Predictive Modeling of Clinical Efficacy for 125I Brachytherapy in Head and Neck Tumors Using Lasso-Logistic Regression.

Abstract

Background: In view of the differences in the clinical efficacy of ¹²⁵I radioactive particle brachytherapy for head and neck tumors, this study aims to systematically analyze the key factors affecting its efficacy, and build a reliable prediction model to provide a scientific basis for clinical precise evaluation and personalized treatment plan formulation.

Methods: Retrospective analysis of 174 patients (2020-2024) divided into training (n=122) and validation (n=52) sets. Efficacy was assessed using RECIST criteria. Lasso Logistic regression identified independent factors, and a nomogram model was constructed and evaluated.

Results: The study confirmed that patients' age, tumor stage, tumor diameter, particle implantation dose and serum tumor marker level were independent factors affecting the clinical efficacy (P<0.05). The nomogram prediction model has excellent performance, and the c-index values in the training set and the validation set are 0.867 and 0.725, respectively, showing good discrimination ability; The results of calibration curve showed that the predicted value was in good agreement with the actual value, and the average absolute errors of the two groups were 0.114 and 0.133, respectively; In Hosmer lemeshow test, the training set χ² =7.422 (P=0.491), the validation set χ² =12.086 (P=0.147), suggesting that the model fitting effect is ideal; The area under the ROC curve in the training set and the validation set was 0.860 (95% CI:0.767-0.953) and 0.750 (95% CI:0.501-0.999), respectively, which showed high sensitivity and specificity.

Conclusion: The model effectively predicts ¹²⁵I brachytherapy outcomes, aiding clinical evaluation and supporting precision treatment for head and neck tumors.