AI Finds 75% of Missed Self-Harm Cases in Medical Records

Researchers at the University of New Mexico just solved a problem that's been hiding in plain sight for years. When they used artificial intelligence to scan 1.3 million veterans' medical records, they found something shocking: traditional diagnosis codes only catch about 25% of patients with self-harm history.

The numbers tell a striking story. While diagnosis codes flagged self-harm in just 1.85% of patients, the AI found evidence in medical notes showing the real number was closer to 7.9%. That means thousands of veterans were getting treatment without doctors knowing their full mental health history.

The team developed a tool called PULSNAR that uses machine learning to read between the lines. It scans enormous medical files, some running up to 500,000 lines long, looking for patterns that human doctors simply don't have time to find. The AI learned from known cases and then spotted similar warning signs buried deep in other patients' records.

This matters because past self-harm is one of the strongest predictors of future suicide risk. When doctors miss this history, they can't provide the targeted care that could save a life. The problem got worse after COVID-19 increased isolation and mental health struggles across the country.

The hidden information isn't lost because doctors are careless. Medical notes are massive and complex, and clinicians are stretched thin between patient care and paperwork. Critical details about depression, PTSD, or substance abuse often get documented in notes but never make it into the summary codes that appear on problem lists.

The Ripple Effect

This technology could reshape mental healthcare from reactive to predictive. Hospitals could identify high-risk patients before a crisis hits, giving intervention teams a chance to step in early. Suicide prevention programs would finally have accurate data to guide their efforts.

The breakthrough extends beyond self-harm detection. The same AI approach could help track other underreported mental health conditions, creating better public health data and more effective treatment programs. It could help specialists spot patterns in PTSD or opioid use disorder that currently slip through the cracks.

The researchers are clear this tool isn't ready to work alone in clinical settings yet. It needs human oversight and careful privacy protections to ensure patient data stays secure. But the potential is enormous.

For the first time, we have technology that can read what doctors miss and flag the patients who need help most. In a field where early intervention makes all the difference, that could mean the difference between life and death for thousands of people struggling in silence.