Referral bias among health workers in studies using hospitalization as a proxy measure of the underlying incidence rate.

Contacts with health services like hospitals or general practitioners are usually the only available proxy measure of incidence of disease in studies based on secondary data and differential referrals or care-seeking behavior often bias such proxy measures. In former analyses based on the Occupational Hospitalization Register in Denmark assisting nurses had high Standardized Hospitalization Ratios for many diseases. It was, however, suspected that it fully or partly was due to a referral bias or self-selection to hospital treatment rather than exposures to occupational hazards. The aim of the present study is to evaluate the referral bias hypothesis by comparing hospitalization and mortality data for health workers for a disease category with a high mortality. Cohorts of all gainfully employed 20- to 59-year-old Danes were formed in order to compare Standardized Mortality Ratios and Standardized Hospitalization Ratios of ischemic heart disease (IHD) in occupational groups. The follow-up period was 10 years. For most of the investigated occupational groups a similar disease pattern was found whether hospitalization or death was used as the outcome measure. In "nurse assistants" a statistically significant higher risk was, however, found using hospitalization due to IHD as the end point rather than mortality. Additional analysis did not support the hypothesis that the finding could be explained by differentiated hospitalization due to social factors. The true incidence rates of the disease need not be equally well described by proxy measures such as hospitalization diagnosis or death diagnosis in all occupational groups. Differential access to medical treatment in some groups may lead to bias when hospital data are used as proxy measures for the underlying incidence rates.