We welcome Santhakumaran's article (1) describing some of the
problems and misunderstandings that can arise when adjusting for case-mix
differences between hospitals. In our recent paper (2) we quantified the
bias that is likely to arise when comparing standardised mortality ratios
(SMRs) between one neonatal unit and another. In our paper it was shown
that, using actual observed differences in case-mix, even if two neonatal
units were performing identically for each patient group the ratio of
their SMRs could range from 0.79 to 1.68.
However, this is not to say that the SMR has no role when reporting
of clinical outcomes. First, when case-mix differences are small the
likely bias that occurs when comparing two SMRs is also likely to be
small. Second, the value of the SMR can indicate where intervention (e.g.
training, guidelines) may be the most beneficial. For example, with
Santhakumaran's two hypothetical neonatal units (Table 1 (1)) it seems
entirely reasonable for the hypothetical manager to conclude that
prioritizing intervention in unit A (the unit with the highest SMR) would
result in improved outcomes for more patients than would the same
intervention in unit B, since there are more deaths in unit A than in unit
B.
1 Santhakumaran S. How to adjust for case-mix when comparing outcomes
across healthcare providers Arch Dis Child Educ Pract Ed Published Online
First: 30 September 2013 doi:10.1136/archdischild-2013-303940
2 Evans TA, Seaton SE, Manktelow BN. Quantifying the potential bias
when directly comparing standardised mortality ratios for in-unit neonatal
mortality. PLoS ONE 8(4):e61237
Conflict of Interest:
None declared
We welcome Santhakumaran's article (1) describing some of the problems and misunderstandings that can arise when adjusting for case-mix differences between hospitals. In our recent paper (2) we quantified the bias that is likely to arise when comparing standardised mortality ratios (SMRs) between one neonatal unit and another. In our paper it was shown that, using actual observed differences in case-mix, even if two neonatal units were performing identically for each patient group the ratio of their SMRs could range from 0.79 to 1.68.
However, this is not to say that the SMR has no role when reporting of clinical outcomes. First, when case-mix differences are small the likely bias that occurs when comparing two SMRs is also likely to be small. Second, the value of the SMR can indicate where intervention (e.g. training, guidelines) may be the most beneficial. For example, with Santhakumaran's two hypothetical neonatal units (Table 1 (1)) it seems entirely reasonable for the hypothetical manager to conclude that prioritizing intervention in unit A (the unit with the highest SMR) would result in improved outcomes for more patients than would the same intervention in unit B, since there are more deaths in unit A than in unit B.
1 Santhakumaran S. How to adjust for case-mix when comparing outcomes across healthcare providers Arch Dis Child Educ Pract Ed Published Online First: 30 September 2013 doi:10.1136/archdischild-2013-303940
2 Evans TA, Seaton SE, Manktelow BN. Quantifying the potential bias when directly comparing standardised mortality ratios for in-unit neonatal mortality. PLoS ONE 8(4):e61237
Conflict of Interest:
None declared