Establishment of suicide behavior prediction models for bipolar disorder patients using random forest and backpropagation neural network
Objective To predict suicidal behaviors in patients with bipolar disorder by constructing a machine learning model based on random forest and backpropagation neural network,and provide clinical decision support for the prevention and intervention of patient suicide.Methods From January 2020 to Au-gust 2023,1 005 patients with bipolar disorder were enrolled.The general clinical data and social dysfunc-tion,anxiety,depression scores of all patients were collected.The random forest algorithm was applied for fea-ture selection,and backpropagation neural network model was constructed for evaluating the model's fitting effect and predictive performance.Results There were statistically significant differences in sociodemo-graphic characteristics and physiological and psychological factors among the suicide attempt group(n=293),suicide ideation group(n=332)and non-suicidal group(n=380)of patients with bipolar disorder(P＜0.05).Using the random forest algorithm identified six main predictive variables:educational level,age,free triiodothyronine(FT3),cognitive impairment,hopelessness and psychogenic anxiety.The developed backpropagation neural network model achieved a precision rate of 79.3％,a recall rate of 79.6％,an F1 score of 79.4％,and an AUC of 0.89 on the test set,indicating that the model predictive performance has high accuracy and discriminative power.Conclusion This study developed a machine learning model for predicting suicide in patients with bipolar disorder,which possesses high accuracy and discriminative ability,providing a decision-making basis for the prevention and intervention of suicidal behaviors in patients with bi-polar disorder.
NETL
NSTL
万方数据
