Referral bias in hospital register studies of geographical and industrial differences in health.

INTRODUCTION The Danish National Hospital Register contains four patient types: full-time inpatients, part-time inpatients, outpatients and emergency ward patients. The aim of the present study was to investigate whether results from comparative hospital register studies depend on which patient types we choose to include in the analysis. METHODOLOGY The hospital register was linked to the centralised civil register and the employment classification module. All economically active persons in Denmark aged 20-59 years 1st January 1995 (N = 2,281,480) were followed for six years. We calculated SIRs, first by county then by industry and finally by industry adjusted for county, for a variety of diagnostic groups and for each of the following types of cases: A) full-time inpatients, B) all inpatients, C) all inpatients and outpatients, D) all patients. The ratio between the maximum and the minimum of the four types of SIRs was calculated for each combination of the examined population groups and diseases. A max/min ratio was regarded as a sign of referral bias if it was above 1.2 and statistically significant. RESULTS When calculating SIRs by county 46.7 percent of the max/min ratios signified referral bias. The percentage was 5.5 when calculating SIRs by industry and only 1.7 when they were calculated by industry adjusted for county. CONCLUSIONS Estimates of geographical health differences are often distorted by differences in the health care organisation. Estimates of industrial health differences tend to be robust with a few identifiable exceptions. Standardisation for county will eliminate bias.