Cross-sectional validation of a preoperative multidimensional assessment tool for third molar extraction using machine learning.

Abstract

Background: This study aimed to assess the correlation between a preoperative difficulty assessment form and surgerytime in third molar extractions. Secondary aims included analyzing the relationship between the form’s globalscore and surgical technique, surgeon-perceived difficulty, and postoperative complications. Machine learning modelswere also explored.Material and Methods: A cross-sectional study was conducted on patients requiring a third molar extraction betweenApril 2022 and May 2023, operated by students of the master's degree in Oral Surgery and Implantology (Universityof Barcelona, Spain). Three previously calibrated investigators completed the preoperative difficulty evaluation form,which included clinical, radiological and surgical variables. Descriptive and bivariate analyses were performed usingthe Stata/IC 15.1. Machine learning models were applied to predict surgery time and surgeon-perceived difficulty.Results: A total of 205 patients, 75 males (36.6%) and 130 females (63.4%), with a mean age of 28.5±14.1 years wereincluded; 49 (23.9%) were upper and 156 (76.1%) lower third molars. The global score of the form was significantlycorrelated to surgery time (Spearman's rho=0.640; P<0.001), perceived difficulty (Spearman's rho=0.395; P<0.001)and the surgical technique according to Parant’s classification (P<0.001). Postoperative complications were associatedto higher scores on the difficulty evaluation form (P=0.012).Conclusions: The preoperative assessment form is a valid tool for estimating third molar extraction difficulty. Its globalscore is positively associated with surgery time, perceived difficulty, surgical technique, and postoperative complications.Machine learning models showed better performance at modeling surgery time than surgical difficulty.