Strengths and weaknesses of hospital standardised mortality ratios

When we originally submitted this paper to the BMJ, the manuscript it
included a wide-ranging set of sensitivity analyses around issues of
coding and place of death that we discuss in our narrative account of
HSMRs published here today. For example, we compared HSMRs with and
without adjustment for comorbidity for all English acute trusts rather
than a small subset. The reviewers were generally positive about the
analysis, and it was regarded as potentially important by editors. We were
asked by the editor to split the paper into two parts. However, they
considered that it assumed too high a level of knowledge for BMJ readers.
On the advice of the editors, we have submitted our analysis paper to a
more specialist journal. We have been invited to mention the main findings
in a rapid response alongside today's discussion article. Without wanting
to prejudice other journals regarding our analysis as it undergoes peer
review, we can say that, although increasing secondary diagnosis coding
over time can affect a trust's predicted mortality rate as seen with Mid
Staffordshire, in a given year the effect of coding on HSMRs is more
modest than critics suspect despite occasional exceptions. If HSMRs are
used as a screening tool for high mortality, for example, then whether the
point estimate is 118 or 122 is immaterial.