 25 years! I had some genetics background, having previously worked with geneticists on medical problems, which led my focus to forensic genetics problems.
Describe your path to becoming a forensic statistician.
My career in the area was riddled with difficulty as, being a new field in Italy, it was not recognized by statisticians, which implied that papers in the area were not considered valid for promotion. This is now changing, but it took a lot of effort to obtain this recognition. This means only a few statisticians in Italy have become involved in this exciting area, which is a great pity.
What do you like most about your work in this area?
I really enjoy the interdisciplinary aspect of the area, as it brings together diverse fields such as statistics, genetics, and law, to name a few. It is an exciting field, as new and interesting methodological issues can arise from casework data and technology moves forward. I also enjoy developing probabilistic expert systems models, which have proved to be fundamental for helping to solve complex problems in forensic genetics.
What are the main challenges of being a forensic statistician?
There are many challenges in this field. One of the main ones is trying to communicate what I believe is the logical way to analyze and interpret evidence to forensic analysts and in the legal setting in an easily understandable way.
Understanding that you and Gill described your research on the Daniela Poggiali case in a journal article currently under review, would you provide a short summary that includes the statistical aspects for our readers?
Suspicions about medical murder sometimes arise due to a surprising or unexpected series of events, such as an apparently unusual number of deaths among patients under the care of a particular nurse. There is a statistical challenge of distinguishing event clusters that arise from criminal acts from those that arise coincidentally from other causes.
In the Poggiali case, we examined some of the possible confounders that could explain away the higher death rates when Daniela was on duty.
Simple exploratory data analysis revealed interesting facts. For example, a death registered to have occurred on a particular day is associated with a nurse, even if she is on duty for just a fraction of that day. The actual time of death differs from the recorded time of death. Deaths are recorded more often at 7 a.m. (in the overlap between night and morning shifts), at midnight, and on the hour or half hour. So, a nurse like Daniela who starts her morning shifts early and finishes her night shifts late will be associated with more deaths during hours she was working than hours she was not working.
The prosecution stated the rate of deaths dropped after Daniela was dismissed from the hospital. However, a simple analysis showed admissions dropped considerably then (the hospital had become infamous due to media coverage) and, consequently, deaths diminished!
So, a difference in the mortality rates between different nurses could be due to confounding variables. Even if all the measured confounders are taken into account, no causal effect can be concluded from a mere association in an observational analysis like this.
Later in the case, the defense showed me the analysis based on a simple linear regression between postmortem interval and “vitreous humor” potassium concentration made by the prosecution’s forensic pathologist, who declared potassium chloride poisoning was the cause of a patient’s death. I realized that in making his prediction of postmortem vitreous humor potassium concentration, he did not consider a prediction interval and did not take into account any uncertainty or variability due to various factors.
Considering this, even if the concentration observed at the time of postmortem was above average, it was well within a 95 percent prediction interval.
Causal effects cannot be inferred by just providing a descriptive analysis, which can reveal—at most—an association among the presence of a given nurse and an increased mortality rate. This association is due to many confounding factors that, if not accounted for appropriately, can lead to a misleading conclusion.
We showed how the results of a generalized linear model were in contrast to the findings in the report by the prosecution’s expert witnesses. Using the model, in fact, we showed the increase in the number of deaths had no relation to Daniela’s presence.
A take-home message of the case is association is not causation!
Have you ever worked on a case like Poggiali’s before? If so, tell us a little about it.
I have not worked on similar cases. There are luckily not too many in Italy. I have worked in the investigative phase of several problems concerning DNA mixtures. Recently, I worked on a missing persons case in which the evidence was a DNA mixture presumably from related contributors. Another case concerned the search for the culprit of a brutal murder of an 11-year-old in which, again, the only evidence were DNA mixtures found on her clothing. I also worked on an incomplete paternity recognition of a deceased famous Italian singer in which the evidence consisted of a mixture of DNA.
Essential to success in this area is identifying a statistical flaw in the legal reasoning and then communicating the issue in a way that is understandable to lay people. How do you address this challenge?
I have been involved with a group of statisticians and law scholars for the RSS Statistics and the Law Section on Statistical Issues in Investigation of Suspected Medical Misconduct, which provides advice and guidance on the investigation and evaluation of such cases. This report was prompted by concerns about the statistical challenges such cases pose for the legal system, since statistical evidence is difficult for lay people and even legal professionals to evaluate.
I think it is important to train young students—researchers in legal studies—how to interpret and understand probabilistic-statistical reasoning in court cases. I have been teaching groups of legal students and practitioners, forensic scientists, and investigative police the basic principles underlying our field. Communication and finding a common understandable language to portray statistical concepts, even in complex settings, is a challenge but vital to undertake.
What advice do you have for students and early-career statisticians who might pursue forensic statistics?
I’d like to tell them the following:
- This is an exciting area of research
- The applications can bring on new interesting methodological issues
- The interplay of different disciplines makes it a challenging and thought-provoking area
- They should never take anything for granted in this field, but always go back to first principles and look out for the pitfalls in legal and forensic reasoning
- Most importantly, they can help deliver justice
An earlier version of this story, misstated that Gill was from the UK, he is from the Netherlands.
The article says I’m in the UK. I’m originally English, but I live and work in the Netherlands. Have been there for 50 years. I hold both nationalities.
We are so sorry! Thank you for letting us know, we have corrected this article.
Great interview! Dear Julia, thank you for doing such important work.
Would you be willing to share some of the DNA mixture analyses you have worked on? They sound very interesting.