| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Review |
1 Chicago Medical School, 3333 Green Bay Rd., North Chicago, IL, 60064.
2 Department of Surgery, Advocate Illinois Medical Center, Chicago, IL.
3 Department of Radiology, St. Alphonsus Regional Medical Center, Boise
ID.
Received October 8, 2008;
accepted after revision January 22, 2009.
Address correspondence to C. S. Goodman
(cyle.goodman{at}my.rfums.org).
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. We reviewed MEDLINE articles published from January 1994 through June 2008. The sensitivity, specificity, negative predictive value (NPV), positive predictive value (PPV), and accuracy were calculated for each source and collectively using a meta-analysis.
RESULTS. Of 180 relevant studies, five were included in the meta-analysis. Pooled weighted estimates of sensitivity, specificity, NPV, PPV, and accuracy were 94.90%, 95.38%, 98.62%, 84.51%, and 94.70%, respectively.
CONCLUSION. CT in patients with penetrating abdominal trauma has high sensitivity, specificity, NPV, and accuracy, but has lower PPV in determining the need for laparotomy. It follows that CT is an indispensable tool in predicting the need for laparotomy in these patients but still has room for improvement.
Keywords: abdominal injury • CT • emergency radiology • laparotomy • penetrating trauma • trauma
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
With the increased use of laparotomies in patients with penetrating abdominal injuries, it is not surprising that an increase in the rate of negative laparotomies followed. To combat this trend, the concept of selective management was introduced, in which medical professionals determined whether patients with penetrating abdominal trauma were candidates for nonoperative versus operative management. In the late 1960s and early 1970s, Sir Godfrey Hounsfield and Allan McLeod Cormack independently invented the CT scanner, leading to their shared 1979 Nobel Prize in Medicine.
Since that time, continual advances have been made in CT, and the CT scanner has become an indispensable tool in the process of selective management. Today, the standard of care for patients with penetrating abdominal trauma depends on hemodynamic stability and peritoneal signs [2]. Most patients who are hemodynamically unstable or have signs of peritonitis are taken directly to the operating room. Various algorithms for managing hemodynamically stable patients are available depending on the hospital, but most involve performing CT of the abdomen. However, the incidence of unnecessary laparotomies still ranges from 1.7% to 38%, leading to increased complications, longer hospital stays, and increased costs [3].
Despite technologic and procedural advancements, the question in the medical arena remains: How well do findings on CT images predict which patients with abdominal injuries should undergo laparotomy and which should not? This study set out to answer this question by determining the sensitivity, specificity, negative predictive value (NPV), positive predictive value (PPV), and accuracy of CT in predicting the need for laparotomy in hemodynamically stable patients with penetrating abdominal injuries. In doing so, the appropriate use of CT can be identified, potentially minimizing the rate of unnecessary laparotomies.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Study Selection Criteria
The following inclusion criteria were developed in the meta-analysis
protocol before beginning the search to minimize selection bias. All types of
studies—that is, retrospective, observational, case reports, and so
on—were eligible as long as they contained at least 10 study subjects.
The data in the study had to be collected during or after 1994. The study had
to provide quantitative data on the number of positive and negative CT
examinations and whether laparotomy was performed in each of these groups. A
positive CT study was defined as CT that indicated the need for laparotomy,
such as injury to the retroperitoneal bowel, renal collecting system, or major
vessels. Likewise, a negative CT study was one that did not indicate the need
for laparotomy, such as no clear penetration of the peritoneal cavity, no
visceral injury, no free air or fluid, or no contrast leak. Male and female
patients of all ages and ethnicities were accepted; they must have had a
penetrating abdominal injury and subsequently undergone abdominal CT.
Penetrating abdominal injuries consist of stab wounds, gunshot wounds, or
trauma caused by any object that could have entered the peritoneal cavity or
retroperitoneum and damaged the abdominal contents. Additionally, patients had
to be hemodynamically stable, which was defined as presenting with a systolic
blood pressure 
90 mm Hg after receiving a maximum of 2 L of IV
fluids.
The following exclusion criteria were also incorporated into the meta-analysis protocol. Patients were excluded if they were hemodynamically unstable, exhibited signs of peritonitis, or had any other indication for emergent laparotomy according to the hospital's treatment guidelines. Patients with injuries due to blunt trauma, such as motor vehicle accidents or falls, were also excluded. If a single study was published in multiple publications, the latest published article containing the study data was used.
Study Selection and Quality Control
Two individuals separately performed the article search, study selection,
and study quality evaluation. All of the search and selection protocols were
created beforehand. Any discrepancies were settled at a conference where the
two would state their cases and come to one final decision. After searching
MEDLINE as described, articles were screened for relevance and inclusion
criteria. The sources remaining after the initial screen were then evaluated
with a quality evaluation score sheet modified from Berman and Parker
[4] (Appendix 1). The studies
were required to contain the inclusion criteria discussed. Additionally, a
score of 15 or greater was needed on the quality evaluation form for the study
to be accepted.
