Original title: Development and internal validation of machine learning models for identifying suicidal behavior in inpatients with mood disorders
The researchers used hospital records from people admitted with mood disorders to predict suicidal behavior around the time of admission. They selected 14 useful clues from 65 candidates and compared different prediction methods. A calibrated linear model performed most consistently, but it did not identify every patient with suicidal behavior. The study tested the model within one hospital's records, so it still needs testing elsewhere and in real-time care.
Imagine packing a bag for a trip when you can bring only a few things. It is hard to choose because many items might matter, and some clues about the trip are more useful than others. In this paper, the selected items stand for the patient details used to estimate risk at admission, while the test is whether those choices help identify people with suicidal behavior.
Hospital teams need a clearer way to identify suicidal behavior in people admitted with mood disorders. This model estimates risk at admission, but its results are exploratory and do not establish that it is ready to guide care.
The study is an internal validation using records from one hospital, so it does not establish performance in other hospitals or in prospective clinical use. The authors describe the findings as exploratory and call for standardized outcome measurement and external validation. The reported SHAP contributions are associational, not causal.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Development and internal validation of machine learning models for identifying suicidal behavior in inpatients with mood disorders — Sixiang Liang, Qitong Jiang, Jinhe Zhang et al. — Frontiers in Psychiatry — doi:10.3389/fpsyt.2026.1950186
Abstract licence: CC BY
Explanation generated by Synthesis from the abstract — it is not the original paper.