Bayesian network for predicting mandibular third molar extraction difficulty.

BACKGROUND

This study aimed to establish a model for predicting the difficulty of mandibular third molar extraction based on a Bayesian network to meet following requirements: (1) analyse the interaction of the primary risk factors; (2) output quantitative difficulty-evaluation results based on the patient's personal situation; and (3) identify key surgical points and propose surgical protocols to decrease complications.

METHODS

Relevant articles were searched to identify risk factors. Clinical knowledge and experience were used to analyse the risk factors to establish the Bayesian network. First, the qualitative mechanism knowledge, including the effect of risk factors on the extraction difficulty and the causal relationships between risk factors, was analysed to establish the framework of the Bayesian network. Then, the quantitative knowledge, including the occurrence probability of the parent nodes and the conditional probability table of the nodes with causal relationships, was given by the surgeon experience and calculated using the Dempster-Shafer evidence theory. According to the framework and likelihoods and relationships of risk factors, the Bayesian network model was established.

RESULTS

This Bayesian network model analysed the weight by sensitivity of each risk factor and expressed the interaction relationship among risk factors as well as the effect of risk factors on extraction difficulty quantitatively. This Bayesian network model showed quantitative analysis results for extraction difficulty and key risk factors. The Bayesian network model revealed that the relationship to the inferior alveolar nerve, surgeon experience and patient anxiety were the most important risk factors for extraction difficulty. By integrating these patient-specific risk factors across the entire surgical process, this model could be used during preoperative planning to identify high-risk cases and to optimize resource allocation; during intraoperative management to tailor surgical techniques; and during postoperative follow-up to establish targeted follow-up protocols for high-risk patients. Moreover, this Bayesian network model can flexibly improve inclusion factors and conditional probabilities with the development of relevant research and expert opinions, as well as change states and probabilities of relevant nodes based on actual clinical conditions.

CONCLUSIONS

A model for predicting the difficulty of mandibular third molar extraction was established based on a Bayesian network.