Monitoring and early warning of ovarian cancer using high-dimensional non-parametric EWMA control chart based on sliding window

Abstract Ovarian tumors are one of the common ovarian dysfunctions in women, which to some extent affect women's normal work and life. Although ovarian tumors are usually sensitive to chemotherapy, and typically show initial efficacy with platinum/taxane treatment, the postoperative recurrence rate in advanced patients is very high. There have been a considerable number of researchers working to establish new methods to monitor and warn about the progression and prognosis of this malignant tumor disease. One commonly used approach is to first reduce the dimensionality of the data using methods such as principal component analysis, LASSO, deep learning, etc., to select several features most relevant to malignant tumors. Then, either a one-dimensional control chart is used for multiple monitoring and warning of different indicators, or a multivariate control chart is directly used to monitor and warn about the selected indicators. However, in actual data, different features are not completely independent of each other. Using a one-dimensional control chart for multiple monitoring and warning of different indicators may overlook the interactions between different features. Additionally, reducing the dimensionality of the data may result in the loss of some data information, leading to the omission of details, which may affect the accuracy of the model and result in delayed alarms and poor predictive performance. Therefore, this paper proposes a non-parametric monitoring scheme based on high-dimensional empirical likelihood test and sliding window model of EWMA type, for direct online monitoring and warning of high-dimensional ovarian tumor data. We compared this approach with a multivariate EWMA control chart after dimensionality reduction, and Monte Carlo numerical simulation results showed that the high-dimensional non-parametric EWMA monitoring scheme indeed detects tumor data changes faster and issues alarms more promptly than the dimensionality-reduced multivariate EWMA control chart. Furthermore, we further validated the effectiveness of the high-dimensional non-parametric EWMA monitoring scheme in monitoring and warning using tumor resection data from the Third Affiliated Hospital of Soochow University as an example.