Outcomes of selective nonoperative management of civilian abdominal gunshot wounds: a systematic review and meta-analysis

Background Although mandatory laparotomy has been standard of care for patients with abdominal gunshot wounds (GSWs) for decades, this approach is associated with non-therapeutic operations, morbidity, and long hospital stays. This systematic review and meta-analysis sought to summarize outcomes of selective nonoperative management (SNOM) of civilian abdominal GSWs. Methods We searched electronic databases (March 1966–April 1, 2017) and reference lists of articles included in the systematic review for studies reporting outcomes of SNOM of civilian abdominal GSWs. We meta-analyzed the associated risks of SNOM-related failure (defined as laparotomy during hospital admission), mortality, and morbidity across included studies using DerSimonian and Laird random-effects models. Between-study heterogeneity was assessed by calculating I2 statistics and conducting tests of homogeneity. Results Of 7155 citations identified, we included 41 studies [n = 22,847 patients with abdominal GSWs, of whom 6777 (29.7%) underwent SNOM]. The pooled risk of failure of SNOM in hemodynamically stable patients without a reduced level of consciousness or signs of peritonitis was 7.0% [95% confidence interval (CI) = 3.9–10.1%; I2 = 92.6%, homogeneity p < 0.001] while the pooled mortality associated with use of SNOM in this patient population was 0.4% (95% CI = 0.2–0.6%; I2 = 0%, homogeneity p > 0.99). In patients who failed SNOM, the pooled estimate of the risk of therapeutic laparotomy was 68.0% (95% CI = 58.3–77.7%; I2 = 91.5%; homogeneity p < 0.001). Risks of failure of SNOM were lowest in studies that evaluated patients with right thoracoabdomen (3.4%; 95% CI = 0–7.0%; I2 = 0%; homogeneity p = 0.45), flank (7.0%; 95% CI = 3.9–10.1%), and back (3.1%; 95% CI = 0–6.5%) GSWs and highest in those that evaluated patients with anterior abdomen (13.2%; 95% CI = 6.3–20.1%) GSWs. In patients who underwent mandatory abdominopelvic computed tomography (CT), the pooled risk of failure was 4.1% versus 8.3% in those who underwent selective CT (p = 0.08). The overall sample-size-weighted mean hospital length of stay among patients who underwent SNOM was 6 days versus 10 days if they failed SNOM or developed an in-hospital complication. Conclusions SNOM of abdominal GSWs is safe when conducted in hemodynamically stable patients without a reduced level of consciousness or signs of peritonitis. Failure of SNOM may be lower in patients with GSWs to the back, flank, or right thoracoabdomen and be decreased by mandatory use of abdominopelvic CT scans. Electronic supplementary material The online version of this article (10.1186/s13017-018-0215-0) contains supplementary material, which is available to authorized users.


Background
Mandatory laparotomy has been standard of care for patients with abdominal gunshot wounds (GSWs) for decades. However, this approach is associated with unnecessary, including non-therapeutic (where intra-abdominal injuries are found that do not require intervention) and negative (where no intra-abdominal injuries are found), laparotomies [1][2][3]. Further, mandatory laparotomy in patients with abdominal GSWs has been linked with a 22-41% risk of postoperative complications (e.g., surgical site infections, gastrointestinal ileus, pneumonia, and venous thromboembolism) [4,5] and a 5-to 9-day length of hospital stay [6][7][8].
Selective nonoperative management (SNOM) is frequently conducted in trauma centers in patients with penetrating abdominal trauma who are hemodynamically stable without signs of diffuse peritonitis or evisceration. Although SNOM has been relatively widely adopted for abdominal stab wounds [1,3,9], the concept has not been as embraced for GSWs given the higher associated incidence of visceral and abdominal vascular injuries and the morbidity and mortality associated with missed injuries [9,10]. Therefore, the practice of SNOM for abdominal GSWs remains controversial among some surgeons.
The cornerstone of SNOM lies on the principle of serial physical examinations of patients without a reduced level of consciousness by qualified surgeons or experienced surgical residents. Over the last two decades, there has also been an interest in using abdominopelvic computed tomography (CT) scans as an adjunct when conducting SNOM [11,12]. Proponents suggest that CT scans may better characterize bullet trajectory and have been reported to detect injuries with a sensitivity and specificity exceeding 90% [12]. Opponents argue that it is associated with false negative and positive test results and lacks accuracy in detecting some (e.g., intestinal and diaphragmatic) injuries [13].
There have been no randomized controlled trials to date that have evaluated the outcomes of SNOM versus mandatory laparotomy for management of civilian abdominal GSWs. Moreover, results of cohort studies of this injury management strategy have been variable. To evaluate the potential safety of this approach, the purpose of this systematic review and meta-analysis was to summarize outcomes associated with use of SNOM in cohort studies of civilians with abdominal GSWs. We hypothesized that SNOM of abdominal GSWs would be safe (associated with a small risk of failure, morbidity, and mortality) when conducted in highly experienced trauma centers. We also sought to determine whether variability in reported outcomes of SNOM across cohort studies may be due to differences in study risks of bias, practices of SNOM across trauma centers and countries, or study patient injury patterns (e.g., whether the entry wound of the included patients was predominantly in the anterior abdomen, thoracoabdomen, flank, or back).

