TRAUMA/ORIGINAL RESEARCH

Clinical Prediction Rules for Identifying Adults at Very Low Risk for Intra-abdominal Injuries After Blunt Trauma
James F. Holmes, MD, MPH David H. Wisner, MD John P. McGahan, MD William R. Mower, MD, PhD Nathan Kuppermann, MD, MPH
From the Departments of Emergency Medicine (Holmes, Kuppermann), Surgery (Wisner), Radiology (McGahan), and Pediatrics (Kuppermann), UC Davis School of Medicine, Sacramento, CA; and the Division of Emergency Medicine, UCLA School of Medicine, Los Angeles, CA (Mower).

Study objective: We derive and validate clinical prediction rules to identify adult patients at very low risk for intra-abdominal injuries after blunt torso trauma. Methods: We prospectively enrolled adult patients ( 18 years old) after blunt torso trauma for whom diagnostic testing for intra-abdominal injury was performed. In the derivation phase, we used binary recursive partitioning to create a rule to identify patients with intra-abdominal injury who were undergoing acute intervention (including therapeutic laparotomy or angiographic embolization) and a separate rule for identifying patients with any intraabdominal injury present. We considered only clinical variables readily available with acceptable interrater reliability. The prediction rules were then prospectively validated in a separate cohort of patients. Results: In the derivation phase, we enrolled 3,435 patients, including 311 (9.1%; 95% condence interval [CI] 8.1% to 10.1%) with intra-abdominal injury and 109 (35.0%; 95% CI 29.7% to 40.6%) with intra-abdominal injury requiring acute intervention. In the validation study, we enrolled 1,595 patients, including 143 (9.0%; 95% CI 7.6% to 10.5%) with intra-abdominal injury. The derived rule for patients with intra-abdominal injuries who were undergoing acute intervention consisted of hypotension, Glasgow Coma Scale (GCS) score less than 14, costal margin tenderness, abdominal tenderness, hematuria level greater than or equal to 25 red blood cells/high powered eld, and hematocrit level less than 30% and identied all 44 patients in the validation phase with intra-abdominal injury who were undergoing acute intervention (sensitivity 44/44, 100%; 95% CI 93.4% to 100%). The derived rule for the presence of any intra-abdominal injury consisted of GCS score less than 14, costal margin tenderness, abdominal tenderness, femur fracture, hematuria level greater than or equal to 25 red blood cells/high powered eld, hematocrit level less than 30%, and abnormal chest radiograph result (pneumothorax or rib fracture). In the validation phase, the rule for any intra-abdominal injury present had the following test performance: sensitivity 137 of 143 (95.8%; 95% CI 91.1% to 98.4%), specicity 434 of 1,452 (29.9%; 95% CI 27.5% to 32.3%), and negative predictive value 434 of 440 (98.6%; 95% CI 97.1% to 99.5%). Conclusion: These derived and validated clinical prediction rules can aid physicians in the evaluation of adult patients after blunt torso trauma. Patients without any of these variables are at very low risk for having intraabdominal injury, particularly intra-abdominal injury requiring acute intervention, and are unlikely to benet from abdominal computed tomography scanning. [Ann Emerg Med. 2009;54:575-584.]
Provide feedback on this article at the journals Web site, www.annemergmed.com. 0196-0644/$-see front matter Copyright 2009 by the American College of Emergency Physicians. doi:10.1016/j.annemergmed.2009.04.007

INTRODUCTION
Abdominal trauma is a leading cause of morbidity and mortality.1 Identifying patients with intra-abdominal injuries can be difcult because the abdominal examination does not reliably identify all patients with intra-abdominal injuries.2-6 With improvement in computed tomography (CT) technology, abdominal CT scanning has become the primary method of evaluating hemodynamically stable blunt trauma patients believed
Volume , . : October

to be at risk for intra-abdominal injury at both trauma and nontrauma centers.6-13 Despite the increase in use of abdominal CT scanning in blunt trauma, the indications for abdominal CT in this setting are not well dened.14 Routine abdominal CT scanning of all blunt trauma patients is expensive and impractical in most busy emergency departments (EDs). Furthermore, there is evidence that overuse of CT scanning exposes patients to unnecessary ionizing radiation, potentially leading to lethal malignancies.15-18
Annals of Emergency Medicine 575

Clinical Prediction Rules for Intra-abdominal Injuries

Holmes et al For study purposes, a denitive diagnostic test was dened as any of the following: abdominal CT scan, diagnostic peritoneal lavage, or laparoscopy/laparotomy.7,26,27 We excluded all patients with penetrating trauma, pregnant patients, those presenting in cardiopulmonary arrest, and those patients with blunt torso trauma who did not undergo a denitive diagnostic test. Data Collection and Processing Historical and physical examination ndings for each enrolled patient were recorded onto a standardized data collection form by the treating physician (resident or faculty physician) before knowledge of the results of the diagnostic test. These historical and physical examination ndings included the following: mechanism of injury; initial ED systolic blood pressure; Glasgow Coma Scale (GCS) score; thoracic, abdominal, ank, and pelvic bone examinations for tenderness; clinical evidence of alcohol intoxication; and presence of a distracting painful injury (as determined by the treating physician). The results of the initial chest and pelvic radiographs, as well as laboratory tests (hematocrit level and urinalysis), were also collected. Hypotension was considered present if the initial systolic blood pressure was less than 90 mm Hg. Plain chest radiograph results were considered abnormal if either a rib fracture or a pneumothorax was identied. Plain pelvis radiograph results were considered abnormal if a pelvic fracture was identied. Two physicians (at least 1 an ED faculty member) independently evaluated a convenience sample of 720 enrolled patients to assess interobserver agreement, as measured by the statistic.28 Outcome Measures The 2 outcome measures were patients with intra-abdominal injury who were undergoing acute intervention and patients with any intra-abdominal injury. Patients were considered to have an intra-abdominal injury if any injury was documented to the following structures: spleen, liver, gallbladder, pancreas, kidney, ureter, urinary bladder, gastrointestinal tract, or an intra-abdominal vascular structure. Patients with intraabdominal injuries were considered to have undergone an acute intervention for their intra-abdominal injuries if they underwent a therapeutic laparotomy or angiographic embolization of an injured abdominal organ/vessel. Neither a patient with a normal laparotomy result (ie, no intra-abdominal injury identied) nor a patient with an intra-abdominal injury who underwent a laparotomy that did not have a specic therapy for that intraabdominal injury was considered to have an intra-abdominal injury undergoing acute intervention. Outcome determination was made by an investigator masked to all ED clinical data. Primary Data Analysis We described the study population with simple descriptive statistics and 95% condence intervals (CIs). We then created 2 clinical prediction rules. One prediction rule was derived to
Volume , . : October

