Assessing optimal methods for transferring machine learning models to low-volume and imbalanced clinical datasets: experiences from predicting outcomes of Danish trauma patients.

Introduction: Accurately predicting patient outcomes is crucial for improving healthcare delivery, but large-scale risk prediction models are often developed and tested on specific datasets where clinical parameters and outcomes may not fully reflect local clinical settings. Where this is the case, whether to opt for de-novo training of prediction models on local datasets, direct porting of externally trained models, or a transfer learning approach is not well studied, and constitutes the focus of this study. Using the clinical challenge of predicting mortality and hospital length of stay on a Danish trauma dataset, we hypothesized that a transfer learning approach of models trained on large external datasets would provide optimal prediction results compared to de-novo training on sparse but local datasets or directly porting externally trained models. Methods: Using an external dataset of trauma patients from the US Trauma Quality Improvement Program (TQIP) and a local dataset aggregated from the Danish Trauma Database (DTD) enriched with Electronic Health Record data, we tested a range of model-level approaches focused on predicting trauma mortality and hospital length of stay on DTD data. Modeling approaches included de-novo training of models on DTD data, direct porting of models trained on TQIP data to the DTD, and a transfer learning approach by training a model on TQIP data with subsequent transfer and retraining on DTD data. Furthermore, data-level approaches, including mixed dataset training and methods countering imbalanced outcomes (e.g., low mortality rates), were also tested. Results: Using a neural network trained on a mixed dataset consisting of a subset of TQIP and DTD, with class weighting and transfer learning (retraining on DTD), we achieved excellent results in predicting mortality, with a ROC-AUC of 0.988 and an F2-score of 0.866. The best-performing models for predicting long-term hospitalization were trained only on local data, achieving an ROC-AUC of 0.890 and an F1-score of 0.897, although only marginally better than alternative approaches. Conclusion: Our results suggest that when assessing the optimal modeling approach, it is important to have domain knowledge of how incidence rates and workflows compare between hospital systems and datasets where models are trained. Including data from other health-care systems is particularly beneficial when outcomes are suffering from class imbalance and low incidence. Scenarios where outcomes are not directly comparable are best addressed through either de-novo local training or a transfer learning approach.

Time to CT scan for patients with acute severe neurological symptoms: a quality assurance study.

Emergent brain computed tomography (CT) scan allows for identification of patients presenting with acute severe neurological symptoms in whom medical and surgical interventions may be lifesaving. The aim of this study was to evaluate if time to CT from arrival at the emergency department exceeded 30 min in patients admitted with acute severe neurological symptoms. This was a retrospective register-based quality assurance study. We identified patients admitted to the emergency department with acute severe neurological symptoms between April 1st, 2016 and September 30th, 2020. Data were retrieved from the registry of acute medical team activations. We considered that time to CT from arrival at the emergency department should not exceed 30 min in more than 10% of patients. A total of 559 patients were included. Median time from arrival at the emergency department until CT scan was 24 min (IQR 16-35) in children (< 18 years), 10 min (IQR 7-17) for adults (18-59 years), and 11 min (IQR 7-16) for elders (> 60 years). This time interval exceeded 30 min for 8.2% (95% CI 6.1-10.9) of all included patients, 35.3% of children, 5.9% of adults, and 8.6% of elders. No children died within 30 days. The 30-day mortality was 21.3% (95% CI 16.4-27) in adults, and 43.9% (95% CI 38.2-49.8) in elders. Time from arrival at our emergency department until brain CT scan exceeded 30 min in 8.2% of all included patients but exceeded the defined quality aim in children and could be improved.

A characterization of trauma laparotomies in a scandinavian setting: an observational study.

BACKGROUND: Despite treatment advances, trauma laparotomy continuous to be associated with significant morbidity and mortality. Most of the literature originates from high volume centers, whereas patient characteristics and outcomes in a Scandinavian setting is not well described. The objective of this study is to characterize treatments and outcomes of patients undergoing trauma laparotomy in a Scandinavian setting and compare this to international reports. METHODS: A retrospective study was performed in the Copenhagen University Hospital, Rigshospitalet (CUHR). All patients undergoing a trauma laparotomy within the first 24 h of admission between January 1st 2019 and December 31st 2020 were included. Collected data included demographics, trauma mechanism, injuries, procedures performed and outcomes. RESULTS: A total of 1713 trauma patients were admitted to CUHR of which 98 patients underwent trauma laparotomy. Penetrating trauma accounted for 16.6% of the trauma population and 66.3% of trauma laparotomies. Median time to surgery after arrival at the trauma center (TC) was 12 min for surgeries performed in the Emergency Department (ED) and 103 min for surgeries performed in the operating room (OR). A total of 14.3% of the procedures were performed in the ED. A damage control strategy (DCS) approach was chosen in 18.4% of cases. Our rate of negative laparotomies was 17.3%. We found a mortality rate of 8.2%. The total median length of stay was 6.1 days. CONCLUSION: The overall rates, findings, and outcomes of trauma laparotomies in this Danish cohort is comparable to reports from similar Western European trauma systems.

