Optimizing, diagnostic, and therapeutic strategies using decision-theoretic planning: principles and applications

Area of Science:

• Artificial Intelligence in Medicine
• Clinical Decision Support Systems
• Medical Informatics

Background:

• Decision-theoretic planning offers a novel method for selecting optimal actions.
• Its application to medical decision-making for diagnosis and therapy requires investigation.
• Acute deep venous thrombosis (DVT) management provides a complex case for evaluation.

Purpose of the Study:

• To assess the applicability of decision-theoretic planning to medical decision-making.
• To identify optimal diagnostic and therapeutic strategies for acute DVT.
• To evaluate the DRIPS planning system's efficiency and accuracy.

Main Methods:

• An existing DVT management model with 24 strategies was adapted for the DRIPS system.
• Conditional-probabilistic actions were encoded within an abstraction/decomposition hierarchy.
• A utility function incorporated costs and risks of diagnostic tests and treatments.

Main Results:

• DRIPS processed 312 possible plans, using abstraction to eliminate 44% (136 plans).
• The system identified 'no tests, no treatment' as the most cost-effective strategy, differing from prior results.
• DRIPS revealed an error in the original study's manually constructed decision trees; optimal strategy shifted with higher cost-of-death values.

Conclusions:

• Decision-theoretic planning is a viable and potentially powerful tool for complex medical decisions.
• Inheritance abstraction enhances computational tractability for intricate planning problems.
• Modular data entry in such systems can mitigate errors common in manual decision tree construction.