Triage is the systematic ordering of injured or sick patients for the purpose of allocating treatment resources.Reference Hardern1 In various forms, triage is widely used in daily medical care and is considered a critical component of mass casualty response.Reference Hoey and Schwab2, Reference Kluger, Mayo, Soffer, Aladgem and Halperin3 Its use in mass casualty settings reflects the concern that critically injured patients will experience time-dependent morbidity and mortality without appropriate (although not necessarily definitive) medical or surgical treatment. In this context, triage is the process of finding and prioritizing care for critical patients before the proverbial clock runs out and they suffer irreversible harm from their injuries. Although likely inaccurate, the name commonly given to this time-dependent component of mass casualty triage is “the golden hour.”Reference de Boer and Debacker4, Reference Lerner and Moscati5
Many factors may conspire to prevent the timely delivery of medical treatment to critically injured victims of a mass casualty event, ranging from extrication delays to damage to receiving hospitals. Among these factors, overtriage, which technically is the mislabeling of noncritically injured patients to receive immediate care in a mass casualty setting, but which more commonly has been used as a general metric of emergency department (ED) overcrowding during disasters, has received considerable attention in the trauma and emergency health services literature. In a classic 2002 article analyzing a series of 10 terrorist bombings from 1969 to 1995, Frykberg described a positive linear relationship between overtriage and mortality among critically injured victims of mass casualty events.Reference Frykberg6 Based on this finding, Frykberg states that “overtriage could be as life-threatening as undertriage because of the inundation of overwhelmed medical facilities with large numbers of critical casualties all at once which may prevent the timely detection of that small minority with critical injuries who need immediate treatment and jeopardize their survival.” (p 207)
This concern has been echoed both in the academic literature and in the definitive clinical guideline for trauma response issued by the American College of Surgeons (Resources for the Optimal Care of the Injured Patient, p 13).Reference Hoey and Schwab2,7–Reference Severance9 Several subsequent reports of mass casualty response have postulated alternative relationships between overtriage and clinical outcomes. Rodoplu and colleagues document 2 hospitals' responses to multiple bombings in Istanbul in 2003, in which both had high overtriage rates, but only 1 experienced critical mortality.Reference Rodoplu, Arnold, Yucel, Tokyay, Ersoy and Cetiner10 Aylwin and colleagues report a lack of linear relationship between critical mortality and overtriage in emergency response to the July 7, 2005 bombings in the London Underground subway system.Reference Aylwin, Konig and Brennan11 In both cases the authors note that despite being overtriaged, the hospitals in question retained sufficient capacity to provide high-level care to all of the patients.
The difficulty in predicting when and where a mass casualty event will take place and in capturing the details of the trauma care process in the hectic aftermath of such an event has prevented the conduct of prospective or randomized studies of the role of triage performance and resulting overtriage levels on outcomes of mass trauma care.Reference Cone and MacMillan12 Several recent efforts to clarify these processes have used computer simulation and mathematical programming techniques to overcome these methodological obstacles. Reference Hirshberg, Stein and Walden13–Reference Stevenson, Oakley, Beard, Brennan and Cook15 The simulation model of single-hospital trauma care created by Hirshberg and colleagues confirms the importance of treatment capability (eg, radiological capacity, staffing levels) in determining the quality of trauma care, which may determine the rates of preventable morbidity and mortality.Reference Hirshberg, Stein and Walden13 A linear programming model of triage outcomes developed by Sacco and colleagues explicitly models the differential time-dependent mortality (TDM) of variably injured patients.Reference Sacco, Navin, Fiedler, Waddell, Long and Buckman14 There have been no studies, however, that systematically investigate the relationship between overtriage and critical mortality, taking into account case mix, triage performance, treatment capability, and TDM for mass casualty victims.