Protocol
Our methods were pre-specified in a detailed protocol created according to the Preferred Reporting Items in Systematic Reviews and Meta-Analyses statement [14] and the Meta-analysis of Observational Studies in Epidemiology (MOOSE) proposal (see the completed MOOSE checklist in Additional file 1: digital content S1) [15].

Search strategy
With the assistance of a medical librarian/information scientist, we searched PubMed, Ovid MEDLINE and EMBASE, and the Cochrane Library from their inception to April 1, 2017, without language restrictions. Using a combination of Medical Subject Heading (MeSH) and Emtree terms and keywords, we created search filters covering the themes SNOM, penetrating injury/GSW, and abdomen. These filters were combined in our final database searches using the Boolean operator "AND." Our complete search strategies are shown in Additional file 2: digital content S2. To identify additional studies, we used the PubMed "related articles" feature, hand-searched references of included original and relevant review articles identified during the search, and wrote several authors who had published on the topic.

Selection criteria
Independently and in duplicate, two investigators (ANAR, FAAH) screened the titles and abstracts of citations identified during the search, reviewed potentially relevant articles in full, and decided on study inclusion. We used the following inclusion criteria: (1) study participants were adult (mean age ≥ 16 years) civilian trauma patients with GSWs to the abdomen; (2) some or all of the included patients underwent SNOM of their abdominal GSWs; (3) reported outcomes included SNOM failure, morbidity, mortality, and/or patient hospital length of stay (LOS); and (4) the study used a cohort design. We distinguished cohort studies from case series using the criteria developed by Dekkers et al. and included controlled (which included a comparable control group of patients without diffuse peritonitis or hemodynamic instability that underwent mandatory laparotomy) and uncontrolled cohort studies in the systematic review [16]. We excluded conference abstracts. Study eligibility disagreements were resolved by consensus. Inter-investigator agreement on article inclusion was assessed using kappa (κ) statistics [17]. When the same data were reported across multiple studies, the study with the larger sample size or that provided the most information on SNOM-associated outcomes was included.

Definitions
We subdivided the abdomen into the anterior abdomen, thoracoabdomen, bilateral flanks, and back. We defined the anterior abdomen as the region bounded superiorly by the costal margins, laterally by the anterior axillary lines, and inferiorly by the inguinal creases [18]. The thoracoabdomen was defined as the region enclosed by the nipples (or tips of the scapulae) superiorly and the costal margin inferiorly [18]. We defined the flanks as the region bounded by the costal margins, anterior and posterior axillary lines, and iliac crests [18]. Finally, the back was defined as the region bounded superiorly by the inferior scapular tips, laterally by the posterior axillary lines, and inferiorly by the iliac crests [18].

Data extraction
Three reviewers (ANAR, FAAH, JMB) independently extracted data from included studies in duplicate. Data extracted included (1) study design, temporality, and setting; (2) trauma center and study cohort characteristics [e.g., recruitment period and mean/median patient age and Injury Severity Scale (ISS) score], number of trauma patients and abdominal GSWs assessed per year, and the anatomical regions of the abdomen injured by GSWs; (3) number of patients who underwent SNOM and details regarding SNOM practices (frequency of clinical and laboratory examinations, mandatory versus selective use of CT, duration of observation, and type and duration of follow-up); and (4) outcomes associated with use of SNOM.

