We have been conducting research to develop a system for selecting DPC codes from patient discharge summaries using text mining techniques. The performance demonstrated thus far indicates that the classification system is sufficiently practical for real-world use. However, with the limited dataset used for training, the possibility of overfitting cannot be ruled out, necessitating performance evaluation with new data.In this study, we propose a scheme in which the classification system is constructed using discharge summaries from one fiscal year and evaluated using data from a different fiscal year. Following this scheme, we conducted experiments to test its validity.The results showed that the evaluation metrics obtained through cross-validation were less than those obtained using new data. This finding suggests that constructing the classification system while bias estimation is crucial for its successful implementation in hospital information systems.

We have been conducting research to develop a system for selecting DPC codes from patient discharge summaries using text mining techniques. The performance demonstrated thus far indicates that the classification system is sufficiently practical for real-world use. However, with the limited dataset used for training, the possibility of overfitting cannot be ruled out, necessitating performance evaluation with new data.In this study, we propose a scheme in which the classification system is constructed using discharge summaries from one fiscal year and evaluated using data from a different fiscal year. Following this scheme, we conducted experiments to test its validity.The results showed that the evaluation metrics obtained through cross-validation were less than those obtained using new data. This finding suggests that constructing the classification system while bias estimation is crucial for its successful implementation in hospital information systems.

This paper proposes visualization and simulation methods which estimates clinical indices from the data stored in the hospital information system (HIS). The method is executed as follows. First, we construct DWH where needed variables are extracted from HIS. Second, the indices (nursing needs: Score A, which consists of the histories of monitoring and procedures and Score B, which evaluates of patient’s physical condition) are calculated from extracted data. Third, chronological changes of the indices are visualized. Finally, by using the model of stochastic differential equations, the nursing care needs and its chronological changes during the admission period are estimated. We evaluated the method with the data in HIS (from 2017 to 2022). The results shows that the method correctly simulated the chronological change of nursing care needs.

We conducted an analysis of incident reports from Shimane University School of Medicine Hospital using text mining techniques to efficiently analyze the cases based on the keywords they contained. The analysis targeted the text data of incident reports filed between fiscal years 2017 and 2022, with a particular focus on reports related to personal information. We employed morphological analysis on the collected text data, organizing it word by word to facilitate easier analysis. Subsequently, techniques such as word clouds and cluster analysis were utilized. Based on the results obtained, we conducted a thorough analysis and evaluation of the issues highlighted in the incidents. It was found that incidents related to USB memory often involved them being inadvertently placed in pockets and sent for laundering. Additionally, document-related incidents could be categorized into six distinct clusters.

Last year we proposed the application of order trajectory curve analysis to hospital mangement, which will be useful to analyze the global behavior of hospital activities. This paper proposes multi-dimensional trajectories mining to analyze the temporal characteristics of hospital services. Order Trajectories mining method consists of the following two process. First the similarities between temporal trajectories of selected variables are calculated. Second, similarity-based analysis technique such as clustering and multidimensional scaling are applied. The method was evaluated on data on the number of orders extracted from hospital information system. The results showed that the method discovered several important characteristics of the divisions in the hospital..

The pandemic of COVID-19 reminds us of the basic important principles for prevention of infection: avoid the "Three Cs": closed spaces, crowded places and close contact settings. Outpatient clinics in Japan are typical examples of three Cs, where some kinds of decision support system are required to solve the above situation. This paper proposes data mining based patient navigation support system to prevent the Three Cs. Behind the systems, temporal data mining units plays an important role in providing temporal information to the patients, such as waiting time and human densities in the waiting rooms. It analyzes the data stored in hospital information systems, including patient information, logs of clinical orders. The analysis results show that several aspects of patients’ waiting are visualized by temporal data mining.

