Differential diagnosis of jaw pain using informatics technology

Summary This study aimed to deduce evidence‐based clinical clues that differentiate temporomandibular disorders ( TMD )‐mimicking conditions from genuine TMD by text mining using natural language processing ( NLP ) and recursive partitioning. We compared the medical records of 29 patients diagnosed with TMD ‐mimicking conditions and 290 patients diagnosed with genuine TMD . Chief complaints and medical histories were preprocessed via NLP to compare the frequency of word usage. In addition, recursive partitioning was used to deduce the optimal size of mouth opening, which could differentiate TMD ‐mimicking from genuine TMD groups. The prevalence of TMD ‐mimicking conditions was more evenly distributed across all age groups and showed a nearly equal gender ratio, which was significantly different from genuine TMD . TMD ‐mimicking conditions were caused by inflammation, infection, hereditary disease and neoplasm. Patients with TMD ‐mimicking conditions frequently used “mouth opening limitation” ( P < .001), but less commonly used words such as “noise” ( P < .001) and “temporomandibular joint” ( P < .001) than patients with genuine TMD . A diagnostic classification tree on the basis of recursive partitioning suggested that 12.0 mm of comfortable mouth opening and 26.5 mm of maximum mouth opening were deduced as the most optimal mouth‐opening cutoff sizes. When the combined analyses were performed based on both the text mining and clinical examination data, the predictive performance of the model was 96.6% with 69.0% sensitivity and 99.3% specificity in predicting TMD ‐mimicking conditions. In conclusion, this study showed that AI technology‐based methods could be applied in the field of differential diagnosis of orofacial pain disorders.