Intended for healthcare professionals

Letters

Allocating census data to general practice populations

BMJ 1996; 312 doi: https://doi.org/10.1136/bmj.312.7025.249b (Published 27 January 1996) Cite this as: BMJ 1996;312:249
  1. Glen Scrivener,
  2. David C Lloyd
  1. Social geographer Applied research statistician University of Leeds, Leeds LS2 9NZ

    EDITOR,—We wish to reply to the two letters1 commenting on our paper on allocating census data to a general practice population.2 Paul Aveyard proposes using the method of Bland and Altman rather than correlation to measure agreement between the number of patients aged over 65 obtained from the family health services authority's register and that calculated from the postcodes of patients and the proportion in their census area. Ronan A Lyons and colleagues suggest that better results could have been obtained in their area, where they achieve a high proportion of matches between postcodes and census areas.

    We presented the correlation coefficients since they are what most others in this field (for example, Majeed et al3) have used to justify the use of census data, but we then went on to show, by using a graph and χ2 tests, how misleading a high correlation can be. The technique of Bland and Altman is not appropriate here since it implicitly assumes that neither proportion has higher validity than the other and so would use the average as the best estimate of the true proportion. In this case, the family health services authority's register, whatever its shortcomings, must be regarded as the most accurate available source of information.

    Lyons and colleagues express concern about the number of unmatched postcodes in our study. While this concern may be valid, one of our aims was to investigate how easy or difficult it was to match postcodes in registers of patients with their enumeration district using the postcode to enumeration district directory released with the 1991 census. The number of missing postcodes is an indication of the difficulty that exists. We were looking for a process that could be used straight-forwardly and systematically nationally.

    Both letters express concern at the use of sample surveys as an alternative way of obtaining data on practice populations. We are aware of the problems and envisaged instead a requirement to collect more details from patients at registration and the collection of more information from practices on a regular basis, in the way that data on chronic disease management are becoming available. The low income scheme index, described by Lloyd et al,4 is another example of data collected about the practice rather than estimated from another source.

    Lyons and colleagues express concern about the transfer of resources from more deprived to less deprived areas. If they examine the figure in our paper they will find that this is exactly what the method that they defend achieves since practices at the extremes have predicted values that are closer to the mean than the true value.

    We concede that the decision about whether the estimation of practice characteristics from census variables is “good enough” is ultimately subjective. We believe, however, that the discrepancies that we found with estimating such a gross characteristic as the proportion of elderly people are sufficient to cast grave doubt on the use of characteristics such as permanent sickness, which, because of small numbers, are much more difficult to estimate.

    References

    1. 1.
    2. 2.
    3. 3.
    4. 4.