Background: The coronavirus disease poses an urgent threat to global public health and is characterized by rapid disease progression even in mild cases. In this study, we investigated whether machine learning can be used to predict which patients will have a deteriorated condition and require oxygenation in asymptomatic or mild cases of COVID-19.Methods: This single-center, retrospective, observational study included COVID-19 patients admitted to the hospital from February 1, 2020, to May 31, 2020, and who were either asymptomatic or presented with mild symptoms and did not require oxygen support on admission. Data on patient characteristics and vital signs were collected upon admission. We used seven machine learning algorithms, assessed their capability to predict exacerbation, and analyzed important influencing features using the best algorithm.Results: In total, 210 patients were included in the study. Among them, 43 (19%) required oxygen therapy. Of all the models, the logistic regression model had the highest accuracy and precision. Logistic regression analysis showed that the model had an accuracy of 0.900, precision of 0.893, and recall of 0.605. The most important parameter for 4 predictive capability was SpO2, followed by age, respiratory rate, and systolic blood pressure. Conclusion:In this study, we developed a machine learning model that can be used as a triage tool by clinicians to detect high-risk patients and disease progression earlier.Prospective validation studies are needed to verify the application of the tool in clinical practice.

Ventilator liberation is one of the most critical decisions in the intensive care unit; however, prediction of extubation failure is difficult, and the proportion thereof remains high. Machine learning can potentially provide a breakthrough in the prediction of extubation success. A total of seven studies on the prediction of extubation success using machine learning have been published. These machine learning models were developed using data from electronic health records, 8–78 features, and algorithms such as artificial neural network, LightGBM, and XGBoost. Sensitivity ranged from 0.64 to 0.96, specificity ranged from 0.73 to 0.85, and area under the receiver operating characteristic curve ranged from 0.70 to 0.98. The features deemed most important included duration of mechanical ventilation, PaO2, blood urea nitrogen, heart rate, and Glasgow Coma Scale score. Although the studies had limitations, prediction of extubation success by machine learning has the potential to be a powerful tool. Further studies are needed to assess whether machine learning prediction reduces the incidence of extubation failure or prolongs the duration of ventilator use, thereby increasing tracheostomy and ventilator-related complications and mortality.

Background: Trauma is a serious medical and economic problem worldwide, and patients with trauma injuries have a poor survival rate following cardiac arrest. This study aimed to create a prediction model specific to prehospital trauma care and to achieve greater accuracy with techniques of machine learning.Methods: This retrospective observational study investigated data of patients who had blunt trauma injuries due to traffic accident and fall trauma from January 1, 2018, to December 31, 2019, using the National Emergency Medical Services Information System, which stores emergency medical service activity records nationwide in the United States. Random forest was used to develop a machine learning model. Results:Per the prediction model, the area under the curve of the predictive model was 0.95 and negative predictive value was 0.99. The feature importance of the predictive model was the highest for the AVPU scale (an acronym from "Alert, Verbal, Pain, Unresponsive"), followed by oxygen saturation (SpO2). Among patients who were progressing to cardiac arrest, the cutoff value was 89% for SpO2 in unalert patients.Conclusions: Patients whose conditions did not progress to cardiac arrest could be identified with high accuracy by machine learning model techniques.

No abstract

Background: It is difficult to predict vancomycin trough concentrations in critically ill patients as their pharmacokinetics change with the progression of both organ failure and medical intervention. This study aims to develop a model to predict vancomycin trough concentration using machine learning (ML) and to compare its prediction accuracy with that of the population pharmacokinetic (PPK) model. Methods: A single-center retrospective observational study was conducted. Patients who had been admitted to the intensive care unit, received intravenous vancomycin, and had undergone therapeutic drug monitoring between 2013 and 2020,were included. Thereafter, ML models were developed with random forest, LightGBM, and ridge regression using 42 features. Mean absolute errors (MAE) were compared and important features were shown using LightGBM. Results: Among 335 patients, 225 were included as training data and 110 were used for test data. A significant difference was identified in the MAE by each ML model compared with PPK;4.13 ± 3.64 for random forest, 4.18 ± 3.37 for LightGBM, 4.29 ± 3.88 for ridge regression, and 6.17 ± 5.36 for PPK. The highest importance features were pH, lactate, and serum creatinine. Conclusion: This study concludes that ML may be able to more accurately predict vancomycin trough concentrations than the currently used PPK model in ICU patients.