AI algorithm for suicide risk assessment - Phase 2

A few year ago I did some great work on suicide prediction. Again with the South-West Yorkshire NHS Foundation Trust. Our work developed a risk assessment model tailored to the particular Trust, because we do not believe that the same model will necessarily apply to people in Yorkshire, London, Continental Europe, USA, China etc. Accuracy of our model was a little bit over 70% which is a good result, as all meta-research in suicidology suggests that the traditional risk assessment scores simply do not work. Our results were published, among others, in the Crisis journal:

https://econtent.hogrefe.com/doi/10.1027/0227-5910/a000561

Of course, that was only preliminary work, and the most severe limitation was lack of a control group (ethics...). We had to get around that by seeking to identify those referrals that were close to suicide, and distinguish them to previous referrals. Still, we only used data of people who died by suicide.

I am happy that we are finally moving to the next step, where we will (a) receive more complete data of patients and (b) are expected to receive control group data as well. On this basis, we expect to develop a more robust prediction model that should then be tested in actual clinical practice. Stay tuned for more!

Comments

Post a Comment

Popular posts from this blog

Working frantically on AI in Medicine special issue

I am editing a special issue of Artificial Intelligence in Medicine on Medical Analytics for Healthcare Intelligence, together with my colleagues Elpida keravnou-Papailiou of the University of Cyprus and Tianhua Chen of the University of Huddersfield. We are working hard on it as we have received over 100 submissions to the special issue (I have never experienced anything close to it in my quite long career), a testament to the popularity and importance of this topic. We hope to have paper evaluation and selection completed by the end of April 2020.
Read more
AI for autism diagnosis I am excited to be working with the South-West Yorkshire Partnership NHS Trust   on using knowledge-based AI to support diagnosis of autism in adults. Our works seeks to identify the clear-cut cases that can be safely diagnosed automatically based on the clinical data routinely collected prior to a diagnostic assessment (so no brain scans and other costly and usually unnecessary actions), while complex cases are referred for further assessment by a specialist clinician. The effect is higher productivity per clinician, resulting in reduced costs for the NHS and reduced waiting lists. Initial funding has been obtained by the National Health Service and work commenced in March 2020.
Read more