TheJournal of Emergency Medicme, Vol 10, pp 553-558. 1992

Printed in the USA Copyright 0 1992 Pergamon Press Ltd.

OBJECTIVE DETERMINATION OF THE OPTIMAL RED BLOOD CELL COUNT IN DIAGNOSTIC PERITONEAL LAVAGE DONE FOR ABDOMINAL STAB WOUNDS Michael J. Zappa, MD,* Ann L. Hatwood-Nuss, MD, FACEP,* Robert L. Wears, MD, FACEP, William F. Fallon, MD, FACS* *Divisionof Emergency

Medicine, *Trauma Service, Department of Surgery, University of Florida, Health Science Center-Jacksonville Reprint Address: Ann L. Harwood-Nuss, MD, FACEP, Division of Emergency Medicine, University of Florida Health Science Center-Jacksonville, 655 West 8th Street, Jacksonville, FL 32209

0 Abstract-The purpose of this study was to determine objectively the optimal value or positivity criterion for red blood cell counts in diagnostic peritoneal lavage in stab wounds to the anterior abdomen. Our study group consisted of 91 consecutive adults with abdominal stab wounds who underwent peritoneal lavage. We excluded those patients who met criteria for immediate laparotomy and those with negative stab wound exploration. We divided the patients into two groups based on outcome. Group 1 consisted of those who had undergone laparotomy and had findings that required surgical intervention. Group 2 patients had either undergone laparotomy but had no injury requiring surgical intervention or had no surgery and a benign hospital course and follow-up. Receiver operator characteristic analysis was done on the diagnostic peritoneal lavage RRC counts for both groups. The overlap between the groups was minimal, with 75% of patients in Group 1 having > 120,000 RRC/mm” and 75% of patients in Group 2 having < 486 RRC/mm3 in the lavage effluent. Using the observed probability of 23.1% of patients with abdominal stab wounds requiring surgery, a RRC count of 50,000/mm3 discriminated best those patients who required surgery from those who did not.

years. Routine exploration for stab wounds to the abdomen is associated with a significant negative laparotomy rate and has been largely abandoned (l5,8-13,17-19,21-23,25,30-32,35). The negative laparotomy rate occurs because stab wounds only penetrate the peritoneum in about two-thirds of cases, with less than half resulting in significant visceral damage (19,23,31,32). Selective management for abdominal stab wounds has been adopted in many centers, but does not yet enjoy uniform definition. Although some centers prefer serial examination and observation of the injured patient, many authorities advocate local wound exploration, followed by diagnostic peritoneal lavage (l-5,8-13,17-19,21-23,25, 30-32,35). Diagnostic peritoneal lavage offers busy trauma centers an advantage over serial examination, with comparable accuracy (3,7,12,19,23,30,31,32). Diagnostic peritoneal lavage (DPL) was originally described as a diagnostic test [for detecting] intraabdominal injury in blunt trauma (25). Criteria for positive diagnostic peritoneal lavage were developed initially for blunt trauma and are now well documented as providing an accurate measure of significant injury (1,2,10,12,18,32). In hemodynamically stable patients, selective management of abdominal stab wounds incorporating diagnostic peritoneal lavage is now widely employed and is responsible for a reduction in the negative laparotomy rate from 40% to 7% (l-3,7-12,15,19,23,30-32). However, unlike DPL for blunt trauma, there is lack of agreement regarding the criterion for positivity in the red blood cell count in penetrating trauma. There is a wide range of recommendations for red blood cell count cutoffs, from 1,000 to 100,000/mm3 (l-3,7-12,18,

0 Keywords - diagnostic peritoneal lavage; abdominal stab wound; penetrating abdominal trauma; receiver operating characteristic curve

INTRODUCTION

The management of abdominal stab wounds has undergone considerable refinement during the past 20 Presented at Society for Academic Emergency Medicine, 1990 Annual Meeting, May 21-24,199O.

RECEIVED: 17August1991; ACCEPTED: 9 March 1992

FINALSUBMISSIONRECEIVED: 23 January 1992; 553

0736-4679/92 $5.00 + .OO

M. J. Zappa, A. L. Harwood-Nuss, R. L. Wears, W. F. Fallon

554

19,23,30-32). In our study, we sought to determine the optimal RBC cutoff that identifies significant injury in penetrating abdominal trauma.

