paramedic injury severity perception can aid trauma triage

EMS/ORIGINAL CONTRIBUTION

Paramedic Injury Severity Perception Can Aid

Trauma Triage

From the Oregon Health Division* and the Department of Emergency Medicine, Oregon Health Sciences University, Portland, Oregon.

Received for publication September 19, 199"i. Revision received May 3, 1995. Accepted for publication May !5, t995.

Presented at the I994 Annual Meeting of the National Association of EMS Physicians, Portland, Oregon, August 1994.

Supported in part by grant HRSA #93-953 from the US Department of Hearth and Human Services, Health Resources and Systems Administration.

Copyright © by the American College of Emergency Physicians.

Erik Simmons, MS* Jerris R Hedges, MD, MS* Lisa Irwin, RN, MPA/HA*

Wilhelmine Maassberg, RN*

Howard A Kirkwood Jr, JD, EMT-P*

Study objective: To compare information contained in standard out-of-hospital trauma triage criteria and standard criteria plus advanced emergency medical technician (EMT)injury severity perception for determination of patient need for trauma center evaluation.

Design: Prospective, observational cohort analysis of trauma triage by advanced EMTs.

Participants: 0ut-of-hospital, geographically stratified statewide sample of patients injured in Oregon.

Results: Advanced EMTs provided patient information on demo- graphics, physiologic parameters, injury anatomy and mechanism, premorbid conditions, EMT injury severity perception, and trauma system entry status. A four-point scale was used to grade the injury severity perception. Need for trauma center evaluation was defined as major surgery within 6 hours of hospital arrival, admission to the ICU, death in the hospital, or Injury Severity Scale score of 16 or more. The relative triage information gain with injury severity perception was assessed by use of logistic regression, tree-based models, and receiver operating characteristic (ROC) curves. Of 1,063 patients, 307 (28.9%) warranted trauma center evaluation. With logistic regression modeling, the following standard triage parameters were associated (P<.05) with the need for trauma center evaluation after inclusion of injury severity perception: systolic blood pressure less than 90 mm Hg, abnormal respiratory rate (less than 10 or more than 29), Glasgow Coma Scale score less than 13, penetrating injury (midthigh to head), two or more obvious proximal long- bone fractures, and fall of more than 20 feet. The two largest injury severity perception categories had the greatest odds ratios (20:1 and 167:1). R0C curve areas improved with injury severity perception (.88 versus .83 without; P<.0001 ).

Conclusion: Standard out-of-hospital triage criteria benefit from inclusion of advanced EMT injury severity perception information.

OCTOBER 1995 26:4 ANNALS OF EMERGENCY MEDICINE 46 1

PARAMEDIC INJURY SEVERITY Simmons et al

[Simmons E, Hedges JR, Irwin L, Maassberg W, Kirkwood HA Jr:

Paramedic injury severity perception can aid trauma triage. Ann Emerg Med October 1995;26:461-468.]

INTRODUCTION

It has long been accepted that implementation of a sys- tematic approach to the care of victims of serious trau- matic injury results in improved clinical outcomes. >3 Although it is generally accepted that patients with minor injuries do well regardless of hospital care site'*, the treatment of seriously injured patients in special- ized centers can significantly reduce trauma-related mortality, t-3 To expedite the selective transport of seriously injured patients in the out-of-hospital setting to trauma centers, several criteria for injury severity identification have been proposed and evaluated. >23 Most trauma care systems have adopted criteria based on the patient's presenting physiology, injury anatomy and mechanism, premorbid conditions, and emergency medical technician (EMT) judgment to identify patients

Table 1. Oregon triage criteria.

Triage Criterion Definition

1" 2* 3* 4* 5*

6* 7* 8* 9* 10" 11" 12" 13 14 15 16 17 18 19 20

21 22

*Mandatory.

