Predictability of Lifetime Nonsuicidal Self-injury by Symptoms of Sleep Disorders Using a Neural Network Model

Background: Nonsuicidal self-injury (NSSI), which pertains to self-induced actions that cause harm to the body, comprises repetitive, intentional, and direct actions that do not adhere to social norms and expectations. The present study aimed to predict NSSI by sleep disorders (poor sleep quality, insomnia severity, risk of sleep apnea, and circadian rhythms) utilizing a neural network model. Methods: Using a multi-stage cluster sampling, a group of 400 health-care personnel (70% female, aged 32.5 ± 8.8 years) from the western region of Iran was selected. The subjects completed the Inventory of Statements about Self-Injury, Pittsburgh Sleep Quality Index, Berlin Questionnaire, Insomnia Severity Index, and Munich Chronotype Questionnaire. The neural network model and receiver operating characteristic (ROC) curve were used to examine the association between sleep disorders and NSSI. Results: The use of a neural network architecture resulted in the discovery of a single hidden layer containing three hidden units. Both training and testing models predicted more than 85% of cases with NSSI correctly. The results of area under the ROC curve were completely acceptable (0.834). The findings suggest that the severity of insomnia (0.418) and poor sleep quality (0.333) serve as potent indicators associated with both subgroups, with normalized importance values ranging from 80% to 100%. Conclusion: The neural network algorithm can be effectively employed in predicting NSSI by several sleep disorders. Future research can test the complexity of sleep disorders connected to NSSI comorbid with other psychiatric conditions.