METHODS From January 1987 to December 1989, all patients with stab wounds to the anterior abdomen were identified from the trauma registry at the University of Florida Health Science Center-Jacksonville. The anterior abdomen was defined as that area inferior to the costal margin, anterior to the midaxillary line, and superior to the inguinal ligament. Those patients with abdominal stab wounds were evaluated in the trauma center by the trauma team, consisting of surgical and emergency medicine housestaff and attending physicians. The management protocol during the study period included either immediate laparotomy when indicated or selective management. At our institution, the single, absolute indication for immediate laparotomy was the presence of unstable vital signs during resuscitation. Evisceration or free intraperitoneal air were not absolute indications for immediate laparotomy unless there were other signs of serious injury. Those patients without indications for immediate laparotomy underwent local wound exploration. If there was no evidence of fascial penetration, the wound was irrigated and packed or loosely closed, and the patient was discharged. If the wound exploration identified anterior fascial penetration, peritoneal penetration, or was indeterminate, a peritoneal lavage was performed. A modified open technique for peritoneal lavage was used in all patients. A small, midline incision was made in the infraumbilical region and carried down to the rectus fascia. The fascia was incised and the muscle retracted to expose the posterior fascia. A dialysis catheter with an internal stylet was introduced through the fascia into the abdominal cavity. The stylet was withdrawn and the catheter attached to dialysis tubing and a lo-cc syringe. If the aspiration was negative, Ringer’s lactate (10 cc/kg; max 1 liter) was infused into the abdominal cavity. The lavage effluent was recovered through gravity drainage, and a sample was sent to the laboratory for red blood cell count. Information on the study group was extracted from the trauma registry, operative reports, hospital charts, discharge summaries, and medical record file review of subsequent outpatient visits or hospitalizations. During the study period, criteria for DPL positivity included 3-5 cc of gross blood on aspiration, a lavage effluent count of either 100,000 RBC/mm’ or 500 WBC/mm3. Other standard criteria included the presence of bile, bacte-

ria, intestinal contents, amylase of > 175 IU/lOO mL, or the return of lavage fluid from a chest tube or Foley catheter. All patients who underwent DPL were admitted to the hospital for at least 24 hours of observation. Our study group consisted of 91 adults with stab wounds to the anterior abdomen who received a diagnostic peritoneal lavage. Of the 91 patients who underwent peritoneal lavage, 27 (29.6%) had a positive lavage followed by laparotomy; 64 (70.3%) patients had a negative lavage, were admitted and observed. The operative reports were reviewed (W.F.) to determine the precise nature of the injuries found at the time of laparotomy. We considered only those injuries that required hemostasis, repair, or drainage to be significant. The presence of hemoperitoneum was not considered a true positive finding in the absence of the aforementioned criteria. Based on this definition, we found that 22% (6/27) of those who underwent laparotomy had injuries that did not require surgical intervention. The hospital course and outpatient follow-up for all patients in the study group were reviewed. The follow-up ranged from 3 months to 3 years. Seventy-three patients (80.2%) returned for at least one visit to the hospital. No adverse sequelae were noted. We classified the 91 patients into two groups for analysis: Group 1 (21) consisted of those who had findings that required surgical intervention at the time of the laparotomy. Group 2 (70) included (a) those patients whose injuries were surgically insignificant at the time of laparotomy and (b) those who did not have a laparotomy. A receiver operator characteristic (ROC) curve was generated for the RBC counts of both groups (see Appendix). Criteria such as WBC count, amylase, etc., were not subjected to ROC analysis in our study. The ROC curve permits the analysis of any test whose results yield an ordinal scale of values, such as the RBC counts in DPL (15). To construct an ROC curve for a diagnostic test, the true positive ratio (sensitivity) and false positive ratio (specificity) of the test is determined for each member of a set of possible values for the positivity criterion. In the current study, we determined the true positive (TP) and false positive (FP) ratios of the DPL RBC for the positivity criterion based on the most commonly used RBC criteria: 5,000 RBC/mm’; 10,000 RBC/ mm3; 20,000 RBC/mm3; 50,000 RBC/mm3; 75,000 RBC/mm3; 100,000/mm3; 200,000 RBC/mm3; and grossly positive (l-3,7-12,19,23,30-32). Each pair of TP and FP ratios was plotted on a graph using FP as the abscissa and TP as the ordinate. The points can be directly connected to form the ROC curve, or a smooth curve can be fitted to the data using a variety

DPL in Abdominal Stab Wounds

555

Table 1. False Positive and Falee Negative Results for Individual RBC Cutoffs

of techniques (6,28). To facilitate generalization of the results to the larger population of patients with abdominal stab wounds, we chose to fit a smooth curve to our data as opposed to analyzing each point individually (28).

