Machine learning for outcome predictions of patients with trauma during emergency department care

Development of clinical decision support for patients older than 65 years with fall-related TBI using artificial intelligence modeling

BACKGROUND

Older persons comprise most traumatic brain injury (TBI)-related hospitalizations and deaths and are particularly susceptible to fall-induced TBIs. The combination of increased frailty and susceptibility to clinical decline creates a significant ongoing challenge in the management of geriatric TBI. As the population ages and co-existing medical conditions complexify, so does the need to improve the quality of care for this population. Utilizing early hospital admission variables, this study will create and validate a multinomial decision tree that predicts the discharge disposition of older patients with fall-related TBI.

METHODS

From the National Trauma Data Bank, we retrospectively analyzed 11,977 older patients with a fall-related TBI (2017-2021). Clinical variables included Glasgow Coma Scale (GCS) score, intracranial pressure monitor use, venous thromboembolism (VTE) prophylaxis, and initial vital signs. Outcomes included hospital discharge disposition re-categorized into home, care facility, or deceased. Data were split into two sets, where 80% developed a decision tree, and 20% tested predictive performance. We employed a conditional inference tree algorithm with bootstrap (B = 100) and grid search options to grow the decision tree and measure discrimination ability using the area under the curve (AUC) and calibration plots.

RESULTS

Our decision tree used seven admission variables to predict the discharge disposition of older TBI patients. Significant non-modifiable variables included total GCS and injury severity scores, while VTE prophylaxis type was the most important interventional variable. Patients who did not receive VTE prophylaxis treatment had a higher probability of death. The predictive performance of the tree in terms of AUC value (95% confidence intervals) in the training cohort for death, care, and home were 0.66 (0.65-0.67), 0.75 (0.73-0.76), and 0.77 (0.76-0.79), respectively. In the test cohort, the values were 0.64 (0.62-0.67), 0.75 (0.72-0.77), and 0.77 (0.73-0.79).

CONCLUSIONS

We have developed and internally validated a multinomial decision tree to predict the discharge destination of older patients with TBI. This tree could serve as a decision support tool for caregivers to manage older patients better and inform decision-making. However, the tree must be externally validated using prospective data to ascertain its predictive and clinical importance.

Development and validation of risk prediction model for adverse outcomes in trauma patients

BACKGROUND

The prognosis of trauma patients is highly dependent on early medical diagnosis. By constructing a nomogram model, the risk of adverse outcomes can be displayed intuitively and individually, which has important clinical implications for medical diagnosis.

OBJECTIVE

To develop and evaluate models for predicting patients with adverse outcomes of trauma that can be used in different data availability settings in China.

METHODS

This was a retrospective prognostic study using data from 8 public tertiary hospitals in China from 2018. The data were randomly divided into a development set and a validation set. Simple, improved and extended models predicting adverse outcomes were developed, with adverse outcomes defined as in-hospital death or ICU transfer, and patient clinical characteristics, vital signs, diagnoses, and laboratory test values as predictors. The results of the models were presented in the form of nomograms, and performance was evaluated using area under the receiver operating characteristic curve (ROC-AUC), precision-recall (PR) curves (PR-AUC), Hosmer-Lemeshow goodness-of-fit test, calibration curve, and decision curve analysis (DCA).

RESULTS

Our final dataset consisted of 18,629 patients (40.2% female, mean age of 52.3), 1,089 (5.85%) of whom resulted in adverse outcomes. In the external validation set, three models achieved ROC-AUC of 0.872, 0.881, and 0.903, and a PR-AUC of 0.339, 0.337, and 0.403, respectively. In terms of the calibration curves and DCA, the models also performed well.

CONCLUSIONS

This prognostic study found that three prediction models and nomograms including the patient clinical characteristics, vital signs, diagnoses, and laboratory test values can support clinicians in more accurately identifying patients who are at risk of adverse outcomes in different settings based on data availability.

A pre-trained language model for emergency department intervention prediction using routine physiological data and clinical narratives

INTRODUCTION

The urgency and complexity of emergency room (ER) settings require precise and swift decision-making processes for patient care. Ensuring the timely execution of critical examinations and interventions is vital for reducing diagnostic errors, but the literature highlights a need for innovative approaches to optimize diagnostic accuracy and patient outcomes. In response, our study endeavors to create predictive models for timely examinations and interventions by leveraging the patient's symptoms and vital signs recorded during triage, and in so doing, augment traditional diagnostic methodologies.

