This study provides a distinctive national perspective on LIB. The large patient population enabled us to estimate the mortality associated with LIB, and to identify multiple adjusted predictors of in-hospital death. An estimated 3.9% of patients with a discharge diagnosis consistent with LIB died while hospitalized in 2002. Factors associated with significantly increased in-hospital mortality were advanced age, intestinal ischemia, comorbid illness, onset of bleeding after hospitalization for a separate disease process (secondary bleeding), coagulopathy, hypovolemia, transfusion(s) of packed red blood cells, and male gender. Patients diagnosed with hemorrhoids, colorectal polyps, and anal fissures were at a decreased risk of in-hospital mortality.
Most prior studies of LIB have assessed relatively small cohorts from individual academic centers, often focusing on specific diagnoses or severely bleeding patients. For these reasons, mortality estimates for LIB have ranged broadly, and predictors of inhospital mortality have been difficult to define. Longstreth studied the epidemiology of LIB in a cohort of 219 patients.6
In this population-based study, the in-hospital mortality was 3.6%, almost identical to what we found; however this estimate was based on only 8 in-hospital deaths. Das and colleagues developed both multiple logistic regression models and artificial neural networks to predict in-hospital mortality, recurrent bleeding and intervention for bleeding control in patients with LIB, 4
and again, the ability to accurately identify predictors of in-hospital mortality was limited by the small numbers of deaths: 4 (3%) in the training group and 4 (6%) and 8 (6%) in the validation groups. When Velayos et al. analyzed predictors of adverse in-hospital outcome in LIB other than severe bleeding, only 3 patients (3%) died;18
ongoing or recurrent bleeding was the only independent predictor of adverse outcome.
Our findings emphasize the importance of underlying health status in determining outcomes of LIB. We found that age, burden of comorbid illness, and bleeding after admission for a separate process were strong and consistent predictors of in-hospital mortality. In addition, intestinal ischemia was strongly related to in-hospital mortality in all analyses, presumably due to underlying atherosclerosis and/or precipitating hypotensive episodes as well as subsequent bowel infarction and acidosis. Indeed, intestinal ischemia was significantly associated with a secondary diagnosis of bleeding (p<0.001). In contrast, diverticular bleeding, a common source of severe bleeding that is not necessarily associated with systemic illness, was not associated with the risk of mortality. Nevertheless, hypovolemia, anticoagulation, and the need for blood transfusion were also independent predictors suggesting that bleeding severity is also important.
Male gender was a significant predictor of in-hospital mortality in this study. Male gender was also associated with poor outcomes in a study of LIB by Das and colleagues. 4
In our study, comorbid illness may partially explain the higher mortality in men compared to women (mean comorbidity score 2.6 vs. 2.3, p<0.001). However, men tended to be younger than women (mean age 67.3 vs. 70.9 years, p<0.001), and were less likely to have intestinal ischemia (5.1% vs. 7.7%, p<0.001). Men underwent more diagnostic tests than women (55.2% versus 52.2%, p=0.01), suggesting that underutilization of therapeutic procedures did not play a role.
We found that patients undergoing diagnostic tests were less likely to die than those who did not. These patients were presumably more likely to have had a confirmed source of LIB, and the lower mortality in this subgroup may reflect the low mortality in LIB compared to other disorders. Patients with a diagnostic test were also on average healthier than patients without diagnostic testing (lower comorbidity scores, rates of hypovolemia, coagulopathy and secondary bleeding). The lower mortality in patients undergoing diagnostic testing could also reflect a therapeutic and/or diagnostic benefit of these interventions. Receipt of endoscopic bleeding control was uncommon (302 patients), and was not a significant predictor when added to the model (adjusted HR 0.67, 95% CI 0.31–1.44, p=0.30). However, the lack of statistical significance may reflect a Type 2 error related to an inadequate number of patients with endoscopic hemostasis.
We found no significant differences in mortality according to hospital characteristics. Hospital characteristics have influenced outcomes including death in a variety of other conditions, particularly those that involve complex procedural interventions,12, 14
although a large study of academic versus community hospitals found no differences in mortality with upper gastrointestinal bleeding.19
The lack of influence of hospital characteristics on mortality in LIB may indicate that interventions to control LIB do not reduce mortality in the majority of patients, as bleeding stops spontaneously in 80% of cases.7
The primary strengths of this study relate to the large, nationally representative patient population of more than 9,000 patients with LIB (representing an estimated 227,022 patients nationwide) across a wide spectrum of hospitals and geographic locations. This scope enabled the analysis and identification of multiple predictors of inhospital mortality including hospital characteristics. Thus, our results are likely to be generalizable across a range of locations and practice settings, except federal hospitals which are not included in the NIS database.
This study also has several limitations. NIS data lack the clinical detail available in other study designs. Such data are useful for denoting the severity of underlying disease including bleeding, and for confirming the presence of LIB and the cause of death. Nonetheless, the c-statistic for the logistic model of mortality was 0.80, suggesting that our analysis discriminated effectively between patients who survived and those who died. The reliance on ICD-9 codes to identify patients with LIB is challenging because many potential sources lack codes that specifically indicate bleeding,6, 20
and our estimate of the number of patients discharged with LIB is likely to be upwardly biased. Although our case identification algorithm attempted to exclude individuals with chronic or occult bleeding, upper gastrointestinal bleeding, or conditions associated with bleeding in the absence of bleeding, patients with these diagnoses may have been included in our cohort. Likewise, inaccuracies in coding may have led to inclusion of patients without LIB. In support of our patient identification algorithm, the breakdown of sources of bleeding in the current study is very similar to those reported in the recent literature.6, 9, 10, 20
In addition, we performed a secondary analysis in patients with diverticular bleeding and found similar results.
In this large, nationwide study we found that the all-cause in-hospital mortality rate in LIB is low (3.9%). The mortality rate of 1.8% in patients undergoing a diagnostic test suggests that the actual mortality rate in LIB may be even lower. Increasing age and comorbid illness, intestinal ischemia, bleeding after admission for another disorder, coagulopathy and hypovolemia were the strongest predictors of in-hospital mortality. These findings suggest that aggressive supportive care and management of comorbid conditions are more likely to improve in-hospital mortality rates in LIB than early, potentially therapeutic interventions. Because in-hospital deaths are uncommon in LIB, efforts focused on improving other in-hospital outcomes and preventing recurrent events in the patients who survive will likely have the greatest impact.