GP Databases, Record Linkage

When describing databases for pharmacoepidemiology, a distinction can be made between two types of databases. The first are created using record-linkage techniques: different (health) information sources are linked either through probabilistic linkage and these can be used to establish drug exposure and disease status. Using information from more than one information source in this way provides the opportunity for data validation.

The second type comprises databases using a single information source, for instance, those containing only primary care data, or those following up a specific cohort with the aim to conduct epidemiological research. Finally, disease registries exist that sometimes are used for drug safety research. Examples of the different types of data source will be presented.

Most of these databases have been set up with other aims than to carry out research and this has implications for the information available for research.

Strengths and limitations: Some databases' limitations arise due to what is asked at the point of data entry, whereas other databases are limited due to what is deemed relevant by the user. To an extent, the availability of data depends on the health care system in which they are collected. In addition, most databases lack information on over the counter (OTC) drug use and on important confounders for drug safety research such as ethnicity, disease severity and life style factors.

Apart from information on exposure, outcome, and potential confounders, sufficient data to detect clinically and statistically significant associations is needed, hence the database needs to cover a sufficiently large population. What constitutes ‘a sufficiently large population’ is a function of both frequency of the exposure and frequency of the outcome studied. Examples will be given of the contribution of healthcare databases to drug safety studies.