Technology, Machine Learning, and Bias in Emergency Care
Permanent URI for this collectionhttps://hdl.handle.net/10125/107491
Browse
Recent Submissions
Item type: Item , Machine Learning Models for Point-of-care Ultrasound Education and Training: National Cost Savings and Expert Time Reduction(2024-01-03) Driver, Lachlan; Duggan, Nicole M.; Brower, Chares; Ebnali, Mahdi; Baymon, Da’Marcus; Wagner, Alexei; Dias, Roger; Samir, Anthony E.; Kapur, Tina; Goldsmith, Andrew J.; Baugh, Christopher W.Point-of-care ultrasound (POCUS) serves as a valuable diagnostic tool for healthcare providers. It enhances diagnostic accuracy and patient outcomes while reducing Emergency department (ED) length-of-stay and expenses. Nonetheless, barriers such as access to instructors and the costs of training novices impede widespread POCUS implementation. One alternative is artificial intelligence (AI) guided image acquisition tools. This study explores the potential national cost savings of employing AI acquisition software to teach POCUS to residents. A Monte Carlo simulation estimated the hours and costs of attending physician time needed for traditional versus AI-guided ultrasound education. The findings suggest that incorporating AI-guidance in ED resident ultrasound education could save $5.3 million annually in costs nation-wide. This cost-effective method holds the potential to maintain or even enhance quality of education while alleviating financial constraints. Investing in AI technology for medical education has the potential for improved patient care and streamlined workflows in healthcare environments.Item type: Item , Head Motion Analysis is an Objective Measure of Point-of-Care Ultrasound Image Acquisition Competency(2024-01-03) Walsh, Carrie; Duggan, Nicole M.; Dias, Roger; Goldsmith, Andrew; Ebnali, Mahdi; Schwid, Madeline; Fischetti, Chanel; Driver, Lachlan; Stegeman, Joseph; Bernier, Denie; Selame , Lauren; Plevek, PhillipBackground: POCUS education and competency milestones are required for emergency medicine residency graduation. Currently, POCUS competency is assessed using OSCEs. This approach is non-standardized, subjective, and resource intensive. In this pilot study we determine the ability of head motion analysis to differentiate between novice-and-expert-level POCUS performance. Methods: Fifteen emergency medicine physicians, eight novice POCUS users and nine experts, performed cardiac and FAST exams. Head motion was tracked using Muse2-headband with accelerometer and gyroscope sensors. Fellowship-trained experts observed all exams and independently recorded OSCE scores. Results: Experts scored higher in OSCEs than novices in both examinations (p<0.00001). Experts demonstrated less head motion distribution in the X,Y and Z-directions, with significant differences (p<0.001) between expert and novice groups. Conclusions: Head-motion metrics can differentiate novice-and expert-level ultrasonographers, which could offer objective competency assessments for new POCUS learners. Additional studies are needed to identify minimum threshold values for defining competency based on these metrics.Item type: Item , Hospital Delirium is Associated with Lower Mean Activity Counts: Secondary Analysis of a Large Cohort Study of ICU Patients(2024-01-03) Southerland, Lauren; Peng, Jing; Boyer, Edward; Brummel, NathanHospital delirium is a dangerous condition characterized by confusion and altered consciousness. Hypoactive delirium, the most common type of delirium, results in decreased spontaneous movement and is easily missed by hospital staff. We evaluated the use of wrist accelerometers to detect an association with delirium in intensive care unit patients. We found that daily mean activity count was lower in patients with delirium, even controlling for age and mechanical ventilation status. This suggests that accelerometers could be a good biosensor to assist hospital staff with delirium detection and management.Item type: Item , Artificial Intelligence for End Tidal Capnography Guided Resuscitation: A Conceptual Framework(2024-01-03) Nassal, Michelle; Sugavanam, Nithin; Aramendi, Elisabete; Jaureguibeitia, Xabier; Elola, Andoni; Panchal, Ashish; Ulintz, Alexander; Wang, Henry; Ertin, EmreArtificial Intelligence (AI) and machine learning have advanced healthcare by defining relationships in complex conditions. Out-of-hospital cardiac arrest (OHCA) is a medically complex condition with several etiologies. Survival for OHCA has remained static at 10% for decades in the United States. Treatment of OHCA requires the coordination of numerous interventions, including the delivery of multiple medications. Current resuscitation algorithms follow a