The Weekly Mind Reader: AI-Driven Suicide Prediction

Despite a few dubious forays of artificial intelligence (AI) into the mental health space, the technology demonstrates some potential in psychiatry. A new The Journal of Clinical Psychiatry paper suggests that leveraging an AI model known as natural language processing (NPL) can help predict suicide ideation.

Analysing Risk

The study used a new NPL algorithm called REACH-VET to identify and assess suicide risk among veterans.

To develop the algorithm, the researchers analyzed medical encounter notes from more than 4,500 US veterans who died by suicide. All subjects had at least one VHA encounter. Using a method called Term Frequency–Inverse Document Frequency (TFIDF), they measured the importance of  keyword terms within the context of the entire medical record. 

The findings suggest that REACH-VET has the potential to enhance suicide prevention efforts within the VHA healthcare system. The NLP-based models outperformed the standard EHR model and improved the accuracy of suicide predictions by 19 percent. Specifically, these models were 6 times more effective in pinpointing patients in the highest risk category (top 0.1 percent) compared to other methods. 

However, the researchers noted that the majority of VHA patients that die by suicide do not fit within the high-risk tier. As such, even REACH-VET fails to detect risk among the preponderance of patients who go on to die by suicide.

Fighting the Battle

The researchers said that they conceived this study in response to their concern about high suicide rates among veterans. Nationally, suicide rates for the average population stood at 13.5 per 100,000 in 2020. For veterans, it was 31.7 per 100,000, according to a Veterans Affairs report. 

“Responding to this concern, the Veterans Health Administration (VHA) has substantially invested in suicide prevention, including establishing the Veterans Crisis Line, staffing designated suicide prevention specialists at each medical center, and establishing suicide prediction and surveillance metrics, helping ensure that individuals receive targeted preventative services,” the researchers wrote. They said they hope their study makes a significant contribution to improving the mental health and well-being among veterans. 

IN OTHER PSYCHIATRY AND NEUROLOGY NEWS THIS WEEK

• When first- and second-line treatments have failed to yield symptom relief in treatment-resistant premenstrual dysphoric disorder, consider the use of GnRHa with hormone add-back.
• Lifetime suicidal ideation and behavior predicts the risk of recent suicidal ideation and behavior.
• Two cases of misdiagnosis in subacute cerebral infarction make the argument for revisiting magnetic resonance imaging gadolinium contrast enhancement.
• Viral sensation Ren sings with heartbreaking clarity about his medical struggles after his Lyme disease was misdiagnosed as bipolar and other disorders. He’s inspiring others to share their own mental health journeys. 
• Donald Triplett was the first person ever to be diagnosed with autism. Known as “Case 1”, he died last week at the age of 89. 
• Monday was World Psychedelics Day, as our Tweet of the Week highlights.  In a new video, Scott T. Aaronson, MD, the Chief Science Officer for the Institute for Advanced Diagnostics and Therapeutics at The Retreat at Sheppard Pratt, explained how these drugs are creating a revolution in psychiatry.

Happy #WorldPsychedelicsDay! To celebrate the progress of the psychedelic movement, join @PsychedelicsDay for their annual series of lively and informative discussions with psychedelic leaders. https://t.co/hXYG8jmxHP pic.twitter.com/Nq57bWkSt5

— MAPS (@MAPS) June 20, 2023

NEW AT CME INSTITUTE

Click to earn free accredited CME credit.

• Patient-Specific Considerations, the GABA Pathway, and New Clinical Trial Data on Neuroactive Steroids in MDD and PPD, a JCP journal supplement worth 1.50 AMA PRA Category 1 Credits™.

• The Bipolar Disorder Education Collection, worth up to 13 AMA PRA Category 1 Credits™.