Primary and secondary outcomes
The primary outcome was the risk of failure of SNOM. We defined failure as the conduct of laparotomy on a patient undergoing SNOM for an abdominal GSW during their hospital admission. Secondary outcomes included therapeutic and unnecessary laparotomy among patients who failed SNOM, in-hospital mortality, reported morbidities associated with SNOM, and hospital LOS. Unnecessary laparotomy was defined as either negative (where no injury was identified during laparotomy) or non-therapeutic (where an injury was found during laparotomy that did not require surgical intervention).

Risk of bias assessment
The risk of bias of the included studies was assessed by two independent investigators with graduate training in epidemiology (ANAR, FAAH) using a modified version of the Quality in Prognosis Studies (QUIPS) tool [19]. This tool includes 24 decision items that cover five quality domains of interest, including (1) patient selection, (2) study attrition, (3) prognostic factor measurement, (4) outcome measurement, and (5) statistical analysis and reporting. We scored the adequacy of reporting for each item of the above five QUIPS domains as "yes," "no," or "unclear." This scoring led to the overall judgment of low, moderate, or high risk of bias per quality domain. Disagreements regarding study risk of bias were resolved by consensus.

Statistical analyses
We calculated study estimates of the risk of failure of SNOM and therapeutic and unnecessary laparotomy, reported morbidities, and in-hospital mortality associated with use of SNOM in patients with abdominal GSWs. We determined standard errors and 95% confidence intervals (CIs) for these estimates using Clopper-Pearson exact methods [20]. We applied a continuity correction of 1 to estimates with a zero numerator or denominator to estimate their standard error [21].
DerSimonian and Laird random-effects models were used to calculate pooled estimates of the risk of outcomes across the included studies [22]. Summary mean and median LOS across studies were calculated by weighting these estimates by study sample sizes. Heterogeneity in pooled estimates was assessed by calculating I 2 inconsistency and Q statistics and conducting tests of homogeneity (p value < 0.10 considered significant given the low power of these tests) [23]. The I 2 statistic represents the percentage of variation between studies due to factors other than chance. I 2 statistics of > 25%, > 50%, and > 75% were considered to represent low, moderate, and high degrees of heterogeneity, respectively [24]. A test of homogeneity p value < 0.10 was considered to indicate more heterogeneity than would be expected between studies due solely to chance [24].
In the presence of at least low inter-study heterogeneity, we conducted stratified meta-analyses and meta-regression to determine whether our pooled risk estimates varied across a number of study-level covariates selected a priori. Covariates of interest included study setting (USA, South Africa, or other) and temporality (prospective versus retrospective), study patient injury patterns [anterior abdominal, right thoracoabdominal, isolated renal or hepatic/right upper quadrant (RUQ), back, or flank GSWs], and reported SNOM practices (serial physical examinations done by surgeons or surgical residents and mandatory versus selective use of abdominopelvic CT scans).
We examined for evidence of small study effects potentially due to publication bias by creating funnel plots and conducting Begg's funnel plot asymmetry test [25]. We used the Duval and Tweedie "trim and fill" method to evaluate the potential influence of publication bias on our pooled estimates [26,27]. Using this method, small outlying studies were first "trimmed" (removed until funnel plots were symmetric). The remaining study results were then used to re-estimate the theoretically unbiased center of the plot before it was "filled" (the missing outlying study results and their theoretical counterparts were replaced around the re-estimated center), permitting calculation of a publication bias-adjusted pooled risk estimate [26][27][28][29]. Stata version 13 (Stata Corp., College Station, TX, USA) was used for all analyses.
We found no controlled studies of SNOM that included a comparable control group of patients without diffuse peritonitis or hemodynamic instability that underwent mandatory laparotomy. Instead, all included studies examined outcomes of SNOM in a cohort of patients with abdominal GSWs to the anterior or other regions of the abdomen. Six studies reported outcomes of SNOM in patients with renal GSWs [30][31][32][33][34] while 5 reported SNOM outcomes in patients with hepatic GSWs [35][36][37][38][39]. Another three studies included patients with GSWs to the RUQ [40][41][42], one with GSWs to the back [43], and one with GSWs to the flank [44].

