Use of artificial intelligence to support prehospital traumatic injury care: A scoping review

Background

Artificial intelligence (AI) has transformative potential to support prehospital clinicians, emergency physicians, and trauma surgeons in acute traumatic injury care. This scoping review examines the literature evaluating AI models using prehospital features to support early traumatic injury care.

Methods

We conducted a systematic search in August 2023 of PubMed, Embase, and Web of Science. Two independent reviewers screened titles/abstracts, with a third reviewer for adjudication, followed by a full-text analysis. We included original research and conference presentations evaluating AI models-machine learning (ML), deep learning (DL), and natural language processing (NLP)-that used prehospital features or features available immediately upon emergency department arrival. Review articles were excluded. The same investigators extracted data and systematically categorized outcomes to ensure consistency and transparency. We calculated kappa for interrater reliability and descriptive statistics.

Results

We identified 1050 unique publications, with 49 meeting inclusion criteria after title and abstract review (kappa 0.58) and full-text review. Publications increased annually from 2 in 2007 to 10 in 2022. Geographic analysis revealed a 61% focus on data from the United States. Studies were predominantly retrospective (88%), used local (45%) or national level (41%) data, focused on adults only (59%) or did not specify adults or pediatrics (27%), and 57% encompassed both blunt and penetrating injury mechanisms. The majority used machine learning (88%) alone or in conjunction with DL or NLP, and the top three algorithms used were support vector machine, logistic regression, and random forest. The most common study objectives were to predict the need for critical care and life-saving interventions (29%), assist in triage (22%), and predict survival (20%).

Conclusions

A small but growing body of literature described AI models based on prehospital features that may support decisions made by dispatchers, Emergency Medical Services clinicians, and trauma teams in early traumatic injury care.

A New Time-Window Prediction Model For Traumatic Hemorrhagic Shock Based on Interpretable Machine Learning

ABSTRACT

Early warning prediction of traumatic hemorrhagic shock (THS) can greatly reduce patient mortality and morbidity. We aimed to develop and validate models with different stepped feature sets to predict THS in advance. From the PLA General Hospital Emergency Rescue Database and Medical Information Mart for Intensive Care III, we identified 604 and 1,614 patients, respectively. Two popular machine learning algorithms (i.e., extreme gradient boosting [XGBoost] and logistic regression) were applied. The area under the receiver operating characteristic curve (AUROC) was used to evaluate the performance of the models. By analyzing the feature importance based on XGBoost, we found that features in vital signs (VS), routine blood (RB), and blood gas analysis (BG) were the most relevant to THS (0.292, 0.249, and 0.225, respectively). Thus, the stepped relationships existing in them were revealed. Furthermore, the three stepped feature sets (i.e., VS, VS + RB, and VS + RB + sBG) were passed to the two machine learning algorithms to predict THS in the subsequent T hours (where T = 3, 2, 1, or 0.5), respectively. Results showed that the XGBoost model performance was significantly better than the logistic regression. The model using vital signs alone achieved good performance at the half-hour time window (AUROC = 0.935), and the performance was increased when laboratory results were added, especially when the time window was 1 h (AUROC = 0.950 and 0.968, respectively). These good-performing interpretable models demonstrated acceptable generalization ability in external validation, which could flexibly and rollingly predict THS T hours (where T = 0.5, 1) prior to clinical recognition. A prospective study is necessary to determine the clinical utility of the proposed THS prediction models.

Out-of-hospital Circulatory Measures to Identify Patients With Serious Injury: A Systematic Review

OBJECTIVE

The objective was to systematically identify and summarize out-of-hospital measures of circulatory compromise as diagnostic predictors of serious injury, focusing on measures usable by emergency medical services to inform field triage decisions.

METHODS

We searched Ovid MEDLINE, CINAHL, and the Cochrane databases from 1996 through August 2017 for published literature on individual circulatory measures in trauma. We reviewed reference lists of included articles for additional relevant citations. Measures of diagnostic accuracy included sensitivity, specificity, and area under the receiver operating characteristic curve (AUROC). Indicators of serious injury included resource need, serious anatomic injury, and mortality. We pooled estimates when data permitted.

RESULTS

We identified 114 articles, reporting results of 111 studies. Measures included systolic blood pressure (sBP), heart rate (HR), shock index (SI), lactate, base deficit, and HR variability. Pooled out-of-hospital sensitivity estimates were sBP < 90 mm Hg = 19% (95% confidence interval [CI] = 12% to 29%), HR ≥ 110 beats/min = 28% (95% CI = 20% to 37%), SI > 0.9 = 37% (95% CI = 22% to 56%), and lactate > 2.0 mmol/L = 74% (95% CI = 48% to 90%). Pooled specificity estimates were sBP < 90 mm Hg = 95% (95% CI = 91% to 97%), HR ≥ 110 beats/min = 85% (95% CI = 74% to 91%), SI > 0.9 = 85% (95% CI = 72% to 92%), and lactate > 2.0 mmol/L = 62% (95% CI = 51% to 72%). Pooled AUROCs included sBP = 0.67 (95% CI = 0.58 to 0.75), HR = 0.67 (95% CI = 0.56 to 0.79), SI = 0.72 (95% CI = 0.66 to 0.77), and lactate = 0.77 (95% CI = 0.67 to 0.82). Strength of evidence was low to moderate.

CONCLUSIONS

Out-of-hospital circulatory measures are associated with poor to fair discrimination for identifying trauma patients with serious injuries. Many seriously injured patients have normal circulatory measures (low sensitivity), but when present, the measures are highly specific for identifying patients with serious injuries.

