A population-based case-control study for examining early life influences on geographical variation in adult mortality in England and Wales using stomach cancer and stroke as examples

Abstract

Geographical variation in mortality is influenced by factors operating in early life and in adulthood. The relative contributions of these factors may be examined by comparing the extent to which adult mortality is related to places of residence in early life and at death. We describe a population-based case-control design, in which all deaths are used as cases and the Office for National Statistics (ONS) Longitudinal Study (LS) survivors are used as controls. Cases were all deaths from stomach cancer and stroke in England and Wales 1993-1995 amongst people born between January 1930 and September 1939 and for whom place of enumeration in 1939 could be imputed from the first three characters of their National Health Service number. Controls were all LS members born in the same period, enumerated in the 1991 census, resident in England and Wales in mid-1994 and for whom place of enumeration in 1939 could be similarly imputed. Logistic regression was used, adjusting for birth year, sex and social class. A previous mapping exercise by ONS generated comparable geographical units (counties) for 1939 enumeration and area of residence in 1991 or at death. 'Non-migrant' (i.e. 1939 'county' the same as county in 1991 or at death) case:control ratios were calculated to indicate background mortality risk in counties, with adjustment for imprecision using Bayesian smoothing methods. These ratios were then used in modelling risk for inter-county migrants. There were 2590 stomach cancer and 7778 stroke deaths and 28,400 men and 28,180 women as controls. For men, 64%, 61% and 67% of stomach cancer deaths, stroke deaths and controls respectively could be assigned a county of enumeration in 1939. The corresponding percentages for women were 76%, 72% and 75%. For stomach cancer, after adjustment for county of enumeration in 1939, a significant association with the non-migrant case:control ratio for county of residence in 1991 or at death was observed (P= 0.010), indicating an association between current area of residence and stomach cancer mortality. There was no evidence of an independent effect of county of enumeration in 1939. For stroke, there was a highly significant trend in relation to 1939 county (P = 0.0004)and a less significant association with county of residence in 1991 or at death(P = 0.016). The method described is able to detect the effect of place of residence in early life on geographical variation in adult mortality and will be useful for investigating specific characteristics of areas of enumeration in 1939 in relation to subsequent risk of mortality from a range of diseases.