Systolic blood pressure less than 90 mm Hg Respiratory rate less than 10 or more than 29 Glasgow Coma Scale score less than 13 Penetrating wound, midthigh to head Burns to more than 15% of total body surface or to face,

feet, hands,or genitalia (in conjunction with other injuries) Amputation, proximal wrist or ankle Spinal cord injury/paralysis Flail chest Two or more obvious proximal long-bone fractures Death of same-car occupant Ejection from enclosed passenger space Complex extrication lasting more than 20 minutes Fall of 20 feet or more Pedestrian hit at 20 mph or more thrown at least 15 feet Vehicle reliever Motorcycle, all-terrain vehicle, or bicycle crash Significant impact or intrusion Age less than 12 or more than 60 years Hostile environment (eg, extremes of heat or cold) Medical illness (eg, chronic lung disease, heart failure,

renal failure.) Presence of intoxicants Pregnancy

with potentially life-threatening injuries for triage to a trauma center. 24

Physiologic criteria alone are too insensitive to serve as the sole guide for the identification of injured patients who have been deemed likely to benefit from trauma center evaluation. 5-9,16 Mechanism of injury criteria suggesting a high-energy transfer and criteria based on premorbid factors lack specificity when used in isolation for identifying patients at risk for life-threatening injuries. 9,15,It,Is The role of EMT clinical impression has received limited evaluation. Prior studies have suggested that EMT judgment has results similar to those seen with more tradi- tional triage instruments in the prediction of death or need for immediate surgery, s,t<22

In Oregon, the decision to enter a patient into the trauma system is based on the presence of any one or more of 12 mandatory and 10 discretionary criteria that may be coupled with EMT concern. For example, a patient who exhibits physiologic decompensation must be transported to a designated trauma center. However, a patient who merely experiences a high-energy transfer is not necessarily transported to a trauma center unless the EMT in charge decides that the patient might benefit from a trauma center evaluation. In those regions of the state where the only possible receiving hospital is a participating trauma center, the out-of-hospital enrollment of a patient into the trauma system by the EMT results in additional resources being available on the patient's arrival at the hospital.

We undertook this prospective study to determine the sensitivity and specificity of Oregon triage criteria in the identification of the injured patient in need of trauma center evaluation and to ascertain whether advanced EMT injury severity perception adds significant information to the standard trauma system triage criteria.

MATERIALS AND METHODS

We performed a prospective, observational cohort analysis of trauma triage criteria with a geographically stratified sample of injured patients in Oregon. The study was exempt from institutional review board review, and all data were maintained at the Oregon Health Division and subject to state-legislated confidentiality protection.

Oregon has a fully implemented, statewide inclusive trauma system. Forty-seven of the 63 currently licensed acute care hospitals in Oregon are categorized as trauma hospitals; of the 16 hospitals that do not participate m the system, all but 2 are excluded from participation by the designation process used in the greater Portland metro-

462 ANNALS OF EMERGENCY MEDICINE 26:4 OCTOBER 1995

PARAMEDIC INJURY SEVERITY Simmons et ~.I

politan area) The Oregon Trauma Registry (OTR) collects data on all trauma system patients. Input to the registry is provided in most cases by means of computerized data entry at each trauma hospital.

All injured patients who were treated by participating EMS providers and who required transportation to a hospital or who were pronounced dead at the scene in Oregon between October 1, 1992, and August 31, 1993, were eligible for enrollment. Any patient with an obvious isolated single-extremity fracture, an isolated hip fracture due to a fall, or minimal external injuries who did not meet any triage criteria was excluded, as was any patient who did not have a properly completed out-of-hospital data form or for whom incomplete information or no reformation was obtained from the receiving hospital.

Providers were chosen to geographically represent patients and emergency medical service (EMS) agencies participating in the Oregon Trauma System. Except for the Portland metropolitan area, one EMS agency was chosen on the basis of volume and location in each trauma system area. In the Portland metropolitan area, where ap- proximately 40% of the state's population lives, four private ambulance agencies that transport patients and two fire departments that are first-response agencies participated in the study. All of the out-of-hospital care providers that participated in this study were advanced EMTs (ie, EMT-IIIs or EMT-paramedics [EMT-Ps]).

Patient information was collected through the use of a supplemental data sheet completed by the EMTs. In- formation was obtained on patient presenting physiology, injury anatomy and mechanism, premorbid conditions, EMT injury severity perception, and whether the patient was entered into the trauma system. Mandatory and discretionary triage criteria for trauma system entry in Oregon are shown in Table 1. EMTs were instructed to denote all triage criteria that applied when completing the data sheets. Information on the EMT'S injury-severity perception was collected with a four-point scoring system similar to that of Emerman et a122 (Table 2).