The role of point of care ultrasound in prehospital critical care: a systematic review.

BACKGROUND: In 2011, the role of Point of Care Ultrasound (POCUS) was defined as one of the top five research priorities in physician-provided prehospital critical care and future research topics were proposed; the feasibility of prehospital POCUS, changes in patient management induced by POCUS and education of providers. This systematic review aimed to assess these three topics by including studies examining all kinds of prehospital patients undergoing all kinds of prehospital POCUS examinations and studies examining any kind of POCUS education in prehospital critical care providers. METHODS AND RESULTS: By a systematic literature search in MEDLINE, EMBASE, and Cochrane databases, we identified and screened titles and abstracts of 3264 studies published from 2012 to 2017. Of these, 65 studies were read in full-text for assessment of eligibility and 27 studies were ultimately included and assessed for quality by SIGN-50 checklists. No studies compared patient outcome with and without prehospital POCUS. Four studies of acceptable quality demonstrated feasibility and changes in patient management in trauma. Two studies of acceptable quality demonstrated feasibility and changes in patient management in breathing difficulties. Four studies of acceptable quality demonstrated feasibility, outcome prediction and changes in patient management in cardiac arrest, but also that POCUS may prolong pauses in compressions. Two studies of acceptable quality demonstrated that short (few hours) teaching sessions are sufficient for obtaining simple interpretation skills, but not image acquisition skills. Three studies of acceptable quality demonstrated that longer one- or two-day courses including hands-on training are sufficient for learning simple, but not advanced, image acquisition skills. Three studies of acceptable quality demonstrated that systematic educational programs including supervised examinations are sufficient for learning advanced image acquisition skills in healthy volunteers, but that more than 50 clinical examinations are required for expertise in a clinical setting. CONCLUSION: Prehospital POCUS is feasible and changes patient management in trauma, breathing difficulties and cardiac arrest, but it is unknown if this improves outcome. Expertise in POCUS requires extensive training by a combination of theory, hands-on training and a substantial amount of clinical examinations - a large part of these needs to be supervised.

[Advanced life support for cardiac arrest beyond the algorithm].

In an advanced emergency medical service all parts of the advanced life support (ALS) algorithm can be provided. This evidence-based algorithm outlines resuscitative efforts for the first 10-15 minutes after cardiac arrest, whereafter the algorithm repeats itself. Restoration of spontaneous circulation fails in most cases, but in some circumstances the patient may benefit from additional interventional approaches, in which case transport to hospital with ongoing cardiopulmonary resuscitation is indicated. This paper has summarized treatments outside the ALS algorithm, which may be beneficial, but are not supported by firm scientific evidence.

Effect of ultrasound training of physicians working in the prehospital setting.

BACKGROUND: Advances in technology have made ultrasound (US) devices smaller and portable, hence accessible for prehospital care providers. This study aims to evaluate the effect of a four-hour, hands-on US training course for physicians working in the prehospital setting. The primary outcome measure was US performance assessed by the total score in a modified version of the Objective Structured Assessment of Ultrasound Skills scale (mOSAUS). METHODS: Prehospital physicians participated in a four-hour US course consisting of both hands-on training and e-learning including a pre- and a post-learning test. Prior to the hands-on training a pre-training test was applied comprising of five videos in which the participants should identify pathology and a five-minute US examination of a healthy volunteer portraying to be a shocked patient after a blunt torso trauma. Following the pre-training test, the participants received a four-hour, hands-on US training course which was concluded with a post-training test. The US examinations and screen output from the US equipment were recorded for subsequent assessment. Two blinded raters assessed the videos using the mOSAUS. RESULTS: Forty participants completed the study. A significant improvement was identified in e-learning performance and US performance, (37.5 (SD: 10.0)) vs. (51.3 (SD: 5.9) p = < 0.0001), total US performance score (15.3 (IQR: 12.0-17.5) vs. 17.5 (IQR: 14.5-21.0), p = < 0.0001) and in each of the five assessment elements of the mOSAUS. CONCLUSION: In the prehospital physicians assessed, we found significant improvements in the ability to perform US examinations after completing a four-hour, hands-on US training course.

Effect of prehospital ultrasound on clinical outcomes of non-trauma patients--a systematic review.

BACKGROUND: Advances in technology have made prehospital ultrasound (US) examination available. Whether US in the prehospital setting can lead to improvement in clinical outcomes is yet unclear. OBJECTIVE: The aim of this systematic review was to assess whether prehospital US improves clinical outcomes for non-trauma patients. METHOD: We conducted a systematic review on non-trauma patients who had an US examination performed in the prehospital setting. We searched MEDLINE, EMBASE, the Cochrane Central Register of Controlled Trials and the ISI Web of Science and the references of the included studies for additional relevant studies. We then performed a risk of bias analysis and descriptive data analysis. RESULTS: We identified 1707 unique citations and included ten studies with a total of 1068 patients undergoing prehospital US examination. Included publications ranged from case series to non-randomized, descriptive studies, and all showed a high risk of bias. The large heterogeneity between the different studies made further statistical analysis impossible. CONCLUSION: There are currently no randomized, controlled studies on the use of US for non-trauma patients in the prehospital setting. The included studies were of large heterogeneity and all showed a high risk of bias. We were thus unable to assess the effect of prehospital US on clinical outcomes. However, consistent reports suggested that US may improve patient management with respect to diagnosis, treatment, and hospital referral.

