Automated stratification of trauma injury severity across multiple body regions using multi-modal, multi-class machine learning models

OBJECTIVE

The timely stratification of trauma injury severity can enhance the quality of trauma care but it requires intense manual annotation from certified trauma coders. The objective of this study is to develop machine learning models for the stratification of trauma injury severity across various body regions using clinical text and structured electronic health records (EHRs) data.

MATERIALS AND METHODS

Our study utilized clinical documents and structured EHR variables linked with the trauma registry data to create 2 machine learning models with different approaches to representing text. The first one fuses concept unique identifiers (CUIs) extracted from free text with structured EHR variables, while the second one integrates free text with structured EHR variables. Temporal validation was undertaken to ensure the models' temporal generalizability. Additionally, analyses to assess the variable importance were conducted.

RESULTS

Both models demonstrated impressive performance in categorizing leg injuries, achieving high accuracy with macro-F1 scores of over 0.8. Additionally, they showed considerable accuracy, with macro-F1 scores exceeding or near 0.7, in assessing injuries in the areas of the chest and head. We showed in our variable importance analysis that the most important features in the model have strong face validity in determining clinically relevant trauma injuries.

DISCUSSION

The CUI-based model achieves comparable performance, if not higher, compared to the free-text-based model, with reduced complexity. Furthermore, integrating structured EHR data improves performance, particularly when the text modalities are insufficiently indicative.

CONCLUSIONS

Our multi-modal, multiclass models can provide accurate stratification of trauma injury severity and clinically relevant interpretations.

Experiential Learning of Interdisciplinary Care Skills in Surgery Assessed From Student Reflections

INTRODUCTION

Interdisciplinary healthcare collaboration improves patient outcomes, increases workplace satisfaction, and reduces costs. Our medical school utilizes an experiential learning tool for teaching interprofessionalism known as the Longitudinal Patient Project (LPP). Medical students are directed to identify a surgical patient to establish continuity with by observing them throughout preoperative, intraoperative, and postoperative periods, and follow-up with the patient after their procedure. Students then write reflections on their LPP experience. This study examines the LPP as an interprofessionalism teaching tool using qualitative analysis of student reflections.

METHODS

NVivo 12 was used to code reflections. One researcher coded reflections for subject, depth, temporality, and confidence. Depth was assessed using Mezirow's Critical Reflection Theory, with students receiving titles of "content," "process," or "premise" reflectors based on the deepest level of reflection exhibited. Confidence was assessed by labeling reflective statements as "concrete" or "verbal." Data were coded by a second researcher for validation. Consensus was reached, the remainder of the dataset was updated to reflect codebook changes, and trends were assessed.

RESULTS

Inter-rater agreement was 83%. All students achieved "content" level reflection. Ninety-seven percent of students reached "process" reflection. Ninety-three percent of students reached "premise" reflection. Students provided more concrete indicators of knowledge gained from the LPP than from prior experiences. Subjects included communication, team dynamics, patient impact, and student experience. Increased depth and breadth of reflection on communication and team dynamics were observed from the LPP.

CONCLUSIONS

The LPP illustrates the importance of interdisciplinary care in surgery. Future iterations should emphasize the impact on patients and their families.

Which hospital-acquired conditions matter the most in trauma? An evidence-based approach for prioritizing trauma program improvement

BACKGROUND

Prevention of hospital-acquired conditions (HACs) is a focus of trauma center quality improvement. The relative contributions of various HACs to postinjury hospital outcomes are unclear. We sought to quantify and compare the impacts of six HACs on early clinical outcomes and resource utilization in hospitalized trauma patients.

METHODS

Adult patients from the 2013 to 2016 American College of Surgeons Trauma Quality Improvement Program Participant Use Data Files who required 5 days or longer of hospitalization and had an Injury Severity Score of 9 or greater were included. Multiple imputation with chained equations was used for observations with missing data. The frequencies of six HACs and five adverse outcomes were determined. Multivariable Poisson regression with log link and robust error variance was used to produce relative risk estimates, adjusting for patient-, hospital-, and injury-related factors. Risk-adjusted population attributable fractions estimates were derived for each HAC-outcome pair, with the adjusted population attributable fraction estimate for a given HAC-outcome pair representing the estimated percentage decrease in adverse outcome that would be expected if exposure to the HAC had been prevented.

