lancet-header
Preprints with The Lancet is part of SSRN´s First Look, a place where journals and other research experts identify content of interest prior to publication. These preprint papers are not peer-reviewed and are posted here as part of a 12-month trial. Authors have either opted in at submission to The Lancet family of journals to post their preprints on Preprints with The Lancet, or submitted directly via SSRN. The usual SSRN checks and a Lancet-specific check for appropriateness and transparency have been applied. These papers should not be used for clinical decision making or reporting of research to a lay audience without indicating that this is preliminary research that has not been peer-reviewed. For more information see the Comment published in The Lancet, or visit The Lancet´s FAQ page, and for any feedback please contact preprints@lancet.com

A Machine Learning Approach to Identifying Future Suicide Risk

15 Pages Posted: 12 Nov 2018

See all articles by John Pestian

John Pestian

Cincinnati Children's Hospital Medical Center, Department of Pediatrics, Division of Biomedical Informatics

Daniel Santel

Cincinnati Children's Hospital Medical Center, Department of Pediatrics, Division of Biomedical Informatics

Michael Sorter

Cincinnati Children's Hospital Medical Center, Department of Pediatrics, Division of Psychiatry

Ulya Bayram

University of Cincinnati - Department of Electrical Engineering and Computer Science; Cincinnati Children's Hospital Medical Center, Department of Pediatrics, Division of Biomedical Informatics

Brian Connolly

Cincinnati Children's Hospital Medical Center, Department of Pediatrics, Division of Biomedical Informatics

Tracy Glauser

Cincinnati Children's Hospital Medical Center, Department of Pediatrics, Division of Neurology

Melissa DelBello

University of Cincinnati, College of Medicine, Department of Psychiatry & Behavioral Neuroscience

Suzanne Tamang

Stanford University, School of Medicine, Center for Population Health Sciences, Department of Biomedical Data Science

Kevin Cohen

University of Colorado at Denver, School of Medicine, Computation Bioscience Program

More...

Abstract

Background It is nearly impossible to predict when someone will die from suicide, but with early intervention many deaths are preventable. Clinical trials using machine learning algorithms have been able to use patient language to compute the likelihood that someone is suicidal at a particular point in time. Here we determine if the language characteristics associated with suicide risk are persistent 30 days after discharge.

Methods Multiple hospital base emergency departments and outpatient clinics were used to enroll subjects (n=253) into one of two groups: suicidal or control. Their responses to standardized instruments and interviews designed to harvest thought markers were recorded and analyzed with machine learning algorithms. Subjects were re-interviewed approximately 30 days later, and the language in their interviews was analyzed to determine the presence of suicidal ideation.

Outcomes The results show that the language characteristics used to classify suicidality at the initial encounter are still present in the participants’ speech 30 days later (AUC = 0·89 (95% CI: 0·85-0·95) with p < 0·0001) and that this also holds in the inverse case; classifiers trained on second interviews could identify the cohort that produced the first interviews (AUC = 0·85 (95% CI: 0·81–0·90) with p < 0·0001).

Interpretation This approach explores the stability of suicidal interviews 30 days after it is recorded. It does so with computational innovations and well established computational linguistic methods. The results show that the thoughts exhibited by a patient’s language are still valid for machine learning 30-days after first disclosed, but the initial correlation with the standard measures is not. This can be useful when seeking decision support for follow-up care.

Funding Statement: Cincinnati Children’s Hospital Medical Center, Innovation Fund.

Declaration of Interests: The authors have no competing interests to declare.

Ethics Approval Statement: A prospective clinical trial was conducted between October 2013 and March 2015 (Institutional Review Board (#2013-3770) approved).

Keywords: Natural Language Processing, Machine Learning, Acoustics, Suicide, Decision Support Techniques

Suggested Citation

Pestian, John and Santel, Daniel and Sorter, Michael and Bayram, Ulya and Connolly, Brian and Glauser, Tracy and DelBello, Melissa and Tamang, Suzanne and Cohen, Kevin, A Machine Learning Approach to Identifying Future Suicide Risk (November 1, 2018). Available at SSRN: https://ssrn.com/abstract=3279211

John Pestian (Contact Author)

Cincinnati Children's Hospital Medical Center, Department of Pediatrics, Division of Biomedical Informatics ( email )

3333 Burnet Avenue
Cincinnati, OH 45229
United States

Daniel Santel

Cincinnati Children's Hospital Medical Center, Department of Pediatrics, Division of Biomedical Informatics

3333 Burnet Avenue
Cincinnati, OH 45229
United States

Michael Sorter

Cincinnati Children's Hospital Medical Center, Department of Pediatrics, Division of Psychiatry

3333 Burnet Avenue
Cincinnati, OH 45229
United States

Ulya Bayram

University of Cincinnati - Department of Electrical Engineering and Computer Science

2901 Woodside Drive
Cincinnati, OH
United States

Cincinnati Children's Hospital Medical Center, Department of Pediatrics, Division of Biomedical Informatics

3333 Burnet Avenue
Cincinnati, OH 45229
United States

Brian Connolly

Cincinnati Children's Hospital Medical Center, Department of Pediatrics, Division of Biomedical Informatics

3333 Burnet Avenue
Cincinnati, OH 45229
United States

Tracy Glauser

Cincinnati Children's Hospital Medical Center, Department of Pediatrics, Division of Neurology

3333 Burnet Avenue
Cincinnati, OH 45229
United States

Melissa DelBello

University of Cincinnati, College of Medicine, Department of Psychiatry & Behavioral Neuroscience

CARE/Crawley Building, Suite E-870
3230 Eden Avenue
Cincinnati, OH
United States

Suzanne Tamang

Stanford University, School of Medicine, Center for Population Health Sciences, Department of Biomedical Data Science

291 Campus Drive
Li Ka Shing Building
Stanford, CA
United States

Kevin Cohen

University of Colorado at Denver, School of Medicine, Computation Bioscience Program

Building 500 - 13001 E. 17th Place, Campus Box C29
Aurora, CO 80045
United States

Click here to go to TheLancet.com

Go to TheLancet.com

Paper statistics

Abstract Views
299
Downloads
53