End Points and Data Extraction
The principal objective was to calculate sensitivity, specificity, NPV,
PPV, and accuracy of CT. To accomplish this, the following specific data from
each study were extracted and recorded: the total number of hemodynamically
stable patients with penetrating abdominal injury, total number of CT
examinations performed, number of positive and negative CT examinations,
number of therapeutic laparotomies after positive and negative CT
examinations, number of nontherapeutic laparotomies after positive and
negative CT examinations, number of negative laparotomies after positive and
negative CT examinations, nontherapeutic laparotomy rate, and the percentage
of avoided laparotomies in stable patients. Laparotomy was considered negative
when no injury was found, nontherapeutic when injuries requiring no
intervention were discovered, and therapeutic when injuries were identified
and repaired on exploration.
Next, 2 x 2 tables were set up comparing the CT results with the laparotomy results. True-positives were patients with a positive CT examination and a therapeutic laparotomy. False-positives were patients with a positive CT examination and a nontherapeutic, negative, or no laparotomy. True-negatives were patients with a negative CT examination and a nontherapeutic, negative, or no laparotomy. False-negatives were patients with a negative CT examination and a therapeutic laparotomy. These 2 x 2 tables were used to calculate the sensitivity, specificity, NPV, PPV, and accuracy of CT. Additionally, researchers collected data regarding the mean age of the patients, the percentage of male and female patients, and year that the data were obtained.
Statistical Analysis
The study's primary goal with respect to statistical analysis was to
summarize the available data and to explain any variability that may exist
among studies. The protocol for data analysis was set up before data
collection to limit false-positive conclusions. We analyzed the following
effect sizes independently: sensitivity, specificity, NPV, PPV, and accuracy.
The Mantel-Haenszel method (fixed-effects model) and DerSimonian and Laird
method (random-effects model) were used to calculate the weighted mean effect
sizes from the pooled data. The 95% CI and inverse variance weight were
calculated using the modified Wald method
[5]. Wilson's practical
meta-analysis was used as a guide to calculate the mean effect size, standard
error of the mean effect size, 95% CI, Cochran's Q, and the random-effects
variance component [6]. The
individual effect sizes, mean effect sizes, and 95% CIs were used to create a
forest plot to visually summarize the data using software (Prism 5, GraphPad
Software).
Heterogeneity across studies was assessed by interpreting Cochran's Q with an alpha value of 0.10 as the cutoff point, which correlated to a critical chi-square value of 7.78. To find out if study characteristics contributed to any heterogeneity, we calculated Pearson's correlation coefficient (r) by comparing the weighted effect sizes with patient age, percentage of males and females, and year that data were collected. Pearson's r was converted to t to test for significance of the Pearson's product-moment correlation coefficient. The critical t value was determined to be 2.353 for p < 0.05 according to the appropriate degrees of freedom.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
The sensitivity of CT in detecting the need for laparotomy is summarized in Figure 1, which contains source data as well as weighted mean sensitivity of the pooled data with 95% CIs. The weighted mean sensitivity was 95.38% (95% CI, 90.41–100%) with Cochran's Q equal to 3.47 (p = 0.48) according to the fixed-effects model and 94.90% (89.15–100%) with Cochran's Q equal to 3.13 (p = 0.54) according to the random-effects model.
|
|
The weighted mean NPV ranged from 98.70% (95% CI, 97.02–100%) with Cochran's Q equal to 2.39 (p = 0.66) to 98.62% (96.37–100%) with Cochran's Q equal to 2.28 (p = 0.68) using the fixed-effects model and random-effects model, respectively (Table 2 and Fig. 3).
|
|
The PPV was calculated as 83.29% (95% CI, 76.67–89.91%) with a Cochran's Q of 6.75 (p = 0.15) using the fixed-effects model and 84.51% (75.08–93.94%) with a Cochran's Q of 3.89 (p = 0.42) by means of the random-effects model (Fig. 4). The fixed-effects model produced an accuracy of 95.18% (95% CI, 93.13–97.22%), whereas the random-effects value was 94.70% (95% CI, 91.29–98.10%) with Cochran's Q equal to 8.31 (p = 0.08) and 4.22 (p = 0.38), respectively (Table 3 and Fig. 5).
|
|
|
Table 4 summarizes the correlation between study characteristics and weighted effect sizes by listing the Pearson's r values and t values that were calculated. The critical t value for p < 0.05 was 2.353. The only study characteristic that produced t values greater than 2.353 was the year that data were collected. When comparing the median year that data were collected before 2008 with weighted sensitivity, the Pearson's r value was –0.85 with a t value of 2.76 (p = 0.035). All other correlations between study characteristics and weighted effect sizes yielded t values less than 2.353.