To test the hypothesis that overtriage has a consistent relationship to increased critical mortality after mass casualty incidents (MCI), we created a simulation model that includes 3 essential components of mass casualty care: the number and distribution of patients by casualty type, the triage process, and the treatment capability of the trauma care system. If this model were to show that the relationship between overtriage and outcomes were not consistent, then we would investigate the value of other candidate metrics for describing trauma system surge after MCI. We consider these matters from a trauma systemwide or regional, rather than a single-hospital, perspective, principally to understand the role of systemwide capability constraints on outcomes at various overtriage levels.Reference Sampalis, Denis and Lavoie16
METHODS
Model Structure
We constructed a discrete event simulation queuing model of mass casualty care with inputs for event size, patient type (ie, critical vs noncritical), triage test performance, trauma system treatment capability, time requirements for hospital-based evaluation and treatment, and TDM. To highlight the role of overtriage in outcomes in a parsimonious model, we considered only 1 of the several potential queuing-related delays in patient care during an MCI, namely the potential delay after arrival at the receiving hospital due to the unavailability of treatment resources. For simplicity, these resources are represented here by an available and appropriately staffed trauma bay linked, if needed, to an available, staffed operating room; other resources such as radiographic tests are not explicitly modeled, but the time required to conduct these tests is included in treatment time estimates.Reference Hirshberg, Stein and Walden13 We did not model patient extrication or transportation time from the site of the MCI to the hospital or the time required to apply the triage test, which commonly is reported to take ≤1 minute.Reference Cook17 Triage is, in this modeling environment, an instantaneous, test-based ordering of patients that determines the sequence of evaluation once patients reach the ED of a receiving hospital. Figure 1 provides annotated descriptions of the model structure and Table 1 lists the quantitative assumptions underlying each model component, described here.
FIGURE 1 Simulation modules and underlying logic of the model. A, Creation and triage of simulated mass casualty patient population; B, hospital-based treatment of simulated mass casualty patient population
TABLE 1 Model Assumptions
Event Size
We examined the triage and treatment of trauma patients resulting from MCI ranging in size from 50 to 1000 patients; we present detailed results for a baseline event size of 100 total patients both for clarity of presentation and because of the large number of real-life events with total casualties in that range.Reference Frykberg6 The proportion of critically to noncritically injured patients (defined by an Injury Severity Score [ISS] ≥15) varies around a baseline value of 25%, consistent with reports of recent mass casualty events in New York, Turkey, London, and Mumbai.Reference Rodoplu, Arnold, Yucel, Tokyay, Ersoy and Cetiner10,Reference Aylwin, Konig and Brennan11,Reference Cushman, Pachter and Beaton18,Reference Deshpande, Mehta and Kshirsagar19 Patients are generated simultaneously at the start of each simulation, reflecting an instantaneous injury-causing event such as a bombing.
Triage Test Performance
The model treats triage as a diagnostic test the function of which is to correctly label patients as critical or noncritical, thereby dictating treatment prioritization (ie, with critical patients receiving immediate stabilization and, possibly, damage control surgery and noncritical patients receiving rapid evaluation and expedited management). Although a large proportion of mass casualty victims will self-transport to health care facilities, we assume that arrivals at an emergency department are not treated in a first come-first served manner but rather are assigned a priority at some point before treatment.Reference Auf der Heide20 The current version of this model does not distinguish whether this priority score is applied in the field (ie, by emergency medical service first responders) or by hospital-based clinicians (typically junior or senior surgical staff located in the ED receiving bay); the increased complexity caused by events such as surges of walk-in patients unfolding over time will be addressed in future models.Reference Hirshberg8
Diagnostic test characteristics commonly are summarized using receiver operating characteristic (ROC) curves describing the relationship of sensitivity and specificity. We found high-quality published sensitivity and specificity data for the most commonly used triage protocol, START, in 1 peer-reviewed article, but this provided only 1 point on a potential ROC curve at 85% sensitivity and 86% specificity.Reference Garner, Lee, Harrison and Schultz22 To examine the theoretical range in test performance for START or a START-like triage protocol, we constructed a hypothetical ROC curve manually (see the Technical Appendix data supplement for details) and used the subsequent pairings of sensitivity and specificity for our examination of the impact of triage test performance on overtriage and outcomes.Reference Hoey and Schwab2,Reference Hirshberg8,Reference Cook17,Reference McNeil, Weber, Harrison and Hellman21,Reference Garner, Lee, Harrison and Schultz22
Overtriage
Overtriage may arise from 1 of 2 mechanisms: increasing the sensitivity of a triage test applied to a fixed patient population (ie, with fixed total casualty size and case mix) or decreasing the prevalence of critical patients in a population subjected to a given triage test (ie, with fixed sensitivity and specificity). The second mechanism may be subdivided into cases varying the proportion of critical casualties among a fixed total number of casualties (eg, an enclosed vs open-air bombing) and cases in which the number of critical casualties remains fixed but the total number of casualties changes (eg, events with fewer or more walk-ins). Table 2 shows 3 cases that illustrate overtriage arising from each of these mechansims, which are described in more detail below.
