Randomness in the RAWP formula: the reliability of mortality data in the allocation of National Health Service revenue.

Abstract

A corner-stone of the recommendations by the Resource Allocation Working Party (RAWP)1 for the allocation of the non-psychiatric service budget was the introduction of standardised mortality ratios (SMRs) as proxy measures of regional variations in morbidity. The appropriateness of this measure has been heavily criticised but it remains in the present allocation method. However, the introduction of seemingly objective measures such as SMRs may give a misleading impression of accuracy. Fluctuations in death rates unrelated to the general force of mortality or to factors relevant to health services require that each SMR should be considered as a sample taken from a range of possible SMRs which are distributed about the true value. The problem of the reliability of SMRs is accentuated by the use of SMRs disaggregated by condition groups corresponding to the chapter divisions of the International Classification of Diseases (ICD) and this is one of the chief difficulties in implementing the RAWP recommendations at subregional level. The working party acknowledged this problem in its report but claimed (p 40) that 'condition-specific SMRs can be used at area and district level with some loss of reliability due to small numbers involved. Sensitivity tests indicate, however, that the results will still be closer in most cases to the notionally 'correct' result than would be the case if another measure such as overall SMRs were applied'. The sensitivity tests referred to have not been published but several health authorities have carried out their own statistical analyses. Such a lack of quantitative support is not confined to RAWP. There is an obvious deficiency of data in many criticisms of it. On the question of the reliability of SMRs calculated on small populations, Geary2 and Ferrer et al3 do support their views with statistical analyses, but these are used to establish the possibility that problems exist rather than to provide estimates of the scale of the problems. In this paper we provide an empirical basis for assessing the importance of the statistical significance of SMRs in resource allocation decisions. The method used centres on the calculation of the variance of SMRs and the implications of excluding st tistically insignificant deviations of SMRs from the averag of 100. To place the discussion in its appropriate policy context we present first a brief review of current national and regional allocation policy.