Editors Capsule Summary

What is already known on this topic Many trauma patients get abdominal computed tomography (CT) scans but do not have an injury. What question this study addressed Whether a clinical decision rule can be developed and validated to predict patients at low risk of abdominal injury. What this study adds to our knowledge Patients with none of the following are unlikely to have an abdominal injury: Glasgow Coma Scale score less than 14, abdominal or costal margin tenderness, hematuria, low hematocrit level, femur fracture, or abnormal chest radiograph result. How this might change clinical practice These criteria lend credence to those who prefer clinical judgment to very liberal CT scanning, with its radiation exposure.

Thus, appropriate selection of injured patients for abdominal CT scanning would provide more efcient, cost-effective, and safe patient care. Clinical prediction rules have previously been developed to assist clinicians in determining the need for radiographic evaluation of injured patients after head and cervical spine trauma.19-23 Previous studies suggest that a clinical prediction rule for abdominal imaging may be feasible because these studies have identied several clinical variables associated with an increased risk of intra-abdominal injury.3,24,25 Despite the apparent need for an instrument to assist clinicians with decisionmaking about abdominal CT scanning,14 a clinical prediction rule for identifying adult patients with blunt torso trauma who are at risk for intra-abdominal injury has not yet been developed. The objective of this study was to derive and validate clinical prediction rules to identify adult patients at very low risk for intra-abdominal injury after blunt torso trauma. We hypothesize that we could derive and validate clinical prediction rules that identify a group of patients at sufciently low risk for intra-abdominal injury that abdominal CT imaging could be obviated.

MATERIALS AND METHODS

We conducted a prospective observational cohort study at an urban Level I trauma center. The study was approved by the study sites institutional review board. We enrolled adult patients, 18 years of age or older, with blunt torso trauma who underwent a denitive diagnostic test to determine the presence or absence of intra-abdominal injury. 576 Annals of Emergency Medicine

Holmes et al identify patients with intra-abdominal injury who were undergoing acute intervention, and a second prediction rule was derived to identify patients with any intra-abdominal injury. We developed the clinical prediction rules by using binary recursive partitioning with Classication and Regression Trees (CART) software (Salford Systems, San Diego, CA). Binary recursive partitioning is a nonparametric technique that classies subjects according to clinical variables for the outcome of interest by using a treelike structure with prediction nodes.29 This analytic technique is considered preferable to multiple logistic regression when the goal is to develop a clinical prediction rule with high sensitivity.30 The structure of the tree allows the clinician to determine the risk of the patient according to the presence/absence of the clinical variables in the tree. In building the trees, we used the Ginni splitting function in CART. We set the misclassication costs for missing an intra-abdominal injury undergoing acute intervention at 500:1 (relative cost of 500 for misclassifying 1 patient with intraabdominal injury who was undergoing acute intervention to that of misclassifying 1 patient without intra-abdominal injury who was undergoing acute intervention). In the second prediction rule, we set the misclassication costs for missing any intra-abdominal injury at 100:1 (relative cost of 100 for misclassifying 1 patient with any intra-abdominal injury to that of misclassifying a patient without an intra-abdominal injury). The cost for intra-abdominal injury undergoing acute intervention (500:1) was set higher than the cost for any intraabdominal injury (100:1), following clinical sensibility about the greater need to identify those patients undergoing an intervention for their intra-abdominal injury as opposed to patients with intra-abdominal injuries that are simply observed without any specic therapy. In the derivation phase, predictor variables were selected for inclusion according to a priori determined associations with intra-abdominal injury3-5,10,24,25,31-39 and acceptable interrater reliability. The latter was dened as a statistic with a lower bound of the 95% CI greater than or equal to 0.40, which indicates at least moderate agreement between observers.28 In addition, we excluded any physical examination variables that were missing from the data collection more than 5% of the time. The variables considered in the prediction rules included the following: hypotension, costal margin tenderness, abdominal tenderness, abdominal seatbelt sign (a linear area of ecchymosis/abrasion over the abdomen because of the seatbelt), GCS score less than 14, clinical evidence of alcohol intoxication, distracting painful injury, initial hematocrit level less than 30%, hematuria greater than or equal to 25 red blood cells/high powered eld, abnormal chest radiograph result (rib fracture or pneumothorax), femur fracture, and pelvic fracture on initial pelvic radiograph.3-5,10,24,25,31-39 After derivation of the clinical prediction rules, we then validated the rules in a separate cohort of patients (validation phase). Patients in the validation phase were enrolled in the same manner as those in the derivation phase. In addition, we
Volume , . : October