TABLE 2 Mass Casualty Overtriage Scenarios
Application of any diagnostic test will lead to type I (false positive) and type II (false negative) classification errors.Reference Hardern1 Using different scoring cutpoints or thresholds on a particular triage protocol (ie, changing the sensitivity and specificity of the test) will change the balance of type I and II errors. Increasing the sensitivity of triage (eg, lowering the threshold for some clinical or physiological value) will lead to more type I errors, which, independent of case mix and casualty load, will result in calling more patients with noncritical injuries “critical” (ie, increase overtriage). The opposite holds true for increasing the specificity of triage, which will lead to calling more critically injured patients “noncritical” (ie, undertriage). Although it is possible to increase both sensitivity and specificity (eg, by devising an improved triage protocol), for any given protocol there will always be a tradeoff of improved sensitivity at the expense of specificity, and vice versa.
An axiom of evidence-based medicine is that prevalence of disease dictates the clinical consequences of diagnostic test performance.Reference Hardern1 At any given level of triage test performance, therefore, reducing the number of truly critical patients among a fixed total casualty load will increase the likelihood of false positive results, thereby increasing overtriage. Conversely, increasing the number of noncritical casualties around a fixed number of critical casualties (ie, increasing the total event size by adding only noncritical casualties) will also lead to an increase in false positive results, thus increasing overtriage. Unlike standard diagnostic tests for which the triad of sensitivity, specificity, and prevalence of disease are sufficient to fully define test performance, however, triage test performance in the context of the unified trauma system also hinges on the total number of cases, which reflects the total workload on a defined regional health care system. In particular, these 2 methods of increasing overtriage by changing the prevalence of the target condition are not truly equivalent because they change the “critical” workload (specifically the treatment time of all of the critical patients plus those noncritical patients who are listed as critical) resulting from the triage process. Thus, in 1 case overtriage results in a reduction in the critical work (ie, fewer critical patients who make up the bulk of the processing time, whose aggregate decrease in treatment time will outweigh the impact of a larger number of noncritical patients), whereas in another case there is an increase in the critical work (due to more noncritical patients triaged to the critical category and no change in the number of patients with truly critical injuries).
Hospital Treatment Capability
Mass casualty trauma victims with critical injuries may require nonoperative stabilization procedures and operative management of intrathoracic, intraabdominal, or intracranial injuries.Reference Hirshberg8 Ideally, these patients are rapidly assessed and treated to the extent possible in the ED (eg, via placement of airway, tube thoracostomy, transfusion, radiographic evaluation, exploratory laparotomy) and then, if necessary, are transferred to an open, staffed operating room for expedited surgical care. We therefore defined the unit of critical patient treatment capability as the combination of an open, staffed ED trauma bay linked to an unoccupied, staffed operating room. We based critical and noncritical patient treatment times (from hospital door through completion of operative management) on data used by Hirshberg and colleagues in their single-hospital simulation study.Reference Hirshberg, Scott, Granchi, Wall, Mattox and Stein23 They estimated that critical patients require on average 175 minutes of combined ED and operating room time and that noncritical patients require on average 11 minutes of ED evaluation time. We report data having used normal distributions for these treatment times with means of 175 and 11 minutes and SDs of 30 and 6 minutes, respectively. We investigated the impact of changing the distributions, means, and SDs of these values on outcomes.
