Study: AI Helps Identify Patients at High Risk for Repeat Penetrating Trauma
Artificial intelligence can help predict which patients with penetrating trauma are most likely to return with repeat injury in the future, a new study published online October 18 by JAMA Surgery reveals. Identifying people at higher risk could allow physicians to know when to take more aggressive preventive action.
“Inpatient identification of patients who are likely to be readmitted allows physicians to arrange for additional care, such as psychiatric consultation, to take place while the patient is still under close supervision,” said Joshua Parreco, M.D., general surgery chief resident at the University of Miami Miller School of Medicine Palm Beach campus.
Encouraged by evidence in other disciplines that machine learning, also known as artificial intelligence, or AI, can discern subtle features of human behavior to help predict future actions, Parreco and Rishi Rattan, M.D., assistant professor in the DeWitt Daughtry Family Department of Surgery, studied its potential role in trauma.
The researchers identified people who survived nonelective penetrating trauma injuries in the Nationwide Readmissions Database in 2013 and 2014. Of the 63,678 patients, 1,229 (1.9 percent) experienced readmission for re-injury. Suicide and self-inflicted injury by cutting and piercing were the most common E-codes — the “external” code for injuries caused by something outside the health care system — accounting for 34 percent of admissions and 63 percent of readmissions.
AI compared favorably to a number of other statistical prediction methods the investigators used to confirm its accuracy.
“The most surprising finding was the diagnostic abilities of the machine learning classifiers with this database,” Parreco said. “The database we used is comprised of billing data and lacks the fine-grained clinical detail that is typically necessary to make these types of predictions.”
A potential limitation of the study involves use of the administrative database, which could introduce institutional bias or contain coding errors.
“Regardless, this study demonstrated that machine learning may be useful for developing highly accurate predictive models for re-injury after penetrating trauma using a large database of readily available data,” the researchers wrote in the Research Letter.
Although AI technology is primarily in the research and development phase, future electronic medical record systems could incorporate these predictive models to identify these high-risk patients in real time, Parreco said. Also, as a follow-up to the current findings, “The next step would be to prove the usefulness of this technology in a clinical trial.”