In the realm of psychiatric populations, machine learning has become a valuable tool for predicting suicide. Previous studies have grouped psychiatric and non-psychiatric populations together, leading to limitations in identifying specific risk factors for suicide within distinct populations. To address this issue, a comprehensive review of machine learning studies focusing solely on psychiatric clinical populations was conducted.
The review involved a systematic search of literature on PubMed, EMBASE, and Scopus following PRISMA guidelines up to November 17, 2022. Only original research that utilized machine learning techniques to predict suicide attempts or assess suicide risk in psychiatric populations were included. A total of 1032 studies were screened, with 81 meeting the inclusion criteria for further analysis.
Commonly used features in these studies included clinical and demographic characteristics. When comparing algorithms, random forest, support vector machine, and convolutional neural network models consistently outperformed others in terms of accuracy. Most of the studies reported an accuracy rate of over 70%, with factors like previous attempts, the severity of the disorder, and medications being significant predictors of suicide risk.
Despite the promising results, machine learning algorithms for suicide prediction still face challenges. These include the lack of neurobiological and imaging data, as well as the absence of external validation samples. Addressing these limitations could pave the way for the integration of machine learning models into clinical practice, ultimately aiding in the reduction of suicide rates.
Historically, predicting suicide has been a difficult task, with a lack of effective methods for anticipating individual suicides or categorizing patients based on suicide risk. Globally, suicide is a significant cause of premature death, particularly among young people between the ages of 15-29, making it the second leading cause of mortality in this age group after traffic accidents. In the age range of 15-44, suicide ranks as the third most common cause of death.
Despite efforts to identify risk factors and implement interventions, recent research suggests that current approaches are insufficient. Many individuals who attempt suicide have sought help from healthcare professionals prior to their attempts, highlighting a potential window for intervention. The challenge in predicting suicidal behaviors stems from the lack of definitive psychiatric biomarkers and the limited predictive power of individual risk factors.
Suicide risks are complex and are influenced by a combination of environmental and trait variables. While factors such as mental disorders, prior suicide attempts, early trauma, and negative life events are recognized as risk factors for suicide, traditional methods for predicting suicide have shown limited efficacy. Even high-risk populations like those with depression struggle with accurate suicide prediction.
In conclusion, the quest for an effective method to predict suicides and stratify patients based on suicide risk remains an ongoing challenge in psychiatric populations. Traditional risk factors have their limitations, and biomarkers for suicide risk assessment are yet to be established. Ongoing research in machine learning holds promise for improving the accuracy of suicide prediction, potentially revolutionizing suicide prevention efforts in the future.
