Negative Surgical Exploration in Patients Suspected GIP: Trend, Preoperative Predictors, Etiologies, and Outcomes

BACKGROUND AND OBJECTIVE: Despite the rapid improvement of clinical science and imaging technology including computed tomography, the entity of negative surgical exploration in suspected gastrointestinal perforation (N-GIP) remains existing. However, few studies have focused on this issue and most studies are case reports. We undertook this study to investigate the rates of N-GIP, and explore a set of possible preoperative predictors associated with N-GIP. METHODS: This was a retrospective study performed at the department of general surgery in our treatment center. All patients included were suspected GIP cases, aged 14 years and over, and underwent emergency surgery between 2009 and 2019. A predictive multivariable model of the presence of N-GIP was developed using logistic regression analysis. RESULTS: A total of 973 patients were identied and 30 (3.1%) were found to have no evidence of perforated gastrointestinal tract. The mean age of patients was 59.74 (range 14–97) years and 67.2 percent of the patients were males. The rates of N-GIP did not have a signicant change over time (P =0.212 for trend). In multivariable analysis, absence of generalized peritonitis, duration of abdominal pain > 19.6 hours, and neutrophil-to-lymphocyte ratio < 3.80 were signicant predictors of N-GIP. N-GIP was more common in patients with gastrointestinal tumors and foreign bodies. Five patients (16.7%) in N-GIP group experienced complications and the 90-day mortality rate was 6.7%. CONCLUSION: The rates of N-GIP did not change signicantly over time. N-GIP was associated with the absence of generalized peritonitis, duration of abdominal pain > 19.6 hours, and neutrophil-to-lymphocyte ratio < 3.80. inclusion criteria for our study. Percentage of neutrophils, NLR were frequently decreased and signs of peritonitis were less serious for patients with N-GIP. The N-GIP group had statistically signicant higher percentage of lymphocytes and longer duration of abdominal pain than patients with GIP. We found no statistically signicant differences between the N-GIP and GIP groups for age, sex, ASA grade, shock, temperature and white blood cell count.


Introduction
Gastrointestinal perforation (GIP) is de ned as the destruction of integrity of the gastrointestinal tract. GIP is usually presented as an acute abdominal condition, with localized or generalized peritonitis. A large percentage of all hospital admissions for emergency-surgery in General Surgery Department are secondary to perforation of abdominal viscera (1). GIP is a surgical emergency that carries high risk for mortality (30 ~ 50%) in spite of improvements in antibiotics, radiographic imaging, resuscitation therapy, and surgical technique (2,3). Many individuals who suffer from GIP develop SIRS or severe sepsis accompanied by hemodynamic compromise, coagulopathy, acidosis, and hypothermia (4,5). As a lifethreatening disease, (especially in elderly patients), GIP usually needs emergency surgery and active rescue in surgical ICU. GIP can occur secondary to various conditions such as iatrogenic and traumatic causes, neoplasms, in ammation, or peptic ulcers. In general, many factors may contribute to GIP, including older age, diabetes, glucocorticoid therapy, antecedent diverticulitis, and usage of bevacizumab, tofacitinib, tocilizumab, non-steroidal anti-in ammatory drugs (NSAIDs) or other biologic agents (6)(7)(8).
Successful treatment requires a complete understanding of the pathophysiology, anatomy, and microbiology of the disease process and thorough knowledge of the therapy, such as antibiotics, source control, physiologic support and resuscitation (2). Laparotomy and laparoscopic surgery for patients clinically suspected of having GIP are the most effective and important treatment methods (9,10). However, such procedures are rarely performed in the absence of perforation upon macroscopic or microscopic examination, a circumstance called as negative surgical exploration in suspected GIP (N-GIP). N-GIP may lead to a palpable degree of morbidity and mortality, an appreciable prolongation of hospital stay, and complications because of postoperative infection. It is also a main contributor to medical con ict and healthcare costs.
Radiographic evaluation of GIP traditionally began with upright and supine abdominal radiographs.
Currently, helical CT has become a preferred tool to assess patients suspected of having GIP (11). Oral or intravenous contrast agents are used to facilitate identi cation of perforation when feasible. Accurate and con rmative preoperative diagnosis of GIP remains challenging despite proper history taking, clinical examination, and improvement in new imaging techniques like ultrasound imaging and computed tomography. Preoperative diagnosis of GIP is very crucial to minimize the morbidity and mortality in densely populated countries like China. As management errors can be signi cantly reduced by correct preoperative diagnosis, clinical acumen plays a pivotal role in the diagnosis and management of GIP and ensures de nitive per-operative surgical treatment, which in turn minimizes negative laparotomy or laparoscopy. Thus, surgeons need to build up diagnostic acumen and decision-making in the management of GIP.
Regardless to the development of evidence-based guidelines and medical technology, it remains unclear whether a reduction in rates of N-GIP has been achieved over time. Until recently, far too little attention has been drawn to this issue and the overall problem of N-GIP has remained refractory. The aim of the present study was thus to analyze the characterization of its prevalence and clinical predictors, summarize the causes of N-GIP so as to provide objective basis for surgeons' decision-making.

