Routine health insurance data for scientific research: potential and limitations of the Agis Health Database

Respondents in an epidemiologic survey had fewer psychotropic prescriptions than nonrespondents: an insight into health-related selection bias using routine health insurance data

“…First, we analyzed nonresponse not only solely according to traditional sociodemographic background variables but also according to various morbidity and health care utilization indicators, including mental health proxies. Second, our study was register based: we analyzed nonresponse using accurate and rather exhaustive information [37] on health care utilization from the MGEN reimbursement claim database obtained independently of the response status and not from a separate supplementary survey of ''willing nonrespondents.'' Indeed, many previous analyses of health disparities by participation status have been limited by the fact that they contrasted respondents to ''participating nonrespondents'' (e.g., persons who did not accept to participate at first but then did accept to respond a shortened interview/questionnaire), possibly repeating the same nonparticipation biases.…”

“…In addition to their longitudinal availability, health insurance data represent large practicebased information not subject to recall bias. As they are based on financial reimbursement, the information is generally reliable [37].…”

Monitoring drug safety with registries: useful components of postmarketing pharmacovigilance systems

“…In contrast to the tightly controlled patient selection approach adopted by clinical trials, registries provide a more representative picture of the range of patients receiving a drug, their additional medications, and existing medical conditions [10,11]. For monitoring the safety of medications used in the treatment of Rheumatoid arthritis (RA), a number of registries have been established that aim to provide data across the full range of medicated patients.…”

“…Registries, such as the ARAD and the BSRBR have taken steps to recruit control samples of patients whereas other registries have initiated data-linkage efforts with other existing sources, such as the Haemostasis Registry's link with ASCTS. Health insurance databases offer another useful resource for reliably capturing large numbers of patients, monitoring the course of disease, assessing drug utilization, and following patient groups over long time horizons [11]. Because of a lack of clinical detail in these databases, linkage to clinical registries is becoming increasingly useful, facilitating a linked-data source which minimizes selection bias, allows identification of a control group, and contains detailed clinical data for sufficient risk adjustment.…”

Validity of chronic drug exposure presumed from repeated patient interviews varied according to drug class

“…There are several sources of data available for drug exposure assessment in pharmacoepidemiological research including patient interviews and computerized databases recording reimbursement claims but all are limited in a certain way in their ability to reflect true patient exposure. Whereas reimbursement claims databases remain attractive because they totally avoid memory bias and allow an easier accessibility to large data sets [3], they suffer from a lack of data regarding over-thecounter (OTC) drugs and also that reimbursement does not necessarily mean use. Information obtained through wellconducted patient interviews may be exhaustive with regard to the use of OTC drugs or self-medication [4], yet when…”