Furthermore, by choosing our study period, we have ensured no systematic changes in coding because the ICD-10 coding system has been in continuous use in HES from 1995 to present. This, of course, does not exclude variation in rates of coding over the study period affecting our

estimates. For example, if the potential error in coding was systematically changing over time with increased coding of patients’ comorbidity rather than patients having more comorbidity, then clearly that could bias our results. However, the different trends ABT-199 nmr in comorbidity for variceal and nonvariceal bleed admissions and different trends in mortality in different age and comorbidity strata suggest that there was no systematic change in comorbidity coding over the time period of our study. Under-reporting of the comorbidities in the Charlson index may have resulted in incomplete adjustment for comorbidity. However, although the alternative Elixhauser index assessed almost twice the number of comorbidities, it did not alter the adjustment of comorbidity in the

model. Comorbidity adjustment by either index increased the magnitude of the mortality reduction, and, therefore, any residual confounding in this regard would only, we believe, cause an underestimate of the real mortality trend in our study. A PubMed search, to October 2010, found the largest comparable population-based study for nonvariceal hemorrhage mortality trends used a Canadian hospital discharge database with ICD-10 and ICD-9 codes. However, it identified less than one-third of the number of bleeds used for Enzalutamide price this study (n = 142,363) and was not able to identify a reduction in case fatality for nonvariceal hemorrhage between 1993 Carnitine palmitoyltransferase II and 2003.3 The researchers adjusted for changes in age but not for changes in comorbidity. They

also only identified deaths that occurred before discharge. The low mortality identified in this study (3.5%) is similar to other North American20 and Mediterranean1 and 21 studies but is much lower than other European studies.2, 22 and 23 However, a study of Medicare patients in the United States found that the proportion being managed as outpatients varied between states from 18.6% to 45.3%.24 These differences in practice would lead to differences in inpatient study populations and confound comparisons with countries such as England where outpatient management is not routine. Although the most recent report from the US National Inpatient Sample showed a 23% reduction in upper gastrointestinal hemorrhage mortality from 1998 to 2006 (n = unreported because only extrapolated estimates from the 20% sample are provided),20 this was a global figure for the reduction seen at the end of the study rather than year on year, and it did not distinguish variceal and nonvariceal hemorrhage. Another report from the US National Inpatient Sample noted an adjusted reduction in variceal hemorrhage from 18% to 12%.