METHODS

Focusing on four key areas-medication dispensing, vital interventions, laboratory testing, and emergency radiology exams, the study employed Natural Language Processing (NLP) and seven advanced machine learning techniques. The research was centered around the innovative use of BioClinicalBERT, a state-of-the-art NLP framework.

RESULTS

BioClinicalBERT emerged as the superior model, outperforming others in predictive accuracy. The integration of physiological data with patient narrative symptoms demonstrated greater effectiveness compared to models based solely on textual data. The robustness of our approach was confirmed by an Area Under the Receiver Operating Characteristic curve (AUROC) score of 0.9.

CONCLUSION

The findings of our study underscore the feasibility of establishing a decision support system for emergency patients, targeting timely interventions and examinations based on a nuanced analysis of symptoms. By using an advanced natural language processing technique, our approach shows promise for enhancing diagnostic accuracy. However, the current model is not yet fully mature for direct implementation into daily clinical practice. Recognizing the imperative nature of precision in the ER environment, future research endeavors must focus on refining and expanding predictive models to include detailed timely examinations and interventions. Although the progress achieved in this study represents an encouraging step towards a more innovative and technology-driven paradigm in emergency care, full clinical integration warrants further exploration and validation.

Prediction of mortality among severely injured trauma patients A comparison between TRISS and machine learning-based predictive models

BACKGROUND

Given the huge impact of trauma on hospital systems around the world, several attempts have been made to develop predictive models for the outcomes of trauma victims. The most used, and in many studies most accurate predictive model, is the "Trauma Score and Injury Severity Score" (TRISS). Although it has proven to be fairly accurate and is widely used, it has faced criticism for its inability to classify more complex cases. In this study, we aimed to develop machine learning models that better than TRISS could predict mortality among severely injured trauma patients, something that has not been studied using data from a nationwide register before.

METHODS

Patient data was collected from the national trauma register in Sweden, SweTrau. The studied period was from the 1st of January 2015 to 31st of December 2019. After feature selection and multiple imputation of missing data three machine learning (ML) methods (Random Forest, eXtreme Gradient Boosting, and a Generalized Linear Model) were used to create predictive models. The ML models and TRISS were then tested on predictive ability for 30-day mortality.

RESULTS

The ML models were well-calibrated and outperformed TRISS in all the tested measurements. Among the ML models, the eXtreme Gradient Boosting model performed best with an AUC of 0.91 (0.88-0.93).

CONCLUSION

This study showed that all the developed ML-based prediction models were superior to TRISS for the prediction of trauma mortality.

Validation of the China mortality prediction model in trauma based on the ICD-10-CM codes

The China mortality prediction model in trauma, based on the International Classification of Diseases, Tenth Revision, Clinical Modification lexicon (CMPMIT-ICD-10), is a novel model for predicting outcomes in patients who experienced trauma. This model has not yet been validated using data acquired from patients at other trauma centers in China. This retrospective study used data retrieved from the Peking University People's Hospital discharge database and included all patients admitted for trauma between 2012 and 2022 for model validation. Model performance was categorized into discrimination and calibration. In total, 23,299 patients were included in this study, with an overall mortality rate of 1.2%. CMPMIT-ICD-10 showed good discrimination and calibration, with an area under the curve of 0.84 (95% confidence interval: 0.82-0.87) and a Brier score of 0.02. The performance of the CMPMIT-ICD-10 during validation was satisfactory, and the application of the model will be scaled up in future studies.

Descriptive Analysis of Trauma Admission Trends before and during the COVID-19 Pandemic

INTRODUCTION

Traumatic injuries are a significant global health concern, with profound medical and socioeconomic impacts. This study explores the patterns of trauma-related hospitalizations in the Lublin Province of Poland, with a particular focus on the periods before and during the COVID-19 pandemic.

AIM OF THE STUDY

The primary aim of this research was to assess the trends in trauma admissions, the average length of hospital stays, and mortality rates associated with different types of injuries, comparing urban and rural settings over two distinct time periods: 2018-2019 and 2020-2021.

METHODS

This descriptive study analyzed trauma admission data from 35 hospitals in the Lublin Province, as recorded in the National General Hospital Morbidity Study (NGHMS). Patients were classified based on the International Classification of Diseases Revision 10 (ICD-10) codes. The data were compared for two periods: an 11-week span during the initial COVID-19 lockdown in 2020 and the equivalent period in 2019.

RESULTS

The study found a decrease in overall trauma admissions during the pandemic years (11,394 in 2020-2021 compared to 17,773 in 2018-2019). Notably, the average length of hospitalization increased during the pandemic, especially in rural areas (from 3.5 days in 2018-2019 to 5.5 days in 2020-2021 for head injuries). Male patients predominantly suffered from trauma, with a notable rise in female admissions for abdominal injuries during the pandemic. The maximal hospitalization days were higher in rural areas for head and neck injuries during the pandemic.