RBC cutoff 5k 10k 20k 50k 75k 1OOk 200k Grossly +

RESULTS The distribution of the RBC counts for Groups 1 and 2 is shown in Figure 1. In Group 1, 16 patients (75%) had greater than 120,000 RBC/mm3 whereas 54 patients (75%) of Group 2 had fewer than 486 RBC/ mm3. The minimal overlap between the two groups identified a cutoff point that discriminated between those with and those without significant injury. The false positive and false negative results for the isolated individual RBC cutoffs are shown in Table 1. The ROC curve generated from our data is seen in Figure 2. The area under the ROC curve is 0.94, indicating that the RBC count is a good discriminator between Group 1 (required surgery) and Group 2 (did not require surgery). There were six laparotomies performed on the basis of DPL results that we would classify as false positive laparotomies. The mean length of hospitalization for these patients was 6.6 days, with one patient developing a postoperative pneumonia (17%).

% OF 100

False positives

False negatives

10

1

6

1 1 4 5 5 6 9

: 6 5 4 4

The follow-up period on this subgroup ranged from 3 to 12 months, and no cases of bowel obstruction were identified in the short term. The slope of an ROC curve for any test is equal to the likelihood ratio of the test at that point (13). For example, a likelihood ratio of 3 for a given RBC count would mean that RBC count is 3 times more likely to occur in Group 1 than in Group 2. It can be shown that the value of the likelihood ratio that minimizes misclassification (that is, diagnostic error) is equal to one divided by the odds of disease (3). The odds of disease in our study were 21170, which implies that the likelihood ratio, or slope of the ROC curve at the point of minimum misclassification, is equal to 70/21 or 3.33 (90% confidence interval 2.17 to 5.25). The slope of the fitted ROC curve is closest

Group Group

CASES

1: significant injury 2: no significant injury

_

80 -

60 -

(5k

50,OOO/mm’were found to have significant visceral injuries). A cutoff of 50,000 RBC/mm3 resulted in six false negative lavages. However, only 2 of the 6 injuries were significant (lacerations of the stomach and diaphragm). A gall bladder laceration was present in one patient, but this was manifested by bile-stained lavage fluid. In 1983, Oreskovich and Carrico published a prospective study on stab wounds to the anterior abdomen; of 572 patients in his study group, 236 received a DPL followed by laparotomy (23). The authors defined significant intra-abdominal injury as one requiring hemostasis, repair, or drainage. Of the 72 with DPL RBC counts greater than 100,000/mm3, 94% had positive laparotomies, 60 with solid organ Table 2. Variables Influencing Selection of RBC Cutoff 1. Incidence of disease 2. Morbidity and mortality from missed injuries 3. Morbidity and mortality from negative laparotomies 4. cost 5. Resource constraints

DPL in Abdominal Stab Wounds

557

Table 3. Types of Significant Injury (21 Patients)

Solid organ Hollow organ Vascular Diaphragm

< 50k

> 50k

Grossly +

Total

2 1 0 1

5 2 2 2

6

13 8 5 7

z 4

Note: 10 patients had injuries in more than 1 category.

and 6 with hollow organ injuries. Fifteen patients had RBC counts between 50,000 and 100,000/mm3: 12 were found to have positive laparotomies due to solid organ injuries; the remaining 3 had no visceral injury. The last group consisted of 82 patients with RBC counts between 1,000 and 50,000/mm3. In this group, although 47 (57%) had no visceral injury at the time of laparotomy, 35 (40%) had positive laparotomies, with hollow organ injuries accounting for the majority (21/35). The authors conclude that if the RBC count is greater than 50,000/mm3, a laparotomy is indicated. RBC counts between 1000 and 50,000/mm3 suggest the presence of hollow organ injury, while counts less than 1,000 RBC/mm3 justify nonoperative management. Both Galbraith and Oreskovich have comparable definitions of positivity and sufficient data to permit ROC analysis. We generated ROC curves for their data, with a minor modification in nomenclature to bring the definition of surgically significant injury into conformity with ours. Figure 3 represents the ROC curves from these studies superimposed onto our ROC curve. The three ROC curves are nearly identical, with no statistical difference in area under the three curves (29). Selection of the RBC cutoff point can be influenced by two variables: the prevalence of disease and the value associated with false positive and false negative diagnoses (Table 2). The prevalence of significant injury in our work was 23.1%, comparable to that seen in other studies on abdominal stab wounds (11). However, if this level was found to be higher, the optimal cutoff point could be shifted to the right on the ROC curve (lower RBC count). The second variable is more difficult to define: the value of false positive and false negative diagnosis. We attempted to find the RBC value that would minimize diagnostic errors; this places equal value on

