Thomas: building Bayesian statistical expert systems to aid in clinical decision making

Area of Science:

• Biostatistics
• Artificial Intelligence in Medicine
• Decision Analysis

Background:

• Classical statistical analysis separates data analysis from decision-making.
• Bayesian frameworks offer integrated data-analytic and decision-making capabilities for biostatistical expert systems.
• Existing systems lack dynamic model construction and user-belief integration.

Purpose of the Study:

• To present a Bayesian framework for biostatistical expert systems that integrates data analysis and decision-making.
• To enable systems to make recommendations on decision-analytic grounds.
• To facilitate dynamic statistical model construction and updating based on user beliefs and study data.

Main Methods:

• Developed a knowledge-based system architecture integrating statistical and domain knowledge.
• Reinterpreted traditional statistical concerns, replacing statistical significance with pragmatic clinical thresholds.
• Created a prototype system (THOMAS) for interpreting randomized clinical trial results.

Main Results:

• The proposed architecture allows for dynamic model construction and updating.
• The system enables interaction at a semantic level suitable for clinical users.
• The THOMAS prototype demonstrated effective interpretation of clinical trial data for physician decision-makers.

Conclusions:

• A Bayesian framework can effectively integrate data analysis and decision-making in biostatistical expert systems.
• Dynamic model updating and user-centric interaction enhance clinical utility.
• This approach facilitates more informed clinical decision-making based on evidence.