Evaluating the Diagnostic Accuracy of a Novel Bayesian Decision-Making Algorithm for Vision Loss

Area of Science:

• Ophthalmology
• Medical Informatics
• Artificial Intelligence

Background:

• Current diagnostic tools for acute vision loss rely on static flowcharts, limiting dynamic, stepwise patient evaluation.
• There is a need for improved diagnostic accuracy and efficiency in managing acute vision loss, especially by non-specialists.

Purpose of the Study:

• To evaluate the diagnostic accuracy of a novel dynamic Bayesian algorithm for acute vision loss.
• To compare the algorithm's performance against traditional diagnostic methods used by emergency medicine and primary care providers.

Main Methods:

• A prospective study involving 79 participants with acute vision loss in Windsor, Canada.
• Data collection via a questionnaire on ocular symptoms and findings by non-ophthalmologists.
• Comparison of the algorithm's differential diagnosis against a gold-standard diagnosis provided by an ophthalmologist.

Main Results:

• The algorithm achieved a diagnostic accuracy of 70.9%, significantly higher than the 30.4% accuracy of referring providers.
• Accuracy increased to 88.6% when including the top three diagnoses suggested by the algorithm.
• In urgent cases, the algorithm's sensitivity was 94.4% and specificity was 76.0% for the top diagnosis.

Conclusions:

• The dynamic Bayesian algorithm substantially enhances diagnostic accuracy for acute vision loss.
• The algorithm effectively utilizes clinical data gathered by non-ophthalmologists, improving patient care pathways.
• This AI-driven approach offers a promising advancement over static diagnostic flowcharts for vision loss.