Predicting suicidal ideation at subsequent clinical encounter in multiple sclerosis patients

Area of Science:

• Neurology
• Psychiatry
• Health Informatics

Background:

• Persons with multiple sclerosis (PwMS) have a 30% lifetime risk of suicidal ideation (SI).
• Currently, no validated predictive tools exist to identify PwMS at risk for SI.
• This gap necessitates the development of methods for early risk detection.

Purpose of the Study:

• To develop and validate predictive models for suicidal ideation (SI) in persons with multiple sclerosis (PwMS).
• To utilize low-burden self-report measures and electronic health record (EHR) data for model development.
• To create accessible tools for integrating SI risk prediction into routine clinical practice.

Main Methods:

• A retrospective cohort study design was employed, analyzing data from 5694 PwMS.
• Prediction models were constructed using least absolute shrinkage and selection operator (LASSO) and association testing.
• Models were developed for various subgroups, including sex-specific and diagnosis-specific cohorts, with internal validation.

Main Results:

• Of 5694 PwMS, 578 reported SI at follow-up.
• Key predictors identified included PHQ-9 item 6 (feelings of worthlessness) and the time between clinical encounters.
• The developed models demonstrated high predictive accuracy, with an area under the curve (AUC) ranging from 0.73 to 0.77.

Conclusions:

• Nomograms and an online application were created to facilitate proactive SI risk prediction in PwMS.
• These tools offer a framework for integrating SI risk assessment into routine care.
• Further out-of-sample cross-validation is recommended to confirm model robustness.