Information on patient outcome was collected with a data sheet completed by the medical records department of the destination hospital. These data included information on outcome, surgical requirements in the first 6 hours of admission, need for ICU monitoring, length of hospital stay, transfer for higher care level, Injury Severity Scale (ISS) score, and final diagnosis. A copy of the patient's hospital discharge summary was also collected. All data were entered into a computer database, where they were reviewed by a nurse trauma system coor- dinator for completeness.

An injured patient was considered to be at risk for major trauma and in need of trauma-center evaluation (TCE+) if any of the following conditions were noted on hospital follow-up: major surgery within 6 hours of hospital arrival, ICU admission within 3 days of hospital arrival, ISS score of 16 or more, and death.

With TCE+ as the response variable, we used several statistical techniques to measure the predictive strength of the triage criteria alone and in combination with injury severity perception. Two triage criteria had to be excluded from part or all of the analysis. First, all six patients with the amputation criterion marked were TCE+. Because numeric regression algorithms break down in such circumstances, the amputation criterion was not included in the logistic-regression models. Second, in no case received was the pregnancy criterion marked; this criterion was also excluded from analysis.

We used logistic regression modeling to develop statistical models with which to predict TCE outcome and to generate TCE+ odds ratios (ORs) for specific triage criteria. 25 Stepwise logistic regression models were developed with each triage criterion individually and as multivariate models of all criteria and all criteria plus injury severity perception. All triage criteria and the injury severity perception variable were coded for ORs with the absence of the criterion or the first category as reference. Analysis of deviance, partial residual plots, and other tools were used to assess and compare the fitted models. 26

Table 2. EMT injuu severity perception categories.

Category Name Description

1 Minor Patient transported as a precaution; may need admission. 2 Acute Non-fife-threatening. Patient will likely need diagnostic studies or ED wound care; probably needs admission. 3 Life-threatening Patient obviously needs admission; probably needs surgery or admission to ICU. 4 Needed resuscitation CPR, intubation, or bag-valve-mask ventilation performed in the field.

OCTOBER 1999 28:4 ANNALS OF EMERGENCY MEDICINE 4 6 3

P A R A M E D I C INJURY SEVERITY Simmons et al

Tree-based modeling uses recursive partitioning to repeatedly split the response variable (TCE+ versus TCE-) according to the best choice of predictor variable at each step. 27 Tree models can model interaction among variables more generally than regression models, and they are often easier to interpret. The reduction m deviance can be plotted against tree size (ie, the number of terminal nodes) to yield a cost-complexity picture of how much information is gained by each split of the model. In this study, classification trees were used to model TCE both with all triage criteria and with all triage criteria plus injury severity perception. These models provide an alternative to the logistic regression models constructed with the same variables. We also created a tree model with injury severity perception as the response and all triage criteria as predictors. This model was used to gain information about the relationship between injury severity perception and various combinations of triage criteria.

Receiver-operator characteristic (ROC) curves graphically display the amount of information imparted by a model. One constructs an ROC curve by plotting sensitivity against specificity loss (defined as 1-specificity) for a range of estimated response probabilities. For this study, sensitivity equates to the proportion of TCE+ patients correctly predicted by a model (ie, a true positive) and specificity loss to overtriage where a patient is predicted by the model to need TCE, but did not meet any TCE+ criteria (ie, a false positive). We constructed ROC curves for several of the logistic regression and classification tree models by computing the sensitivity and specificity loss for TCE+ ranging from 0 to 1 by increments of .01. The area under each curve was estimated with the trapezoidal rule. xs We obtained the dif- ference in sensitivity between two models for any given rate of overtriage by comparing their ROC curves. 29

Table 3. EMT injury severity perception by need for trauma center evaluation.

Life- Needed Perception Minor Acute Threatening Resuscitation Total

TCE- 289 366 75 1 731 TCE+ 13 59 160 59 291

Total 302 425 235 60 1,022

In 41 cases, injury severity perception was net noted. The one patient with an injury severity perception of "needed resusc;tation" and with TCE- was extremely intoxicated.

