Risk factors affecting the Difficulty of Mandibular Third Molar extraction

Based on these 14 articles, risk factors affecting the difficulty of extracting mandibular third molars were abstracted and are shown in Table 1. The risk factors were weight [5,6,7, 14, 15], mouth opening [16], cheek flexibility [16], oblique ridge [16], age [5,6,7, 14, 15, 17,18,19,20], root morphology [5,6,7, 14,15,16, 20,21,22,23,24], number of roots [6, 16, 19, 23], molar angulation [6, 7, 14, 15, 17,18,19,20, 23, 24], impaction depth [6, 14, 17, 18, 21,22,23], ramus relationship available [6, 7, 21,22,23], relationship to the inferior alveolar nerve (IAN) [7, 14, 18, 19], patient anxiety [18] and surgeon experience [14, 22, 24].

Table 1 Risk factors affecting the difficulty of mandibular third molar extraction from 14 articles

Banuls et al. proposed that 3 to 5 experts should be invited if there are 10 to 20 risk factors [25]. As 13 risk factors were obtained, three experienced surgeons were invited as experts to analyse their integral interactions towards to the target difficulty factor, thus constructing the multiple hierarchy framework of the Bayesian model and providing the occurrence probability of the parent nodes and the conditional probability table of the nodes with causal relationships. This approach was used to establish the Bayesian network for predicting the difficulty of mandibular third molar extraction.

The target node “surgery difficulty” is influenced by these 13 risk factors. The “Sensitivity analysis” function of Netica software can provide information to evaluate the influence. The column “Mutual info” of sensitivity analysis (Table 2) indicates how much the beliefs of the target node could be influenced by a single finding at each of the other nodes in the net. Therefore, it can be used to evaluate the effect of the influence of the risk factors on the extraction difficulty.

Table 2 Sensitivity analysis of risk factors influencing extraction difficultyA model for predicting mandibular third molar extraction difficulty based on a bayesian network

A model for predicting mandibular third molar extraction difficulty based on a Bayesian network is shown in Fig. 2. The first-level risk factors were operation field, extraction resistance, relationship to the IAN, patient anxiety and surgeon experience. The operation field had second-level risk factors, which were weight, mouth opening, cheek flexibility and oblique ridge. Extraction resistance had second-level risk factors, which were age, root morphology, number of root, molar angulations, depth impaction and ramus available. The change in any risk factor would influence the surgery difficulty.

Fig. 2

A model of predicting the difficulty of mandibular third molar extraction based on a Bayesian network

An example of model application

There was an example about how to use the Bayesian network model in our clinical practice. The difficulty threshold should be obtained. Two surgeons (1 year out of residency) assessed surgery difficulties of 30 consecutive patients separately (60 patients in total). Two variables were used to measure the extraction difficulty: the probability of surgery difficulty based on the Bayesian network model before surgery, and the surgery results at the end of the procedure (completion, not completion and referred), as shown in Table 3. According to Table 3, the difficulty threshold for extraction of mandibular molars was 50.0%. That is, if the difficulty of extraction was below 50.0%, the surgery could be performed; otherwise, the surgery plan should be adjusted to decrease the extraction difficulty probability.

At the preoperative planning stage, the patient’s personal situation and the surgeon’s experience should be input as parent nodes. Then, the final quantitative analysis result of extraction difficulty and the key risk factors were clearly shown by the Bayesian network model. Surgeons could compare the extraction difficulty with the threshold. If the extraction difficulty was higher than the threshold, surgeons should adjust the surgical plan carefully according to the key risk factors to decrease the extraction difficulty, thereby reducing the risk of complications. Assuming a patient with normal weight, mouth opening and cheek flexibility and had a horizontal oblique ridge, age > = 24 years old, unfavourable root morphology, number of roots > 1, horizontal direction of impaction, depth of impaction at Level C, ramus relationship available at III, root contact with the mandibular canal, the patient was very anxious. The analysis results of the patient’s extraction difficulty obtained from the Bayesian network model are shown in Fig. 3a.

Fig. 3

An example of model application. (a) The model for assessing the difficulty of mandibular third molar extraction based on a Bayesian network for patients in the example. (b) The model of assessing the difficulty of mandibular third molar extraction based on a Bayesian network after adjusting “Surgeon experience”. (c) The model for predicting the difficulty of mandibular third molar extraction based on a Bayesian network after adjusting “Patient anxiety”

Table 3 The “very difficult” state based on the bayesian model and the surgery results

Due to the probability of being in a “very difficult” state being 68.1%, which exceeded the acceptable threshold for mandibular molar extraction difficulty (50.0%), it was necessary to adjust the surgical plan carefully. Among all factors, “surgeon experience” and “patient anxiety” could be adjusted. When adjusting the “Surgeon experience” to a surgeon with rich experience to perform the surgery, the Bayesian network model provided the evaluation results of extraction difficulty, as shown in Fig. 3b. At this point, the probability of being in a “very difficult” state decreased from 68.1 to 45.8%. Within the acceptable threshold for extraction difficulty, surgery could be performed.

In addition, measures could be taken to further alleviate the patient’s anxiety, such as verbal comfort or considering sedation. Assuming that the patient’s anxiety level was adjusted from ‘severe’ to ‘mild’, the Bayesian network model provided the difficulty of extraction, as shown in Fig. 3c. After adjusting the patient’s emotion, the probability of a “very difficult” state further decreased from 45.8 to 41.1%. Using appropriate methods to help patients alleviate anxiety could further reduce the difficulty of surgery.

Moreover, when the probability of the “very difficult” state is high, minimally invasive tooth extraction tools or power instruments should be used for bone removal or tooth sectioning to decrease the surgery difficulty. A coronectomy could be considered to decrease the risk from the close relationship between roots and IAN.

At the intraoperative management stage, surgeons could communicate with patients about the case condition, treatment plan, and complications and perform the surgery using appropriate methods. For those cases that cannot be adjusted to result in extraction difficulty below the threshold, referral should be considered.

At postoperative follow-up stage, high-risk patients should be more carefully monitored than low-risk patients to assess the risk of complications.