The reports therefore compare different populations from each time period, and, although a number of weighting procedures are used, the estimates remain susceptible RG7420 nmr to selection bias. One smaller study from Wales (n = 24,421) used the same ICD-10 definitions as our study and also found an overall reduction in case fatality but did not report variceal and nonvariceal hemorrhage mortality trends separately or trends in different age and comorbidity strata.10 Other nonvariceal hemorrhage studies from Spain (n = 17,663),1 The Netherlands (n = 1720),2 Greece (n = 1304),21 France (n = 1165),23

and Italy (n = 1126)22 did not identify reductions in nonvariceal inpatient mortality. Although these were large studies, they may have been underpowered to detect a change, and none of them adjusted the trends in case fatality for changes in comorbidity. Furthermore, none of these studies identified deaths that occurred after discharge. The remainder of the studies contained less than 1000 patients and therefore could not provide accurate estimates of mortality trends. For variceal hemorrhage, the largest study on mortality after hospitalization because of varices (n = 12,281; compared with 14,682 for this study) did not differentiate between hemorrhage and nonhemorrhage admissions.26 The next largest

study (n = Ipilimumab chemical structure 1475) compared variceal hemorrhage mortality between control groups in randomized trials 1960–2000 and showed a similar reduction in mortality.27 However, these control groups were from different geographical populations with different study exclusion criteria. Comparisons were therefore susceptible to selection bias. Other studies of trends in variceal hemorrhage mortality contained less than 1000 patients. The other finding of note in our study in relation to variceal hemorrhage is the small proportion of overall hemorrhages that they represent. In the context of the increasing burden of liver disease28 and an apparent increase in variceal hemorrhage in the recent BSG audit,8 a higher proportion might have been expected. Our finding, however, was similar

to that from the 1993 BSG audit (4%) and to other studies.9 and 29 It is possible that some of the variceal HSP90 hemorrhages in our study may have been incorrectly coded to esophageal hemorrhage, but a sensitivity analysis, assuming the most likely misclassification of all esophageal hemorrhage codes being miscoded variceal bleeds, did not alter the adjusted reduction in mortality. The previous difficulties in detecting a reduction in mortality might imply that we are reaching the point where mortality becomes unavoidable because of age and comorbidity. However, because the mortality in our study continued to improve right up to the end of the study period, improvements in management would appear to be continuing to have an impact on mortality following gastrointestinal hemorrhage. The reasons for the reduction in mortality we have observed are likely to be complex.