Time-dependent Mortality
To estimate the impact of treatment delays on patient mortality, we created 3 TDM curves based on published reports. These represent late critical mortality (97% survival over the first 6 hours followed by a linear 18%/hour decline in survival to a baseline of 5% survival, based on data from Garner et al); linear critical mortality (a linear decline of 12%/hour to the 5% survival baseline, based on data from Sacco et al); and rapid critical mortality (exponentially decreasing survival due to a 57% increase in mortality every 10 minutes waiting in the treatment queue, based on data from Sampalis et al). Reference Garner, Lee, Harrison and Schultz14,16,22Table 1 summarizes the main assumptions and mortality curve formulas used in the model.
Modeling Software and Replication Parameters
The model was created using the Arena 9.0 discrete event simulation software package (Rockwell Software, Sewickley, PA). A total of 10,000 replications of the baseline case were performed to evaluate the precision of the model output. Results are presented with 95% confidence intervals based on the half-widths generated by 500 iterations of each case, representing the interval in which 95% of results from individual model replications may be expected to fall.
RESULTS
Baseline Scenario Results
The baseline case consists of an MCI involving 100 total patients of whom 25 are critically injured, with a triage process that is 85% sensitive and 86% specific, in a treatment environment that consists of 6 available trauma bays with mean treatment times of 175 minutes (30-minute SD) for critical patients and 11 minutes (6-minute SD) for noncritical patients, with linear time-dependent mortality. When run for 500 iterations with these inputs, the model estimates a critical mortality rate of 44.10% (±1.24% 95% confidence interval); the critical mortality rate estimated with 10,000 iterations is 43.33% ± 0.28%, a 77.4% reduction in variance. Mean time for completion of treatment of all 25 critically injured patients is 7.49 hours and for noncritically injured patients, the mean time is 5.43 hours.
Effect of Overtriage Due to Triage Test Performance
To demonstrate the impact of triage test performance on overtriage and outcomes, we ran the model at overtriage levels ranging from 10% to 59%, corresponding to ROC values generated for the START triage protocol (range of sensitivity and specificity 53%–98%; see Technical Appendix for details). Figure 2 shows model outputs over this range of overtriage for the baseline scenario stratified by treatment capability and choice of TDM curve. Treatment capability is represented in 2 ways: by number of available and staffed trauma bays and by the ratio of critical patients to available trauma bays, noted as the critical surge to capability ratio (CSCR).
FIGURE 2 Critical mortality due to overtriage resulting from varying triage test performance (N = 100, 25% critical, baseline treatment time, ±95% confidence interval)
The results in Figure 2 demonstrate that in general the difference in outcomes produced by varying triage performance across most levels of treatment capability is minimal. The average range from minimal to maximal predicted mortality for each stratum is 2.76%, equivalent to a difference in <1 additional saved life out of 25 critical patients. The largest effect of triage performance is evident under the exponential TDM assumption at the highest treatment capability level (CSCR = 1), where the maximum range between the highest (at 10% overtriage) and lowest (at 40% overtriage) mortality estimates is 9.96%, equivalent to a difference in outcome for 2 to 3 out of 25 patients.
Insofar as there are discernible differences between overtriage rates at each stratum, Figure 2 is notable for the flat or U-shaped relationship between overtriage and outcomes at a number of combinations of treatment capability and mortality assumption. The U-shaped relationship, best seen in the scenarios that produce aggregate critical mortality between 10% and 60%, show that overtriage rates in the 33% to 50% range (corresponding to a triage sensitivity of 85%–95%) produce superior results (ie, lower critical mortality) in moderately resource-constrained environments over a range of TDM assumptions. At the extremes of treatment capability (CSCR = 1 under the late and linear mortality assumptions, 2.1 under the late assumption, and 12.5–25 under all 3 assumptions), the curves flatten, indicating no particular advantage for a specific overtriage level.