Clinical Prediction Rules for Intra-abdominal Injuries queried the physicians providing care to these patients about their perceived risk of intra-abdominal injury after clinical evaluation and before abdominal CT. Physician-perceived risk of intra-abdominal injury was categorized as less than 1%, 1% to 5%, or greater than 5%. We calculated the sensitivity, specicity, and positive and negative predictive values with 95% CIs of the 2 prediction rules when applied to patients in the validation phase. Relative risk ratios were calculated for individual variables identied in the prediction rules. The a priori sample size for the derivation study was calculated as 299 patients with intra-abdominal injury, assuming a desired sensitivity of the rule for intra-abdominal injury of 100%, with a desired lower bound of the 95% CI of 99.0%. The sample size for the validation study was set at 100 patients with intra-abdominal injury, consistent with previous literature for validation of clinical prediction rules.40

RESULTS
From March 2002 to April 2004, we enrolled 3,435 patients into the derivation phase of the study. Of these 3,435 patients, 311 (9.1%; 95% CI 8.1% to 10.1%) were identied as having an intra-abdominal injury. Of the 311 patients with intraabdominal injuries, 109 (35.0%; 95% CI 29.7% to 40.6%) underwent an acute intervention (therapeutic laparotomy or angiographic embolization). Interventions included the following (mean intervention/patient 1.6): angiographic embolization 6, splenectomy/splenorrhaphy 40, Avitene/Surgifoam to bleeding livers/spleen 21, liver packing 13, nephrectomy 4, mesentery repair 22, gastrointestinal perforation resection 21, gastrointestinal serosal repair 19, gastrointestinal perforation repair 10, pancreas drainage/resection 9, and other 12. Of the 3,435 patients enrolled, 1,546 (45%) had no abdominal tenderness and a GCS score greater than or equal to 14, including 72 (4.7%) patients with intra-abdominal injury. We enrolled 81% of eligible patients. Of the 806 eligible patients not enrolled, 69 (8.6%; 95% CI 6.7% to 10.7%) had intra-abdominal injuries. From May 2004 to April 2005, 1,595 patients were enrolled into the validation phase of the study. Of these 1,595 patients, 143 (9.0%; 95% CI 7.6% to 10.5%) were identied as having intra-abdominal injuries. Of the 143 patients with intraabdominal injuries, 44 (31%; 95% CI 23% to 39%) underwent acute interventions. The characteristics of the patients in both the derivation and validation phases of the study are presented in Table 1. The results of the interrater reliability for the clinical variables evaluated for possible consideration for entry into the prediction rule are presented in Table 2. Most of these were considered for inclusion into the prediction rules; however, several, including ank tenderness, abdominal distention, and peritoneal irritation, were not considered further because of insufcient interrater reliability.
Annals of Emergency Medicine 577

Clinical Prediction Rules for Intra-abdominal Injuries

Table 1. Clinical characteristics of the entire study population categorized by derivation and validation phases.
Clinical Variable Mean age ( 1 SD) Age range, y Male sex (%) MVC (%) Fall (%) Assault (%) Auto vs pedestrian (%) Other mechanism (%) Abdominal CT performed (%) Diagnostic peritoneal lavage performed (%) Laparotomy performed (%)
MVC, Motor vehicle crash.

Holmes et al

Derivation Phase, n 3,435 39.6 16.8 18-101 2,138 (65) 1,759 (51) 466 (14) 407 (12) 286 (8) 517 (15) 3,381 (98.4) 57 (1.7) 130 (3.8)

Validation Phase, n 1,595 40.3 17.5 18-95 1,038 (65) 827 (52) 194 (12) 147 (9) 136 (9) 291 (18) 1,566 (98.2) 9 (0.6) 42 (2.6)

Table 2. Interobserver agreement on the presence or absence of potential predictor variables for the prediction rule (n 720 patients).
Predictor Variable GCS score 14 Intoxication Abdominal tenderness Costal margin tenderness Abdominal seatbelt sign Distracting painful injury Flank tenderness Abdominal distention Peritoneal irritation (95% CI) 0.83 (0.75-0.90) 0.69 (0.62-0.76) 0.67 (0.60-0.74) 0.66 (0.59-0.74) 0.49 (0.42-0.57) 0.49 (0.41-0.56) 0.43 (0.36-0.50)* 0.25 (0.18-0.31)* 0.24 (0.17-0.31)*

* Measurements of these variables considered insufcient for consideration for inclusion in the clinical prediction rules.

None of the physical examination variables were missing in more than 5% of the cases (all eligible variables missing 3%). The derived clinical prediction rule (Figure 1) for intraabdominal injury undergoing acute intervention consisted of the following 6 variables: hypotension (systolic blood pressure 90 mm Hg), GCS score less than 14, costal margin tenderness, abdominal tenderness, hematuria level greater than or equal to 25 red blood cells/high powered eld, and hematocrit level less than 30%. The derived clinical prediction rule (Figure 2) to identify patients with intra-abdominal injury consisted of the following 7 variables: GCS score less than 14, costal margin tenderness, abdominal tenderness, femur fracture, hematuria level greater than or equal to 25 red blood cells/high powered eld, hematocrit level less than 30%, and abnormal chest radiograph result (rib fracture or pneumothorax). The relative risk ratios for all variables identied in either rule are presented in Table 3. The test performance of the clinical prediction rule for identifying patients with intra-abdominal injuries undergoing acute intervention in the validation phase is presented in Table 4. The specicity was such that if the clinical prediction rule was applied and abdominal CT scans were not obtained for those patients with negative prediction rule results, nearly one third of the abdominal CT scans in this study would be avoided. The 578 Annals of Emergency Medicine