Risk of bias assessment
One study included 12,707 patients that sustained abdominal GSWs between 2002 and 2008 and were included in the National Trauma Data Bank (NTDB) [62], which contains data collected from approximately 900 trauma centers in the USA. This study likely included duplicate patients from other American studies published during the same period [9,12,31,32,37,44,60,63,66]. Further, a number of other studies used the same data source to identify patients, but looked at different outcomes and/or patient populations [33,39,40,43,45,50,59,67]. However, it was not possible to determine the amount of overlap or duplication of patients with certainty between these studies, and therefore, the influence of overlap was explored in a post hoc sensitivity analysis described below.
Hospital length of stay Twenty-nine studies (70.7%) reported the mean hospital LOS among patients who underwent SNOM, which varied from 2 to 20 days. The weighted average hospital LOS was 5.9 days. However, in patients with isolated abdominal GSWs without any associated extra-abdominal injuries, the weighted median hospital LOS was 2.3 days. Conversely, the weighted median hospital LOS in patients who failed SNOM or developed complications was 10.1 days. Table 4 details results of stratified meta-analyses and meta-regression of variables associated with failure of SNOM for abdominal GSWs. There was no difference in the pooled risk of SNOM failure between prospective and retrospective studies or those conducted in the USA, South Africa, or other countries. Risks of failure of SNOM were lowest in studies that evaluated patients with right thoracoabdomen (3.4%; 95% CI = 0-7.0%; I 2 = 0%; homogeneity p = 0.45), flank (7.0%; 95% CI = 3.9-10.1%), and back (3.1%; 95% CI = 0-6.5%) GSWs and highest in those that evaluated patients with anterior abdomen (13.2%; 95% CI = 6.3-20.1%) GSWs. Estimates of the pooled risk of failure in studies where serial physical examinations were reportedly done by attending trauma surgeons were approximately one-third that of those reported in studies where physical examinations were done by surgical residents. When studies were divided according to the policy of CT use, the pooled estimate of SNOM failure in studies of patients undergoing selective abdominopelvic CT was approximately double that of the pooled estimate of SNOM failure in studies of patients undergoing mandatory CT. A sensitivity analysis excluding the results of studies that reported potentially overlapping patient outcome data (including those that may have overlapped with those recruited into the study that analyzed patients included in the NTDB) yielded a similar pooled estimate of the risk of failure of SNOM for abdominal GSWs (6.7%, 95% CI = 2.5-10.9%, I 2 = 94.4%, homogeneity p < 0.001) to the pooled estimate that included data from all studies reporting data on this outcome.

Publication bias
Inspection of the funnel plots of the reported risks of failure of SNOM versus the standard error across the included studies revealed that smaller studies may have reported higher risks of failure than larger studies (Fig. 4). Further, Begg's test was significant (p = 0.008) for funnel plot asymmetry. However, using Trim and Fill methods, the publication bias-adjusted summary estimate of the risk of failure of SNOM was unchanged (7.2%; 95% CI = 4.2%-10.1%) from the unadjusted estimate (7.0%; 95% CI = 3.9-10.1%), providing evidence that publication bias likely had little influence on the pooled results.

