Choi DH, et al. J Korean Med Sci. 2021 Jan;36(5):e44. https://doi.org/10.3346/jkms.2021.36.e44

Objectives: Medical artificial intelligence (AI) has recently attracted considerable attention. However, training medical AI models is challenging due to privacy-protection regulations. Among the proposed solutions, federated learning (FL) stands out. FL involves transmitting only model parameters without sharing the original data, making it particularly suitable for the medical field, where data privacy is paramount. This study reviews the application of FL in the medical domain.Methods: We conducted a literature search using the keywords "federated learning" in combination with "medical," "healthcare," or "clinical" on Google Scholar and PubMed. After reviewing titles and abstracts, 58 papers were selected for analysis. These FL studies were categorized based on the types of data used, the target disease, the use of open datasets, the local model of FL, and the neural network model. We also examined issues related to heterogeneity and security.Results: In the investigated FL studies, the most commonly used data type was image data, and the most studied target diseases were cancer and COVID-19. The majority of studies utilized open datasets. Furthermore, 72% of the FL articles addressed heterogeneity issues, while 50% discussed security concerns.Conclusions: FL in the medical domain appears to be in its early stages, with most research using open data and focusing on specific data types and diseases for performance verification purposes. Nonetheless, medical FL research is anticipated to be increasingly applied and to become a vital component of multi-institutional research.

Objective We evaluated the performance of diastolic shock index (DSI) and lactate in predicting vasopressor requirement among hypotensive patients with suspected infection in an emergency department.Methods This was a single-center, retrospective observational study for adult patients with suspected infection and hypotension in the emergency department from 2018 to 2019. The study population was split into derivation and validation cohorts (70/30). We derived a simple risk score to predict vasopressor requirement using DSI and lactate cutoff values determined by Youden index. We tested the score by the area under the receiver operating characteristic curve (AUC). We performed a multivariable regression analysis to evaluate the association between the timing of vasopressor treatment and 28-day mortality.Results A total of 1,917 patients were included. We developed a score, assigning 1 point each for the high DSI (≥2.0) and high lactate (≥2.5 mmol/L) criteria. The AUCs of the score were 0.741 (95% confidence interval [CI], 0.715–0.768) at hypotension and 0.736 (95% CI, 0.708–0.763) after initial fluid challenge in the derivation cohort and 0.676 (95% CI, 0.631–0.719) at hypotension and 0.688 (95% CI, 0.642–0.733) after initial fluid challenge in the validation cohort, respectively. In patients with scores of 2 points, early vasopressor therapy initiation was significantly associated with decreased 28-day mortality (adjusted odds ratio, 0.37; 95% CI, 0.14–0.94).Conclusion A prediction model with DSI and lactate levels might be useful to identify patients who are more likely to need vasopressor administration among hypotensive patients with suspected infection.

Introduction The purpose of this study was to construct a prediction model for endotracheal tube depth using neck CT images. Methods A retrospective image review was conducted that included patients who had undergone neck CT. Using sagittal neck CT images, we calculated the length between upper incisor and mid-trachea and then derived the model via regression analysis. The model was validated externally using chest radiographs of patients who had undergone endotracheal intubation. We compared performance of our model with that of other methods (Broselow tape and APLS formula) via Bland-Altman analysis and the percentage of estimations within 10% of the measured values. Results A total of 1111 children were included in this study. The tube depth obtained from CT images was linearly related to body weight (tube depth (cm)=5.5+0.5×body wt (kg)) in children younger than 1 year and to height (tube depth (cm)=3+0.1×height (cm)) in children older than 1 year. External validation demonstrated that our new model showed better agreement with the desired tube depth than Broselow tape and APLS formula. The mean differences in children younger than 1 year were 0.61 cm and −1.24 cm for our formula and Broselow tape, respectively. The mean differences in children older than 1 year were −0.43 cm, −1.98 and −1.64 cm for our formula, Broselow tape and APLS formula, respectively. The percentages of estimates within 10% of the measured values were 52.7% and 35.8% for our formula and Broselow tape in children younger than 1 year, respectively, and 54.3%, 33.8% and 37.2% for our formula, Broselow tape and APLS formula in children older than 1 year, respectively (P<0.01). Conclusion Our new formula is useful and more accurate than the currently available methods.

Background: An incentive spirometer (IS) is a medical device used to help patients improve the functioning of their lungs. It is provided to patients who have had any surgery that might jeopardize respiratory function. An incentive spirometer plays a key role in the prevention of postoperative complications, and the appropriate use of an IS is especially well known for the prevention of respiratory complications. However, IS utilization depends on the patient’s engagement, and information and communication technology (ICT) can help in this area. Objective: This study aimed to determine the effect of mobile ICT on the usage of an IS (Go-breath) app by postoperative patients after general anesthesia. Methods: For this study, we recruited patients from April to May 2018, who used the Go-breath app at a single tertiary hospital in South Korea. The patients were randomly classified into either a test or control group. The main function of the Go-breath app was to allow for self-reporting and frequency monitoring of IS use, deep breathing, and active coughing in real time. The Go-breath app was identical for both the test and control groups, except for the presence of the alarm function. The test group heard an alarm every 60 min from 9 am to 9 pm for 2 days. For the test group alone, a dashboard was established in the nurse’s station through which a nurse could rapidly assess the performance of multiple patients. To evaluate the number of performances per group, we constructed an incentive spirometer index (ISI). Results: A total of 44 patients were recruited, and 42 of them completed the study protocol. ISI in the test group was 20.2 points higher than that in the control group (113.5 points in the test group and 93.2 points in the control group, P=.22). The system usability scale generally showed almost the same score in the 2 groups (79.3 points in the test group and 79.4 points in the control group, P=.94). We observed that the performance rates of IS count, active coughing, and deep breathing were also higher in the test group but with no statistically significant difference between the groups. For the usefulness “yes or no” question, over 90% (38/42) of patients answered “yes” and wanted more functional options and information. Conclusions: The use of the Go-breath app resulted in considerable differences between the test group and control group but with no statistically significant differences. Trial Registration: ClinicalTrials.gov NCT03569332; https://clinicaltrials.gov/ct2/show/NCT03569332 (Archived by WebCite at http://www.webcitation.org/74ihKmQIX).

Intussusception is common among children at the pediatric emergency department (ED) with acute abdomen. Diagnosis and treatment delay remain a challenge. This study aimed to evaluate the impact of intussusception clinical pathways (CPs) implementation, including bedside point-of-care ultrasonography, on patient management in a pediatric ED. In January 2017, an intussusception management protocol was implemented for children with symptoms of intussusception. We retrospectively examined the charts of patients diagnosed with intussusception during the preprotocol (January 2015 to December 2016) and postprotocol (January 2017 to January 2019) periods and compared their outcomes. A total of 106 and 108 patients were included in the preprotocol and postprotocol groups, respectively. After CP implementation, the median door-to-ultrasonography time decreased from 66.5 (range: 13, 761) to 54 (20, 191) minutes; meanwhile, door-to-reduction time decreased from 121.5 (37, 1077) to 80.5 (40, 285) minutes; the median ED length of stay decreased from 440 to 303.5 minutes; and finally, admission rate increased from 18.9% to 40.7% (P < .01). There was no between-group difference in the rates of complications, readmission, emergency surgery, or reduction failure. The implementation of an intussusception CP decreased time-to-diagnosis, time-to-treatment, and ED length of stay estimates among children screened using point-of-care ultrasonography. The present findings suggest that the implementation of an intussusception CP may improve the efficiency of time and resource use.