test performance of the clinical prediction rule for identifying patients with any intra-abdominal injury on the patients in the validation phase is presented in Table 5. Similarly, the prediction rule displayed excellent sensitivity and, if strictly applied, would result in a similar reduction in abdominal CT scanning. In addition, the prediction rules had similar test performance characteristics in the derivation sets. The characteristics of the 6 patients with intra-abdominal injury in the derivation phase and the 6 patients with intraabdominal injury in the validation phase who were not identied by either prediction rule are presented in Table 6. None of the patients who had negative results for both prediction rules underwent acute intervention for their intraabdominal injury. All were observed in the hospital and then subsequently discharged safely to home. In the validation phase, we queried the physicians providing care to these patients about their perceived risk of intraabdominal injury after clinical evaluation and before CT. Data were available for 1,453 of 1,595 (91%) patients, including 135 of those with intra-abdominal injury. In 8 (5.9%) of 135 patients with intra-abdominal injury, physicians assessed the patient to have a less than 1% risk of intra-abdominal injury, and 31 of 135 (23.0%) were considered to have a 1% to 5% risk of intra-abdominal injury. Assuming a physician-estimated risk of less than 1% as negative for intra-abdominal injury and a physician risk estimate greater than or equal to 1% as positive for intra-abdominal injury, physician estimation had the following test characteristics: sensitivity 127 of 135 (94%; 95% CI 89% to 97%), specicity 365 of 1,318 (28%; 95% CI 25% to 30%), positive predictive value 127 of 1,080 (12%; 95% CI 9.9% to 13.8%), and negative predictive value 373 of 385 (98%; 95% CI 96% to 99%). One patient considered at less than 1% risk by physician suspicion for intra-abdominal injury underwent a therapeutic laparotomy for a gastrointestinal injury.

LIMITATIONS
We did not evaluate all potentially clinically important variables in this study. For example, we did not assess the utility of liver function tests41 because these are not routinely obtained
Volume , . : October

Holmes et al

Clinical Prediction Rules for Intra-abdominal Injuries

Figure 1. The decision rule for IAI undergoing acute intervention in the derivation phase. Each box represents the number of patients with and without IAI who were undergoing acute therapy, given the particular nding. IAI, Intra-abdominal injury; rbc/hpf, Red blood cells/high powered eld; HCT, head computed tomography.

in the evaluation of adult trauma patients. It is possible that several of the patients with hepatic injuries missed by the prediction rules would have been identied by these laboratory tests. We did not enroll all eligible patients. The rate of intraabdominal injury was similar, but we did not capture further data on missed eligible patients for comparison. The study was primarily designed to identify a population of adult trauma patients in whom abdominal CT scanning would be of very low utility. Therefore, we studied only patients who underwent a denitive diagnostic study and therefore cannot comment on patients who did not receive denitive diagnostic testing. These prediction rules should be viewed as assistive and not directive for abdominal CT scanning of all patients who have one of the high-risk variables present.42 The implications and risks associated with the presence of only 1 of the variables in the rules is different from the risk associated with a different variable or combinations of variables in the rule. Furthermore,
Volume , . : October

we excluded variables with unacceptably low measurements. Inclusion of these variables may have resulted in a different rule, but inclusion of variables with unacceptable measurements violates principles of decision rule development.30 The measurements include an assessment by a resident physician and a faculty physician. The reliability of the resident examination may not be equal to that of a faculty emergency physician. The prevalence of intra-abdominal injury in this study was similar to that of previous studies (including multicenter studies), suggesting similar rates of CT utilization among US trauma centers.3,43 As with previous studies, we considered abdominal CT scanning to be a reference standard test for the identication of intra-abdominal injury, although it is likely that the CT scan is not a perfectly accurate diagnostic test. This may have resulted in some patients without intra-abdominal injury being misclassied as having intra-abdominal injury because of a falsely positive abdominal CT scan result. If the
Annals of Emergency Medicine 579

Clinical Prediction Rules for Intra-abdominal Injuries

Holmes et al

Figure 2. The decision rule for any IAI in the derivation phase. Each box represents the number of patients with and without IAI, given the particular nding. CXR, Chest radiograph.

Table 3. Relative risk ratios for the clinical variables identied in prediction rule and intra-abdominal injury (derivation phase).
Clinical Variable Systolic blood pressure 90 mm Hg GCS score 14 Costal margin tenderness Abdominal tenderness Femur fracture Hematuria level 25 red blood cells/high powered eld Hematocrit level 30% Abnormal chest radiograph result
*Diagnostic test not obtained for all subjects.

IAI Present 37/311 71/311 161/311 177/311 38/311 122/300* 36/311 87/311 *

IAI Absent 72/3,124 363/3,124 1,100/3,124 1,326/3,124 130/3,124 270/2,729* 95/3,124 224/3,069*

Relative Risk (95% CI) 4.1 (3.1-5.5) 2.0 (1.6-2.6) 1.9 (1.5-2.3) 1.7 (1.4-2.1) 2.7 (2.0-3.7) 5.1 (4.1-6.2) 3.3 (2.4-4.5) 3.8 (3.1-4.8)

580 Annals of Emergency Medicine

Volume , . : October

Holmes et al

Clinical Prediction Rules for Intra-abdominal Injuries

Table 4. Test performance of the clinical prediction rule to identify patients with intra-abdominal injury who were undergoing acute intervention in the derivation and validation phases.
Derivation Phase (95% CI) Sensitivity Specicity Positive predictive value Negative predictive value 109/109 824/3,326 109/2,611 824/824 100% (97.2, 100%) 24.8% (23.3, 26.3%) 4.2% (3.4, 5.0%) 100% (99.6, 100%) Valiation Phase (95% CI) 44/44 476/1,551 44/1,119 476/476 100% (93.4, 100%) 30.7% (28.4, 33.1%) 3.9% (2.9, 5.2%) 100% (99.4, 100%)

The clinical prediction rule for IAI undergoing acute intervention consists of hypotension, GCS score level greater than or equal to 25 red blood cells/high powered eld, and hematocrit level 30%.

