Introduction
Gastrointestinal perforation (GIP) is one of the most common acute abdominal diseases in general surgery. Untreated GIP may lead to life-threatening complications, including acute diffuse peritonitis, sepsis, and septic shock [
1‐
4]. Clinically, GIP above the Treitz ligament is defined as upper gastrointestinal perforation (uGIP), and the lesion distal to the Treitz ligament is defined as lower gastrointestinal perforation (lGIP). Causes are diverse, gastroduodenal ulcer and gastrointestinal tumors commonly cause uGIP, whereas lGIP is often associated with intestinal tumors, Crohn’s disease, ulcerative colitis, and intestinal diverticulosis [
5]. Therefore, it’s important to identify the perforation site.
Currently, imaging technologies, such as abdominal plain film, computed tomography (CT), ultrasound, and magnetic resonance imaging (MRI), alongside patient history and physical examination, are vital for GIP diagnosis. While abdominal plain film offers limited information with a 50–70% diagnostic accuracy, CT’s accuracy reaches 90%, despite requiring higher radiation doses [
6]. Ultrasound sensitivity and specificity depend on operator expertise and are susceptible to gas interference. Although MRI lacks ionizing radiation, its application in GIP remains infrequent due to time constraints [
7]. Traditional laboratory markers, like procalcitonin (PCT) and C-reactive protein (CRP), reflect inflammation levels, aiding in prognosis differentiation between uGIP and lGIP [
8,
9]. Those studies suggest that the etiology, complications, and degree of inflammation burden vary in different perforation sites, which may be reflected by the corresponding levels of laboratory markers. In this study, the integration of clinical features, laboratory markers, and CT imaging was used to construct an effective multi-omics prediction model for the identification of GIP sites.
Discussion
GIP is one of the common surgical acute abdomen-related conditions, which can be fatal in severe cases. It has various etiologies, locations, treatment modalities, and prognoses. The preoperative judgment of the perforation site is very important for the surgical plan. Currently, surgeons mainly rely on the patient history and physical examination combined with imaging techniques to diagnose GIP.
The occurrence of GIP often presents with signs of peritoneal irritation, including abdominal tenderness, rebound pain, and abdominal muscle tension [
10,
11]. Chemical peritonitis caused by gastric fluid, bile, and pancreatic fluid entering the peritoneal cavity usually results in significant abdominal pain due to uGIP in the early stage. However, the irritating symptoms of chemical peritonitis caused by lGIP may be non-significant owing to the differences in bowel contents and flora. On the other hand, bacterial peritonitis can be predominant in the early stage [
12,
13]. In the univariate regression analysis, the duration, tenderness, rebound tenderness, and tension were significant. Whereas in the multivariate regression, only the duration and rebound tenderness were significant. This may be attributed to the false positive peritoneal stimulation signs in the patients with tension or involuntary contraction of the abdominal muscles, or false negative peritoneal stimulation signs due to the reduced reactivity caused by increasing age, infirmities, poisoning, coma, sedation, and analgesia [
11]. CT has been widely used in the diagnosis of GIP in recent years because of its advantages of high-density and spatial resolution compared with abdominal radiographs, among which the distribution of free air outside the lumen is the most important sign [
14,
15]. Due to the anatomical characteristics of the omentum, mesangium, ligaments, and organs in the abdominal cavity, the free gas caused by GIP in different sites has a certain distribution rule. However, the specificity and sensitivity of CT imaging were different. Celik et al. reported that the free air around the portal vein, liver, and stomach is of great significance in gastroduodenal perforation with sensitivities of 86.4%, 54.2%, and 84.7%, and specificities of 40.3%, 69.3%, and 48.4%, respectively [
16]. Cho et al. also reported that the presence of free air around the portal vein indicates a higher possibility of uGIP [
17]. In this study, the transverse mesocolon was used as the boundary. Accordingly, the free gas was classified as above the transverse mesocolon (in the subphrenic space, around the falciform ligament of the liver, around the round ligament, and around the hilar of the liver) or below the transverse mesocolon (around the small mesocolon, ascending colon, sigmoid colon, and mesocolon). The distribution of free gas above and below the transverse mesentery had a significant effect in discriminating the localization of uGIP and lGIP (P
UFA<0.001 and P
UFA + LFA<0.001, respetctively), which was similar to the aforementioned results.
The acquisition of laboratory parameters was relatively simple, fast, and inexpensive. In this study, four laboratory indicators, namely MC, MPV, ALB, and FIB, were included in the prediction model.
Monocytes and/or macrophages are important non-specific immune cells in the body. GIP causes severe peritoneal cavity inflammation, leading to the depletion of defense mechanisms such as peritoneal macrophages, neutrophils, and complement aggregation. It also aggravates the inflammatory load and coagulation load of the body [
12,
13]. In this study, MC (OR, 0.353;
P = 0.012) was found to be an important predictor of upper and lower digestive tract perforation, and the prediction model showed that MC level in uGIP is higher than that in lGIP.
