Construction of a Clinical Prediction Model for the Risk of Iatrogenic Vertical Root Fracture Using Multiple Clinical Indicators

It contains clinical and radiographic data from 80 cases of endodontically treated molar or premolar teeth. Data were collected between January 2013 and December 2024 at the Department of Stomatology, Tianjin Nankai Hospital.The dataset comprises 47 variables covering the following dimensions: (1) patient demographics (age, sex); (2) tooth anatomy (tooth type, number of roots, number of canals, root curvature); (3) treatment-related factors (type of endodontic treatment, retreatment history, apical surgery, canal shaping technique, filling technique, post placement); (4) coronal restoration status (restoration type, restoration quality); (5) radiographic features (CBCT-assessed root canal filling density, apical filling extent, perirradicular hypodensity, J-shaped hypodensity, etc.); and (6) clinical signs (periodontal pocket, sinus tract, tenderness to percussion, tenderness to palpation, tooth mobility, etc.).

The outcome variable is VRF presence.Data preprocessing: Categorical variables were one-hot encoded; continuous variables (age, crown-to-root ratio, root curvature angle) were Z-score normalized. To address class imbalance (22 VRF-positive vs. 58 VRF-negative cases), SMOTE oversampling was applied to the training set prior to machine learning feature selection, generating a balanced dataset (116 cases). Three machine learning algorithms (XGBoost, Random Forest, LASSO) were used for feature selection, and a final risk score model was constructed using Firth-penalized logistic regression.The dataset is provided as a single Excel file (Training_Set_80.xlsx) containing 80 rows (one per case) and 47 columns.