Because these results indicate that treatment capability has a much larger effect on outcomes than triage characteristics, we reanalyzed the data in Figure 2 to show the relationship between CSCR and outcomes by TDM assumption (Fig 3). The 3 curves in Figure 3 show the primary importance of treatment capability and the secondary role of mortality assumption in determining outcomes, with only a minor effect of overtriage level due to triage test performance.
FIGURE 3 CSCR and choice of TDM assumption have a greater impact on critical mortality than variations in overtriage due to triage test performance (N = 100, 25% critical, baseline treatment time, confidence intervals not shown)
Effect of Overtriage Due to Change in Prevalence of Critical Casualties
Having found that overtriage due to triage test performance has a minimal effect on outcomes, we investigated the impact of increasing overtriage by reducing the prevalence of critical casualties. As shown in Figure 4, we found 2 distinct relationships between prevalence-related overtriage and critical mortality. Critical mortality declines when higher overtriage results from decreasing the number of critical patients among a fixed total number of casualties. For example, predicted mortality falls from 65.6% to 8.2% if the number of critically injured patients in a 100-person MCI decreases from 50 to 5, corresponding to an increase in overtriage from 14.2% to 75.7% (under baseline assumptions). In contrast, critical mortality increases with higher overtriage resulting from increasing the total casualty load while holding the number of critical casualties constant. Figure 4 shows that predicted mortality increases from 40.9% to 59.6% as the number of noncritical patients increases from 25 to 975 with a fixed critical patient load of 25, corresponding to an increase in overtriage from 14.4% to 86.7%.
FIGURE 4 Overtriage due to reduced prevalence of critical casualties may increase or decrease critical mortality depending on etiology. (Base case with 6 available trauma bays, linear TDM, and baseline treatment times. 95% confidence intervals are shown for critical mortality [Y axis] and overtriage [X axis] for the baseline 6 trauma bay scenarios.)
The positive and negative relationships between overtriage and mortality depending on the etiology of overtriage hold under a wide range of treatment capability assumptions, as illustrated in Figure 4 by the similar orientation of curves for each scenario with 4 or 12 as opposed to 6 trauma bays. Figure 5 reanalyzes the data in Figure 4 to clarify the relationship between critical patient load, treatment capability, and outcomes. The 3 scenarios in which the number of critical patients varies display a logarithmic association of critical mortality to CSCR, whereas the scenarios in which only the number of noncritical patients varies appear “stacked” at levels of CSCR that correspond to the ratio of 25 critical patients to 12, 6, and 4 available trauma bays (ie, CSCR = 2.1, 4.2, and 6.3). Because the last 3 scenarios could represent the phenomenon of increasing noncritical walk-ins around a fixed critical casualty load, we sought to clarify further the incremental effect of increasing overtriage in this fashion. Figure 6 breaks out these 3 scenarios by showing the total (left axis) and incremental (right axis) increase in critical mortality with increasing numbers of noncritical patients. This figure shows that over a range of treatment capabilities, the detrimental impact of walk-ins on outcomes declines with total patient load, so that the 101st patient evaluated in the ED but who is not critically injured has a larger negative impact on critical outcomes than the 1001st patient.
FIGURE 5 Data from Fig 4 reanalyzed to display the relationship between critical mortality and CSCR for scenarios with different prevalence of critical casualties by varying either the number of critical patients (open figures) or number of noncritical patients (filled figures). (Base case with 6 available trauma bays, linear TDM, and baseline treatment times, ±95% confidence intervals shown.)
FIGURE 6 Impact of increasing noncritical patients (walk-ins) on critical mortality across different treatment capability levels, showing both overall mortality (left axis, closed figures) and the incremental increase in mortality (right axis, open figures) attributable to each additional walk-in at various MCI sizes. (Base case with 6 available trauma bays, linear TDM, and baseline treatment times, ±95% confidence intervals shown.)
