SIR, You've led me astray!

Background: The Standardized Infection Ratio (SIR) is an indirectly standardized morbidity ratio (SMR) originally used to compare the crude surgical site infection (SSI) incidence rate in any one hospital to an expected occurrence based on procedure-specific rates from a universe of hospitals participating in the National Healthcare Safety Network (NHSN). More recently, the US Department of Health & Human Services promoted use of SIR as a metric for inter-hospital comparison in its Action Plan to Prevent Healthcare Associated Infections.  A “shifting base distortion” is long recognized as an inherent bias in the indirect SMR calculation, leading to cautions about its use if comparing more than two groups at a time.  This advocated metric, if used to compare numerous hospitals with different populations simultaneously, may lead to distorted comparisons and conclusions regarding the rates of SSI incidence in these hospitals.
Objective: Evaluate extent of bias introduced by “shifting base” inherent in SIR.
Methods: Hypothetical & real data are examined by relative risk and by direct & indirect SMR across a range of varying population compositions. The extent to which the 3 metrics agree is noted.
Results: Public advocates demand to see infection rates; SIR presents a ratio of rates, not the actual rates, so does not address our current understanding of information demands.  People anticipating a particular type of hospital service may want to know infection rates in the units serving that service - stratified rather than crude rates may be more pertinent to them and SIR obscures this level of detail.
Using hypothetical & real data, the SIR was calculated for hospitals across a range of population compositions and SSI incidence density rates. The resulting measurements of SIR could be manipulated greatly by variations in population numbers and showed that hospitals could have different SIR values that eliminated, amplified, or reversed the differences that were apparent in the other SSI metrics. Also, because SIR compresses this comparison into one number, important details may be lost.
It is impossible to tell whether an indirect SMR is distorted, or estimate direction & magnitude of bias, just by looking at SMR numbers. When population compositions are similar among groups (where standardization is least needed) there should be little distortion; when population compositions differ appreciably (where standardization or stratification is most needed) shifting-base distortion is maximized.
Conclusions: SIR has the potential to obscure important details when hospital population compositions differ appreciably and in situations where standardization may be most needed. In addition, because public disclosure advocates ask to see infection rates, SIR, which is a ratio of rates, does not address the information demands by the public.  Stratified rates are easier to comprehend, so better suited for public display inherent in the work of public reporting programs.