Skip to main content
eScholarship
Open Access Publications from the University of California

Prediction Model for 30-day Outcomes Among Emergency Department Patients with Lower Gastrointestinal Bleeding

  • Author(s): Ramaekers, Rosa
  • Perry, Jeffrey
  • Leafloor, Cameron
  • Thiruganasambandamoorthy, Venkatesh
  • et al.
Abstract

Introduction: There are currently no robust tools available for risk stratification of emergency department (ED) patients with lower gastrointestinal bleed (LGIB). Our aim was to identify risk factors and develop a preliminary model to predict 30-day serious adverse events among ED LGIB patients.

Methods: We conducted a health records review including adult ED patients with acute LGIB. We used a composite outcome of 30-day all-cause death, recurrent LGIB, need for intervention to control the bleeding, and severe adverse events resulting in intensive care unit admission. One researcher collected data for variables and a second researcher independently collected 10% of the variables for inter-observer reliability. We used backward multivariable logistic regression analysis and SELECTION=SCORE option to create a preliminary risk-stratification tool. We assessed the diagnostic accuracy of the final model.

Results: Of 372 patients, 48 experienced an adverse outcome. We found that age ≥75 years, hemoglobin ≤100 g/L, international normalized ratio ≥2.0, ongoing bleed in the ED, and a medical history of colorectal polyps were statistically significant predictors in the multivariable regression analysis. The area under the curve (AUC) for the model was 0.83 (95% confidence interval, 0.77-0.89). We developed a scoring system based on the logistic regression model and found a sensitivity 0.96 (0.90-1.00) and specificity 0.53 (0.48-0.59) for a cut-off score of 1.

Conclusion: This model showed good ability to differentiate patients with and without serious outcomes as evidenced by the high AUC and sensitivity. The results of this study could be used in the prospective derivation of a clinical decision tool.

Main Content
Current View