Machine Learning: An Approach in Identifying Risk Factors for Coercion Compared to Binary Logistic Regression.

Area of Science:

• Psychiatry
• Machine Learning
• Clinical Research

Background:

• Coercive measures in psychiatry remain prevalent despite known negative effects.
• Predicting the risk of coercion is crucial for improving patient care and safety.

Purpose of the Study:

• To identify risk factors associated with coercive measures in involuntarily hospitalized patients.
• To evaluate the accuracy of machine learning algorithms in predicting the risk of experiencing coercion.

Main Methods:

• Analysis of risk factors in 393 involuntarily hospitalized patients using chi-square and Mann Whitney U tests.
• Training and testing of machine learning algorithms (logistic regression, SVM, decision trees) using five-fold cross-validation.
• Comparison of machine learning model performance against binary logistic regression.

Main Results:

• A Support Vector Machine (SVM) model with 8 admission risk factors achieved 69% accuracy (60% sensitivity, 78% specificity, AUC 0.74).
• A logistic regression model with 18 post-discharge risk factors reached 75% accuracy (71% sensitivity, 79% specificity, AUC 0.82).
• Machine learning models demonstrated comparable or superior predictive accuracy to binary logistic regression.

Conclusions:

• Clinical and demographic variables are valuable for estimating coercion risk in psychiatric patients.
• Machine learning algorithms offer a promising approach for predicting coercion, with good to excellent predictive power.
• Further research analyzing more variables with machine learning can enhance the prevention of coercive situations.