RESULTS

A total of 529,856 patients requiring 5 days or longer of hospitalization were included. The incidences of HACs were as follows: pneumonia, 5.2%; urinary tract infection, 3.4%; venous thromboembolism, 3.3%; surgical site infection, 1.3%; pressure ulcer, 1.3%; and central line-associated blood stream infection, 0.2%. Pneumonia demonstrated the strongest association with in-hospital outcomes and resource utilization. Prevention of pneumonia in our cohort would have resulted in estimated reductions of the following: 22.1% for end organ dysfunction, 7.8% for mortality, 8.7% for prolonged hospitalization, 7.1% for prolonged intensive care unit stay, and 6.8% for need for mechanical ventilation. The impact of other HACs was comparatively small.

CONCLUSION

We describe a method for comparing the contributions of HACs to outcomes of hospitalized trauma patients. Our findings suggest that trauma program improvement efforts should prioritize pneumonia prevention.

LEVEL OF EVIDENCE

Therapeutic/Care Management; Level IV.

Comparison and interpretability of machine learning models to predict severity of chest injury

Objective

Trauma quality improvement programs and registries improve care and outcomes for injured patients. Designated trauma centers calculate injury scores using dedicated trauma registrars; however, many injuries arrive at nontrauma centers, leaving a substantial amount of data uncaptured. We propose automated methods to identify severe chest injury using machine learning (ML) and natural language processing (NLP) methods from the electronic health record (EHR) for quality reporting.

Materials and Methods

A level I trauma center was queried for patients presenting after injury between 2014 and 2018. Prediction modeling was performed to classify severe chest injury using a reference dataset labeled by certified registrars. Clinical documents from trauma encounters were processed into concept unique identifiers for inputs to ML models: logistic regression with elastic net (EN) regularization, extreme gradient boosted (XGB) machines, and convolutional neural networks (CNN). The optimal model was identified by examining predictive and face validity metrics using global explanations.

Results

Of 8952 encounters, 542 (6.1%) had a severe chest injury. CNN and EN had the highest discrimination, with an area under the receiver operating characteristic curve of 0.93 and calibration slopes between 0.88 and 0.97. CNN had better performance across risk thresholds with fewer discordant cases. Examination of global explanations demonstrated the CNN model had better face validity, with top features including “contusion of lung” and “hemopneumothorax.”

Discussion

The CNN model featured optimal discrimination, calibration, and clinically relevant features selected.

Conclusion

NLP and ML methods to populate trauma registries for quality analyses are feasible.

Comparison of the Perspectives of Medical Students and Residents on the Surgery Learning Environment

BACKGROUND

The learning environment plays a critical role in learners' satisfaction and outcomes. However, we often lack insight into learners' perceptions and assessments of these environments. It can be difficult to discern learners' expectations, making their input critical. When medical students and surgery residents are asked to evaluate their teachers, what do they focus on?

MATERIALS AND METHODS

Open-ended comments from medical students' evaluations of residents and attending surgeons and from residents' evaluations of attendings during the 2016-2017 academic year were analyzed. Content analysis was used, and codes derived from the data. A matrix of theme by learner role was created to distinguish differences between medical student and resident learners. Subthemes were grouped based on similarity into high-order themes.

RESULTS

Two overarching themes were Creating a positive environment for learning by modeling professional behaviors and Intentionally engaging learners in training and educational opportunities. Medical students and residents made similar comments for the subthemes of appropriate demeanor, tone and dialog, respect, effective direct instruction, feedback, debriefing, giving appropriate levels of autonomy, and their expectations as team members on a service. Differences existed in the subthemes of punctuality, using evidence, clinical knowledge, efficiency, direct interactions with patients, learning outcomes, and career decisions.

CONCLUSIONS

Faculty development efforts should target professional communication, execution of teaching skills, and relationships among surgeons, other providers, and patients. Attendings should make efforts to discuss their approach to clinical decision making and patient interactions and help residents and medical students voice their opinions and questions through trusting adult learner-teacher relationships.

Number of rib fractures thresholds independently predict worse outcomes in older patients with blunt trauma

BACKGROUND

There have been conflicting reports regarding whether the number of rib fractures sustained in blunt trauma is associated independently with worse patient outcomes. We sought to investigate this risk-adjusted relationship among the lesser-studied population of older adults.

METHODS

A retrospective review of the National Trauma Data Bank was performed for patients with blunt trauma who were ≥65 years old and had rib fractures between 2009 and 2012 (N = 67,695). Control data were collected for age, sex, injury severity score, injury mechanism, 24 comorbidities, and number of rib fractures. Outcome data included hospital mortality, hospital and intensive care unit durations of stay, duration of mechanical ventilation, and the occurrence of pneumonia. Multiple logistic and linear regression analyses were performed.