Methods
The approval of the study was obtained from the Hospital Ethical Committee before starting. We performed a retrospective analysis of clinical suspected GIP cases that underwent an emergency surgery on the same admission to Zhongshan hospital of Fudan university between 2009 and 2019. Patients who either had an elective surgical exploration or were younger than 14 years of age at the time of operation were excluded from this study.
The preoperative data recorded were as follows: demographic information (age and sex), clinical symptoms at presentation (abdominal pain) and clinical signs (sign of peritonitis), duration of abdominal pain, reports from abdominal US, abdominal radiographs, CT, MRI, Endoscope. The white blood cell count (WBC), percentage of neutrophils, percentage of lymphocytes, neutrophil-to-lymphocyte ratio (NLR), Creactive protein (CRP), procalcitonin (PCT), D-dimer, and preoperative vital signs (temperature, shock), ASA grades were also collected. Operative factors such as surgical approach and operative ndings were obtained as well by reviewing of the operative notes. Postoperative data included any complications recorded within the 90-day postoperative period and histopathological ndings.
Considering laboratory results and vital signs uctuating of patients during the course of medical attention, we used the highest recorded value before surgical exploration for our analysis, or "peak preoperative" value, rather than using presented values at certain times.
In this study, N-GIP was de ned as a macroscopically and/or microscopically scrutinized normal Gastrointestinal tract in a patient who underwent emergency operation based on clinical diagnosis and imaging suspicious for GIP.

Statistical Analysis
Descriptive statistics are reported as means and standard deviation, median with interquartile range, or as otherwise indicated. Differences in continuous and categorical data between groups were analyzed by means of Student's t-test, Chi-square test and Mann-Whitney U-test respectively. Rates of outcomes were tested for changes over time using χ2 test for trend.
Multivariable model predicting the presence of N-GIP was performed with demographic and clinical factors found to be statistically signi cant in univariable analysis in a machine learning model by gradient boosting decision tree model (catboost) from our model development cohort (70% of patients). A validation set of the remaining 30% of patients was built to test the model. After the model training was completed, the SHAP value of each feature was calculated to re ect the importance of the feature, and the relationship between the feature and the outcome variable was explored. The SHAP value can guide the grouping of continuous variables, which can be incorporated into the logistic regression model.
The predictors of N-GIP were identi ed using a binary logistic regression modelling technique. Odds ratios (OR) and 95% con dence intervals (CI) were determined in the regression models. Overall model t was assessed by estimating the overall likelihood ratio χ2, the Hosmer-Lemeshow goodness-of-t test, and the C-statistic to measure the discrimination of the model.
A signi cance level of α = 0·05 was used for all tests. All P values were two-sided. All demographic and clinical data of each patient were analyzed by SPSS, version 20.0 for Windows (IBM, Analytics, Armonk, NY, United States) and Python programming language, version 3.6 (Python Software Foundation, https://www.python.org/).