CONCLUSIONS

The study highlights significant disparities in trauma care between urban and rural areas and between the pre-pandemic and pandemic periods. It underscores the need for healthcare systems to adapt to changing circumstances, particularly in rural settings, and calls for targeted strategies to address the specific challenges faced in trauma care during public health crises.

[Application and prospect of machine learning in orthopaedic trauma]

Objective

To review the current applications of machine learning in orthopaedic trauma and anticipate its future role in clinical practice.

Methods

A comprehensive literature review was conducted to assess the status of machine learning algorithms in orthopaedic trauma research, both nationally and internationally.

Results

The rapid advancement of computer data processing and the growing convergence of medicine and industry have led to the widespread utilization of artificial intelligence in healthcare. Currently, machine learning plays a significant role in orthopaedic trauma, demonstrating high performance and accuracy in various areas including fracture image recognition, diagnosis stratification, clinical decision-making, evaluation, perioperative considerations, and prognostic risk prediction. Nevertheless, challenges persist in the development and clinical implementation of machine learning. These include limited database samples, model interpretation difficulties, and universality and individualisation variations.

Conclusion

The expansion of clinical sample sizes and enhancements in algorithm performance hold significant promise for the extensive application of machine learning in supporting orthopaedic trauma diagnosis, guiding decision-making, devising individualized medical strategies, and optimizing the allocation of clinical resources.

Emergency Department Pediatric Readiness Among US Trauma Centers: A Machine Learning Analysis of Components Associated With Survival

OBJECTIVE

We used machine learning to identify the highest impact components of emergency department (ED) pediatric readiness for predicting in-hospital survival among children cared for in US trauma centers.

BACKGROUND

ED pediatric readiness is associated with improved short-term and long-term survival among injured children and part of the national verification criteria for US trauma centers. However, the components of ED pediatric readiness most predictive of survival are unknown.

METHODS

This was a retrospective cohort study of injured children below 18 years treated in 458 trauma centers from January 1, 2012, through December 31, 2017, matched to the 2013 National ED Pediatric Readiness Assessment and the American Hospital Association survey. We used machine learning to analyze 265 potential predictors of survival, including 152 ED readiness variables, 29 patient variables, and 84 ED-level and hospital-level variables. The primary outcome was in-hospital survival.

RESULTS

There were 274,756 injured children, including 4585 (1.7%) who died. Nine ED pediatric readiness components were associated with the greatest increase in survival: policy for mental health care (+8.8% change in survival), policy for patient assessment (+7.5%), specific respiratory equipment (+7.2%), policy for reduced-dose radiation imaging (+7.0%), physician competency evaluations (+4.9%), recording weight in kilograms (+3.2%), life support courses for nursing (+1.0%-2.5%), and policy on pediatric triage (+2.5%). There was a 268% improvement in survival when the 5 highest impact components were present.

CONCLUSIONS

ED pediatric readiness components related to specific policies, personnel, and equipment were the strongest predictors of pediatric survival and worked synergistically when combined.

Artificial intelligence and machine learning for hemorrhagic trauma care

Artificial intelligence (AI), a branch of machine learning (ML) has been increasingly employed in the research of trauma in various aspects. Hemorrhage is the most common cause of trauma-related death. To better elucidate the current role of AI and contribute to future development of ML in trauma care, we conducted a review focused on the use of ML in the diagnosis or treatment strategy of traumatic hemorrhage. A literature search was carried out on PubMed and Google scholar. Titles and abstracts were screened and, if deemed appropriate, the full articles were reviewed. We included 89 studies in the review. These studies could be grouped into five areas: (1) prediction of outcomes; (2) risk assessment and injury severity for triage; (3) prediction of transfusions; (4) detection of hemorrhage; and (5) prediction of coagulopathy. Performance analysis of ML in comparison with current standards for trauma care showed that most studies demonstrated the benefits of ML models. However, most studies were retrospective, focused on prediction of mortality, and development of patient outcome scoring systems. Few studies performed model assessment via test datasets obtained from different sources. Prediction models for transfusions and coagulopathy have been developed, but none is in widespread use. AI-enabled ML-driven technology is becoming integral part of the whole course of trauma care. Comparison and application of ML algorithms using different datasets from initial training, testing and validation in prospective and randomized controlled trials are warranted for provision of decision support for individualized patient care as far forward as possible.