Birthweight-specific rates can be useful for summarising stillbirth and neonatal mortality in populations but, sometimes, a single summary measure is required to compare several relatively small subpopulations. However, any particular summary has its shortcomings, and various methods have been proposed. We wished to compare mortality between local authorities and between hospitals in the Thames regions and thus required a single summary measure for each subpopulation. It was not obvious in advance which summary to use or whether a single method would work well for both local authorities (a geographical unit) and hospitals.

This study compared six methods of calculating a single summary, three using indirect standardisation to adjust (500 g bands, 10-percentile bands, 10 z-score bands) and three using regression to adjust (mean birthweight, proportion < 2500 g, proportion < 1500 g). The data used were 570 016 births in the Thames Regions, broken down into its 96 local authorities and 65 hospitals. To investigate how well each adjustment had performed, we calculated the rank correlation between the crude and various adjusted mortality rates and mean birthweight, proportion < 2500 g and proportion < 1500 g. This was done separately in the local authorities and hospitals. If a method of adjustment had worked very well, these correlations should be negligible.

For the local authorities, adjustment for proportion < 1500 g gave the lowest correlations. Adjustment for mean birthweight and 500 g-band standardisation did not appear to work so well but gave moderately low correlations. For hospitals, 500 g standardisation gave the lowest correlations. Adjustment for mean birthweight and proportion < 2500 g worked only moderately well. Percentile and z-score adjustment did not work well for local authorities or hospitals. We conclude that several methods appear to work reasonably well for local authorities, whereas for hospitals, 500 g indirect standardisation worked best. Percentile and z-score standardisation did not work well in these subpopulations.