R E S U L T S

Of 1,189 eligible cases, 126 cases (10.6%) were excluded because of missing information; 41 cases missing only injury severity perception reformation were not excluded. The missing cases had demographic data and geographic distribution similar to those of patients with complete data sets. Of the remaining 1,063 cases, 307 (28.9%) were TCE+ and 472 (44.4%) were actually entered into the trauma system in the out-of-hospital setting. Of the 472 trauma system entries, 268 (56.8%) were TCE+. Overall, the EMTs' trauma system entry performance had a sensitivity of 87% and a specificity loss of 27%.

A breakdown of injury severity perception by TCE outcome is shown in Table 3. Logistic regression models were constructed containing each criterion individually and for all criteria both with and without injury severity perception. Many currently used triage criteria did not significantly predict TCE with or without injury severity

Table 4. Logistic regression model with all triage criteria and injury severity perception.

95%CI Variable ~ se(~) OR (o~) for OR P

Intercept -3.4173 .3214 - - - - .0000

Triage criteria 1 2547 ,3964 2,13 .98-4.61 .0571 2 1.3259 .4235 3 .77 1.64-8.60 .0018 3 ,8654 .2640 2 ,38 1.42-3.98 .0011 4 .8877 .3223 2 .43 1.29-4.55 .0059 5 -.5498 1.4527 .58 .03-9.81 .7052 7 -.7701 .5922 .46 ,15-1.47 .1938 8 1.7488 .9873 5 . 7 5 .83-39.41 .0768 9 1.9650 1.0006 7 .13 1.00-50.21 .0498

10 .9993 .7914 2 . 7 2 .58-12.71 .2070 11 .7003 .4548 2.01 .83-4.89 .1239 12 .2473 .5692 1.28 .42-3.89 .6640 13 1.6582 .5926 5,25 1.64-16.67 .0052 14 .7096 .4042 2,03 ,92-4.47 .0794 15 .0685 .3574 1.07 .53-2.15 .8479 16 .4142 .3935 1.51 .70-3.26 .2928 17 .3596 2289 1.43 .91-2.24 .1165 18 -.0210 .2618 .98 .59-1.63 .9360 19 -.2987 .3903 .74 .35-1.59 .4442 20 2767 .4804 1.32 ,51-3.37 .5658 21 -.0023 .2614 1.00 .60-1.67 .9931

Injuryseverity perception category 2 .9964 .3266 2.71 1.43-5.12 .0023 3 2.9998 .3531 10.08 10.05-39.98 .0000 4 5.1159 1.0363 166.65 21.86-1,257.32 .0000

4 6 4 ANNALS OF EMERGENCY MEDICINE 26:4 OCTOBER 1995


perception. In fact, no single criterion predicted more than 44% of the TCE+ patients. However, in these models, injury severity perception appears to be an extremely good predictor of TCE (Table 4). Even after control for the triage criteria currently in use, the injury severity perception ORs for TCE are 20:1 and 167:1 for the largest injury severity perception categories (all ratios are all relative to injury severity perception category-i).

A search for first-order interactions in the model yielded only a few weak interactions that did not radi- cally change the overall fit. These were ignored in the interests of simplicity and interpretability, although an interaction between Glasgow Coma Scale (GCS) score less than 13 and the presence of intoxicants is worth mentioning. When both of these conditions were true, the overall effect in the study population was a decreased probability of TCE+. This result, although intuitive, runs counter to the notion that the more criteria present, the larger the predicted probability of TCE+.

Analysis of deviance indicated that the addition of injury severity perception to the model containing all criteria was highly significant (P<10 -l°) and provided a reduction in deviance of 20% over the model with triage criteria only. By comparison, introduction of all the triage criteria to a model containing only injury severity perception was significant (P<10 -5) but resulted in just an 8% reduction in deviance over the model with injury severity perception alone.

A quantile-quantile plot of the predicted probabilities from the two models showed that the model with injury

severity perception tended to classify many patients with TCE+ at a higher probability than the model without injury severity perception. These results suggest that injury severity perception is a powerful predictor of TCE+; in fact, injury severity perception appears to be the most powerful predictor considered, stronger than all the triage criteria combined.