RESULTS

Sustaining ≥5 rib fractures was associated with increased intensive care unit admission (odds ratio: 1.14, P < .001) and hospital duration of stay (relative duration: 105%, P < .001). Sustaining ≥7 rib fractures was associated with an increased incidence of pneumonia (odds ratio: 1.32, P < .001) and intensive care unit duration of stay (relative duration: 122%, P < .001). Sustaining ≥8 rib fractures was associated with increased mortality (odds ratio: 1.51, P < .001) and duration of mechanical ventilation (relative duration: 117%, P < .001).

CONCLUSION

In older patients with trauma, sustaining at least 5 rib fractures is a significant predictor of worse outcomes independent of patient characteristics, comorbidities, and trauma burden.

The Sumatra-Andaman Earthquake and Tsunami of 2004: The Hazards, Events, and Damage

The Sumatra-Andaman Earthquake and subsequent Asian Tsunami of 26 December 2004 affected multiple countries in the Indian Ocean and beyond, creating disasters of a scale unprecedented in recorded history. Using the Conceptual Framework and terminology described in the Disaster Health Management: Guidelines for Evaluation and Research in the Utstein Style, the hazard, events, and damage associated with the Earthquake and Tsunami are described. Many gaps in the available information regarding this event are present. Standardized indicators and reporting criteria are necessary for research on future disasters and the development of best practice standards internationally.

Finite-element analysis of hepatic cryoablation around a large blood vessel

Cryoablation is a minimally invasive ablation technique for primary and metastatic hepatic tumors. Inadequate freezing around large blood vessels due to the warm blood flow can lead to local recurrence, and thus, necessitates close application of a cryoprobe to the large blood vessels. In this study, we constructed a perfusion model with an ex vivo bovine liver and ablated the tissue around a large blood vessel with one or two cryoprobes applied to the side of the vessel. The finite-element computer model developed in our previous study was modified to include a blood vessel and its convective heat transfer to the vicinity of the blood vessel. We compared the predicted simulation results to those acquired from this ex vivo perfusion model. The results indicate that blood vessels act as a heat source and generate steep temperature profiles in the area next to the large blood vessel. After validation, the maximum allowable distance between the cryoprobe and the large blood vessel for successful cryoablation was presented. The results of this study should be considered when placing cryoprobes in the vicinity of large blood vessels.

An electrode array that minimizes blood loss for radiofrequency-assisted hepatic resection

Hepatic resection is currently the standard treatment for liver cancer. During hepatic resection part of the liver containing the tumor is surgically removed. This type of surgery is accompanied by high blood loss of approximately 0.6-1.35 L. Blood loss is associated with increased complication rates, prolonged hospital stay, and reduced patient survival, especially when transfusion is required. Other researchers have suggested using radiofrequency (rf) or microwave ablation to coagulate a tissue slice before resection to reduce blood loss, but conventional devices typically take several hours. We developed a device consisting of a linear array of blade-shaped, 1 cm wide radiofrequency (rf) electrodes 1.5 cm apart. Bipolar rf power is applied between pairs of adjacent electrodes, leading to high tissue temperatures between the electrodes that promote coagulation of large vessels (>3 mm) in the resection plane. Rapid switching of applied power between pairs of adjacent electrodes allows simultaneous heating and coagulation of the entire resection plane within 3-6 min. In seven in vivo trials in a porcine model, resection along a plane pre-coagulated with the device resulted in little (<20 mL) to no blood loss, while coagulating all vessels (up to 4.5 mm diameter in this study). Average treatment time (from placement of the device to transection) was 6.8+/-0.5 min when four electrodes were used, and 11.3+/-1.2 min when 5-7 electrodes were used. This device may reduce blood loss related morbidity during resection and reduce treatment time by coagulating all vessels in the resection plane.

Current status of liver tumor ablation devices

The liver is a common site of disease for both primary and metastatic cancer. Since most patients have a disease that is not amenable to surgical resection, tumor ablation modalities are increasingly being used for treatment of liver cancer. This review describes the current status of ablative technologies used as alternatives for resection, clinical experience with these technologies, currently available devices and design rules for the development of new devices and the improvement of existing ones. It focuses on probe design for radiofrequency ablation, microwave ablation and cryoablation, and compares the advantages and disadvantages of each ablation modality.