Sensitivity Analyses
We performed analyses to test the robustness of our results with respect to the queuing and processing time assumptions. Running the model under the baseline conditions with a nonprioritized, first come-first served queue (ie, equivalent to a noninformative triage test) produced a 13% increase in critical mortality, equivalent to 1 additional death of a critically injured patient. In contrast, using an exponential vs normal distribution for processing times led to a 16% reduction in critical mortality. Both noncritical and critical patient treatment times had a positive linear impact on critical mortality. In the baseline scenario, each 10-minute increase in noncritical treatment time raises the critical mortality rate by approximately 1.5%, with small variation depending on critical treatment time (range, 1.6% if critical treatment time is reduced from the baseline of 175 minutes to 2 hours; 1.3% if critical treatment time is increased to 4 hours). Each 10-minute lengthening in critical treatment time increases critical mortality by 1.2% (range, 1.2% if noncritical treatment time is 10 minutes; 1.1% if noncritical treatment time is 50 minutes). Changing the SDs of critical and noncritical treatment times yielded minor changes in outcomes; the combination of a large (1 hour) SD in critical treatment and small (6 minute) SD in noncritical treatment yielded the lowest critical mortality (43.3% ± 1.2%), whereas the combination of small (30 minute) SD for critical and large (30 minute) SD for noncritical treatment produced the highest mortality (46.0% ± 1.3%).
DISCUSSION
This simulation model of MCI response captures 2 essential features of trauma care, the expectation that critically injured patients will deteriorate over time and the assumption that there is a limited capacity to treat all patients. Even in such a parsimonious model, we found that overtriage has a complex nonlinear relationship to critical mortality, raising questions about its usefulness as a descriptor of trauma system patient load. We found 3 distinct associations between predicted mortality and overtriage: a small and frequently U-shaped relationship when overtriage is the result of variable triage test performance in a given patient population, a negative relationship when overtriage is the result of a decrease in the number of truly critical patients among a fixed total number of casualties, and a positive relationship when overtriage is the result of a dilution of a fixed number of critical patients among an increasing number of noncritical patients.
Our results highlight the importance of considering treatment capability when discussing the impact of overtriage. In each case the dominant driver of outcomes is the relative balance between the number of critically injured patients and available treatment capacity, a relationship that we propose capturing in a new metric for reporting mass casualty response, the CSCR. A high-CSCR event is one in which the number of critical patients outstrips available treatment resources, whereas a low-CSCR event is one in which resource availability is less likely to play a major role in determining outcomes. Engineers and clinicians who conceptualize surge treatment in terms of throughput will recognize that the CSCR is 1 step toward defining the resource utilization of a trauma system in a single number.
The correlation between overtriage and critical mortality has become a guiding principle in the evaluation and design of mass casualty trauma care systems7; however, the very concepts of trauma triage and critical mortality remain poorly defined and inadequately studied. For example, the definition of critical mortality used to assess mass casualty outcomes has been variably defined as the mortality rate among patients presenting with a high ISS (typically ≥15 or 16), completion of nonorthopedic operative management within 24 hours of admission, or admission to an intensive care unit. Reference Frykberg18,24,25 “Expectant” patients, referring to those who have such devastating injuries that they are not expected to survive without heroic (and resource-intensive) efforts, are often excluded from calculation of critical mortality in studies of overtriage, even though the proportion of expectant patients may be dictated by the very triage and treatment processes under study (ie, patients with salvageable injuries may become expectant due to delays in operative management).Reference Cushman, Pachter and Beaton18
Overtriage itself has been used as a proxy for discussion of trauma and critical care resource allocation during mass casualty events, whereby high overtriage rates have been taken to imply misallocation of scarce resources away from those patients who truly need it.Reference Severance9 Our results suggest that this characterization is overly simplistic. Even under the commonly held “golden hour” assumption of clinical decline after critical injury, higher overtriage may be beneficial, as demonstrated by the results for the exponential TDM assumption at the highest treatment capability level (Fig 2, CSCR = 1), showing that predicted outcomes may improve as overtriage rises and correspondingly as undertriage falls. To our knowledge, this is the first report quantifying nonlinear effects of overtriage in a mass casualty response and the first simulation model to document positive effects of overtriage at the trauma-system level.
