Development and international validation of logistic regression and machine-learning models for the prediction of 10-years molar loss

To develop and validate logistic regression and artificial-intelligence based models for prognostic prediction of molar survival in periodontally-affected patients.Clinical and radiographic data from 4 different centers across 3 continents (2 in Europe, 1 in USA, and 1 in China) including 515 patients and 3157 molars were collected and used to train and test different types of machine-learning algorithms for their prognostic ability of molars over 10 years. The following models were trained: Logistic Regression, Support Vector Machine, K-Nearest Neighbors, Decision Tree, Random Forest, Artificial Neural Network, Gradient Boosting and Naive Bayes. In addition, different models were aggregated by means of Ensembled Stacking method. The primary outcome of the study was related to the prediction of overall molar loss in patients after active periodontal treatment.The general performance in the external validation settings (aggregating 3 cohorts) revealed that the Ensembled model that combined Neural Network and Logistic Regression showed the best performance among the different models for the prediction of overall molar loss with an AUC = 0.726. The Neural Network showed the best AUC = 0.724 for the prediction of periodontitis-related molar loss. In addition, the Ensembled model showed the best calibration performance.Through a multicenter collaboration, both prognostic models for the prediction of molar loss were developed and externally validated. An Ensembled model showed the best performance in terms of both discrimination and validation, it is made freely available to clinicians for widespread use in clinical practice.