We observed a sepsis rate of 10.5% after GI endoscopy in cirrhotic patients waiting for LT. Decompensated cirrhosis (Child-Pugh B and C classifications) and HH were important risk factors in the experience cohort. Drainage of HH and routine antibiotic prophylaxis before endoscopic procedures in the validation cohort reduced the incidence of sepsis to 1.3% with significant improvement regarding the Child-Pugh B and C categories. Bacterial infections are diagnosed in 40% of in-hospital cirrhotic patients and lead to a four-fold increase in mortality [20]. The literature also pointed out that sepsis and bacterial infection are recognized as distinct stages in the progression of chronic liver disease that speed up the organ failure and contribute to the high mortality [20].
We used the definition of sepsis from the criteria of SIRS for documented or suspected infection [15, 16], which has high sensitivity and low specificity compared to the new sepsis definition based on organ dysfunction using Sequential Organ Function Assessment (SOFA) [21]. Nevertheless, Julian et al. reported that SIRS is associated with mortality and organ failure with relative risk of 3.2 and 3.5, respectively, and that it is not inferior to the SOFA criteria. These results were obtained in a large prospective study comprising 8871 patients [21]. Moreover, even in the absence of bacterial infection, the occurrence of SIRS in cirrhotic patients has prognostic relevance, which might aggravate portal hypertension, renal failure, and hepatic encephalopathy, thereby contributing to multi-organ failure and mortality [22]. Therefore, the high sensitivity of SIRS has fundamental clinical importance for identifying sepsis early and applying effective treatment to minimize complications in cirrhotic patients who have high infection risk.
The severity of liver disease is associated with increased risks of bacterial infection and correlated with mortality after infection [23]. Numerous variables were associated with mortality after bacterial infection in a meta-analysis by Arvaniti et al. [23]. There were five variables related to the severity of cirrhosis (CTP score, PT, bilirubin, albumin, and MELD score) and three related to cirrhosis-associated complications (encephalopathy, GI hemorrhage, and HCC). Similar to their findings, several parameters in our study associated with liver function were found to be relevant to post-endoscopy sepsis in the univariate analysis (HCC, MELD score, Child score, Child-Pugh classification, ascites, HH, platelet count, PT, and serum albumin and bilirubin levels). However, only the CTP score, Child-Pugh classes B and C, HH, and no use of prophylactic antibiotics were risk factors for sepsis in the multivariate analysis (Table 2).
The MELD score consists of three variables: serum International Normalized Ratio (INR), total bilirubin level, and creatinine level. However, the CTP score is calculated using five variables: ascites, encephalopathy, PT, and serum levels of bilirubin and albumin. The CTP score uses more clinically significant parameters (albumin, bilirubin, PT, and ascites) than the MELD score (bilirubin, INR) to predict the endoscopy-associated sepsis, which makes using the CTP score for prediction more favorable. Moreover, our data showed that decompensated cirrhosis (Child-Pugh B and C) is more predisposing for sepsis compared with Child-Pugh A compensated liver cirrhosis, and the cutoff point of the CTP score was over 8.5 (Figure 2), which also corresponds to Child-Pugh classes B and C.
Cirera et al. reported that the prevalence of bacterial translocation significantly increased according to the Child-Pugh classification, with 3.4% for Class A, 8.1% for Class B, and 30.8% for Class C patients (x2= 6.106, P < 0.05). This may explain the occurrence of sepsis or bacteremia following GI endoscopy correlating with the severity of cirrhosis in our results [24]. Another two variables, platelet count and HCC, also reflect the complexity of liver cirrhosis. Thrombocytopenia is a marked feature of chronic liver disease and cirrhosis, especially in patients with hepatitis B and C infections compared to other causes of chronic liver disease [25]. In our cohort, the LT candidates who also had HCC had less severe liver dysfunction than other patients without HCC (CTP score in HCC group: 7.0; non-HCC group: 9.2; p <0.001). This may explain the presence of HCC being associated with lower risk of infection.
In our study, the occurrence of HH was 3.6% (23/642) during 2008 to 2016, which is slightly lower than the incidence rate of 5-10% reported by other studies [18]. This difference could have occurred because our patients were LT candidates, including HCC groups without portal hypertension or cirrhotic change. In the total of 23 cases presenting with HH, only 11 patients (47.8%) also had massive ascites greater than one liter and showed poor response to diuretics. In the univariate and multivariate analyses, HH was a significant risk predisposing factor for sepsis associated with GI endoscopy (OR= 4.69, P= 0.016), but none of our patients presented with pulmonary symptoms or empyema or pneumonia.