14, costal margin tenderness, abdominal tenderness, hematuria

Table 5. Test performance of the clinical prediction rule to identify patients with any intra-abdominal injury in the derivation and validation phases.
Derivation Phase (95% CI) Sensitivity Specicity Positive predictive value Negative predictive value 305/311 818/3,124 305/2,611 818/824 98.1% (95.8, 99.3%) 26.2% (24.6, 27.8%) 11.7% (10.5, 13.0%) 99.3% (98.4, 99.7%) Validation Phase (95% CI) 137/143 434/1,452 137/1,115 434/440 95.8% (91.1, 98.4%) 29.9% (27.5, 32.3%) 11.9% (10.1, 13.9%) 98.6% (97.1, 99.5%)

The prediction rule for any IAI consists of GCS score 14, costal margin tenderness, abdominal tenderness, femur fracture, hematuria level greater than or equal to 25 red blood cells/high powered eld, hematocrit level 30%, and abnormal chest radiograph result (pneumothorax, rib fracture).

Table 6. Characteristics of patients with intra-abdominal injuries in the derivation and validation phases not identied by either of the clinical prediction rules.
Age, y 20 20 64 35 18 54 35 26 30 53 55 51 Mechanism High speed MVC Assault with club 9-Foot fall 30-Foot fall MVC, low speed MVC MVC, low speed MVC MVC, rollover 8-Foot fall MVC, rollover Fall Injury Spleen Kidney Liver Liver Liver Liver Liver Spleen Liver Spleen/kidney Spleen Kidney Hemo* Yes No No No No No No No Yes No Yes No Other Clinical Findings L1 burst fracture Intoxicated, abdominal abrasions T12 burst fracture L1 burst fracture Back/ank pain Gluteal hematoma Tibial fracture Concussion, GCS score 15 T8 Fx, L1, L2 transverse process Fx Intracranial hemorrhage, GCS score 15 Rib fractures, L1 transverse process Fx L1, L2 transverse process Fx Hospital Course Observation Observation Observation Observation Observation Observation Observation Observation Observation Observation Observation Observation

Fx, Fracture. *Hemoperitoneum identied by abdominal CT or at laparotomy.

abdominal CT scan has a theoretical specicity of 99%, we would expect 1% of patients in this study (ie, 50 patients) to be misclassied as having an intra-abdominal injury because of false-positive abdominal CT scan results. It is therefore possible that several of the patients categorized with intra-abdominal injury according to CT, but not identied by the prediction rules, did not truly have an intra-abdominal injury. Because the prediction rules do not identify all patients with intraabdominal injury, strict application would result in some patients with intra-abdominal injury not being identied. These patients missed by the decision rules were all observed and none underwent specic therapy. The implications of missing a few injuries that do not require specic therapy is unclear. However, some may argue that counseling patients with minor intra-abdominal injuries about return to sports activity and other contact activity would be important.
Volume , . : October

Finally, we did not include bedside abdominal ultrasonography for consideration in the creation of this prediction rule. Abdominal ultrasonography is used with increasing frequency in the evaluation of patients with blunt trauma. However, its use in the United States has been primarily restricted to trauma centers, and it is not routinely available or used in all EDs.12 Creating a prediction rule in which a technology is not available to all physicians evaluating trauma patients would limit its applicability.42 We did, however, model a prediction rule for patients with intraabdominal injury when the ultrasonographic result was included as a predictor variable. This analysis resulted in a prediction rule that included the following variables: positive ultrasonographic examination result, hypotension, GCS score less than 14, costal margin tenderness, abdominal tenderness, and hematuria level greater than or equal to 25 red blood cells/high powered eld.
Annals of Emergency Medicine 581

Clinical Prediction Rules for Intra-abdominal Injuries Thus, the variables included in the models are similar, regardless of the inclusion of abdominal ultrasonographic results in the model.

Holmes et al scanning in 12% of their studied population. That study, however, did not collect urine data, relied heavily on laboratory testing, and used logistic regression to identify risk factors associated with intra-abdominal injury, all of which may have impaired the utility and clinical feasibility of applying the nal rule in practice.22,30 A prospective study conducted by Salim et al6 suggested that some adults with blunt torso trauma but without apparent clinical ndings of intra-abdominal injury may, in fact, have intra-abdominal injuries. Among the 1,000 patients in that study considered not to have obvious signs of injury, 49 (4.9%) nevertheless had intra-abdominal injuries. Twenty-one (2.1%) of these patients, however, had depressed levels of consciousness, making the abdominal examination results unreliable. In the current study, the clinical variable GCS score less than 14 applied to the Salim et al6 study population would have identied these 21 patients. Of the remaining 28 patients with intra-abdominal injuries but with normal mental status and without obvious signs of injury in the study, how many had other ndings that the current study would suggest are predictive of intra-abdominal injury (costal margin tenderness, femur fracture, hematuria, abnormal initial chest radiograph result, etc) was not described. Furthermore, had we implemented a denition of without obvious signs of injury similar to that in the Salim et al6 study, 4.7% of our patients with intra-abdominal injuries in the derivation phase would be considered to be without signs of injury. Several other studies have focused on special populations with specic clinical characteristics in attempts to determine the appropriate use of CT scanning after blunt abdominal trauma. These studies have identied a variety of associations with intraabdominal injury, including the importance of the abdominal seatbelt sign,38,53 costal margin injury,34,37 femur fractures,31 and distracting injuries.33 In our study, we considered distracting painful injury separately from femur fractures and thoracic injuries as candidate variables. Our subsequent recursive partitioning analysis did not nd the variable distracting painful injuries useful in indentifying patients with intra-abdominal injuries but did nd that femur fractures and thoracic injuries were very important predictors. The most important variable in our prediction rule, located at the top of the prediction tree (Figures 1 and 2) and having the highest relative risk for intra-abdominal injury, however, was the presence of signicant hematuria. Although hematuria, including microscopic hematuria, is a recognized independent risk factor for intra-abdominal injury in children,35,54 this variable is more controversial in the evaluation of adult trauma patients. Many ED clinicians and trauma surgeons simply use the urinalysis as a screen for genitourinary trauma and frequently do not consider the urinalysis as a screen for other intra-abdominal injuries. However, given the close proximity of other intra-abdominal organs to the kidneys, it is not surprising that hematuria is a marker for other intra-abdominal injuries. In fact, one previous study has suggested that microscopic
Volume , . : October