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The calculation of specificity and accuracy with the fixed-effects model produced Cochran's Q values greater than the critical value. In this instance, it was concluded that heterogeneity did indeed exist between the studies. Addressing this heterogeneity, study characteristics were compared with weighted effect sizes to analyze between study variability. Additionally, researchers fit the data to the random-effects model. The only study characteristic to show a significant correlation with weighted effect sizes was the year that data were collected. The more recent the CT examinations were conducted, the higher the sensitivity. This seems reasonable given that CT has been continually improving in quality over the years. None of the study characteristics that were examined could account for the heterogeneity found between studies when specificity was computed. Thus, it was assumed that the excess variability across effect sizes was derived from random differences across studies. Researchers, therefore, concluded that it is more appropriate to use the values calculated using the random-effects model rather than the fixed-effects model.
The sensitivity, specificity, NPV, and accuracy were all found to be
about 95%, whereas the PPV was approximately 85%. The lower PPV could possibly
be explained by how a positive CT was defined in this study. Findings of free
intraperitoneal fluid, air, or both were considered as positive CT findings in
the articles analyzed for this meta-analysis; therefore, it was necessary in
this study to include these as positive CT images. In the trauma setting,
however, free intraperitoneal fluid or air alone does not necessarily indicate
a need for laparotomy. Including these findings as positive CT images had the
potential to increase the number of false-positives and subsequently
negatively affect the specificity and PPV of CT.
The results of this study show that CT is an indispensable tool in predicting the need for laparotomy but still has room for improvement. As mentioned, unnecessary laparotomy rates for trauma range from 1.7% to 38% [3]. According to Demetriades and Velmahos [12], the overall incidence of early complications in patients who underwent unnecessary laparotomy for penetrating injuries was 10.6%. The overall late complication rate in this setting has been reported at 2.4% [13].
Increased cost and length of hospital stay have been reported for patients who underwent unnecessary laparotomy. According to one study, the average hospital charge for nonoperatively managed patients with abdominal gunshot wounds was $8,595, whereas patients who underwent unnecessary surgery paid an average of $18,123 [14]. The proper use of CT has the possibility to decrease the number of unnecessary laparotomies and thereby reduce complication rates, hospital stays, and hospital costs. Some critics of selective nonoperative management might argue that these patients are at increased risk of having missed injuries and that the missed findings have the potential for increased complications. In a review of 728 patients with penetrating injuries, Demetriades et al. [2] found a 3.4% incidence of delayed diagnosis with delayed treatment leading to no deaths and a morbidity that was comparable to patients receiving an early operation.
In this meta-analysis, researchers did not have access to the type of CT scanner used in each study (e.g., single-detector, multidetector, helical). Furthermore, the scanning protocols used, such as the use of contrast material and thickness of slices, were not made available. These data were requested from the authors of the sources; however, some authors did not respond and some of the responding authors could not accurately produce these data. Therefore, assessing whether differences in outcomes could be explained by use of specific types of CT scanners or scanning protocols was not possible.
Another potential limitation of our study is that the studies selected for this meta-analysis represent a combination of stab and gunshot wound patients. The trajectory of a sharp object in a stab wound tends to be linear, whereas the trajectory of a bullet in a gunshot wound is usually nonlinear. This difference leads to gunshot wounds behaving more aggressively and likely causing injuries to multiple organs. More aggressive and obvious wounds are more likely to be described as positive and are less likely to be missed. Thus, in theory, gunshot wounds should produce a higher true-positive rate, and stab wounds should produce a higher false-negative rate. Additionally, stab wounds tend to be treated more conservatively, whereas gunshot wounds are treated more aggressively. Many surgeons will perform laparotomies on patients with gunshot wounds despite the negative CT findings. This leads to an increased number of laparotomies performed on patients with gunshot wounds compared with patients with stab wounds.
The follow-up period varied from article to article. Some subjects were observed for less than 24 hours and were subsequently lost to follow-up. Others were followed for 44 months. The reported mean follow-up ranged from 7 days to 10 months between articles. If these patients were not adequately followed, then the number of true-negatives, false-positives, or both could be inflated. However, the articles did state that most patients were followed up for at least a couple of days, which is enough time to detect most injuries requiring laparotomy.
As with any meta-analysis, this research has the potential to contain biases despite researchers' efforts to optimize protocols before accumulating data and conducting analyses. Publication bias is always a possibility because studies that show a significant effect are more likely to be published than studies with nonsignificant findings. In this study, language bias is likely because only sources written in English were used in an attempt to protect against unreliable translations of articles published in other languages. Additionally, this analysis involved some small studies, which tend to have more dramatic results than large studies.
In light of the possibility for bias, this meta-analysis is just one step in the process of determining CT's potential in triaging patients with penetrating abdominal injury. As CT scanners are advancing and the quality of CT images is improving, CT's full capabilities must be explored. A next step could entail setting up a multicenter prospective research project with a standardized protocol. Such a study would minimize potential biases and produce more reliable results. Furthermore, the accuracy of today's CT could be identified and compared with imaging of the past. Technology is continually improving, and medical professionals must continue to adapt their practices to maximize patient benefits.
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
