Delirium misdiagnosis risk in psychiatry: a machine learning-logistic regression predictive algorithm

Area of Science:

• Psychiatry
• Medical Informatics
• Data Science

Background:

• Delirium is a common diagnosis in Consultation-Liaison Psychiatry (CLP).
• Studies indicate frequent misdiagnosis of delirium before CLP referral.
• Multivariate analyses of delirium misdiagnosis factors are limited.

Purpose of the Study:

• Quantify delirium misdiagnosis rates upon CLP referral.
• Develop a predictive classifier for delirium misdiagnosis.
• Identify key variables influencing delirium misdiagnosis.

Main Methods:

• Retrospective observational study at Alfred Hospital.
• Data collected over 5 months on factors contributing to misdiagnosis.
• Machine Learning-Logistic Regression model developed to classify delirium diagnosis accuracy.

Main Results:

• 35 out of 74 new cases (47%) were misdiagnosed.
• Predictive algorithm achieved 79% ROC AUC.
• Classification accuracy was 72%, with 77% sensitivity and 67% specificity.

Conclusions:

• Delirium is often misdiagnosed in hospital environments.
• Machine learning-logistic predictive classifiers show promise for healthcare applications.
• Improved diagnostic accuracy for delirium is achievable with predictive modeling.