Two tree-based models were constructed in a fashion similar to that of the logistic regression models. The first model contained only the triage criteria; the second model contained these criteria along with injury severity perception. The injury severity perception variable was chosen before all triage criteria and was split into two groups defined by injury severity perception categories 1 and 2 and injury severity perception categories 3 and 4. The strength of injury severity perception as a predictor of TCE is illustrated by the fact that injury severity perception was used as the splitting variable in the first three nodes of the tree.

In an effort to find out which triage criteria were associated with injury severity perception, we also constructed a tree model using injury severity perception as the response and all triage criteria as predictors. The criteria most predictive of high injury severity perception were GCS less than 13, systolic blood pressure less than 90 mm Hg, respiratory rate less than 10 or more than 29, and penetrating wound (midthigh to head). Other associations included flail chest, two or more proximal long-bone fractures, significant impact, ejection from a vehicle, fall of 20 feet or more, and vehicle rollover.

Figure 1. A, ROC plot for logistic regression model with all triage criteria; B, ROC plot for logistic regression model with cdl triage criteria and EMT irzju~:y severity perception.

Sensitivity 1 .0 -

. 8 - ~

/

, 6 - - / ¥ ÷ /

t . 4 - I

f . 2 - ?

I I ] / / /

.0

. 0 -

A

,~ //" / //"

/ / / / / / /

//" /,/

/ / / / /

// ///////"

,// / //"

//" ///"

Area Estimate: .8351 I I

.2 .4 .6 1-Specificity

I.F,"

I

.8 I

1.0

* Actuarsystem pedormance

1.0

.8

Sensitivity . ~ . . . . . # /

q¢ . . ,+ . i 1 1 1 ~ 4 " ~ /////"

q[.4 / / / // //

,~_ /'/ / / ] ,"

/ / / / /

/ 4- //" 4- /" 4- / / /

-- / , /

I / / / / ] /" r~ /

,/ I

.0

Area Estimate: .8921 I I I I I

.2 ,4 .6 .8 1.0 1-Specificity

O C T O B E R 1995 26:4 ANNALS OF EMERGENCY MfiDICINE 4 6 5

PARAMEDIC INJURY SEVERITY 5*mmons et al

ROC curves were constructed for both logistic regression models and also for both tree-based models. The ROC curve area for the logistic regression model without injury severity perception was .84 versus .89 with injury severity perception (P<.0001). The inclusion of injury severity perception brought both logistic regression (Figure 1A and B) and tree-based model ROC curves (Figures 2A and B) close to the actual performance of the EMTs in the study For the logistic regression model, the specificity loss at an operating point of 80% TCE+ sensitivity is roughly 50% less with injury severity perception (ie, injury severity perception provided a reduction in overtriage from 30% to 15%).

D I S C U S S I O N

Presuming that a seriously injured patient can be identi- fied with a high degree of accuracy, the patient who meets triage criteria in an established trauma system is taken directly to a designated trauma hospital for trauma team evaluation. In the Portland metropolitan area, where only 2 of 16 hospitals are designated trauma centers, patients who do not meet such criteria are taken to the hospital of their choice when that option exists. In a rural area with only one receiving hospital, triage criteria are used to guide specific trauma responses. Furthermore, ma W EMS systems and state health divisions specifically monitor the outcomes of the patients entered into the trauma system at considerable expense. Hence it is critical for a tool that identifies the patient as %ntered" into a trauma system to

have an appropriate balance between the ability to recognize serious injury (sensitivity) while avoiding overtriage of patients without serious injury into the trauma system (specificity loss). In the latter case, the triage tool could adversely affect the health care environment by increasing the cost of delivering and monitoring trauma care for patients subsequently found to have minor injuries. 3°

We performed a critical analysis of existing statewide trauma triage criteria in Oregon and addressed the role of advanced EMT judgment (injury severity perception). We found (1) that standard criteria for trauma hospital triage are inadequate when taken individually; (2) although combinations of criteria can be reasonably discriminating, EMTs use additional cues to fine-tune their decision- making for trauma triage; and (3) that EMT injury severity perception adds to the accuracy of trauma triage.