Discussion
This systematic review of 41 studies involving 22,847 civilians with abdominal GSWs is the first to comprehensively meta-analyze outcomes associated with use of SNOM in this patient population. Our findings suggest that highly experienced trauma centers have safely treated greater than 90% of the patients included in these studies nonoperatively. Importantly, in all studies, only hemodynamically stable patients without a reduced level of consciousness and no signs of peritonitis were selected for SNOM. SNOM may be more successful among patients with GSWs to the right thoracoabdomen, flank, or back than the anterior abdomen and when serial physical examinations are performed by attending trauma surgeons. This practice also appears to be more successful in patients who undergo mandatory abdominopelvic CT scans and in those with injuries proven to involve the kidney on imaging.
Although SNOM has been linked with development of atelectasis, GSW infections, and biliary fistulae in patients with GSWs to the liver, these complications are uncommon and appear to be less frequent than that historically reported after mandatory laparotomy in similar patient populations [69][70][71].
As most of the included studies were conducted at high-volume, level 1, academic trauma centers in the USA, our results may not be generalizable to centers without the necessary structures and processes to manage these patients. Many of the centers conducting research on SNOM described dedicated, monitored areas to observe and examine patients with abdominal GSWs in the initial hours after injury (and therefore these systems may be required to assure timely rescue of patients who fail SNOM). Of those who failed SNOM, the most frequent reasons for delayed laparotomy included development of peritonitis or worsening abdominal tenderness, each of which was mostly commonly detected within the first 24-48 h of admission through close clinical monitoring. It is interesting that 28% of patients who failed SNOM across the included studies underwent unnecessary laparotomy, which suggests that success of SNOM may potentially benefit from studies focused on creating appropriate indications for delayed operation during SNOM [72].
Our findings identified that mandatory use of abdominopelvic CT was associated with a failure rate of SNOM that was approximately half of that reported for selective use of CT scanning. Further, the number of unnecessary laparotomies in patients who failed SNOM was higher in studies that used a selective policy of CT scanning. These findings suggest that use of CT may confer a higher degree of confidence to surgeons managing patients with an equivocal clinical exam or concerning bullet trajectory.  Whereas CT may be argued to be of minimal help to patients with anterior abdominal GSWs who are clearly hemodynamically stable and have no abdominal tenderness, there are situations in which CT will provide much-needed, additional information to guide surgical decision-making. These include patients with suspected tangential abdominal wounds, back or flank GSWs who may have retroperitoneal injuries, or RUQ GSWs who may have isolated hepatic trauma. These findings may support the level 2 recommendation made by the Eastern Association for the Surgery of Trauma in their 2010 guideline that abdominopelvic CT be strongly considered in patients undergoing SNOM of penetrating abdominal wounds (especially those with GSWs to the flank, back, and RUQ) [13].
Common complications of SNOM, as reported across the included studies, were frequently relatively minor and may be divided into those that are thoracic (pneumothorax, hemothorax, or pleural effusion) and abdominal (GSW infection and biliary fistulae in patients with GSWs to the liver). As the risk of complications in patients undergoing mandatory laparotomy has been reported to be as high as 22-41% [4,5], the frequency of complications in patients undergoing SNOM in experienced trauma centers (13% across 11 studies) may be similar or even lower than mandatory laparotomy. However, when patients develop complications after SNOM for abdominal GSWs, their mean hospital LOS appears to increase from approximately 2 (in patients without associated extra-abdominal injuries) or 6 (in patients with associated extra-abdominal injuries) to 10 days. In particular, SNOM of liver GSWs was associated with a risk of biliary fistulae (6.3%) and pulmonary complications (15.9%).
This systematic review has several limitations. First, we identified no controlled studies comparing outcomes of conducting SNOM versus mandatory laparotomy in hemodynamically stable patients with abdominal GSWs without diffuse peritonitis. Second, although we were unable to determine the amount of duplication of included patient outcome data with certainty, a sensitivity analysis excluding the results of studies that reported potentially overlapping data yielded a similar pooled estimate of the failure of SNOM. Third, few studies examined associations between the use of angioembolization/endovascular interventions and SNOM failure or provided details of whether reported complications were directly attributable to a failure of SNOM. Fourth, some studies did not report their methods of conducting SNOM. Thus, the management of each case was ultimately dictated by attending surgeons, which likely introduced unmeasured variability in practice and outcomes. Finally, some studies included patients in the SNOM group that only had tangential and superficial abdominal GSWs without peritoneal breach, which likely improved the reported outcomes of SNOM.

Conclusion
In conclusion, this systematic review and meta-analysis of 41 cohort studies involving 22,847 patients with abdominal GSWs (of whom 6777 patients underwent SNOM) suggests that SNOM is safe when conducted in experienced trauma centers. The practice may be especially useful when serial physical examinations are performed by attending trauma surgeons and in patients with GSWs involving the right thoracoabdomen, flank, or back instead of the anterior abdomen and those proven to involve the kidney on imaging. The most frequent reasons for delayed laparotomy in patients undergoing SNOM include development of peritonitis or worsening abdominal tenderness, each of which is likely to be detected within the first 24-48 h of admission across hospitals with the necessary experience and/or resources to ensure timely rescue of patients who fail SNOM. Mandatory use of abdominopelvic CT may increase the success of SNOM, potentially by increasing clinician confidence when managing patients with an equivocal clinical examination or identifying/characterizing retroperitoneal/isolated hepatic injuries in those with GSWs to the flank, back, or RUQ.