DISCUSSION
In this study, we derived and validated clinical prediction rules with high sensitivity and negative predictive value for identifying adult patients with and without intra-abdominal injury after blunt torso trauma. The clinical variables in these rules are routinely collected as part of the assessment of patients with blunt torso trauma who present to the ED and are thus readily available when the decision to perform abdominal CT scanning is considered. In addition, these variables have acceptable levels of interobserver agreement and thus can be reliably assessed by emergency physicians. These clinical prediction rules identied nearly all patients with intraabdominal injuries and, more important, identied all patients in both the derivation and validation phases with intraabdominal injuries who were undergoing acute intervention. These data provide scientic evidence to assist clinicians in their decisions about CT imaging of the abdomen in patients with blunt torso trauma. Initial identication of patients at risk for intra-abdominal injury may be difcult, and delayed diagnosis is associated with increased morbidity and mortality.44-46 Most adult patients with blunt torso trauma, however, do not have intra-abdominal injuries; therefore, routine imaging of all patients is not warranted because there are disadvantages to excessive abdominal CT utilization.18 Furthermore, abdominal CT scanning is a source of substantial radiation risk to the trauma patient.47 Reducing abdominal CT use would have several potential benets, including decreasing patient ED length of stay, patient care costs, and associated risks (radiation-induced malignancy, aspiration of oral contrast, intravenous contrast reactions, and contrast-induced nephropathy).16,48-50 If the clinical prediction rules derived and validated in this study were strictly applied to the studied patient population, nearly one third of abdominal CT scans in this study would have been obviated. Previous research studying indications for abdominal CT scanning after blunt torso trauma has identied multiple variables associated with the presence of intra-abdominal injury but has had methodological limitations, including the use of inadequate analytic techniques or retrospective data.3,24,25 In the current study, we followed standard and accepted methodological recommendations for deriving and validating clinical prediction rules.30,51,52 The single largest prospective study previously conducted to identify those blunt trauma patients not requiring abdominal CT suggested that patients with GCS scores greater than 13, normal abdominal ultrasonographic examination results, normal chest and pelvic radiograph results, and normal laboratory testing results (hematocrit level, WBC count, and lactate, amylase, and serum glutamic oxaloacetic transaminase levels) do not require abdominal CT imaging.3 Application of these ndings would have obviated the need for abdominal CT 582 Annals of Emergency Medicine

Holmes et al hematuria is a very important predictor of intra-abdominal injury in adults and is useful as a screening tool.10 Another study suggests hematuria is an important predictor of intraabdominal injury in those adult trauma patients with normal abdominal ultrasonographic examination results.55 We created prediction rules for patients with intraabdominal injuries undergoing acute intervention and for patients with any intra-abdominal injuries. Not surprisingly, the 2 rules are similar. However, hypotension was identied as a risk factor for patients with intra-abdominal injuries who were undergoing acute intervention but not for patients with any intra-abdominal injuries. This is not surprising because it is intuitive that hypotension after blunt torso trauma would be a marker for high-severity injury. We would advocate considering patients with any of the variables identied by either of these 2 rules to be at nonnegligible risk for intra-abdominal injury and therefore that appropriate evaluation (abdominal CT, ultrasonography, or serial abdominal examinations) be implemented according to the clinical scenario and resources available. To assist with the decision of how aggressively to evaluate for intra-abdominal injury, a patients particular risk for intra-abdominal injury, as well as intra-abdominal injury undergoing acute intervention, can be determined by reviewing the decision trees. In conclusion, clinical prediction rules consisting of hypotension, GCS score less than 14, costal margin tenderness, abdominal tenderness, femur fracture, hematuria level greater than or equal to 25 red blood cells/high powered eld, hematocrit level less than 30%, and abnormal chest radiograph result (rib fracture or pneumothorax) can aid physicians in the evaluation of adult patients with blunt torso trauma. Patients without any of the high-risk variables are at very low risk of intra-abdominal injury and are unlikely to benet from abdominal CT scanning.
Supervising editor: Judd E. Hollander, MD Author contributions: JFH was responsible for the study concept and supervision, analyzing and interpreting the data, and drafting the article. JFH, WRM, and NK were responsible for the study design. JFH, DHW, and JPM acquired the data. All authors were responsible for critical revision of the article for important intellectual content. JFH and NK provided statistical expertise and obtained funding. JH takes responsibility for the paper as a whole. Funding and support: By Annals policy, all authors are required to disclose any and all commercial, nancial, and other relationships in any way related to the subject of this article that might create any potential conict of interest. See the Manuscript Submission Agreement in this issue for examples of specic conicts covered by this statement. Funded in part by the Society for Academic Emergency Medicine Research Training Grant. Publication dates: Received for publication September 8, 2008. Revisions received January 8, 2009, and March 28, Volume , . : October

Clinical Prediction Rules for Intra-abdominal Injuries

2009. Accepted for publication April 15, 2009. Available online May 19, 2009. Presented at the SAEM annual meeting, Chicago, IL, May 2007. Address for reprints: James F. Holmes, MD, MPH, UC Davis Medical Center, Department of Emergency Medicine, 2315 Stockton Blvd, PSSB 2100, Sacramento, CA 95817-2282; 916-734-1533, fax 916-734-7950; E-mail jfholmes@ucdavis.edu.

