Abstract
OBJECTIVE: This study aimed to enhance the accuracy of prognosis assessment for reduction mammaplasty by improving a swarm intelligence optimization algorithm and to develop an intelligent prediction system to support clinical decision-making. METHODS: This study enrolled 224 patients who underwent reduction mammaplasty at Xijing Hospital between January 14, 2018, and February 4, 2023, and 137 patients who underwent the same procedure at Plastic Surgery Hospital between January 14, 2018, and May 1, 2020, constituting the training set. Ninety-two patients who underwent reduction mammaplasty at Plastic Surgery Hospital between May 2, 2020, and February 4, 2023, were defined as the test set. Data collection encompassed preoperative anatomical parameters, intraoperative procedural characteristics, and postoperative follow-up outcomes. Prognostic indicators included postoperative complications and the BRQS score. Guided by the Improved Secretary Bird Optimization Algorithm (ISBOA), the optimization algorithm was integrated with an AutoML framework to achieve fully automated optimization spanning from feature selection to model parameter configuration. A classification model was employed to predict the occurrence of postoperative complications, while a regression model was used to predict patient satisfaction at 1 year postoperatively. RESULTS: The ISBOA algorithm significantly outperformed other algorithms in stability, convergence speed, and avoidance of local optima. The AutoML framework achieved an ROC-AUC of 0.9369 and a PR-AUC of 0.8856 for complication prediction (test set), and an R(2) of 0.9165 for quality-of-life prediction (test set). SHAP analysis identified key features influencing complications and quality of life. Decision Curve Analysis (DCA) demonstrated that the AutoML model possessed high net benefit and stability across various threshold probabilities. The developed clinical decision support system could rapidly generate prediction results, aiding physicians in formulating personalized treatment plans. CONCLUSION: This study successfully constructed a prognosis assessment and intelligent prediction system for reduction mammaplasty based on an improved swarm intelligence optimization algorithm. The results indicate that the ISBOA algorithm exhibits significant advantages in global optimization performance and convergence efficiency. The AutoML model demonstrated excellent performance in predicting complications and assessing quality of life, with its clinical utility further validated by DCA. The developed clinical decision support system provides physicians with a convenient decision-making tool, promising to enhance the scientific rigor and efficiency of medical decision-making and offering a substantial opportunity for improving prognosis quality.