Discussion
Hemorrhage is the leading cause of preventable death after trauma. This study demonstrates the difficulties clinicians face in promptly identifying bleeding trauma patients and highlights the consequences of delayed diagnosis. We established that the initial clinical examination of an injured patient, even when performed by an experienced trauma clinician, demonstrated only a moderate ability to detect major hemorrhage. Notably, the number of cases where MH was either missed or overdiagnosed was roughly equal, translating to diagnostic errors in about 1 in 10 injured patients. The mechanism and site of injury influenced the likelihood of a missed diagnosis of major hemorrhage, with penetrating or abdominal wounding patterns associated with underdiagnosis. Overdiagnosis was more likely to occur with hypotension, polytrauma, and clinician diagnostic uncertainty. Crucially, a missed diagnosis of major hemorrhage carried severe repercussions, resulting in a mortality rate three times that of the broader study population. This underscores the inherent limitations of relying solely on clinical examination in the prehospital setting for detecting potential hemorrhage.
The main implication of this paper is that occult hemorrhage after trauma is difficult to identify, even for expert clinicians. Therefore, a broader approach that emphasizes an astute assessment of risk, rather than sole reliance on diagnostic accuracy, becomes paramount. In situations where potential torso injuries exist, the emphasis should not be on absolute diagnosis but rather on the overarching risk of occult bleeding.16 This perspective is important in the prehospital setting, especially when a patient’s consciousness is compromised, as abdominal examination has proved unreliable under these conditions.17 When faced with a potential risk of occult torso hemorrhage, even if signs are minimal, clinicians should still prioritize critical therapeutic elements of care such as rapid progression to definitive care, early tranexamic acid administration, and a lower threshold for major hemorrhage protocol activation.
Another important implication is the apparent need for decision support. A sensitivity analysis of the diagnostic accuracy subgroups showed a clear gradient between those patients who were obviously unwell and obviously well. The most obviously unwell were those correctly diagnosed with MH, followed by those patients in whom MH was missed, those who were overdiagnosed, and those who were correctly identified as not having MH, who were the most well. Thus, the clinical gestalt of the expert prehospital clinicians was correct for the vast majority of patients, but the difficulties with diagnosis and decision-making lie in the patients that fall between these two easy-to-identify groups. In these cases, decision support that could help discriminate the difficult-to-identify patients would be valuable and is likely to result in a survival benefit, as non-compressible torso hemorrhage is a significant cause of preventable mortality in trauma patients.2 Point-of-care ultrasound,18 point-of-care blood tests,19 remote decision support with telemedicine,20 21 and risk prediction clinical decision support systems22–24 may help clinicians to correctly classify their patients. Our findings suggest that diagnostic tools that address bleeding risk in torso trauma caused by a penetrating mechanism may be especially helpful in preventing missed hemorrhage. A number of studies have shown that massive transfusion can be accurately predicted using data readily available early after injury.25 26 However, predicting 10 units of packed red blood cells given within 24 hours is problematic: these cut-offs are arbitrary, there may be treatment bias (units transfused and units needed may differ), as well as survivor bias.24 27 28 Survivor bias can be partially mitigated by using different thresholds, such as the CAT.29 Future prediction models should avoid dichotomous thresholds, predict transfusion needs, and focus on the first hours after injury.24
Efforts to improve diagnostic accuracy should not merely focus on the issue of sensitivity, though, given the equivalency of the absolute numbers of patients exposed to a false-positive misclassification. Since clinician uncertainty indicated a propensity for over-triage, specificity might be improved by training prehospital providers to recognize their uncertainty, maintain awareness of likely bias, and incorporate confirmatory steps to resolve diagnostic fidelity. Decision support is difficult to implement in the dynamic, uncertain, information-poor circumstances of prehospital trauma systems.30 The American College of Surgeons Committee on Trauma (ACS COT) recommends that prehospital triage protocols perform at benchmark under-triage levels of <5% and over-triage levels of <50%.31 It is a challenge to achieve these benchmarks, even for advanced systems,32 and the problem of prehospital identification of severely injured patients is well described.33 34 Enhanced measures to re-triage patients on hospital reception, and to stand-down the MHPA—or conversely to initiate it—as soon as possible in patients once their true state is known, may reduce the risks associated with misclassification.