Accurate reporting on mass casualty response requires the completion of 2 difficult tasks: the assessment of the number of critically injured patients presenting for care and assessment of the availability of treatment resources at the time that those critical patients present. As noted by recent reports on the London and Mumbai terrorist attacks of 2005 and 2006, these are highly dynamic assessments requiring both a new vernacular and new graphical representations of trauma system activation and utilization.Reference Aylwin, Konig and Brennan11, Reference Deshpande, Mehta and Kshirsagar19 We believe that the CSCR may prove to be a useful metric in this regard because it has the potential to help standardize the assessment of trauma system capacity across mass casualty events of varying size.
The need for an improved theoretical understanding of the genesis and impact of overtriage in mass casualty care is reflected in real-world practices: mass casualty triage and transportation strategies vary widely across the globe, with Israel and the United Kingdom representing ends of the overtriage spectrum. Whereas Israeli emergency medical services put a premium on rapid transport of all patients to hospital EDs for evaluation and treatment with a minimum of field-based differentiation among casualty types, in the United Kingdom field triage and management of mass casualty victims are extensive and may extend to the air transport of trauma surgeons to the incident scene before definitive patient transportation. Reference Segell11,26,27 If the relationship between overtriage and critical mortality were positive and linear, then an Israeli-type “scoop-and-run” approach would predictably increase the risk of poor patient outcomes for large MCIs. Our model is so highly abstract that caution must be exercised in applying it to any specific real-world scenario. However, it does provide evidence that for a wide range of scenarios defined by treatment capacity and patient mix, increasing overtriage mix led to improved critical care outcomes.
As with any modeling project, this study has a number of limitations relating to the accuracy and realism of model inputs and structure. Despite the prevalence of trauma and specially designated systems for trauma care, there is remarkably little peer-reviewed data on the TDM of critically injured patients. Of the 3 mortality curves used here (linear, late, and exponential), we had to supply the precise shape of the late mortality curve. We use an acutely abbreviated definition of evaluation and treatment of critical injuries (ie, only damage control surgery, not definitive treatment) because of our interest in investigating mortality at the “front end” of MCIs and because critical mortality in this model is the result of delays due to lack of timely access to surgical resources.Reference Frykberg25, Reference Liang, Shih and Shih28 This means that other essential aspects of trauma care, such as radiological procedures, are not explicitly modeled.
In the interest of clarifying the role of triage on critical mortality, we also made a number of simplifying assumptions about critical care (eg, normally distributed treatment times) that may have biased our results. One of the values in developing computer models is that additional features may be added in the future to increase the realism of this simulation.Reference Connelly and Bair29 For example, we do not consider realistic factors such as delays in field extraction, ambulance transportation time, hospital delays caused by factors other than queuing for ED or operating room availability, or more fine-grained distinctions between patient types aside from critical or noncritical. We hope to address these and other factors as more detailed information becomes publicly available, and to attempt to validate a predicted outcome against actual MCI responses.
Overtriage, in this modeling framework, turns out to be an intermediary measure that does not have a consistent relationship with increased critical mortality. In contrast, the ratio of critical patient load and treatment capability tracks nonlinearly but consistently with outcomes, and may prove to be a more useful metric of trauma system response. It is our hope that this model will be 1 of many building blocks to improve the effectiveness of mass casualty trauma systems.
Authors' Disclosure
Dr Hupert and Mr Hollingsworth were supported under contract no. 290-00-0013 from the Department of Health and Human Services, Agency for Healthcare Research during the initial development of the model.
Acknowledgment
The authors thank Dan Hanfling, MD, of Innova Health System, Aaron Bair, MD, of University of California-Davis, and Nicholas Cagliuso, MPH, and Daniel Meisels of New York-Presbyterian Hospital and Healthcare System, all of whom provided valuable clinical insights that contributed to the model presented here.