Currently, the Oregon trauma system can be activated for a patient on the basis of the presence of any single trauma triage criterion (Table 1), independent of the presence of others. The univariate logistic regression models suggest that this approach is inadequate and does not match actual EMS system performance. Hence it is likely that advanced EMTs use more than single criteria for trauma triage.

The criterion for amputation may be a good predictor of TCE+, but we could not evaluate it in this study because of its infrequent occurrence and its complete concordance with TCE+. The sensitivities of the other individual criteria seem to naturally fall into three groups: criteria with sensitivity greater than 20%, criteria with sensitivity from

Figure 2. A, ROC plot for tree model with all triage criteria; B, ROC plot for tree model w~th all triage criteria and EMT injury severity perception.

1.0

.8

.2

.0

A

Sensitivity

/m' ..+-" /,

,/ I+ / / / /

/ //" /,

/ / / ,/ /

m / /

I / / Jr / / / ~_ / / / / /

/ / / / / / ,

///"

,,," Area Estimate: .8577 I I I

.0 .2 .4 .6 .8 1-Specificity

~ - - ~ " Sensitivity

1.0

4" .8 /-

.6 /

.4 4 l/l/ 4[ //"

/,// ,2 / /

f /m /

/ .0 /" Area Estimate: .9061

I I I 1 0 .0 .2 .4 .6

B 1-Specificity

* Actua[ s y s t e m p e r f o r m a n c e

/, / /

/ / /. m / /,

/ , /

/,

,/ / /

t I .8 1.0

//" / /

/,.

4 6 6 ANNALS OF EMERGENCY MEDICINE 26:4 OCTOBER 1995


5% to 10%, and criteria with sensitivity less than 3%. Only the physiologic status criteria (systolic blood pressure less than 90 mm Fig, respiratory rate less than 10 or greater than 29 per minute, GCS score less than 13) and penetrating-injury mechanism (midthigh to head) fell in the first group; 12 criteria fell in the third group. Hence more than half of the criteria could not identify the need for trauma center evaluation more than 3% of the time. Review of individual cases suggests that more precise defi- nitions may improve the predictive power of some criteria. For example, the "spinal cord injury" criterion was commonly marked by our EMTs in the setting of a possible spinal cord injury mechanism, not just when paralysis was present:

The predictive power of models combining criteria was substantially better than for the univariate models. However, modeling the triage process with the standard triage criteria could not produce a multivariate model that performed as well as actual system performance. Several criteria that possessed univariate significance became insignificant in the multivariate models. The lack of TCE predictive power for premorbid conditions confirms their weakness as individual triage criteria.

By several techniques (logistic regression model analysis of deviance, tree-based modeling, and ROC curve area analysis) the addition of injury severity perception was found to strengthen triage model performance. In the full logistic regression model, injury severity perception was the most significant variable in the regression. It had the largest ORs of any variable and provided a large reduction in deviance over the model with triage criteria only. Many of the previously significant criteria were weakened by the addition of injury severity perception. This may have been because injury severity perception is a good surrogate for several current criteria.

The tree model using injury severity perception as the response demonstrated that the criteria most strongly predictive of TCE+ also were the criteria associated with injury severity perception. This implies that the criteria may have influenced an EMT's injury severity perception through repeated use. If the tree model is thought of as a surrogate for the EMT's decision process, injury severity perception is judged mainly on the basis of the presence of decreased GCS score, penetrating wounds (midthigh to head), decreased systolic blood pressure, significant impact or passenger-space intrusion, fall of 20 feet or more, and passenger ejection. The first splits (decision points) in the tree are all measurable or visi- ble conditions that have been shown above to be highly predietive of the need for TCE.

These findings extend earlier studies assessing the impact of EMT judgment as a method of triage, s,1<22 Ornato et al 8 found that EMT impression using a five- point scale was more sensitive and more specific than the CRAMS Score 5,6 or Trauma Score 9,14 in predicting the need for immediate surgical intervention. Hedges et a116 found that EMT-P impression on a three-point scale complemented but did not surpass 11 standard triage instruments. Emerman et a122 found that using either EMT estimates of mortality probability on a continuous 0-to-100 scale or of injury severity on a discrete fou> point scale, they could predict either mortality or need for immediate surgery similar to the CRAMS Score, Trauma Score, and Prehospital Index. lO,ll Their EMT impression of mortality was most associated with GCS score components, respiratory effort, and systolic blood pressure. 22 More recently, Esposito et aP 1 found that a binary provider "gut feeling" variable was a weak predictor of an ISS score of 16 or more but added valuable high-yield information to the cases of those patients who met mechanism of injury triage criteria.

