Effectiveness of endoscopy in patients admitted to the intensive care unitwith upper GI hemorrhage

Background: Esophagogastroduodenoscopy (EGD) is generally indicated for the management of patients admitted to intensive care units (ICUs) with upper gastrointestinal (GI) hemorrhage but its impact in community practice has n ot been measured. Thus, the effectiveness of 3 EGD factors, viz., accurate initial diagnosis, performance within 24 hours of admission (early EGD), an d appropriate intervention, was examined. Methods: Records of 214 patients admitted to the ICU of 10 metropolitan hos pitals with upper GI hemorrhage were reviewed. Unadjusted and severity-adju sted associations of the 3 EGD factors with length of hospital stay, length of ICU stay, readmission to ICU, recurrent bleeding, surgery, and death we re evaluated. Results: Inaccurate diagnosis occurred in 10% of patients at initial EGD an d was associated with significant increases in risk of recurrent bleeding ( 70% vs. 11%, p < 0.001), rate of surgery (20% vs. 4%, p < 0.05), length of hospital stay (median 7.5 vs. 5 days, p < 0.005), length of ICU stay (media n 4 vs. 2 days, p < 0.005), and rate of readmission to ICU (20% vs. 0.6%, p < 0.001). These associations persisted after adjusting for severity of ill ness. Early EGD performed in 82% of patients was associated with significan t severity-adjusted reductions in hospital (-33%: 95% CI [-45%, -18%]) and ICU (-20%: 95% CI [-24%, -3%]) stay. Appropriate intervention at initial EG D, performed in 84% of patients, was associated with reductions in severity -adjusted length of ICU stay (-18%: 95% CI [-32%, 0%]) and rate of recurren t bleeding (odds ratio = 0.37, 95% CI [0.13, 1.06]). Conclusions: Early, accurate EGD with appropriate therapeutic intervention is effective as practiced in the community and is associated with improved outcomes for patients with upper GI hemorrhage admitted to the ICU. Inaccur ate diagnosis at initial EGD is uncommon but has a significant adverse asso ciation with all outcome measures.