K Walsh1. 1. MRC Environmental Epidemiology Unit, Southampton.
Abstract
AIM: This study set out to show how well samples from general practice registers compare with census data, to describe those characteristics of the population and of the register that influence the response to postal surveys, and to demonstrate how general practice records can be used to assess non-response bias. METHOD: The data for this study were obtained from a large postal survey about low back pain among the general adult population aged 20-59 years in eight areas of the United Kingdom, using general practice age-sex registers as the sampling frame. RESULTS: The overall response rate was 59%. In the areas chosen, general practice registers yielded samples of size and age-sex composition close to that predicted from national census data. Responses were more likely to be obtained from women, from older age groups and from practices where the sample lists had been inspected for errors. The use of computerized registers and a letter of recommendation from the general practitioner had no effect on the response rate. Inspection of the general practice records of subsamples of respondents and non-respondents to determine consultation rates suggested that there was little response bias in respect of the subject of the survey. CONCLUSION: General practice registers can provide a suitable sampling frame for epidemiological purposes. Inaccuracies in the register can be reduced to some extent by careful inspection, but an irreducible minimum remain. Information held in general practice records can be useful in assessing response bias in health surveys.
AIM: This study set out to show how well samples from general practice registers compare with census data, to describe those characteristics of the population and of the register that influence the response to postal surveys, and to demonstrate how general practice records can be used to assess non-response bias. METHOD: The data for this study were obtained from a large postal survey about low back pain among the general adult population aged 20-59 years in eight areas of the United Kingdom, using general practice age-sex registers as the sampling frame. RESULTS: The overall response rate was 59%. In the areas chosen, general practice registers yielded samples of size and age-sex composition close to that predicted from national census data. Responses were more likely to be obtained from women, from older age groups and from practices where the sample lists had been inspected for errors. The use of computerized registers and a letter of recommendation from the general practitioner had no effect on the response rate. Inspection of the general practice records of subsamples of respondents and non-respondents to determine consultation rates suggested that there was little response bias in respect of the subject of the survey. CONCLUSION: General practice registers can provide a suitable sampling frame for epidemiological purposes. Inaccuracies in the register can be reduced to some extent by careful inspection, but an irreducible minimum remain. Information held in general practice records can be useful in assessing response bias in health surveys.