Comparison of rheumatological diagnoses by a Bayesian program and by physicians

Area of Science:

• Rheumatology
• Medical Informatics
• Artificial Intelligence in Medicine

Background:

• Accurate diagnosis of rheumatic disorders is crucial for effective patient management.
• Existing diagnostic methods can be complex and time-consuming.
• Decision support systems offer potential to enhance diagnostic accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate a Bayesian decision support system for diagnosing rheumatic disorders.
• To compare the performance of different methods for optimizing diagnostic weights within the system.
• To assess the system's diagnostic accuracy against predictions made by experienced rheumatologists.

Main Methods:

• A Bayesian decision support system was constructed using evidential weights derived from patient data.
• Weights were calculated using logarithm of likelihood ratios from 1,000 rheumatology clinic patients.
• Four methods, including mathematical approaches and expert-adapted weights, were tested to improve system performance.

Main Results:

• The system's performance was evaluated on 570 independent cases.
• Expert-adapted weights yielded optimal results: 65% sensitivity and 96% specificity.
• These results closely matched rheumatologists' predictions (64% sensitivity, 98% specificity).

Conclusions:

• A Bayesian decision support system, particularly with expert-adapted weights, demonstrates high diagnostic performance for rheumatic disorders.
• The system's accuracy is comparable to that of experienced physicians.
• This approach holds promise for improving the diagnostic process in rheumatology.