In the experience cohort, HH was not drained in all patients, and the sepsis rate was 38.4% (5/13). However, in the validation cohort, HH was recognized as a risk factor for post-endoscopy sepsis and poor response to diuretics and fluid restriction. Thus, all patients with HH received antibiotic prophylaxis and chest pigtail drainage prior to endoscopy procedures, so no patient experienced sepsis. In the 77 cases of HH reviewed by Ricardo et al., the average time from presentation of HH to death was 368 days. Thus, the outcome was extremely poor in these groups except for those undergoing trans-jugular intrahepatic portosystemic shunt or LT [14]. Moreover, when analyzing the cause of mortality in the same group of patients, sepsis was the most common etiology (37%) [14]. This result correlates with our findings and indicates that HH is a marker of ESLD associated with high infection and mortality rate. Thus, drainage of HH and early LT are recommended to prevent complication.
Technical factors may have been related to septic complications after GI procedures. Several studies concluded that factors have not been associated with bacteremia, such as the underlying bowel pathology, duration of the procedure, or performance of endoscopic biopsies or polypectomy [26]. In our study, neither biopsy nor polypectomy in upper and lower GI endoscopy induced sepsis (Table 3).
Our patients received combined EGD and CFS, and it could not be differentiated whether sepsis came from the upper or lower GI tracts. In prospective studies, upper and lower GI endoscopy was associated with bacteremia rates of 4%, and the observed bacteremia usually was short lived and not caused infectious adverse events [27, 28]. Although reports of infectious sequelae are rare, there are several case reports of bacteremia and mortality following colonoscopy in cirrhotic patients [10, 11, 29]. The prospective study by Josep et al. [30] concluded that lower intestinal endoscopy does not induce bacteremia in cirrhotic patients in the absence of GI bleeding and recommended against the routine antibiotic prophylaxis. That study recruited 58 cirrhotic patients with 28 and 21 patients categorized as Child-Pugh B and C status, respectively. Compared with previous the study, our study enrolled more decompensated liver cirrhotic patients, including 259 Child-Pugh B and 159 Child-Pugh C patients out of a total of 642 LT candidates. Moreover, all of our patients stayed in the hospital after the procedure for close monitoring of their clinical condition and to obtain a more precise record of vital signs.
For screening, adequate bowel preparation is required to ensure colonic cleaning. A solution with PEG is safer than oral sodium phosphate for low risk of renal injury, especially in cirrhotic patients who have the potential for hepatorenal syndrome. Liver cirrhosis was a strong predictor for poor bowel preparation, probably due to ascites and general weakness, which make patients unable to tolerate too much fluid intake [31]. Therefore, we used a low volume of PEG (2 L) rather than a standard volume (4 L) for colon cleaning. Although the cause of post endoscopic sepsis was unknown in our study, bowel preparation does not enhance bacterial translocation or even sepsis according to several studies [32].
The risk of pulmonary aspiration during sedative EGD has been reported and may be up to 3.9% [33]. Concerning this potential adverse event, chest X-ray was performed for every sepsis patient in our study, but none of them was diagnosed with aspiration pneumonia. This may be explained by the minimal residual gastric food after fasting for more than eight hours and the stand-by of specialized anesthesia doctors and assistants providing sufficient sedation and continuous saliva suction.
According to the ASGE guidelines for GI procedures from 2015, prophylactic antibiotic administration is recommended for percutaneous endoscopic gastrostomy, endoscopic ultrasound-guided fine needle aspiration, endoscopic retrograde cholangio-pancreatography, and patients with cirrhosis admitted for GI bleeding, and which should receive antibiotic therapy with third-generation cephalosporin [9]. Based on this, we used third-generation cephalosporin (ceftazidime, ceftriaxone, or flomoxef) as a prophylactic antibiotic agent. Flomoxef was used more often in our practice for the additional coverage of anaerobic infection compared with ceftazidime and ceftriaxone.
The four positive blood cultures in the experience cohort indicated Micrococcus, S. aureus, E. coli, and A. sobria bacteremia. In the validation cohort with overall antibiotic prophylaxis, there were only four cases of post-endoscopic sepsis, and the culture revealed negative results. An association between bacteremia and antibiotics was also evident in our study (0% vs. 1.3% in the groups with and without antibiotic prophylaxis, respectively; P=0.032).
Limitations of the study
This study had several limitations. Firstly, the retrospective data may have impacted the identification of some confounding factors. Secondly, the study was conducted in a single medical center, and sepsis associated with GI endoscopy may vary in different hospitals. Thirdly, because EGD and CFS were performed together, it was difficult in differentiate sepsis coming from single and combined procedures. Finally, the study recruited only 642 patients. Therefore, a large, prospective, randomized, controlled trial is required to study the incidence of sepsis after GI endoscopy in cirrhotic patients.