Nurse-administered early warning score system can be used for emergency department triage.

INTRODUCTION: Studies have shown that early warning score systems can identify in-patients at high risk of catastrophic deterioration and this may possibly be used for an emergency department (ED) triage. Bispebjerg Hospital has introduced a multidisciplinary team (MT) in the ED activated by the Bispebjerg Early Warning Score (BEWS). The BEWS is calculated on the basis of respiratory frequency, pulse, systolic blood pressure, temperature and level of consciousness. The aim of this study is to evaluate the ability of the BEWS to identify critically ill patients in the ED and to examine the feasibility of using the BEWS to activate an MT response. MATERIAL AND METHODS: This study is based on an evaluation of retrospective data from a random sample of 300 emergency patients. On the basis of documented vital signs, a BEWS was calculated retrospectively. The primary end points were admission to an intensive care unit (ICU) and death within 48 hours of arrival at the ED. This study was registered at clinicaltrials.gov (NCT01243021). RESULTS: A BEWS ≥ 5 is associated with a significantly increased risk of ICU admission within 48 hours of arrival (relative risk (RR) 4.1; 95% confidence interval (CI) 1.5-10.9) and death within 48 hours of arrival (RR 20.3; 95% CI 6.9-60.1). The sensitivity of the BEWS in identifying patients who were admitted to the ICU or who died within 48 hours of arrival was 63%. The positive predictive value of the BEWS was 16% and the negative predictive value 98% for identification of patients who were admitted to the ICU or who died within 48 hours of arrival. CONCLUSION: The BEWS is a simple scoring system based on readily available vital signs. It is a sensitive tool for detecting critically ill patients and may be used for ED triage and activation of an MT response.

Quality of care using a multidisciplinary team in the emergency room.

INTRODUCTION: Bispebjerg Hospital has implemented a multidisciplinary team reception of critically ill and severely injured patients at the Emergency Department (ED), termed emergency call (EC) and trauma call (TC). The aim of this study was to describe the course, medical treatment and outcome for patients received by this multidisciplinary team and to evaluate the quality of acute medical treatment of these patients. MATERIAL AND METHODS: A retrospective evaluation was made of all ECs and TCs registered during a six-month period. Information on sex, age, interventions at the ED, time spent at the ED and outcome measures (admission, Intensive Care Unit (ICU) admission and death) were obtained. The quality of the acute medical treatment during the ED stay and the first 48 hours of admission were evaluated by senior consultants from the departments receiving the patients. RESULTS: A total of 150 ECs and 47 TCs were included. The median time spent at the ED was 65 minutes for ECs and 95 minutes for TCs. In EC patients a median of eight interventions were performed at the ED, while a median of five interventions were performed in TC patients. A total of 137 EC patients were admitted to hospital including 32 patients admitted to the ICU. In all, 49 EC patients died during admission. Forty percent of TC patients were discharged to their homes. Only one trauma patient died and none were admitted to the ICU. The acute medical treatment was found to be satisfactory in 87% of EC patients and 96% of TC patients. CONCLUSION: A multidisciplinary team reception ensures early initiation of diagnostic procedures and treatment, short ED stays and admission to relevant departments in critically ill and severely injured patients.

Low compliance with a validated system for emergency department triage.

INTRODUCTION: Bispebjerg Hospital has introduced a triage system at the Emergency Department (ED) based on "primary criteria" and a physiological scoring system named the Bispebjerg Early Warning Score (BEWS). A BEWS is calculated on the basis of five vital signs which are accessible bedside. Patients who have a "primary criterion" or a BEWS ≥ 5 are presumed to be critically ill or severely injured and should be received by a multidisciplinary team, termed the Emergency Call (EC) and Trauma Call (TC), respectively. The aim of this study was to examine compliance with this triage system at Bispebjerg Hospital. MATERIAL AND METHODS: Retrospective evaluation of the triage of a random sample of 300 ED patients. ED medical charts were searched for "primary criteria", documentation of vital signs and a BEWS score. If a BEWS score had not been calculated, this was done retrospectively by the author. An evaluation was made to determine whether ECs or TCs had been correctly activated. RESULTS: In 47 patients, all five vital signs for calculation of a BEWS had been documented. A BEWS had been calculated in 22 patients. Nine patients had a TC activation criterion, and in all these cases a TC was activated. A total of 48 patients had an EC activation criterion, but an EC had only been activated in 24 patients. Among the 24 patients for whom an EC had not been activated, eight had a "primary criterion" and 16 patients had a retrospective BEWS ≥ 5. CONCLUSION: The triage system is not being used systematically and documentation of vital signs is insufficient at Bispebjerg Hospital. As a consequence, many patients who are presumed to be critically ill are not allocated to an EC. Initiatives have been taken to raise compliance with the system.