Finite-element analysis of ex vivo and in vivo hepatic cryoablation

Cryoablation is a widely used method for the treatment of nonresectable primary and metastatic liver tumors. A model that can accurately predict the size of a cryolesion may allow more effective treatment of tumor, while sparing normal liver tissue. We generated a computer model of tissue cryoablation using the finite-element method (FEM). In our model, we considered the heat transfer mechanism inside the cryoprobe and also cryoprobe surfaces so our model could incorporate the effect of heat transfer along the cryoprobe from the environment at room temperature. The modeling of the phase shift from liquid to solid was a key factor in the accurate development of this model. The model was verified initially in an ex vivo liver model. Temperature history at three locations around one cryoprobe and between two cryoprobes was measured. The comparison between the ex vivo result and the FEM modeling result at each location showed a good match, where the maximum difference was within the error range acquired in the experiment (< 5 degrees C). The FEM model prediction of the lesion size was within 0.7 mm of experimental results. We then validated our FEM in an in vivo experimental porcine model. We considered blood perfusion in conjunction with blood viscosity depending on temperature. The in vivo iceball size was smaller than the ex vivo iceball size due to blood perfusion as predicted in our model. The FEM results predicted this size within 0.1-mm error. The FEM model we report can accurately predict the extent of cryoablation in the liver.

Dielectric properties of human normal, malignant and cirrhotic liver tissue: in vivo and ex vivo measurements from 0.5 to 20 GHz using a precision open-ended coaxial probe

Hepatic malignancies have historically been treated with surgical resection. Due to the shortcomings of this technique, there is interest in other, less invasive, treatment modalities, such as microwave hepatic ablation. Crucial to the development of this technique is the accurate knowledge of the dielectric properties of human liver tissue at microwave frequencies. To this end, we characterized the dielectric properties of in vivo and ex vivo normal, malignant and cirrhotic human liver tissues from 0.5 to 20 GHz. Analysis of our data at 915 MHz and 2.45 GHz indicates that the dielectric properties of ex vivo malignant liver tissue are 19 to 30% higher than normal tissue. The differences in the dielectric properties of in vivo malignant and normal liver tissue are not statistically significant (with the exception of effective conductivity at 915 MHz, where malignant tissue properties are 16% higher than normal). Also, the dielectric properties of in vivo normal liver tissue at 915 MHz and 2.45 GHz are 16 to 43% higher than ex vivo. No statistically significant differences were found between the dielectric properties of in vivo and ex vivo malignant tissue (with the exception of effective conductivity at 915 MHz, where malignant tissue properties are 28% higher than normal). We report the one-pole Cole-Cole parameters for ex vivo normal, malignant and cirrhotic liver tissue in this frequency range. We observe that wideband dielectric properties of in vivo liver tissue are different from the wideband dielectric properties of ex vivo liver tissue, and that the in vivo data cannot be represented in terms of a Cole-Cole model. Further work is needed to uncover the mechanisms responsible for the observed wideband trends in the in vivo liver data.

Taser Dart-to-Heart Distance That Causes Ventricular Fibrillation in Pigs

Electromuscular incapacitating devices (EMDs), such as Tasers, deliver high current, short duration pulses that cause muscular contractions and temporarily incapacitate the human subject. Some reports suggest that EMDs can kill. To help answer the question, "Can the EMD directly cause ventricular fibrillation (VF)?", ten tests were conducted to measure the dart-to-heart distance that causes VF in anesthetized pigs [mass = 64 kg +/- 6.67 standard deviation (SD)] for the most common X26 Taser. The dart-to-heart distance that caused VF was 17 mm +/- 6.48 (SD) for the first VF event and 13.7 mm +/- 6.79 (SD) for the average of the successive VF events. The result shows that when the stimulation dart is close enough to the heart, X26 Taser current will directly trigger VF in pigs. Echocardiography of erect humans shows skin-to-heart distances from 10 to 57 mm (dart-to-heart distances of 1-48 mm). These results suggest that the probability of a dart on the body landing in 1 cm2 over the ventricle and causing VF is 0.000172.

A floating sleeve antenna yields localized hepatic microwave ablation

We report a novel coaxial antenna for hepatic microwave ablation. This device uses a floating sleeve, that is, a metal conductor electrically isolated from the outer connector of the antenna coaxial body, to achieve a highly localized specific absorption rate pattern that is independent of insertion depth. This floating sleeve coaxial dipole antenna has low power reflection in the 2.4-GHz IMS band. Ex vivo experiments confirm our numerical simulation results. Index Terms-Ablation, coaxial aperture antennas, finite element methods, floating sleeve, microwave heating.