Given the increasing incidence of cancer, the need for preventive methods and optimal disease management is more crucial than ever. Home care is one effective way to improve the quality of life for cancer patients. This type of care reduces treatment costs and increases patient comfort and psychological support. A significant amount of valuable information is contained in the patients' electronic health records, but because it is embedded in unstructured narrative documents that are not easily analyzable, it remains inaccessible. In addition to its direct effects, cancer can lead to multiple side effects and complications. Two significant complications in cancer patients are cachexia and venous thromboembolism. These complications can complicate treatment and pose serious risks to patients. Therefore, predicting and managing these complications is of high importance. The aim of this research is to identify risk factors using literature review and expert opinion, and to extract them from text using text mining to structure and extract information from electronic health records. This study develops an intelligent expert system using data mining and text mining techniques to predict two major complications, cachexia and thrombosis, in cancer patients. The designed expert system has been eva‎luated using one-year clinical data collected from the MAA Cancer Institute, which provides home care services for cancer patients. The main objective of this research was to increase the accuracy of predicting these complications by analyzing and extracting relevant features from clinical texts with the help of expert opinions. To date, text mining has not been used in the literature on home care and the complications of cachexia and thrombosis. Using the intelligent expert system, 27 features were extracted from visit texts to predict complications. The use of these text-extracted features along with structured features improved the prediction of cachexia and thrombosis complications by an average of 0.86% and 1.46%, respectively, based on accuracy, and 2.95% and 14.06%, respectively, based on F-measure, compared to when text-extracted features were not used. Additionally, comparing the influential factors used in this study with those considered in previous research showed an improvement in the prediction of cachexia and thrombosis complications by 25.84% and 322.82%, respectively, based on the F-measure. Overall, modeling based on textual and structured features improved by an average of 8.50% and 174.33%, respectively, based on the F-measure compared to modeling based solely on structured features and factors introduced in the literature. Finally, the model output has been developed into an operational dashboard for user use.

صبا صارمي نيا

مهدي خاشعي آشياني , مهدي مهنام