One of the most important task of data mining in hospital is to discover structured knowledge about decision making, which is useful for management of clinical process. However, most of the data in hospital information are stored without classification labels or meaning of clinical actions. Thus, unsupervised learning techniques are required for analysis. This paper proposes a method which induces a clinical pathway by using sample and attribute clustering of the histories of nursing orders stored in hospital information system. The method consists of the following five steps: first, frequencies of nursing orders are extracted from hospital information system as a dataset in which row and column represents nursing orders and days. Second, orders are classified into several groups by using sample clustering. Then, attributes clustering is applied to the data for feature selection. Fourth, for each sample and attribute clustering, the number of clusters are obtained from the sequence of the height values and according to the results of attribute clustering, the original dataset is decomposed into subtables. Then, the second to fourth steps will be repeated in a recursive way until the grouping of attributes (days) are stable. Finally, a new pathway will be constructed from all the induced results. The proposed method was evaluated on datasets extracted from a hospital information system. The experiment results show that the method is useful for construction of a clinical pathway when the distribution of length of stay is uni-modular.

Two of the most important roles of hospital information system (HIS) is to transfer clinical orders issued by doctors and nurses to other division and to store results of executed orders. Thus, the numbers of issued and executed orders will reflect the clinical activities in large hospitals. This paper proposes a visualization technique, called order trajectory analysis which visualize the temporal sequences of the number of orders. Then, clustering is applied to the order trajectory in order to show the similarities between clinical divisions.

This paper proposes a method which classifies discharge summaries stored in hospital information system, which consists of the following four steps. First, a term matrix of the set of summaries is induced by morphological analysis (RMecab). Next, correspondence analysis is applied to the term matrix and numerical values of two dimensional coordinates are assigned to each keyword and each concept. By measuring the euclidean distance between categories and keywords, keywords are ordered. Then, keywords are selected as attributes according to the rank, and training examples for classifiers will be generated. Finally, learning methods are applied to the training examples. Experimental validation shows that random forest achieved the best performance and deep learning (multiple layer perceptron) is the second best.

This paper proposes construction of clinical process plan from nursing order histories and discharge summaries stored in hospital information system. First, the system extracts subgrouping from clinical cases with the same Diagnostic Procedure Combination code (DPC) by mixture model clustering. Subgroups give different types of diseases with different temporal evolution. Then, classification models of each subgroup are constructed by the analysis of discharge summaries to capture the meaning of each subgroup. Finally, cases are classified by using the classification model and a clinical pathway is generated for each new subgroup. The proposed method was evaluated on the datasets extracted hospital information system, whose results show that plausible clinical pathways were obtained, compared with previously introduced methods.

Electronic patient records (EPR) are rich in texts, where almost all the decision making processes of medical staff are written. Thus, mining in EPR is important for acquision of decision making process and diagnosis. In this paper, as a first step, we focus on text mining for discharge summaries, which include the compact explanation for the patient's admission. a record of her complaints, physical findings, laboratory results and radiographic studies while hospitalized; a list of changes in her medications at discharge; and recommendations for follow up care. Text mining process consists of the following four processes: first, morphological analysis is applied to a set of summaries and a term matrix is generated. Second, correspond analysis is applied to the classification labels and the term matrix and generates two dimensional coordinates. By measuring the distances between categories and the assigned points, ranking of key words will be generated. Then, keywords are selected as attributes according to the rank, and training examples for classifiers will be generated. Finally, learning methods are applied to the training examples. Experimental validation shows that random forest achieved the best performance and the second best was the deep learner with a small difference, but decision tree methods with many keywords performed only a little worse than neural network or deep learning methods.

This paper proposes an active mining process for improvement of quality of clinical process by using service logs in a hospital information system. First, datasets of temporal change of the number of orders are extracted from service logs stored in hospital information system. Then, since datasets of temporal change can be viewed as time-series of a statistic, clustering can be applied to the data. By using the groups obtained, datasets of command sequences are extracted from the logs and sequence mining process is applied. The results of sequence mining are interpreted with the results of clustering and hypothesis will be generated. The results show that the method improved the clinical process and waiting time in outpatient clinic.