Public Health (2007) 121, 905–906 Commentary Are hospital league tables calculated correctly? A commentary Paul Aylin à , Alex Bottle Dr Foster Unit at Imperial College London, Department of Primary Care and Social Medicine, Imperial College, 1st Floor, Jarvis House, 12 Smithfield St. London EC1A 9LA, UK Available online 7 September 2007 The paper by Julious et al. 1 suggests that the standardised mortality ratio (SMR) is a form of standardisation that does not allow a valid compar- ison of different populations. The authors compare indirectly standardised SMRs with directly standar- dised comparative mortality figures (CMFs), and highlight the differences. In their discussion, they explain that SMRs are invalid because a different denominator applies in the calculation of the SMR for each unit. They conclude by saying that SMRs should be avoided for any form of league table or between-group comparison. This argument is not new and the authors have made the point previously 2 using a different data set. Their paper suggests that SMRs are not valid for comparisons between different populations. In fact, the main assumption required when calculat- ing SMRs via the indirect method is proportionality, i.e. that the rate ratio between the area and the standard for each stratum is assumed to be the same, otherwise the SMR will not be a good summary of all these stratum-specific rate ratios. 3 In the worked example provided, based on local authority data, the stratum-specific rate ratios are not the same for the three age groups. If proportionality is met, then the fact that SMRs for different areas involve standardization to different populations is not a problem. In practice, if proportionality is not met, the resultant bias when using the indirect method is often small (but will be present for some areas). For the example given (which the authors describe as extreme), the difference between the two methods amounts to less than 1.5%, demonstrating empirically (contrary to the paper’s main assertion) that there is little difference between the two methods at this level. In addition, the authors do not provide confidence intervals around the point estimates, which overlap considerably. The suggested alternative of direct standardisation has a larger flaw, which is as follows. In the direct method, expected counts are derived by multiplying the local unit or area’s age- and sex-specific rates by a standard popula- tion. As the authors acknowledge, the CMF will be an imprecise estimate of the rate ratio unless the local area’s rates are well estimated, i.e. not just based on one or two deaths, because otherwise the expected number will be much larger (or much smaller) than it should be. In the indirect method, since the reference rates are based on the whole of England, they will be well estimated for the strata used in the calculation of hospital SMRs since the SMR is the maximum likelihood estimate. The authors provide a description of differences in ranks to support their assertion; however, they make no mention of the inherent uncertainty around each rank, which would eclipse any differ- ences between the two methods. If they had used ARTICLE IN PRESS www.elsevierhealth.com/journals/pubh 0033-3506/$ - see front matter & 2007 The Royal Institute of Public Health. Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.puhe.2006.06.020 à Corresponding author. E-mail address: p.aylin@imperial.ac.uk (P. Aylin).