single strict pathway, regardless of fluctuating cardiac physiology. OHCA resuscitation requires a real-time biomarker that can guide interventions to improve outcomes. End tidal capnography (ETCO2) is commonly implemented by emergency medical services professionals in resuscitation and can serve as an ideal biomarker for resuscitation. However, there are no effective conceptual frameworks utilizing the continuous ETCO2 data. In this manuscript, we detail a conceptual framework using AI and machine learning techniques to leverage ETCO2 in guided resuscitation.Item type: Item , Expert-quality Dataset Labeling via Gamified Crowdsourcing on Point-of-Care Lung Ultrasound Data(2024-01-03) Duggan, Nicole M.; Jin, Mike; Duhaime, Erik; Kapur, Tina; Duran Mendicuti, Maria Alejandra; Hallisey, Stephen; Bernier, Denie; Selame , Lauren; Asgari-Targhi, Ameneh; Fischetti, Chanel; Lucassen, Ruben; Samir, Anthony E.data interpretation. Building such tools requires labeled training datasets. We tested whether a gamified crowdsourcing approach can produce clinical expert-quality lung ultrasound clip labels. 2,384 lung ultrasound clips were retrospectively collected. Six lung ultrasound experts classified 393 of these clips as having no B-lines, one or more discrete B-lines, or confluent B-lines to create two sets of reference standard labels: a training and test set. Sets trained users on a gamified crowdsourcing platform, and compared concordance of the resulting crowd labels to the concordance of individual experts to reference standards, respectively. 99,238 crowdsourced opinions were collected from 426 unique users over 8 days. Mean labeling concordance of individual experts relative to the reference standard was 85.0% ± 2.0 (SEM), compared to 87.9% crowdsourced label concordance (p=0.15). Scalable, high-quality labeling approaches such as crowdsourcing may streamline training dataset creation for machine learning model development.Item type: Item , Assessing the Functionality and Comfort of Chest Heart Rate Monitor Use During Acute Orthopedic Trauma Surgery(2024-01-03) Marquardt, Matthew; Emerson, Angela; Orr, Morgan; Hagen, Joshua; Quatman, CarmenObjective: This study sought to investigate whether wireless chest heart rate (HR) monitors function properly and are comfortable when worn under surgical apparel and lead. Methods: Three participants donned chest heart rate monitors, surgical scrubs, and surgical lead aprons in a simulated operating room. For approximately 40 minutes, they conducted a series of movements that mimics those used during acute trauma surgeries while heart rate metrics (including heart rate variability) data was collected and comfort was evaluated. Results: All chest HR monitors stayed in position and did not produce discomfort in any participants. Additionally, despite their location under surgical lead, the HR monitors successfully transmitted data to the collection hub 98.92% of the time. Conclusions: Chest-worn HR monitors function properly and are comfortable to wear in an operating room environment, opening the possibility for trauma surgeons to use these devices to study their physiologic response to different operations.Item type: Item , Preliminary Feasibility Study of a Simulated Overdose Antidote Delivered by an Unmanned Aerial Vehicle in an Urban Environment(2024-01-03) Ulintz, Alexander; Veliky, Cole; Scott, Derek; Nassal, Michelle; Panchal, Ashish; Wang, Henry; Boyer, Edward; Lyons, MichaelUnmanned aerial vehicles (UAVs; ‘drones’) deliver time-sensitive health care tools to out-of-hospital environments. Many emergency response systems struggle to deliver antidote to victims of opioid overdose before respiratory depression results in morbidity or mortality; thus, UAVs may play a useful role in antidote delivery for out-of-hospital toxicologic emergencies. We tested the feasibility of dropping simulated antidote from a UAV to a bystander in an urban environment, measuring accuracy of drop, ease of recovery, and antidote survivability. A minimum flight altitude of 40m avoided any obstacles to accurately fly to specific coordinates. Simulated antidote drifted an average of 48 feet from the intended target, was discoverable on the ground, and survived the drop. These findings imply that UAV-dropped antidote may be a potential tool in emergency response to opioid overdose. Future research should focus on mechanisms for UAV integration within existing opioid overdose emergency response systems, human-UAV interactions, and payload design.Item type: Item , Introduction to the Minitrack on Technology, Machine Learning, and Bias in Emergency Care(2024-01-03) Jambaulikar, Guruprasad; Goldsmith, Andrew; Boyer, Edward