Machine learning can flag violent patients before they strike
Hospital inpatients who are likely to turn violent can be identified by algorithmic analysis of routine clinical notes stored in electronic health records, according to a study published in JAMA Network Open July 3.
The technique’s accuracy falls from good to fair when an algorithm trained on one patient population is used to assess another. Still, the finding should come as promising news to nurses and other hospital workers who have frequent patient engagements.
Worldwide, an estimated 14% to 20% of patients become physically aggressive during inpatient treatment, while surveys consistently show most healthcare workers face a wrathful patient at some point in their career.
In the present study, Vincent Menger, MSc, of Utrecht University in the Netherlands and colleagues trained and tested the algorithm on more than 6,400 admissions of more than 4,100 patients at two psychiatric hospitals in their country.
To tell whether or not outcomes met the criteria for being deemed violent, they used an aggression scale established specifically for healthcare settings.
Measuring the algorithm’s internal predictive validity using areas under the curve (AUCs), they found it achieved a score of 0.797 for one site and 0.764 for the other.
The scores were significantly lower, albeit not failing, when the algorithm used pretrained data to analyze clinical notes from the opposite site (0.722 and 0.643).
“Internally validated predictions resulted in AUC values with good predictive validity, suggesting that automatic violence risk assessment using routinely registered clinical notes is possible,” the authors concluded. “The validation of trained models using data from other sites corroborates previous findings that violence risk assessment generalizes modestly to different populations.”
In their discussion, Menger et al. predict greater use of machine learning in mental healthcare is around the corner.
“In the near future, we envision that further advancements toward a data-driven psychiatric practice will be made and that EHR data will become an even more valuable asset in supporting important decisions in the clinical process,” they write. “Machine learning approaches have been able to contribute substantially in other fields of medicine, and our study provides evidence that such progress is possible in mental health care as well.”
The study is available in full for free.
Dave P. has worked in journalism, marketing and public relations for more than 30 years, frequently concentrating on hospitals, healthcare technology and Catholic communications. He has also specialized in fundraising communications, ghostwriting for CEOs of local, national and global charities, nonprofits and foundations.