Dynamic Prediction of Outcome for Patients With Ovarian Cancer: Application of a Joint Model for Longitudinal Cancer Antigen 125 Values

Abstract

ObjectiveIn clinical practice, gynecologic oncologists are interested in predicting the prognosis of patients through information from different sources. Recently, the overall survival (OS) of ovarian cancer patients has been widely and intensively researched, and a large number of risk factors have been determined, including the biomarker of cancer antigen 125 (CA-125). For newly diagnosed patients, it is critical to construct effective prognostic models to predict prognosis dynamically by combining their CA-125 values with adjusted clinical factors.Methods/MaterialsA total of 227 ovarian cancer participants entered this study. A 4-step method was used to construct a joint model to examine the association between longitudinal CA-125 measurements and OS time, to explore time-independent predictive factors influencing OS, and to obtain an accurate and credible dynamic prediction of OS for specific subjects.ResultsWe found that CA-125 values were greatly affected by observation time, menarche, Federation International of Gynecology and Obstetrics stage, and ascites at baseline. Similarly, CA-125 values, menopause, Federation International of Gynecology and Obstetrics stage, and surgery state at baseline were selected from the best Cox proportion hazard model and showed a strong correlation with OS. In addition, the analyses presented by the joint model depict that, as time goes by, increasing CA-125 was deemed to be a significant predictor of death.ConclusionsTogether, our results show that a joint model could be highly efficient in clinical consultation and diagnosis for patients newly diagnosed as having ovarian cancer. Longitudinal CA-125 values, which are measured over time, can be used to credibly predict OS after taking all adjusted covariates into account.