A Data-driven Approach to Survival Prediction in Middle-aged and Elderly Primary Ovarian Cancer Patients: an Application of Machine Learning Based on the SEER Database

Ovarian cancer (OC) is one of the primary causes of death from malignant tumors of the female reproductive system, which presents unique challenges due to the fact that, since it usually has no obvious clinical symptoms in the early stages, the majority of cases have already progressed to terminal stages at the time of diagnosis and have a poor prognosis, especially among middle-aged and elderly women. This study leverages advanced data mining techniques and machine learning algorithms to undertake a novel analysis of the relationship between clinical parameters and survival outcomes in primary ovarian cancer patients within these demographics, utilizing the extensive dataset provided by the SEER (Surveillance, Epidemiology, and End Results) database. Innovatively applying both the Cox proportional hazard model and Pearson correlation coefficients to examine multifaceted clinical interactions, this research breaks new ground in understanding how treatment variables correlate with survival rates. Further, this study pioneers the use of a variety of predictive models including Logistic Regression, Decision Tree, Naive Bayes, and an AdaBoost classifier, the latter of which demonstrates superior predictive performance. By adopting a randomized search for hyper-parameter optimization, the AdaBoost model's predictive accuracy for three- and five-year survival rates significantly improved from initial benchmarks of 0.89 and 0.91 to 0.91 and 0.93, respectively. The model's precision, recall, and F1 score metrics all exceeded 0.92, underscoring its robustness and precision. Moreover, ROC curve analysis further highlights the model's enhanced discriminatory power with an average AUC value of 0.94, a notable increase from the baseline of 0.87. This study not only advances our statistical methodologies in oncological prognostication but also provides a more reliable tool for clinicians in the personalized treatment planning for ovarian cancer patients.