The findings of this study are consistent with existing literature. We have previously highlighted how challenging it is to accurately diagnose injuries after major trauma in the prehospital setting.16 A recent systematic review and meta-analysis demonstrated that clinical examination performed prehospital was less likely to identify life-threatening injuries than when performed in-hospital (pooled sensitivity of 46% vs 76%, respectively; p<0.0001).17 For this study, we aimed to understand performance with regard to the identification of patients with major hemorrhage. Our findings broadly align with the landmark PROMMTT (PRospective Observational Multicenter Major Trauma Transfusion)35 trial, which evaluated clinician gestalt regarding MH.7 In this study, patients underwent primary survey in the trauma bay (emergency department resuscitation), at which time clinicians were asked whether the patient was likely to receive a massive transfusion or die from hemorrhage.7 Clinician performance was 65.6% sensitive and 63.8% specific for predicting these endpoints, with NPV and PPV of 34.9% and 86.2%, respectively. PROMMTT differed from our study in setting (trauma bay vs prehospital), endpoint (massive transfusion, rather than CAT) and study design (prospective vs retrospective) though the seniority and experience of PROMMTT clinicians (“trauma attendings”) was similar. PROMMTT attested to the difficulty of predicting patient treatment requirements and outcome, even within a hospital environment. Given the evolution of contemporary trauma systems toward better prehospital care, there is a requirement to understand the accuracy and impact of prehospital as well as resuscitation-bay diagnoses.7 29
This study had several strengths. First, the clinicians who examined patients in our study were very experienced, therefore reducing any bias that may be caused by analyzing findings from inexperienced clinicians. Second, the evaluation occurred prehospital, with no access to advanced diagnostics, therefore minimizing the chance of contamination of clinical assessment results with the results of diagnostic tests. Third, to diminish the risk of survivor bias, we defined major bleeding using CAT instead of massive transfusion (MT). MT, frequently defined as ≥10 units (U) of packed red blood cells (RBCs) within 24 hours, is often used as an endpoint for MH. However, the use of MT risks excluding hemorrhaging patients who die before they can receive 10U RBC, resulting in a survivor bias.27 28 Metrics such as CAT (≥3 RBCs in first 60 min), and resuscitation intensity (RI, total products in first 30 min, including 1 U RBC, 1 U plasma, 1000 mL crystalloid, and 500 mL colloid, each valued at 1 U) have been proposed to counter this bias.27 In a meta-analysis, CAT was found to be a more sensitive predictor of mortality at 24 hours than MT and RI.36 The CAT definition has been expanded to ≥3 units RBC within each 60-minute period within the first 24 hours.13 14 This expanded CAT definition captures patients who have blood products later than on first arrival to the hospital.
There were some limitations of our study. First, we used a retrospective design, which predisposes to information and selection bias, but this was mitigated by evaluating consecutive patients presenting to a MTC. Second, there was some missing data, but attempts were made to corroborate primary data with other sources including paper records and concurrent prospective observational trial data, and any missing data were acknowledged in the tables. Third, we chose the MHPA decision as a useful surrogate for the prehospital diagnosis of MH as, aside from outright clinical error, it is difficult to conceive of circumstances where the latter condition does not result in an MHP activation. Fourth, there is an unspecified risk that the population of patients studied may not be representative of the totality of London Trauma System major trauma patients as it did not include patients who had major hemorrhage to whom LAA were not dispatched, as well as patients who LAA were dispatched to but died prior to admission to the MTC.
Future research should address more nuanced questions, using mixed methods or qualitative methodological approaches to identify the determinants and influences of the MHPA decision, what makes it difficult, and what could make the decision easier. Such research might expand on recent work showing how prehospital trauma clinicians make decisions using incomplete information, alter their judgments as more information becomes available, estimate the likely outcome of alternatives, use heuristics for rapid decision-making, and employ recognition-primed thought processes.37 38