Psychooncology. 2025 May;34(5):e70150. doi: 10.1002/pon.70150.
ABSTRACT
OBJECTIVE: Patients with cancer have a high risk of suicide. However, evidence-based preventive measures remain unclear. This study aimed to investigate suicide prevention strategies for hospitalized patients with cancer by analyzing nationwide patient safety reports using large language models.
METHODS: Data were drawn from patient safety reports collected by the Japan Council for Quality Health Care from 620 hospitals. Reports involving suicides or attempts among patients with cancer were analyzed. BERTopic was used to identify topics in free-text reports, and conditions such as depressive symptoms were labeled using the OpenAI API. Logistic regression was conducted to analyze the relationship between pre-incident conditions and proposed countermeasures.
RESULTS: Among 213 reports, key topics included mental and physical distress, symptom deterioration, nursing records, and post-incident documentation. Over 40% of patients exhibited depressive symptoms, and 30% expressed suicidal ideation. However, fewer received specialized mental care. Notably, over 10% appeared to experience delirium, potentially contributing to the incident. The most frequently suggested countermeasures were mental distress treatment, enhanced medical staff communication, and improved information sharing with families. Logistic regression revealed several associations between proposed countermeasures and pre-incident conditions, including mental health intervention for patients without prior treatment, physical interventions for those in severe pain, and improved staff communication for those with depressive symptoms.
CONCLUSIONS: This study, based on nationwide patient safety reports, highlights critical suicide prevention strategies for hospitalized patients with cancer, many of which align with previously proposed strategies. Additionally, the study provides new insights, such as the need for preventive measures to manage delirium.
PMID:40320591 | DOI:10.1002/pon.70150
AI-Assisted Evidence Search
Share Evidence Blueprint
Search Google Scholar