Alimentary Tract
A simplified prognostic model to predict mortality in patients with acute variceal bleeding

https://doi.org/10.1016/j.dld.2017.11.006Get rights and content

Abstract

Background

Acute variceal bleeding (AVB) is a major cause of death in patients with liver cirrhosis. The aim of this study was to investigate mortality predictors and develop a new simple prognostic model using easily verified factors at admission in AVB patients.

Methods

Between January 2009 and May 2015, 333 consecutive patients with AVB were included. A simplified prognostic model was developed using multiple logistic regression after identifying significant predictors of 6-week mortality. Mortality prediction accuracy was assessed with area under the receiver operating characteristic (AUROC) curve. We compared the new model to existing models of model for end-stage liver disease (MELD) and Child–Pugh scores.

Results

The 6-week overall mortality rate was 12.9%. Multivariate analysis showed that C-reactive protein (CRP), total bilirubin, and the international normalized ratio were independent predictors of mortality. A new logistic model using these variables was developed. This model’s AUROC was 0.834, which was significantly higher than that of MELD (0.764) or Child–Pugh scores (0.699). Two external validation studies showed that the AUROC of our model was consistently higher than 0.8.

Conclusions

Our new simplified model accurately and consistently predicted 6-week mortality in patients with AVB using objective variables measured at admission. Our system can be used to identify high risk AVB patients.

Introduction

Patients with cirrhosis are at risk for developing one or more critical complications, such as ascites, spontaneous bacterial peritonitis, encephalopathy, or hepatorenal syndrome [1], [2], [3], [4]. When these are accompanied by multiple organ failure and culminate in high short-term mortality, this becomes a distinct disease entity known as acute-on-chronic liver failure (ACLF) [5], [6]. ACLF is an increasingly recognized disease entity. It has been postulated that ACLF could allow for early identification of patients at high risk for cirrhosis-related death.

Acute variceal bleeding (AVB) is a major complication of portal hypertension in patients with cirrhosis. Endoscopic ligation therapy combined with vasoactive drugs and prophylactic antibiotics is the current standard of care for AVB patients [7], [8]. Despite advances in diagnosis and management of esophageal and gastric varices, AVB is a main cause of upper gastrointestinal bleeding (UGIB). Mortality remains high in this situation (16%–24%) [7], [9]. Development of sensitive and specific risk prediction models for AVB patients is important for reducing mortality in high-risk patients. Early transjugular intrahepatic portosystemic shunt (TIPS) use in selected high-risk patients reduces mortality [10], [11]. Furthermore, rigorous application of treatments such as restricted blood transfusion or nonselective beta-blockers can improve AVB patient survival [12].

Although there are several AVB prognostic models, they have limited ability to predict patient outcomes. Rockall and Glasgow Blatchford scores are widely used UGIB risk predictors. However, these scores are poor at predicting clinical outcomes of patients with AVB [13]. The Child–Pugh score has subjective components that are inconsistently predictive, such as ascites or encephalopathy [14]. Model for end-stage liver disease (MELD) score is composed of objective variables [15]. However, this model was developed based on patients treated with TIPS more than 10 years ago and so might not apply to AVB patients. The chronic liver failure-sequential organ failure assessment (CLIF-SOFA) score has received the most attention recently and adequately predicts ACLF in chronic liver disease patients [6]. However, this score is complex and includes subjective components.

The aim of the present study was to develop a new and simple prognostic model based on initial objective components in AVB patients. To achieve this aim, we first investigated mortality predictors in AVB patients. Based on this analysis, a new and simple prognostic model was produced using exclusively objective and easily verified factors. Finally, this new model was validated externally to ensure generalizability and implementation in clinical practice.

Section snippets

Study population

A consecutive database that included all patients admitted to Seoul St. Mary’s Hospital, a tertiary care center, for acute gastrointestinal (GI) bleeding from January 2009 to May 2015 was created. This study was approved by Seoul St. Mary’s Hospital Institutional Review Board (KC14RISI0606). From the total GI bleeding patient cohort, those who were found to have cirrhosis and variceal bleeding were identified and selected for this study.

Patients with cirrhosis and acute bleeding from both

Patients features and outcomes

From January 2009 to May 2015, 343 consecutive patients with AVB who were admitted to Seoul St. Mary’s Hospital were identified. After excluding those who were under 18 years of age (n = 1), who did not undergo EGD (n = 2), and those lost to follow-up within 6 weeks from the initial endoscopic exam (n = 7), 333 patients remained and were included.

The demographic characteristics of the studied population are summarized in Table 1. The mean age was 57.5 years (range: 22–98) and there were 247 men.