Patients and Population
From 2009 to 2019, 973 patients were admitted to our department with a preoperative diagnosis of GIP. The mean age at the time of the procedure was 59.74 (range 14-97) years and 67.2 percent of the patients were males. Median ASA grade was II (range I-IV). In all patients, GIP or N-GIP was nally proved by operative and/or histopathological ndings. All patients suffered from abdominal pain. Locations of GIP and sites of N-GIP revealed during surgery and/or in histopathological ndings are displayed in Fig. 2. N-GIP appeared to occur more frequently in duodenum and right-sided colon.

Preoperative Data
Of the 973 patients, 898 (92.3%) patients underwent CT scanning and 68 patients underwent abdominal radiographs (among them 16 patients also underwent CT scanning) during the initial evaluation before surgery, and pneumoperitoneum was found in 890 patients. Other examinations in some patients including ultrasonography, MRI, diagnostic abdominal puncture and ndings in endoscope were similarly helpful to diagnosis preoperationally. According to the outcome, the sensitivity of CT and abdominal radiographs were 95.6% and 92.3%. But in 29 patients with N-GIP who underwent CT scanning, only one patient had surely no indications of GIP on CT images while others all had a pneumoperitoneum, focal wall defect or bowel wall thickening.
As CRP, PCT and D-dimer were applied universally and normally in clinical work in recent years, missing data does exist in patients admitted at earlier time points in the observation period, signi cant rates of missing data were 18.3%, 48.1%, 31.6% respectively. These three variables were excluded from the analysis. Table 1 details the demographic and baseline clinical data of the 973 patients who met the inclusion criteria for our study. Percentage of neutrophils, NLR were frequently decreased and signs of peritonitis were less serious for patients with N-GIP. The N-GIP group had statistically signi cant higher percentage of lymphocytes and longer duration of abdominal pain than patients with GIP. We found no statistically signi cant differences between the N-GIP and GIP groups for age, sex, ASA grade, shock, temperature and white blood cell count. Because the proportion of negative samples is too low, the model training was weighted according to the ratio of positive: negative = 1:12 (i.e., the harm of misjudging one negative sample as positive = the harm of misjudging 12 positive samples as negative), which improves the detection of negative samples. In the test dataset, 284 positive patients were detected 100%, and 3 of 8 negative patients were detected. Figure 3 shows the relationship between the value of each feature (blue is low, red is high) and the outcome variable. The importance of the features is ranked from high to low. Signs of peritonitis is the most important, and shock is the least important.
By calculating the Shap value of each continuous variable exceeding 0 point (Fig. 4) Interestingly, the N-GIP group did not have a signi cantly lower rate of complications and mortality than the GIP group (P = 0.131 and 1). Five of the 30 patients (16.7%) with N-GIP and 277 of the 943 patients (29.4%) with GIP had complications. The 90-day mortality in the N-GIP group and GIP group was 6.7% and 6.5%, respectively.

Etiology of N-GIP
We thoroughly reviewed the whole process of disease progression, treatment and postoperative recovery, searched and studied similar case series and eventually identify the causes of N-GIP (Table 3). N-GIP were most often due to tumor and foreign body of the gastrointestinal tract.