MPV is also one of the commonly used indicators of inflammation. Studies have found that the MPV level in the patients with a positive blood culture is higher than that in the patients with a negative blood culture [
18,
19], suggesting that the increase in MPV level is a sign that the patients are progressing from having local infection to systemic infection, thus, it can be used to evaluate the severity and prognosis of sepsis. The results of this study show that MPV (OR, 5.859;
P = 0.006) is one of the prognostic factors for digestive tract perforation, which may be related to systemic inflammation caused by secondary infection. Excessive inflammation causes over-activation of platelets, which in turn results in over-consumption of platelets. Bone marrow produces more platelets as a compensatory mechanism to supplement the over-consumption of platelets. The platelets that are produced by over-activation undergo changes in their morphology and function, which can be reflected by MPV [
20].
FIB can be used as a molecular marker for hypercoagulability and thrombosis. Over-activated inflammatory reactions and coagulation disorders will result in a vicious coagulation cycle [
21,
22]. Our study showed that lGIP was more likely to occur in the patients with GIP having FIB levels ≥ 4.920 g/L than those having FIB levels < 4.920 g/L. Recently, studies have reported that there are high FIB levels in pancreatic cancer, colorectal cancer, and other gastrointestinal system malignancies [
23,
24], suggesting that the baseline FIB level may be high in patients with tumor-induced lGIP.
ALB, a negative acute phase protein produced by the liver, is a traditional nutritional and inflammatory marker. After abdominal infection, the absorption of endotoxin increases, thus stimulating the production of inflammatory mediators such as TNF, IL-1, and IL-6 in liver macrophages and inhibiting the translation of the ALB transcript, ultimately leading to hypoalbuminemia [
25]. The results of this study showed that ALB (OR, 0.279;
P = 0.006) was an important predictor for distinguishing between uGIP and lGIP, and the ALB level in patients with lGIP was lower than that in patients with uGIP. On the one hand, there are differences in the basic nutrition of these patients. The etiology of uGIP is mainly attributed to ulcerative diseases, but rarely cancer [
3]. In lGIP, small bowel perforation is often caused by intestinal ischemia or inflammatory bowel Crohn’s disease. Colorectal perforation induced by colorectal cancer and diverticulitis is relatively common [
26]. Furthermore, perforation that occurs in colorectal cancer is considered as a late-stage complication, and patients with colorectal cancer mostly have changes in protein metabolism, which are mainly manifested as skeletal muscle atrophy, hypoproteinemia, and other manifestations of cachexia [
27,
28]. On the other hand, the pathophysiological mechanism of infection caused by GIP in different sites varies according to the microenvironment and microflora. Gram-positive
cocci are most frequently detected in patients with uGIP and can play an important pathological role through their virulence factors such as capsular polysaccharide, exotoxin, extracellular enzyme, and adhesin [
29]. However, gram-negative
Bacilli and anaerobic bacteria are often detected in patients with lGIP. They mainly exert their pathological role through their toxic factors such as endotoxin, adhesion, and immunomodulatory molecules [
30].
Nomogram has been widely used for the risk prediction and prognosis assessment of malignancies and chronic diseases [
31‐
35]. The complex multi-factor logistic regression prediction model was transformed into a visual graph through analysis and integration, and was used to predict the GIP. The AUC of the training set was 0.886, and that of the test set was 0.943, suggesting that the model had a high ability to distinguish between uGIP and lGIP. Meanwhile, the calibration curves of both the training set and the test set in this study showed good consistency between the predicted probability of the model and the actual probability. As an emerging method of evaluation and prediction, DCA provides decision evaluation by comparing the size of net benefit values [
36‐
39]. The DCA curves of the training and the test set of the prediction model indicate that the prediction value of the model has good validity. Furthermore, the patients were divided into three groups and the severity of the condition of GIP was evaluated based on whether or not the patient was admitted to the ICU. The number of patients who were admitted to the ICU was the least in the lowest scoring group and the highest in the highest scoring group. Moreover, the length of hospital stay and hospitalization costs in the high-risk group also increased significantly with the increasing scores. Thus, more attention should be paid to patients in high-risk groups when applying this model to predict the perforation site. For critical patients with poor basic condition, damage control surgery should be urgently adopted, and staging surgery is also feasible.
There are some shortcomings in this study. First, it is a single-center retrospective study. A multi-center prospective study is needed to evaluate the actual performance of the prediction model. Secondly, the predictive effectiveness of the column graph prediction model needs to be verified using more data, especially in multi-center large-sample cohorts. Third, the model should indicate contraindications and applicable constraints as far as possible. If data for certain types of patients (such as immunosuppressed patients) are not included, the model may not be applicable in this population. Fourth, there is no data on patient complications, mortality, re-hospitalization rate, post-discharge follow-up records, etc., so the model could not provide a reference for patient prognosis.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.