Can Data Analytics Help People Survive Poison?

By Dr. David Rozier

“We’re exploring how data analysis can help doctors and nurses make fast, dependable, on-the-fly decisions in order to treat their patients.” – Dr. David Rozier

In a hospital emergency room rapid decision making happens every day. It seems the perfect fit for data analytics platforms which can cull information in ways that can offer insights to medical staff. But the emergency room often is ignored when it comes to data science because:

  • Data collected during difficult emergency situations is frequently of low quality for a computer.
  • Given the life and death nature of emergency medical situations, the reliability of data analysis must be high even though it is working on low-quality data.
  • Rule-based systems often are ineffective because of the high diversity of situations.

I’m part of a team of machine learning experts at the Xerox Research Centre Europe in Grenoble, France. We’ve taken up the challenge, and we’re working with a French teaching hospital’s Accident and Emergency (A&E) department. We’re exploring how data analysis can help doctors and nurses make fast, dependable, on-the-fly decisions in order to treat their patients.

Test Case: Accidental Poisoning

Each year, 25,000 children under five-years old are rushed to A&E departments with suspected poisoning, according to the United Kingdom’s Child Accident Prevention Trust. Moreover, a 2015 report from Safe Kids Worldwide points out that nearly half of the 1.34 million calls to poison centers in the United States are for children, and are related to medicine poisoning.

Using anonymized medical data from 200 patients admitted for medicine poisoning in France, learning how to forecast what sort of care each new case will require.

The challenge is difficult because of the very nature of poisoning. The patient is often unconscious or delirious, and the medical staff rarely knows what caused the poisoning. Our team set out to help the staff determine if a patient would likely recover naturally, or if there was a high probability the patient’s condition would worsen and require admittance to the intensive care unit.

If data analytics can help with this real-time sorting, the hospital can increase the quality of patient care and improve how resources are spent on patients.

The initial results are very encouraging. We’ve used data to identify what class of medicine could have caused the poisoning, which helps the medical staff decide on the appropriate level of care.

We’ll expand on these preliminary results in 2016 with a larger study that uses more of our analytics algorithms.

We view this work as an opportunity to use clinical innovation to solve some of the most painful – and costly – issues in society.

David Rozier, an expert in artificial intelligence and scientific computing, is the Healthcare Technology Transfer group manager within XRCE’s Advanced Development Lab.

13 thoughts on “Can Data Analytics Help People Survive Poison?

  1. Dr. Dennis Korneff April 11, 2016 - Reply

    Very interesting work Dr. Rozier. Will you be publishing this work soon, or perhaps already have?

    • David Rozier April 14, 2016 - Reply

      Thank you Dennis.
      This is early stage work, so we have not yet submitted a peer-reviewed publication. I will however be giving an invited talk on the topicat the1st Intl. Workshop on Health Data Management and Mining (HDMM) 2016 in May

  2. Dr. Dennis Korneff April 11, 2016 - Reply

    Very interesting work Dr. Rozier. Another great use of machine learning in medicine.

  3. Alex Goldstein April 11, 2016 - Reply

    I need to help for my children one at 11 y and another one eight y I need to check out for then one 11 y need shock for the school

    • Hi Alex: If you believe your children have been exposed to poison, you should contact your doctor or local emergency room immediately.

      You may also learn more about poison control centers at the American Association of Poison Control Centers’ website:

      Best regards,


  4. dennis korneff April 14, 2016 - Reply

    Dear David,

    Its interesting to me that you will be presenting your findings at the Health Data Management and Mining (HDMM). Is there a website or internet group that I could follow to learn more about this?

    Dennis K

  5. David Rozier April 15, 2016 - Reply

    I guess you found the workshop website:

    Regarding the actual project, no specific communication channel exists today.

    Thank you

  6. dennis korneff April 16, 2016 - Reply

    David, thank you for this information. I will be following the website to check for any and all happenings at the conference.

  7. dennis korneff April 18, 2016 - Reply

    Hello David,

    I wonder if you and colleagues have looked at Decision Trees in your work?

    I am taking a university class now and am using decision tree algorithms to predict hospital admissions. In the case of poisonings, I though my simplistic code below could be a helpful starting point. Please understand that I am a novice at this point but am very curious about using computer algorithms to predict and improve health outcomes. I aspire to become and expert in the future! Any feedback for me would be very valuable! Thank you, Dennis K

    • David Rozier June 2, 2016 - Reply

      Decision trees can indeed be useful to reach a decision once you have all the necessary information (inputs) to navigate your tree.
      Here the difficulty lies more in the extraction of relevant information in the data available. It is only thinly present and strongly mitigated by other signals. Once the relevant information is actually successfully extracted, the decision is fairly easily reached.

  8. Dennis korneff May 29, 2016 - Reply

    Hello Dr Rozier and Mr. Pings,

    I hope you are both well. Though I am no longer working at Xerox, I still curious about the progress of your work with analytics Dr. Rozier. Would you be so kind as to tell me about or inform me about your presentation in Helsinki Finland. Thank you.
    Dennis Korneff

    Ps, I trust the information you spoke about at Helsinki international conference is not proprietary but if it is proprietary then I will be sorry not to be able to see it.

    • David Rozier June 2, 2016 - Reply

      The presentation was very well received in Helsinki, as I exposed the way we worked in our collaboration with a teaching hospital, and as I insisted on the fact that analytics can never be performed by an isolated mathematician but always in collaboration with a domain expert, here a physician.
      The actual final results are still in progress and were not shown.
      Thanks for your interest in our work!

  9. Dennis korneff June 3, 2016 - Reply

    Thank you for your thorough answer. I hope to stay in touch with you on this research.


Your email address will not be published. Required fields are marked *