Discussion
Precise diagnosis of GIP is still challenging despite improvement in clinical examinations and investigative procedures. If the preoperative diagnosis is indecisive it is suggested by clinical experience to operate for exploration immediately rather than waiting and observation. However, such a strategy will certainly raise the unnecessary surgical exploration rate and the morbidity. Most of the previous researches merely focused on reviewing the causes of nonsurgical pneumoperitoneum and numerous studies were case reports (12)(13)(14)(15)(16)(17).
As early as 1983, Roh and his colleagues (18) reported that 14 percent of patients in their study who had abdominal pain and pneumoperitoneum on plain abdominal radiographs underwent an unnecessary laparotomy owing to the fact that pneumoperitoneum was from sources other than a perforated hollow viscus. They believed that the clinical manifestations of their patients who had surgical versus nonsurgical causes of pneumoperitoneum was indistinguishable. In reality, pneumoperitoneum re ects perforation of abdominal viscus in 85-95% of total occurrences. Whereas in 5-15% of cases, pneumoperitoneum does not re ect visceral perforation and arises from another source that does not demand emergency surgery (19). Nonsurgical pneumoperitoneum is de ned by the existence of air in the peritoneal space that is detectable by roentgenogram and either is treated successfully by observation and supportive care alone or leads to a nondiagnostic laparotomy (20). In this study, the rate of nonsurgical pneumoperitoneum was 2.9%. Case reports and small series (21)(22)(23)(24)  16.7% of the total patients. Appendicitis, gastrointestinal in ammation and spontaneously sealed small perforation respectively constituted 10% of the causes. Therefore, we suggest that when duration of abdominal pain is long-lasting, generalized peritonitis is absent, and neutrophil-to-lymphocyte ratio is in a lower level, the possibility of negative surgical exploration should be considered in preoperative assessment for patients suspected GIP, especially in patients with gastrointestinal cancer. We should implement subsequent investigation with additional preoperative diagnostics for selected patients with risk factors for N-GIP, for example, repeated paracentesis judging for presence of GIP. From our perspective, it is indispensable to perform careful and thorough history and physical examination, and to make judgments based on clinical symptoms and signs, laboratory values, imaging and other examinations.
Unexpectedly, our study demonstrated that there were nearly no differences in the rate of complications and mortality between patients who had GIP versus N-GIP. Five patients (16.7%) in the 30 patients with negative nding in operation experienced a complication and two in them later succumbed to postoperative complications. Four patients in N-GIP group underwent laparoscopic surgery, all had no complications in postoperative period. With the rapid development of minimally invasive concept and techniques in recent years, the application of laparoscopy in the diagnosis and treatment of patients with gastrointestinal perforation has become more frequent (28)(29)(30). However, there are few high-quality clinical studies and many researches are biased. When the di culty of the clinical decision making for patients suspected of having GIP is unsolved, laparoscopic exploration may be a simply good management and of great help to surgeons.
A major limitation of this study is that it was retrospective and patients were selected from a single center, which made the evidence level for this study relatively low. The event rate of only 30 patients with N-GIP over a 11-year period is low and re ective of the relative rarity of this presentation in the worldwide context. However, there is a possibility that patients who did not proceed to laparotomy or laparoscopy may have been presented later to another institution, thereby lessening the rate. Moreover, it is likely that the rate of negative surgical exploration in patients with GIP is lower than other areas with incomparable standards of care as well as the availability of CT. A further limitation is that C-reactive protein, procalcitonin and D-dimer were not analyzed, which are regarded as infection index (31)(32)(33) and might be signi cant predictors of N-GIP. At last, as the number of individuals in GIP group was extremely different from that in N-GIP group, robust identi cation of predictors for N-GIP is limited by the sample size of patients who had negative surgical exploration. Thus, future studies with more universal laboratory testing and larger sample size are needed to verify and generalize our ndings.
Despite these limitations, the present study had many strengths. It is the largest retrospective study evaluating the rate of negative surgical exploration in patients with suspected GIP. Second, the data span an interval of 11 years, longer than in most previous investigations. Finally, the model developed had high predictive value and has good generalizability to other patients undergoing laparotomy and laparoscopy for GIP.

Conclusions
The overall rate of negative surgical exploration in patients with suspected GIP was 3.1% with no signi cant decrease over time. Applicating a multivariable logistic regression, we can easily identify predictive factors of N-GIP and help surgeons select a small portion of patients suspected of having GIP to minimize unnecessary emergency surgery, provide a unique opportunity for conservative management in stable patients, and expedite appropriate care.

Declarations
Ethics approval and consent to participate Not applicable.
Consent for publication  The locations of GIP and sites where N-GIP takes place. Undetermined locations in cases of GIP: Due to the special intraoperative conditions, no further exploration is possible to identify the perforation location though gastrointestinal contents were found in abdomen. Undetermined sites in cases of N-GIP: There were no clear indications of gastrointestinal abnormalities with repeating scrutinization.

Figure 3
The relationship between the value of each feature and the outcome variable. The abscissa represents the contribution of the feature to the outcome variable: less than 0 represents a protective effect, and greater than 0 represents a dangerous effect.