Background: Misdiagnosis is prevalent in clinical practice due to incomplete and sometimes inconsistent data, which generate errors in the traditional diagnostic process. Orofacial pain, with its wide range of conditions, poses a considerable diagnostic challenge, particularly for inexperienced clinicians.
Case description: A structured, machine learning-compatible note-taking system was used to document clinical history and examination features from 1,020 patients at an orofacial pain clinic. A naïve Bayesian inference algorithm was used to compute and display the probability of various diagnoses as data were added to the medical record during a clinical encounter. Its accuracy compared favorably with 5 machine learning algorithms for 5 new cases of each of 10 diagnoses varying in their prevalence in the database.
Practical implications: The authors speculated that the key to achieving reasonable concordance was the highly structured electronic medical record, which included disease-defining or unique features of most diagnoses. Extension of these methods to broader clinical domains will require similar attention.
Keywords: Orofacial pain; algorithmic decision support; artificial intelligence; diagnostic accuracy; electronic medical records; structured note taking.
Copyright © 2025 American Dental Association. Published by Elsevier Inc. All rights reserved.