富田 啓介

This study investigated the accuracy of a machine learning algorithm for predicting mortality in patients receiving rapid response system (RRS) activation. This retrospective cohort study used data from the In-Hospital Emergency Registry in Japan, which collects nationwide data on patients receiving RRS activation. The missing values in the dataset were replaced using multiple imputations (mode imputation, BayseRidge sklearn. linear model, and K-nearest neighbor model), and the enrolled patients were randomly assigned to the training and test cohorts. We established prediction models for 30-day mortality using the following four types of machine learning classifiers: Light Gradient Boosting Machine (LightGBM), eXtreme Gradient Boosting, random forest, and neural network. Fifty-two variables (patient characteristics, details of RRS activation, reasons for RRS initiation, and hospital capacity) were used to construct the prediction algorithm. The primary outcome was the accuracy of the prediction model for 30-day mortality. Overall, the data from 4,997 patients across 34 hospitals were analyzed. The machine learning algorithms using LightGBM demonstrated the highest predictive value for 30-day mortality (area under the receiver operating characteristic curve, 0.860 [95 % confidence interval, 0.825-0.895]). The SHapley Additive exPlanations summary plot indicated that hospital capacity, site of incidence, code status, and abnormal vital signs within 24 h were important variables in the prediction model for 30-day mortality.

To efficiently allocate medical resources at disaster sites, medical workers perform triage to prioritize medical treatments based on the severity of the wounded or sick. In such instances, evaluators often assess the severity status of the wounded or sick quickly, but their measurements are qualitative and rely on experience. Therefore, we developed a wearable device called Medic Hand in this study to extend the functionality of a medical worker's hand so as to measure multiple biometric indicators simultaneously without increasing the number of medical devices to be carried. Medic Hand was developed to quantitatively and efficiently evaluate "perfusion" during triage. Speed is essential during triage at disaster sites, where time and effort are often spared to attach medical devices to patients, so the use of Medic Hand as a biometric measurement device is more efficient for collecting biometric information. For Medic Hand to be handy during disasters, it is essential to understand and improve upon factors that facilitate its public acceptance. To this end, this paper reports on the usability evaluation of Medic Hand through a questionnaire survey of nonmedical workers.

Increased fluid overload (FO) is associated with poor outcomes in critically ill patients, especially in acute kidney injury (AKI). However, the exact timing from when FO influences outcomes remains unclear. We retrospectively screened intensive care unit (ICU) admitted patients with AKI between January 2011 and December 2015. Logistic or linear regression analyses were performed to determine when hourly %FO was significant on 90-day in-hospital mortality (primary outcome) or ventilator-free days (VFDs). In total, 1120 patients were enrolled in this study. Univariate analysis showed that a higher %FO was significantly associated with higher mortality from the first hour of ICU admission (odds ratio 1.34, 95% confidence interval 1.15-1.56, P < 0.001), whereas multivariate analysis adjusted with age, sex, APACHE II score, and sepsis etiology showed the association was significant from the 27th hour. Both univariate and multivariate analyses showed that a higher %FO was significantly associated with shorter VFDs from the 1st hour. The significant associations were retained during all following observation periods after they showed significance. In patients with AKI, a higher %FO was associated with higher mortality and shorter VFDs from the early phase after ICU admission. FO should be administered with a physiological target or goal in place from the initial phase of critical illness.

Predicting out-of-hospital cardiac arrest (OHCA) events might improve outcomes of OHCA patients. We hypothesized that machine learning algorithms using meteorological information would predict OHCA incidences. We used the Japanese population-based repository database of OHCA and weather information. The Tokyo data (2005-2012) was used as the training cohort and datasets of the top six populated prefectures (2013-2015) as the test. Eight various algorithms were evaluated to predict the high-incidence OHCA days, defined as the daily events exceeding 75% tile of our dataset, using meteorological and chronological values: temperature, humidity, air pressure, months, days, national holidays, the day before the holidays, the day after the holidays, and New Year's holidays. Additionally, we evaluated the contribution of each feature by Shapley Additive exPlanations (SHAP) values. The training cohort included 96,597 OHCA patients. The eXtreme Gradient Boosting (XGBoost) had the highest area under the receiver operating curve (AUROC) of 0.906 (95% confidence interval; 0.868-0.944). In the test cohorts, the XGBoost algorithms also had high AUROC (0.862-0.923). The SHAP values indicated that the "mean temperature on the previous day" impacted the most on the model. Algorithms using machine learning with meteorological and chronological information could predict OHCA events accurately.

While the development of prehospital diagnosis scales has been reported in various regions, we have also developed a scale to predict stroke type using machine learning. In the present study, we aimed to assess for the first time a scale that predicts the need for surgical intervention across stroke types, including subarachnoid haemorrhage and intracerebral haemorrhage. A multicentre retrospective study was conducted within a secondary medical care area. Twenty-three items, including vitals and neurological symptoms, were analysed in adult patients suspected of having a stroke by paramedics. The primary outcome was a binary classification model for predicting surgical intervention based on eXtreme Gradient Boosting (XGBoost). Of the 1143 patients enrolled, 765 (70%) were used as the training cohort, and 378 (30%) were used as the test cohort. The XGBoost model predicted stroke requiring surgical intervention with high accuracy in the test cohort, with an area under the receiver operating characteristic curve of 0.802 (sensitivity 0.748, specificity 0.853). We found that simple survey items, such as the level of consciousness, vital signs, sudden headache, and speech abnormalities were the most significant variables for accurate prediction. This algorithm can be useful for prehospital stroke management, which is crucial for better patient outcomes.