REFERENCES
1. Minino AM, Heron MP, Murphy SL, et al. Deaths: nal data for 2004. Natl Vital Stat Rep. 2007;55:1-119. 2. Perez FG, OMalley KF, Ross SE. Evaluation of the abdomen in intoxicated patients: is computed tomography scan or peritoneal lavage always indicated? Ann Emerg Med. 1991;20:500-502. 3. Poletti PA, Mirvis SE, Shanmuganathan K, et al. Blunt abdominal trauma patients: can organ injury be excluded without performing computed tomography? J Trauma. 2004;57:1072-1081. 4. Rodriguez A, DuPriest RW Jr, Shatney CH. Recognition of intraabdominal injury in blunt trauma victims. A prospective study comparing physical examination with peritoneal lavage. Am Surg. 1982;48:457-459. 5. Schurink GW, Bode PJ, van Luijt PA, et al. The value of physical examination in the diagnosis of patients with blunt abdominal trauma: a retrospective study. Injury. 1997;28:261-265. 6. Salim A, Sangthong B, Martin M, et al. Whole body imaging in blunt multisystem trauma patients without obvious signs of injury: results of a prospective study. Arch Surg. 2006;141:468-473; discussion 473-475. 7. Grifn XL, Pullinger R. Are diagnostic peritoneal lavage or focused abdominal sonography for trauma safe screening investigations for hemodynamically stable patients after blunt abdominal trauma? A review of the literature. J Trauma. 2007;62:779-784. 8. Heilbrun ME, Chew FS, Tansavatdi KR, et al. The role of negative CT of the abdomen and pelvis in the decision to admit adults from the emergency department after blunt trauma. J Am Coll Radiol. 2005;2:889-895. 9. Holmes JF, Harris D, Battistella FD. Performance of abdominal ultrasonography in blunt trauma patients with out-of-hospital or emergency department hypotension. Ann Emerg Med. 2004;43: 354-361. 10. Richards JR, Derlet RW. Computed tomography and blunt abdominal injury: patient selection based on examination, haematocrit and haematuria. Injury. 1997;28:181-185. 11. Stuhlfaut JW, Anderson SW, Soto JA. Blunt abdominal trauma: current imaging techniques and CT ndings in patients with solid organ, bowel, and mesenteric injury. Semin Ultrasound CT MR. 2007;28:115-129. 12. Jhirad R, Boone D. Computed tomography for evaluating blunt abdominal trauma in the low-volume nondesignated trauma center: the procedure of choice? J Trauma. 1998;45:64-68. 13. Miller LA, Shanmuganathan K. Multidetector CT evaluation of abdominal trauma. Radiol Clin North Am. 2005;43:1079-1095, viii. 14. Garber BG, Bigelow E, Yelle JD, et al. Use of abdominal computed tomography in blunt trauma: do we scan too much? Can J Surg. 2000;43:16-21. 15. Brenner D, Elliston C, Hall E, et al. Estimated risks of radiationinduced fatal cancer from pediatric CT. AJR Am J Roentgenol. 2001;176:289-296.

Annals of Emergency Medicine 583

Clinical Prediction Rules for Intra-abdominal Injuries

16. Brenner DJ, Elliston CD. Estimated radiation risks potentially associated with full-body CT screening. Radiology. 2004;232:735738. 17. Prokop M. Cancer screening with CT: dose controversy. Eur Radiol. 2005;15(suppl 4):D55-61. 18. Brenner DJ, Hall EJ. Computed tomographyan increasing source of radiation exposure. N Engl J Med. 2007;357:2277-2284. 19. Haydel MJ, Preston CA, Mills TJ, et al. Indications for computed tomography in patients with minor head injury. N Engl J Med. 2000;343:100-105. 20. Hoffman JR, Mower WR, Wolfson AB, et al. Validity of a set of clinical criteria to rule out injury to the cervical spine in patients with blunt trauma. National Emergency X-Radiography Utilization Study Group. N Engl J Med. 2000;343:94-99. 21. Mower WR, Hoffman JR, Herbert M, et al. Developing a decision instrument to guide computed tomographic imaging of blunt head injury patients. J Trauma. 2005;59:954-959. 22. Stiell IG, Wells GA, Vandemheen K, et al. The Canadian CT Head Rule for patients with minor head injury. Lancet. 2001;357:13911396. 23. Stiell IG, Wells GA, Vandemheen KL, et al. The Canadian C-spine rule for radiography in alert and stable trauma patients. JAMA. 2001;286:1841-1848. 24. Grieshop NA, Jacobson LE, Gomez GA, et al. Selective use of computed tomography and diagnostic peritoneal lavage in blunt abdominal trauma. J Trauma. 1995;38:727-731. 25. Mackersie RC, Tiwary AD, Shackford SR, et al. Intra-abdominal injury following blunt trauma. Identifying the high-risk patient using objective risk factors. Arch Surg. 1989;124:809-813. 26. Catre MG. Diagnostic peritoneal lavage versus abdominal computed tomography in blunt abdominal trauma: a review of prospective studies. Can J Surg. 1995;38:117-122. 27. Mele TS, Stewart K, Marokus B, et al. Evaluation of a diagnostic protocol using screening diagnostic peritoneal lavage with selective use of abdominal computed tomography in blunt abdominal trauma. J Trauma. 1999;46:847-852. 28. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics. 1977;33:159-174. 29. Brieman L, Friedman JH, Olshen RA, et al. Classication and Regression Trees. Washington, DC: Chapman & Hall; 1984. 30. Stiell IG, Wells GA. Methodologic standards for the development of clinical decision rules in emergency medicine. Ann Emerg Med. 1999;33:437-447. 31. Adili A, Bhandari M, Lachowski RJ, et al. Organ injuries associated with femoral fractures: implications for severity of injury in motor vehicle collisions. J Trauma. 1999;46:386-391. 32. Demetriades D, Karaiskakis M, Toutouzas K, et al. Pelvic fractures: epidemiology and predictors of associated abdominal injuries and outcomes. J Am Coll Surg. 2002;195:1-10. 33. Ferrera PC, Verdile VP, Barteld JM, et al. Injuries distracting from intraabdominal injuries after blunt trauma. Am J Emerg Med. 1998;16:145-149. 34. Holmes JF, Ngyuen H, Jacoby RC, et al. Do all patients with left costal margin injuries require radiographic evaluation for intraabdominal injury? Ann Emerg Med. 2005;46:232-236. 35. Holmes JF, Sokolove PE, Brant WE, et al. Identication of children with intra-abdominal injuries after blunt trauma. Ann Emerg Med. 2002;39:500-509. 36. Reiff DA, McGwin G Jr, Metzger J, et al. Identifying injuries and motor vehicle collision characteristics that together are suggestive of diaphragmatic rupture. J Trauma. 2002;53:11391145. 37. Shweiki E, Klena J, Wood GC, et al. Assessing the true risk of abdominal solid organ injury in hospitalized rib fracture patients. J Trauma. 2001;50:684-688.

