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

Ovarian tumors are a common ovarian dysfunction that affects women's daily lives. Although ovarian tumors are generally sensitive to chemotherapy and initially respond well to platinum/taxane-based treatments, the postoperative recurrence rate remains high in advanced cases. Many researchers are dedicated to developing new methods for monitoring and predicting malignant tumors. Traditional approaches use dimensionality reduction techniques, like principal component analysis and deep learning, to select relevant features, followed by univariate or multivariate control charts for monitoring. However, these methods may overlook interactions between features and dimensionality reduction can result in loss of information, potentially affecting the accuracy of the model and leading to delayed alerts and reduced predictive performance. Therefore, this paper develops a new sliding window EWMA control chart based on high-dimensional empirical likelihood ratio tests. This control chart not only monitors data with unknown underlying distributions but is also applicable to high-dimensional data, allowing for monitoring without dimensionality reduction, thus simplifying the process and avoiding information loss. Monte Carlo results show that this method detects changes in indicators and issues alerts more rapidly than the dimensionality-reduced multivariate EWMA control charts. In addition, we further validated the effectiveness of this method through analysis of a tumor resection data example.