Predicting adverse clinical events leads to preventing patients from severe condition. In the prediction of clinical events by machine learning, not only accuracy but also usability and explainability in a clinical decision support (CDS) system are crucial because they help to understand and trust the model. This research aims to develop a machine learning model to predict clinical events, to implement a CDS system that provides the prediction results and the explanations, and to evaluate the usefulness of the system. An acute heart failure predictive model was developed with records of 475 patients who were hospitalized for congenital heart diseases from 2015 to 2017. We calculated 65 features with sliding window approach from numeric time-series data extracted from electronic health records (EHR) and constructed a random forest model. The acute heart failure events were predicted at AUC=0.88. We developed a CDS system that connects to EHR and PACS, and used the predictive model to prospectively detect the sign of acute heart failure. The prediction accuracy of the prospective evaluation was AUC=0.76. By evaluating the accuracy, usability and explainability, the usefulness of CDS with a machine learning model was shown.

　The face contains facial muscles whose movements create facial expressions. The purpose of this research is to clarify the facial expressions involved during stress in children with severe motor and intellectual disabilities (hereafter, children with SMID) in whom facial expressions are not always easy to understand. At a particular point, we observed the daily life of children with SMID and collected the numerical data on heart rate alteration and facial muscle movement with sensor devices. Based on the fluctuation in the sympathetic nervous system, which was activated due to anxiety and distress, and an increase in heart rate, we focused on the variation in facial muscles and the attitude of the children, and statistically analyzed the observed data. We analyzed three scenes that were considered to be stress in one case (Oshima’s classification 1). The results of multiple regression analysis showed that movement of the facial muscles “Eyes Closed” was a common explanatory variable. When the stress was greater, I was able to explain the movement of the facial muscles to “Outer Brow Raiser”. We could visualize the changes in facial expression in children with SMID, which are subtle and difficult to see with the naked eye.

　【Background】 The National Database of Health Insurance Claims and Specific Health Checkups of Japan (hereinafter, the NDB) is a complete inventory of data relating to healthcare services provided by health insurance in Japan. While patient deaths are described in the NDB, these lack accuracy. To address unspecified deaths, a mortality estimation program was created using machine learning (decision-tree analysis) from data points such as courses of medical treatment and prescription medication, with reference to entries for Nara Prefecture in the National Health Insurance Database of Japan (the “Kokuho Database”; a database of healthcare services provided by health insurance companies in Japan similar to the NDB), which have few omissions. We used these data to track mortality and evaluate postoperative prognoses.

　【Methods】 This study was conducted as part of the evaluation indicator development for health care planning. Four years’ worth of Nara KDB receipts and three years’ worth of NDB receipts were used. Mortality outcomes listed in the master insurer in the KDB were used as teacher data to identify the percentage of correct answers to mortality outcomes in the KDB and the required decision tree practices. Decision tree analysis in R language was used for the analysis. We calculated the mortality rate and standardized mortality ratio (SMR) for patients who underwent total gastrectomy and percutaneous coronary stent placement, both common surgical procedures in internal medicine.

　【Results】 The positive prediction rate of this mortality logic was 96.2％. Among the medical treatments/services most likely to predict mortality were “additional caregiving,” “electrocardiogram monitor and pulse,” and “oxygen inhalation.” Sensitivity was 92.9％ and specificity was 99.7％. Application of this program to three years of data in the NDB confirmed the same trends. The one-year mortality rate after total gastrectomy in 2015 among patients who were over 40 years old was (15.9％, 16.4％; men, women), expected mortality rate divided by one-year mortality rate after surgery was (13.7％, 14.4％), and the SMR was 486,913. Age-specific mortality remained stable until the sixties then gradually increased from 70 years. Excess deaths were more common in patients over 80 years of age who underwent total gastrectomy than in patients who underwent percutaneous coronary stenting.

　【Conclusion】 There was more excess death in patients over 80 years who underwent total gastrectomy than patients who underwent percutaneous coronary stent placement. This means the prognosis for patients who are over 80 is worse. In future, we need to calculate for a variety of diseases, not just these two procedures.