Veterans’ Hidden Struggles: Spotting Unseen Self‑Harm in Health Records

Health records often miss signs of self‑harm, especially among veterans. Because doctors only flag clear cases, the data lacks true “negative” examples. This gap makes it hard to estimate how many people are at risk. Researchers used a special method called Positive and Unlabeled learning. It works well when only confirmed positives are labeled, as in most medical databases. The technique can estimate how many patients actually have a problem and spot those who may be hiding it.In a study of veteran electronic records, the team applied this approach. They found that many veterans who had not been formally diagnosed still showed patterns linked to self‑harm. These hidden cases could benefit from early support and treatment. The findings suggest that current coding practices leave many at risk unnoticed. By treating data as “positive or unlabeled, ” clinicians can better identify those needing help. This shift could improve preventive care and save lives.