Suchergebnisse

It has previously been shown that there is a significant association between mortality from sudden infant death syndrome (SIDS) and rates of long-term migration between local authority districts in England and Wales. This evidence supports the theory that exposure to infections, mediated by population mixing, may be an important factor in many cases of SIDS. In this paper, multilevel modelling is used to examine the variability in SIDS deaths at different geographical scales, namely district, county, and regional levels. Given the population-mixing hypothesis, it is possible that high levels of population mixing in one district will have an effect on the spread of infections in an adjacent district, and the rates for individual districts will not be spatially independent of each other. Factors such as climate varying at regional scale may also be important. A log—linear multilevel model is developed to examine these issues, and the discussion focuses on the methodological issues raised by the analysis such as appropriate multilevel structure, methods of estimation, dispersion of residuals, and significance of parameter estimates.

Postcode addresses from National Health Service patient registers for Norfolk and Suffolk current on census day 1991 were assigned to census wards, and estimates of populations in wards were produced for the total population and for twelve age—sex groups. These were compared with adjusted counts of usual residents from the 1991 Census. Regression analysis was used to identify the characteristics of wards with consistent differences between register estimates and census counts. Patient register estimates were less than census counts for children aged 0–4 years. Patient register estimates exceeded census counts for residents aged 15–44 years (especially males) and over 74 years. Wards with high rates of in-migration and high proportions of residents employed in the armed forces tended to have lower register estimates than census counts. Areas of high population density had higher register estimates than census counts. Taking these effects into account, about half the patient register estimates of total population were within 5% of the equivalent census figure, and 95% of wards had estimates within 20%. The main differences between patient register population estimates and census counts can be explained by delays in the updating of records in family health service authority (FHSA) registers and by undercounting in the 1991 Census. In Norfolk and Suffolk, FHSA registers are an acceptable alternative to the census for population estimation purposes. This supports recent arguments for wider use of population registers and suggests that they may be particularly valuable as a source of intercensal information.