Development and Validation of a Bayesian Network Method to Detect External Beam Radiation Therapy Physician Order...

Area of Science:

• Medical Physics
• Radiation Oncology
• Artificial Intelligence

Background:

• Physician orders in external beam radiation therapy (EBRT) are complex and prone to errors.
• Accurate physician orders are critical for patient safety and effective treatment delivery.
• Existing error detection methods may not adequately address the complexity of modern radiation therapy prescriptions.

Purpose of the Study:

• To investigate the efficacy of a Bayesian network (BN)-based method for detecting errors in EBRT physician orders.
• To develop and validate a probabilistic approach for identifying potential prescription inaccuracies.

Main Methods:

• A total of 4431 EBRT orders were analyzed and categorized into single prescription, concurrent boost, and sequential boost groups.
• Multiple Bayesian networks were trained using Bayesian learning algorithms to model order parameters and disease information.
• A procedure was implemented to select optimal BNs, determine site-specific parameters, and set error detection thresholds.

Main Results:

• The BN method demonstrated high accuracy in error detection across different prescription types.
• For single prescriptions, the true-positive rate (TPR) was 95.72% with a false-positive rate (FPR) of 1.99%.
• Sequential boost cohorts achieved 100% TPR for dose values and other parameters, with low FPRs (9.48% and 4.34%).

Conclusions:

• The probabilistic Bayesian network method offers a highly accurate approach to physician order error detection in EBRT.
• This BN-based strategy surpasses previously reported accuracy levels for complex prescription scenarios.
• Further development of BNs for clinical error detection tools is recommended to augment manual physician order checks and enhance patient safety.