Use of the TREAT 2.0 Model to Create a Web-Based Lung Nodule Calculator for High-Risk Patients.

Background: Indeterminate pulmonary nodules are common in clinical practice. The probability of cancer drives management decisions and depends on age, smoking, size, positron emission tomography avidity, growth, and the clinical setting, among other variables. We developed a web-based calculator for clinician use, using the Thoracic Research Evaluation and Treatment (TREAT) 2.0 model, previously developed by our group. This study evaluated calculator accuracy when the most common missing data patterns in the clinical setting occurred. Methods: A web-based calculator was designed for clinical use of the TREAT 2.0 model. We used a parsimonious modeling approach to select the 8 most important missing data patterns. Model accuracy and discrimination were evaluated using area under the curve (AUC) and Brier scores. Bootstrap internal validation was used for overfit correction. Results: The AUC for the complete cases (all 13 variables) was 0.91(CI: 0.87-0.92), and it was 0.83 (CI: 0.77-0.88) for the training and validation data sets, respectively, with a Brier score of 0.11. The mean AUC for the 8 most common missing variable submodels included in the calculator was 0.79, with a Brier score of 0.14 in the validation data set. Conclusions: The TREAT 2.0 calculator provides reliable risk predictions for indeterminate pulmonary nodules being evaluated in high-cancer prevalence settings while maintaining accuracy with incomplete input of the most common missing variables in clinical practice.