Abstract
Background: The lifetime risk of suicidal ideation (SI) in persons with multiple sclerosis (PwMS) is 30%. No predictive tools are available for SI with empiric support for PwMS.Objective: The objective of the study is to develop predictive models for SI using low-burden self-report tools and electronic health record information.Methods: This is a retrospective cohort study of PwMS who reported no SI (Patient Health Questionnaire-9 (PHQ-9) item 9) at a clinical encounter and had a follow-up visit within a year where the PHQ-9 was completed. Models employing least absolute shrinkage and selection operator and association testing were used to construct prediction models for all PwMS, females, males, and for those with and without prior depression/anxiety-related diagnoses. Internal validation was assessed and nomologs and an online application were constructed.Results: There were 5694 PwMS of which 578 endorsed SI at their follow-up visit. PHQ-9 item 6 (self-worth/self-deprecation) and time between encounters were consistent predictors across models. All models had high predictive capacity (area under the curve (AUC): 0.73-0.77).Discussion: The nomologs and the online application provide a potential framework for integrating proactive SI risk prediction into routine care to identify PwMS at heightened SI risk at the next clinical encounter (https://farrenbriggs.shinyapps.io/suicidal_ideation_predictor/). Out-of-sample cross-validation is needed to support model robustness.