Battlefield Vital Sign Monitoring in Role 1 Military Treatment Facilities: A Thematic Analysis of After-Action Reviews from the Prehospital Trauma Registry

INTRODUCTION

The Prehospital Trauma Registry (PHTR) captures after-action reviews (AARs) as part of a continuous performance improvement cycle and to provide commanders real-time feedback of Role 1 care. We have previously described overall challenges noted within the AARs. We now performed a focused assessment of challenges with regard to hemodynamic monitoring to improve casualty monitoring systems.

MATERIALS AND METHODS

We performed a review of AARs within the PHTR in Afghanistan from January 2013 to September 2014 as previously described. In this analysis, we focus on AARs specific to challenges with hemodynamic monitoring of combat casualties.

RESULTS

Of the 705 PHTR casualties, 592 had available AAR data; 86 of those described challenges with hemodynamic monitoring. Most were identified as male (97%) and having sustained battle injuries (93%), typically from an explosion (48%). Most were urgent evacuation status (85%) and had a medical officer in their chain of care (65%). The most common vital sign mentioned in AAR comments was blood pressure (62%), and nearly one-quarter of comments stated that arterial palpation was used in place of blood pressure cuff measurements.

CONCLUSIONS

Our qualitative methods study highlights the challenges with obtaining vital signs-both training and equipment. We also highlight the challenges regarding ongoing monitoring to prevent hemodynamic collapse in severely injured casualties. The U.S. military needs to develop better methods for casualty monitoring for the subset of casualties that are critically injured.

Determination of Normal Ranges of Shock Index and Other Haemodynamic Variables in the Immediate Postpartum Period: A Cohort Study

OBJECTIVE

To determine the normal ranges of vital signs, including blood pressure (BP), mean arterial pressure (MAP), heart rate (HR) and shock index (SI) (HR/systolic BP), in the immediate postpartum period to inform the development of robust obstetric early warning scores.

STUDY DESIGN

We conducted a secondary analysis of a prospective observational cohort study evaluating vital signs collected within one hour following delivery in women with estimated blood loss (EBL) <500ml (316 women) delivering at a UK tertiary centre over a one-year period. Simple and multiple linear regression were used to explore associations of demographic and obstetric factors with SI.

RESULTS

Median (90% reference range) was 120 (100-145) for systolic BP, 75 (58-90) for diastolic BP, 90 (73-108) for MAP, 81 (61-102) for HR, and 0.66 (0.52-0.89) for SI. Third stage Syntometrine® administration was associated with a 0.03 decrease in SI (p = 0.035) and epidural use with a 0.05 increase (p = 0.003). No other demographic or obstetric factors were associated with a change in shock index in this cohort.

CONCLUSION

This is the first study to determine normal ranges of maternal BP, MAP, HR and SI within one hour of birth, a time of considerable haemodynamic adjustment, with minimal effect of demographic and obstetric factors demonstrated. The lower 90% reference point for systolic BP and upper 90% reference point for HR correspond to triggers used to recognise shock in obstetric practice, as do the upper 90% reference points for systolic and diastolic BP for obstetric hypertensive triggers. The SI upper limit of 0.89 in well postpartum women supports current literature suggesting a threshold of 0.9 as indicating increased risk of adverse outcomes.

A systematic review of the relationship between blood loss and clinical signs

INTRODUCTION

This systematic review examines the relationship between blood loss and clinical signs and explores its use to trigger clinical interventions in the management of obstetric haemorrhage.

METHODS

A systematic review of the literature was carried out using a comprehensive search strategy to identify studies presenting data on the relationship of clinical signs & symptoms and blood loss. Methodological quality was assessed using the STROBE checklist and the general guidelines of MOOSE.

RESULTS

30 studies were included and five were performed in women with pregnancy-related haemorrhage (other studies were carried in non-obstetric populations). Heart rate (HR), systolic blood pressure (SBP) and shock index were the parameters most frequently studied. An association between blood loss and HR changes was observed in 22 out of 24 studies, and between blood loss and SBP was observed in 17 out of 23 studies. An association was found in all papers reporting on the relationship of shock index and blood loss. Seven studies have used Receiver Operating Characteristic Curves to determine the accuracy of clinical signs in predicting blood loss. In those studies the AUC ranged from 0.56 to 0.74 for HR, from 0.56 to 0.79 for SBP and from 0.77 to 0.84 for shock index. In some studies, HR, SBP and shock index were associated with increased mortality.

CONCLUSION

We found a substantial variability in the relationship between blood loss and clinical signs, making it difficult to establish specific cut-off points for clinical signs that could be used as triggers for clinical interventions. However, the shock index can be an accurate indicator of compensatory changes in the cardiovascular system due to blood loss. Considering that most of the evidence included in this systematic review is derived from studies in non-obstetric populations, further research on the use of the shock index in obstetric populations is needed.

Exploiting the existence of temporal heart-rate patterns for the detection of trauma-induced hemorrhage

Unattended hemorrhage is a major source of mortality in trauma casualties. In this study, we explore a set of prehospital heart rate (HR) time-series data collected from 358 civilian casualties to examine whether temporal HR patterns can be used for automated hemorrhage identification. Continuous and reliable HR time series are fragmented into overlapping segments of 128 s, with a 118-s overlap between each two neighboring segments, which are projected into a wavelet coefficient space using the Haar wavelet function. A supervised nearest-neighbor clustering algorithm is developed to explore the existence of temporal HR patterns represented by the wavelet coefficients to discriminate casualties with and without (control) major hemorrhage. The clustering algorithm identifies 162 HR patterns. The most frequent pattern is observed in 11 (23%) hemorrhage and 16 (5%) control patients, which is a significant association (p<0.05, chi-square test). When the top 10 patterns are combined for hemorrhage detection, their sensitivity and specificity are 0.68 and 0.79, respectively, and when the top 20 patterns are used sensitivity increases to 0.77 and specificity decreases to 0.71.