|LETTER TO THE EDITOR
|Year : 2021 | Volume
| Issue : 1 | Page : 86-87
The Effectiveness of machine learning in suicide prediction and prevention
Chidiebere Emmanuel Okechukwu
Department of Public Health and Infectious Diseases, Sapienza University of Rome, Piazzale Aldo Moro 5, 00185 Rome, Italy
|Date of Submission||01-Oct-2020|
|Date of Decision||01-Mar-2021|
|Date of Acceptance||01-Mar-2021|
|Date of Web Publication||16-Mar-2021|
Dr. Chidiebere Emmanuel Okechukwu
Department of Public Health and Infectious Diseases, Sapienza University of Rome, Piazzale Aldo Moro 5, 00185 Rome.
Source of Support: None, Conflict of Interest: None
|How to cite this article:|
Okechukwu CE. The Effectiveness of machine learning in suicide prediction and prevention. MGM J Med Sci 2021;8:86-7
Despite being a leading cause of death, suicide withstands prediction because of its unusual incidence, which hinders its prevention. The application of natural language processing in social media content evaluation and the combination of artificial intelligence techniques with scientifically approved suicide risk assessment methods in the analysis of behavioral and biological variables could be more effective in suicide prevention. Machine learning (ML) has emerged as an avenue for evaluating big datasets to increase the recognition of self-injurious thoughts and behaviors, this can be achieved by analyzing risk variables such as genetic features, depression, quality of sleep, anxiety, mood, cognition, abuse history, substance disorders, neuro-imaging and neuro-anatomical reports, family history, previous hospitalization, unemployment, comorbidity, delinquency, social media use, and voice and text message data. One of the main clinical relevance of ML in suicide prediction and prevention is that it enables early detection of suicidal ideation which is a significant step in suicide prevention, by acting as a platform for continuous assessment of the affected person by clinicians with the aid of various ML techniques for instinctive recognition of the individual’s present mental state about his/her online social activities, interactions, and contents. Furthermore, young people are likely to reveal risk factors for suicide on social media, which they may not want to disclose to their doctors, making the application of ML more feasible. ML cannot be equated to human learning as far as suicide prediction is concerned rather it can be simply designed as a supplementary clinical assessment technique assisting suicide diagnosis and risk verification using existing ML techniques. However, when compared to human learning, investigative data-driven ML could enable the formulation of robust prediction models that depend on associations between variables that are unforeseen by an existing diagnostic guide and are very complex and difficult to be perceived through physician–patient clinical assessment.
Regarding the outcome of the systematic review conducted by Burke et al., they identified five studies that applied ML techniques to predict suicide death, all the analyzed studies used the U.S. veteran or service member data and they applied longitudinal design, the majority of studies analyzed utilized the electronic medical record variables and included between 8 and 979 indicators in their models, most of the studies used several ML techniques to predict suicide death. Burke et al. also evaluated 14 studies that used ML techniques to predict suicidal attempts, the adult sample features varied, and one study assessed an adolescent sample, the entire studies were cross-sectional in design, apart from one longitudinal study, most of the studies used the electronic medical record variables as indicators, the recognized studies included between 16 and 1328 indicators in their models, almost half of the studies used several ML techniques to predict Suicidal attempt. Burke et al. acknowledged one study that used ML techniques to predict recent suicide planning among those with a history of nonsuicidal self-injury, using 62 indicators, mainly showing nonsuicidal self-injury features, the model with the highest performance attained an operating characteristic curve of 0.89, across the ML models, the outcomes advocated that depressive symptoms and the observation of the antisuicide function of nonsuicidal self-injury were the two most significant predictors. Burke et al. further identified ten studies that used ML techniques to predict suicidal ideation, the studies applied numerous samples, mostly from adult primary care patients and municipal adult and adolescent models, most of the studies utilized cross-sectional designs; however, four studies used longitudinal designs, models incorporated between 3 and 62 indicators, and utilized one to four ML methods. Burke et al. additionally recognized three studies that used ML techniques to predict nonsuicidal self-injury, the studies used an undergraduate student sample and included between 1 and 27 indicators in their models, and on average used one ML model. Burke et al. concluded that ML is very useful in suicide prediction and can be used to identify new suicide risk factors in the future.
Generally, the application of ML for the analysis of health information in online platforms has progressed and attained significant height, the information that an individual shared on social media about his/her mental health to seek information and support can be analyzed and can be of good help to clinicians. ML techniques such as Support Vector Machine and Naive Bayes are mostly used by researchers in developing a social media-based surveillance system. The major limitation of ML in suicide prediction and prevention is data deficit, for certain individuals, there are usually insufficient data with social associations, which may manifest as being inactive on social media, and lack of communication with friends and family, in such case it is difficult to follow-up and monitor their mental condition regularly using ML. Moreover, understanding an individual’s suicide intention and predicting the exact time the same individual may likely commit suicide is yet to be achieved using ML techniques.
In conclusion, ML techniques are effective in identifying and classifying individuals at high suicide risk by using analytical and predictive algorithms on large datasets. However, optimizing ML techniques to have strong predictive precision and diagnostic accuracy will enhance the prediction and prevention of suicide. The application of ML in the clinical setting will improve the working efficiency of psychiatrists, clinical psychologists, and mental healthcare workers.
Financial support and sponsorship
Conflicts of interest
There are no conflicts of interest.
| References|| |
Bernert RA, Hilberg AM, Melia R, Kim JP, Shah NH, Abnousi F Artificial intelligence and suicide prevention: A systematic review of machine learning investigations. Int J Environ Res Public Health 2020;17:5929. https://doi.org/10.3390/ijerph17165929
Burke TA, Ammerman BA, Jacobucci R The use of machine learning in the study of suicidal and non-suicidal self-injurious thoughts and behaviors: A systematic review. J Affect Disord 2019;245:869–84.
Yin Z, Sulieman LM, Malin BA A systematic literature review of machine learning in online personal health data. J Am Med Inform Assoc 2019;26:561–76.
Gupta A, Katarya R Social media based surveillance systems for healthcare using machine learning: A systematic review. J Biomed Inform 2020;108:103500.