This study has several limitations. It is based on a geographically stratified sample of advanced EMTs and a sequential sample of patients enrolled by those providers during the study period. Each provider group enrolled cases during the study period until a sample representa- tive of the geographic composition of the Oregon Trauma Registry was obtained. The participating advanced EMTs represented motivated providers active in the state trauma system. We did not assess the triage practices of basic life support EMTs.

Although this study assessed triage practices in a rela- tively large number of cases, some standard triage criteria were present in only a few cases. Limited use of some criteria produces large SEs in fitted-model parameters and may lead to a criterion being labeled "insignificant" as a result of large CI. In studies evaluating a prediction tool, the dataset is often randomly divided into two sub- sets. One half is used to develop a prediction tool, and the other is used to validate the prediction tool. Because our main goal was to assess the i~fformation content of triage criteria rather than to develop a prediction tool, we did not divide our dataset. A second dataset will be needed to validate any revised triage policy suggested by our results.

Future investigations should address whether basic EMTs have similar triage practices and apply clinical judgment to the same extent. The role of clinical experi- ence and geographic practice location on injury severity perception for basic EMTs and advanced EMTs warrants

OCTOBER 1995 26:4 ANNALS OF EMERGENCY MEDIC/NE 46 7

P A R A M E D I C I N J U R Y S E V E R I T Y Simmons et al

evaluation. Knowledge of the impact of these provider variables on injury severity perception would reflect on the reliability of our results to EMT triage practices across the state.

Finally, although we have shown the value of injury severity perception for the fine-tuning of triage decisions, we have not demonstrated that the clinical judgment of EMTs alone is sufficient for triage decisions. It is likely that a complex relationship exists between daily applica- tion of triage criteria by EMTs, clinical feedback to EMTs provided by trauma hospital providers and EMS physician supervisors, and the subsequent development of injury severity perception.

CONCLUSION

Our data suggest that advanced EMTs recognize that significant abnormal physiologic findings and certain injury mechanisms and anatomic considerations are associated with the need for trauma center evaluation. However, advanced EMTs also use judgment when weighing the need for a trauma center in the setting of normal physiologic parameters. Hence EMT trauma triage discretion should be encouraged as a supplement to the more specific predictors of trauma center evaluation (ie, abnormal physiologic parameters, penetrating trauma [midthigh to head], and proximal amputations). The complex interaction of trauma triage criteria use and development of EMT perception of injury severity warrants further investigation.

REFERENCES 1. Shackford SR, Holiingworth-Fridlund P, Cooper GF, etal: The effect of regienalization upon the quality of trauma care as assessed by concurrent audit before and after institution of a trauma system: a preliminary report. J Trauma 1986;26:812-820.

2. Cales RH, Trunkey DD: Preventable trauma deaths: A review of trauma care systems development. JAMA 1985;254:1959-1063.

3. Muilins R J, Veum-Stone J, Holland M, et ak Outcome of hospitalized injured patients after institution of a trauma system in an urban area. JAMA 1994;271:1919-1924.

4. Hedges JR, Osterud HR, Mullins R J: Adult minor trauma patients: Good outcome in small hospitals. Ann Emerg Mad 1 g92; 21:402-406.

5. Gormican SP: CBAMS Scale: Field triage of trauma victims. Ann Emerg Mad 1982;11:132-135.

6. Clemmer TP, Orme JF, Thomas F, et al: Prospective evaluation of the CRAMS scale for triaging major trauma. J Trauma 1985;25:188-191.

7. Kane K, Engelhardt R, Celentrano J, et al: Empirical development and evaluation of prehospital trauma triage instruments. J Trauma 1985;25:482-488.

8. Ornate J, Mlinek EJ Jr, Craren E J, ot at: Ineffectiveness of the trauma score and the CRAMS scale for accurately triaging patients to trauma centers. Ann Emerg Mad 1985;14:1061~1064.

