SCCAP/APA Convention/2017/Novel Approaches to Improve the Identification and Treatment of Suicide Risk

From Wikiversity
Jump to navigation Jump to search

Novel Approaches to Improve the Identification and Treatment of Suicide Risk[edit | edit source]

Session Co-Chairs: Joseph C. Franklin, PhD, Florida State University and Anna Van Meter, PhD, Yeshiva University[edit | edit source]


Although effective treatments for psychopathology and suicidal thoughts and behaviors exist (Das et al., 2016; Hunsley, Elliott, & Therrien, 2014; Mann, Apter, Bertolote, & et al., 2005), the number of people who access available services is dwarfed by the number of people who could benefit. Some populations are more difficult than others to reach with existing interventions; young people tend to be less likely to seek and remain in treatment, relative to older adults (Heflinger & Hinshaw, 2010; McGorry, Bates, & Birchwood, 2013). Related, people who are suicidal are unlikely to seek treatment for their suicidal thoughts/behaviors (Kuo, Gallo, & Tien, 2001); identifying them through other means (e.g., primary care physicians, community screening), so that intervention is possible, is imperative. New methodology can improve the accuracy and reach of assessment for psychopathology and suicidality, and 􀂱􀀃importantly 􀂱􀀃 can make treatment more accessible for people unlikely to utilize traditional, in-person treatment services. This symposium will highlight novel research designed to improve our ability to identify and treat individuals at risk for suicide; Dr. Mortier will describe an international screening program designed to identify college students at future risk for suicidal thoughts and behaviors using a range of risk factors. Dr. Ribeiro will present research demonstrating how data from patients􀂶􀀃health records can be used to accurately predict who is likely to die by suicide. Dr. Auerbach will present two projects designed to (1) assess for psychopathology among young adults and (2) to reduce risk for suicide and other negative outcomes by treating young people with depression using an online CBT intervention.

Predicting suicide death: An accurate and scalable approach to risk detection using machine learning[edit | edit source]

Presenter: Jessica D. Ribeiro, PhD, Florida State University

Co-Authors: Joseph C. Franklin, PhD, Florida State University; Colin G. Walsh, MD, Vanderbilt University

Our ability to predict suicidal thoughts and behaviors has been only marginally better than chance (i.e., AUCs in the 0.50s) since the inception of suicide research over five decades ago (Franklin et al., 2016). This alarming fact helps to explain why suicide rates have not appreciably declined in over a century.

One major methodological pitfall of prior research is that nearly all studies have examined risk factors in isolation (Franklin et al., 2016). Yet, the processes that lead to suicidal thoughts and behaviors are likely highly complex (Ribeiro et al., 2016). Accordingly, accurate prediction of suicidal behaviors likely requires the simultaneous consideration of a multitude of risk factors as well as the complex relationships among those factors. Conventional methods applied in psychology and psychiatry are limited in their ability to address this need, however. Machine learning methods, by contrast, are optimally suited to do so. These methods are designed to develop algorithms that optimize prediction while simultaneously reducing the likelihood of overfitting. Recent efforts applying machine learning to suicide prediction among soldiers support the initial promise of this approach (Kessler et al., 2015).

The objective of this study is to drastically advance suicide death prediction. To this end, we addressed three aims. First, we applied machine learning to predict suicide death among hospital patients. Second, we examined how prediction changed over time from three years to one week before the death. Third, we examined the relative importance of predictors over time. Our sample consisted of 12,000 hospital patients, 502 of whom had died by suicide. Random Forest was the primary predictive modeling framework; bootstrapping was used to correct for optimism. Discrimination performance was good across models (AUC: .90-.94), with modest improvements as suicide death became more imminent.

Our findings represent an initial step toward accurate and scalable suicide risk detection.



Predicting Suicidal Thoughts and Behaviors Among College Students: A Novel Approach[edit | edit source]

Presenter: Philippe Mortier, MD, KU Leuven University, Leuven, Belgium
Co-Authors: Glenn Kiekens, MSc, KU Leuven University; Koen Demyttenaere, MD, PhD, KU Leuven University; Ronny Bruffaerts, PhD, KU Leuven University; Randy P. Auerbach, PhD, McLean Hospital, Harvard Medical School; Pim Cuijpers, MD, PhD, Vrije Universiteit Amsterdam, The Netherlands; Jennifer Greif Green, PhD, Boston University; Ronald C. Kessler, PhD, Harvard Medical School; Alan Zaslavsky, PhD, Harvard Medical School; Matthew K. Nock, PhD, Harvard University

College students are an important subpopulation of young individuals at risk for suicidal thoughts and behaviors (STB). While the campus environment offers unique opportunities to implement and test STB prevention programs, interventions to date have been largely ineffective. In this presentation, we outline important shortcomings of current college student suicide research and explain why basic epidemiological findings cannot be translated to effective interventions. We then propose a number of novel methodological approaches to evaluate risk for college student STB. Findings are presented in the context of the Leuven College Surveys (LCS; KU Leuven university, Belgium), a series of prospective cohort studies of incoming KU Leuven freshmen (N~10,000). Novel approaches include: (1) the use of targeted refusal conversion strategies to obtain representative cohorts of incoming freshmen, (2) the use of specific missing data handling techniques to increase data representativeness, (3) a clear differentiation between students with an onset of STB before or during the college period, (4) the estimation of adverse clinical patterns of STB (i.e., patterns of relapse and persistence), (5) the identification of important population-level risk factors (i.e., highly prevalent risk factors that carry low individual-level risk may be equally important targets for intervention), (6) the evaluation of the cumulative effect of a wide range of distal (e.g., childhood-adolescent traumatic experiences) and proximal risk factors (e.g., past year stressful experiences) when estimating risk for STB, and (7) the evaluation of prediction accuracy (AUC-values and concentration of risk analyses) of prospective STB risk models. The feasibility of these approaches will be consistently supported by realtime LCS findings. We show how these findings contribute to a more fine-grained understanding of STB among college students. Most importantly, we interpret LCS study findings as suitable research evidence for future interventions within a multilevel college student suicide prevention approach.


Other 2017 Resources[edit | edit source]

Retrieved from "https://en.wikiversity.org/w/index.php?title=SCCAP/APA_Convention/2017/Novel_Approaches_to_Improve_the_Identification_and_Treatment_of_Suicide_Risk&oldid=1813357"