Assessing bias in community-based prevalence estimates: towards an unduplicated count of problem drinkers and drug users.

Abstract

General population survey estimates of the overall prevalence of problem drinking and drug use in a community are biased by the exclusion of non-household populations. Estimates based on compiling prevalences in community institutions may also be biased due to over-counting of users of more than one institution. This paper examines prevalence estimates derived from probability samples of problem drinkers in the general population and within alcohol treatment, drug treatment, mental health, criminal justice and welfare agencies in a single US county. Data sets are merged and weighted to reflect a community sample of institutions, and a 17% subset of cases is identified within the institutional samples that are not living in housing units typically included in general population sampling frames. The difference in prevalences of problem drinking in the household and non-household populations is found to be large: 11% and 48%, respectively. Even greater differences are found between estimates of unprescribed weekly drug use (6% and 47%, respectively) and combined problem drinking and weekly drug use (2% and 27%, respectively). This suggests that confining samples to the household population can systematically under-represent the prevalence of problem drinking and drug use. A second source of bias in prevalence is characteristic of studies using records from multiple institutions. When duplication of service use in the five agency samples is considered, it becomes apparent that prevalence may be biased upward due to over-counting of multiple service users.