Holmes et al
38. Velmahos GC, Tatevossian R, Demetriades D. The seat belt mark sign: a call for increased vigilance among physicians treating victims of motor vehicle accidents. Am Surg. 1999;65: 181-185. 39. Wu SR, Shakibai S, McGahan JP, et al. Combined head and abdominal computed tomography for blunt trauma: which patients with minor head trauma benet most? Emerg Radiol. 2006;13: 61-67. 40. Vergouwe Y, Steyerberg EW, Eijkemans MJ, et al. Substantial effective sample sizes were required for external validation studies of predictive logistic regression models. J Clin Epidemiol. 2005;58:475-483. 41. Sahdev P, Garramone RR Jr, Schwartz RJ, et al. Evaluation of liver function tests in screening for intra-abdominal injuries. Ann Emerg Med. 1991;20:838-841. 42. Reilly BM, Evans AT. Translating clinical research into clinical practice: impact of using prediction rules to make decisions. Ann Intern Med. 2006;144:201-209. 43. Livingston DH, Lavery RF, Passannante MR, et al. Admission or observation is not necessary after a negative abdominal computed tomographic scan in patients with suspected blunt abdominal trauma: results of a prospective, multi-institutional trial. J Trauma. 1998;44:273-280; discussion 280-282. 44. Fakhry SM, Brownstein M, Watts DD, et al. Relatively short diagnostic delays ( 8 hours) produce morbidity and mortality in blunt small bowel injury: an analysis of time to operative intervention in 198 patients from a multicenter experience. J Trauma. 2000;48:408-414; discussion 414-415. 45. Miller PR, Croce MA, Bee TK, et al. Associated injuries in blunt solid organ trauma: implications for missed injury in nonoperative management. J Trauma. 2002;53:238-242; discussion 242-244. 46. Clarke JR, Trooskin SZ, Doshi PJ, et al. Time to laparotomy for intra-abdominal bleeding from trauma does affect survival for delays up to 90 minutes. J Trauma. 2002;52:420-425. 47. Winslow JE, Hinshaw JW, Hughes MJ, et al. Quantitative assessment of diagnostic radiation doses in adult blunt trauma patients. Ann Emerg Med. 2008;52:93-97. 48. Brenner DJ. Estimating cancer risks from pediatric CT: going from the qualitative to the quantitative. Pediatr Radiol. 2002;32:228223; discussion 242-244. 49. Cochran ST, Bomyea K, Sayre JW. Trends in adverse events after IV administration of contrast media. AJR Am J Roentgenol. 2001; 176:1385-1388. 50. Parfrey PS, Grifths SM, Barrett BJ, et al. Contrast materialinduced renal failure in patients with diabetes mellitus, renal insufciency, or both. A prospective controlled study. N Engl J Med. 1989;320:143-149. 51. Laupacis A, Sekar N, Stiell IG. Clinical prediction rules. A review and suggested modications of methodological standards. JAMA. 1997;277:488-494. 52. McGinn TG, Guyatt GH, Wyer PC, et al. Users guides to the medical literature: XXII: how to use articles about clinical decision rules. Evidence-Based Medicine Working Group. JAMA. 2000;284: 79-84. 53. Porter RS, Zhao N. Patterns of injury in belted and unbelted individuals presenting to a trauma center after motor vehicle crash: seat belt syndrome revisited. Ann Emerg Med. 1998;32: 418-424. 54. Isaacman DJ, Scarfone RJ, Kost SI, et al. Utility of routine laboratory testing for detecting intra-abdominal injury in the pediatric trauma patient. Pediatrics. 1993;92:691-694. 55. Sirlin CB, Brown MA, Andrade-Barreto OA, et al. Blunt abdominal trauma: clinical value of negative screening US scans. Radiology. 2004;230:661-668.

584 Annals of Emergency Medicine

Volume , . : October