negative laparotomies and missed injuries. In theory, the value of false positives and false negatives may be influenced by the morbidity and mortality from missed injuries and negative laparotomies, cost, and resource constraints. In reality, however, cost is seldom a factor in the decision to operate or observe an individual with an abdominal stab wound. Other constraints on resources exist and will possibly worsen, particularly in urban trauma centers: bed shortages, operating rooms, surgeons, ancillary personnel, and budgets. These, too, must be considered, but are seldom the prime consideration in treatment decisions. The real issue remains: what is the cost of a false negative lavage with delayed recognition of injury? Although concern for missed injuries seems valid, the concern may be more apparent than real. Henneman, Marx, and Moore reported on a large series of DPLs done for both blunt and penetrating trauma (11). Their data on stab wounds reflect 10 false negative initial lavages, all associated with hollow viscus injuries. All 10 patients did well, with a mean hospital stay of 10.2 days. This length of stay and outcome were not significantly different from those with true positive lavage (11). We observed no missed injuries in our study group, though we were using different positivity criterion during that period. In our study group, there were 8 hollow organ injuries; 5 were associated with the return of gross blood on aspiration during the DPL and 2 with RBC counts of >50,000/mm3 (see Table 3). The only hollow organ injury not associated with the RBC criterion for a positive lavage had a WBC count of 1,100/mm3. It remains our policy to observe for 24 hours and to document normal gastrointestinal function after admission and prior to discharge. A missed hollow organ injury could be expected to manifest itself clinically within this time period.

CONCLUSIONS ROC analysis suggests that an RBC count of 50,000 RBC/mm3 as the sole criterion in stable patients with abdominal stab wounds should optimally discriminate between those patients requiring surgery and those not requiring surgery.

REFERENCES 1. Alyono D, Morrow CE, Perry JF. Reappraisal of diagnostic peritoneal lavage criteria for operation in penetrating and blunt trauma. Surgery. 1982;92:751-7.

2. Baker RJ. Peritoneal lavage in blunt and penetrating abdominal trauma. Curr Surg. 1981;149-50. 3. Danto LA, Thomas CW, Gorenbein S, Wolfman EF. Pene-

M. J. Zappa, A. L. Harwood-Nuss, R. L. Wears, W. F. Fallon

558 trating torso injuries: the role of paracentesis and lavage. Am Surg. 1977;43:164-70. 4. DeLacy AM, Pera M, Garcia-Valdecasas JC, et al. Management of penetrating abdominal stab wounds. Br J Surg. 1988; 75:231-3. 5. Demetriades D, Rabinowitz B. Indications for operation in abdominal stab wounds. Ann Surg. 1987;205:129-132. 6. Dorfman DD, Alf E Jr. Maximum likelihood estimation of parameters of signal detection theory and determination of confidence intervals-rating method data. J Math Psychol. 1969;6:487-96. 7. Feliciano DV, Bitondo CG, Steed G, et al. Five hundred open taps or lavages in patients with abdominal stab wounds. Am J Surg. 1984;148:772-7. 8. Galbraith TA, Oreskovich MR, Heimbach DM, et al. The role of peritoneal lavage in the management of stab wounds of the abdomen. Am J Surg. 1980;140:6-4. 9. Griffin JB, Reines HD. Diagnostic criteria for peritoneal lavage in penetrating trauma. Curr Surg. 1983;40:351-3. 10. Gruenberg JC, Brown RS, Talberg JG, et al. The diagnostic usefulness of peritoneal lavage in penetrating trauma: a prospective evaluation and comparison with blunt trauma. Am Surg. 1982;48:402-7. 11. Henneman PL, Marx JA, Moore EE, Cantrill SV, Ammons LA. Diagnostic peritoneal lavage: accuracy in predicting necessary laparotomy following blunt and penetrating trauma. J Trauma. 1990;30:1345-55. 12. Hornyak SW, Shaftan GW. Value of “inconclusive lavage” in abdominal trauma management. J Trauma. 1979;19(5):329-33. 13. Lee WC, Uddo JF, NanceFC. Surgical judgement in the management of abdominal stab wounds. Ann Surg. 1984;199(5):549-54. 14. Lusted LB. Decision-making studies in patient management. NEJM. 1971;284(8):416-24. 15. McAlvanah MJ, Shaftan GW. Selective Conservatism in penetrating abdominal wounds: a continuing reappraisal. J Trauma. 1978;18(3):206-12. 16. McNeil BJ, Keeler E, Adelstein SJ. Primer on certain elements of medical decision making. NEJM. 1975;293(5):211-15. 17. Merlotti GJ, DiJlon BC, Lange DA, et al. Peritoneal lavage in penetrating thoraco-abdominal trauma. J Trauma. 1988;28(1):17-23. 18. Merlotti GJ, Marcet E, Sheaff CM, et al. Use of peritoneal lavage to evaluate abdominal penetration. J Trauma. 1985;25: 228-3 1.