Discussion

Although several prognostic models have been developed to predict outcomes of patients with AVB, these are not widely used because some indicators in the models are subjective and difficult to assess at AVB onset. In the present study, independent predictive factors of 6-week mortality were identified, and a simple yet novel prognostic model calculated from a cohort of patients with AVB was proposed. This model was validated externally in two additional patient series. All model components

Conflict of interests

None declared.

Funding

This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (Ministry of Science, ICT & Future Planning) (NRF-2016R1C1B2015185) and by Leading Foreign Research Institute Recruitment Program through the NRF funded by the Ministry of Science, ICT and Future Planning (NRF-2011-0031644).

Acknowledgement

The authors thank Soung Won Jeong, MD, Associate Professor at Soonchunhyang University College of Medicine, for discussions regarding the data analysis.

References (36)

  • O. Waidmann et al.

    Macrophage activation is a prognostic parameter for variceal bleeding and overall survival in patients with liver cirrhosis

    J Hepatol

    (2013)
  • V. La Mura et al.

    Liver sinusoidal endothelial dysfunction after LPS administration: a role for inducible-nitric oxide synthase

    J Hepatol

    (2014)
  • I. Cirera et al.

    Bacterial translocation of enteric organisms in patients with cirrhosis

    J Hepatol

    (2001)
  • R. Wiest et al.

    Spontaneous bacterial peritonitis: recent guidelines and beyond

    Gut

    (2012)
  • F. Salerno et al.

    Diagnosis, prevention and treatment of hepatorenal syndrome in cirrhosis

    Gut

    (2007)
  • R. Moreau et al.

    Acute-on-chronic liver failure is a distinct syndrome that develops in patients with acute decompensation of cirrhosis

    Gastroenterology

    (2013)
  • S. Augustin et al.

    Effectiveness of combined pharmacologic and ligation therapy in high-risk patients with acute esophageal variceal bleeding

    Am J Gastroenterol

    (2011)
  • A. Monescillo et al.

    Influence of portal hypertension and its early decompression by TIPS placement on the outcome of variceal bleeding

    Hepatology

    (2004)
  • Cited by (17)

    • Risk stratification in acute variceal bleeding: Far from an ideal score

      2021, Clinics
      Citation Excerpt :

      Additionally, advanced cirrhosis is a risk factor for bacterial translocation (increased intestinal permeability, impaired reticuloendothelial function, and decreased complement factor synthesis, allowing passage of bacteria and bacterial components of intestinal origin into the bloodstream, predisposing to infection). Infection can increase portal hypertension due to increased vasoconstrictor production, initiating a vicious cycle (23). In a previous study carried out on our institution from March 2010 to April 2013, among 164 patients with AVB, MELD scores >15 were shown to be strongly associated with death.

    • A partial imputation EM-algorithm to adjust the overestimated shape parameter of the Weibull distribution fitted to the clinical time-to-event data

      2020, Computer Methods and Programs in Biomedicine
      Citation Excerpt :

      When this kind of clinical data is used for time-to-event modeling by pharmacometricians, the final model will also carry the bias. This study was initially motivated by an accidental discovery that the maximum likelihood estimators (MLE) of the shape parameter were overestimated in a cohort study [3] based on the Weibull survival model (Fig. 1). In that study, approximately 400 patients with upper gastrointestinal bleeding visited the emergency room, and their time to second bleeding event was recorded on daily basis.

    • Validation of a Machine Learning Model That Outperforms Clinical Risk Scoring Systems for Upper Gastrointestinal Bleeding

      2020, Gastroenterology
      Citation Excerpt :

      For example, the largest study of ML in UGIB used 2380 patients and found that a neural network model had improved performance over the full Rockall score (which included endoscopic findings) for 30-day mortality on internal validation only, but this study had no external validation of the model and no assessment of more clinically relevant pre-endoscopic clinical risk scores or the clinically important composite outcome.42 For mortality, other ML models have a trend toward better performance than clinical risk scores, although only 3 studies compared ML models to clinical risk scores designed to assess risk in gastrointestinal bleeding, and only 1 of them compared to a pre-endoscopic clinical risk score, which is the appropriate comparator for risk stratification of gastrointestinal bleeding.42–44 Two studies examined the prediction of mortality on external validation with ML models compared to liver-specific Child-Pugh and Model for End-Stage Liver Disease scores (designed to predict mortality in all patients with cirrhosis rather than in those presenting with gastrointestinal bleeding), and only 1 study found improved performance.45

    View all citing articles on Scopus
    View full text