9. Long WB, Bachulis BL, Hynes GO: Accuracy and relationship of mechanisms of injury, trauma score, and injury severity score in identifying major trauma. Am J Surg 1988;151:581-584.

10. KeeNer J J, Beer L J, Malafa SA, et al: Prehospital index: A scoring system far field triage trauma victims. Ann Emerg Mad 1986:15:178-182.

11. Koehler J J, Malafa SA, Hilies]and J, et el: Prehospital index: A multicenter trial. Ann Emerg Mad 1986;15:625~626.

12. West JG, Murdock MA, Baldwin LC, et al: A method for evaluating field triage criteria. J Trauma 1986;26:855-659.

13. Rhodes M, Perline R, Aronson J, et al: Field triage for on-scone helicopter transport. J Trauma 1986;26:963-968.

14. Morris JA Jr, Auerbach PS, Marshall GA, et al: The trauma score as a triage tool in the pre~ hospital setting. JAMA 1986;266:1319-1325.

15. Lowe DK, Oh GR, Noely KW, Peterson CG: Evaluation of injury mechanism as a criterion in trauma triage. Am J Surg 1986;152:6- 9.

16. Hedges JR, Feero S, Moore B, et el: Comparison of prehospital triage instruments in a semirural population. J Emerg Mad 1987;5:197-208.

17. Knapp R, Yangi A, Kallsen G, et al: Mechanism of injury and anatomic iniury as criteria for prehospital trauma triage. Ann Emorg Mad 1988;17:895-902.

18. Kreis DJ Jr, Fine EG, Gomez GA, et al: A prospective evaluation of field categorization of trauma patients. J Trauma 1988;28:995-999.

19. Baxt WG, Berry CC, Epperson MD, et ai: The failure of prehospital trauma prediction rules to cbssify trauma patients accurately. Ann Emerg Mad 1989;18:1-8.

20. Baxt WG, Jones G, Fertlage D: The trauma triage rule: A new, resource-based approach to the prehospital identification of major trauma victims. Ann Emerg Mad 1990;19:1401-1406.

21. Smith JB, Bartholomew M J: Trauma index revisited: A better triage tool. Cfit Care Mad 1990;18:174-180.

22. Emerman CL, Shade B, Kubincanek J: A comparison of EMT judgment and prehospital trauma triage instruments. J Trauma 1991;31:1369-1375.

23. Emerman CL, Shade B, Kubincanek J: Comparison of the Baxt rule with ether trauma triage rules. Ann Emerg Mad 1992;21:644.

24. Eastman AB, Lewis FR, Champion HR, Mattox KL: Regional trauma system design: critical concepts. Am J Surg 1987;154:79-87.

25, Hosmer DW Jr, Lemeshow S: Applied Logistic Regression. New York, John Wiley & Sons, 1989.

26. Chambers JM, Hustle T J: Statistical Models. Pacific Grove, California, Wadsworth, 1992.

27. Breiman L, Friedman JH, OIshen RA, etal: Classification and Regression tree& Pacific Grove, California, Wadsworth, 1984.

28. Hanley JA, McNeil B J: The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 1982;143:29-36.

29. Hanley JA, McNeil B J: A method of comparing the areas under receiver operating character- ]stic curves derived from the same cases. Radiology 1983;148:839-843.

30. Heft WS, Tinkoff GH, Lucko JF, et el: Impact of minimal injuries on a Levet I trauma center. J Trauma 1992;33:408-412.

31. Esposito T J, Offnor P J, Jurkovich G J, et al: Do prehospital trauma center triage criteria identify major trauma victims? Arch Surg 1995;130:171-178.

Reprint no. 47/1/67480 Address for reprints:

Jerris R Hedges, MD, MS

Department of Emergency Medicine

Oregon Health Sciences University

3181 Southwest Sam Jackson Park Road

UHN-52

Portland, Oregon 97201-3098

503-494-7008

Fax 503-494-7689

46 8 ANNALS OF EMERGENCY MEDICINE 26:4 OCTOBER 1995

paramedic injury severity perception can aid trauma triage

Documents