Record linkage in a regional mental health planning study: accuracy of unique identifiers, reliability of sociodemographics, and estimating identification error.

Assembling information about individuals over time allows health managers and researchers to describe the progression of diseases, the care history of individuals and the sequences of care episodes that potentially result in improving individuals' health status. However, current mental health statistics generally focus on sets of events rather than groups of individuals making it impossible to distinguish between two different persons being admitted and the same person being admitted twice. Accurate figures on treatment prevalence cannot be generated and multiservice users across time or across agencies will inflate the statistics used to plan needed services. The capacity to link consistently defined bits of information together is critical to developing a reliable information system. This article examines the adequacy of using unique identifier codes to accomplish linkage by focusing on one example of record linkage that incorporates mental health information from both community and institutional sectors in one region of Ontario, Canada. Findings indicate that unique "cradle to grave" identifiers do not guarantee accuracy if manual transcription is involved.