19. Moore EE, Marx JA. Penetrating abdominal wounds: rationale for exploratory laparotomy. JAMA. 1985;253:2705-8. 20. Moore JB, Moore EE, Thompson JS. Abdominal injuries associated with penetrating trauma in the lower chest. Am J Surg. 1980;140:724-30, 21. Miller FB, Cryer HM, Chilikuri S, et al. Negative findings on laparotomy for trauma. South Med J. 1989;82:1231-4. 22. Nance FC, Wennar MH, Johnson LW, et al. Surgical judgement in the management of penetrating wounds of the abdomen. Ann Surg. 1974;179:639-45. 23. Oreskovich MR, Carrico CJ. Stab wounds of the anterior abdomen. Ann Surg. 1983;98:411-19. 24. Peterson WW, Birdsall TG, Fox WC. Theory of signal detectability. Tram IRE Professional Group in Information Theory. 1954;PGlT-4:171-212. 25. Robin AP, Andrews JR, Lange DA, et al. Selective management of anterior abdominal stab wounds. J Trauma. 1989;29: 1684-9. 26. Root HD, Hauser CW, McKinley CR, et al. Diagnostic peritoneal lavage. Surg. 1965;57:633-7. 27. Shaftan GW. Indications for operation in abdominal trauma. Am J Surg. 1960;99:657-64. 28. Stenson H. SIGNAL: Supplementary Module for SYSTAT. Evanston, IL: SYSTAT, Inc.; 1988. 29. Swets JA, Pickett RM. Evaluation of diagnostic systems: methods from signal detection theory. New York: Academic Press; 1982. 30. Thal ER. Evaluation of peritoneal lavage and local exploration in lower chest and abdominal stab wounds. J Trauma. 1977; 17642-S. 31. Thompson JS, Moore EE, Van Duzer-Moore S, et al. The evolution of abdominal stab wound management. J Trauma. 1980;20:478-84. 32. Thompson JS, Moore EE. Peritoneal lavage in the evaluation of penetrating abdominal trauma. Surg Gynecol Obstet. 1981; 153:861-3. 33. Weigelt JA, Kingman RG. Complications of neaative lanarotomy-for trauma.Am J Surg. 1988;156:544-7. 34. Weinstein MC. Finebera HV. Clinical decision analvsis. Philadelphia, PA: WB Saunders; 1980. 35. Wilder JR, Lofti MW, Jurani P. Comparative study of mandatory and selective surgical intervention in stab wounds of the abdomen. Surgery. 1971;69:546-9.

APPENDIX Receiver operator characteristic curve analysis was developed in the context of electronic signal detection (1). However, since it is based solely on statistical and information-theoretic considerations, its usefulness is independent of any particular problem domain. It has been applied in a wide variety of areas, including the evaluation of medical diagnostic systems (2). ROC analysis can be performed on any medical diagnostic test (or combination of tests) for which the results can be unambiguously ordered. Careful reflection will indicate that the ROC curve will be fixed at the points (0,O or lower left and 1,l or upper right) that represent the naive alternatives of calling every result negative or every result positive, respectively. If the diagnostic test contributes no information, its ROC curve will be the straight line diagonally connecting these two points. A test that contributes perfect information will have 100% true positives and no false positives and, thus, will occupy the point (1,0) in the upper left corner. The area under the ROC is a general, criterion-independent measure of the test’s discriminating ability. It can be thought of as the probability

of correctly ranking an unknown normal-abnormal pair on the basis of the test result. The slope of the fitted RGC curve can be used to determine the value of the positivity criterion that provides maximum information (that is, maximum discrimination between the alternatives). ROC analysis offers four significant advantages over other methods of evaluating diagnostic tests (2): 1. It supplies a pure index of accuracy that can be used to compare diagnostic methods, independent of the particular criteria and biases used to determine positivity for those methods. 2. It supplies estimates of the probabilities of true positive, true negative, false positive, and false negative results for any of the possible positivity criteria that might be selected. 3. It can estimate the positivity criterion at which discriminating ability is maximized. 4. It is objective in that different observers analyzing the same data set will obtain the same results.