Abstract
Pharmacy dispensing databases provide a comprehensive source of data on medicines use free from many of the biases inherent in administrative databases. There are challenges associated with using pharmacy databases however. This paper describes the methods we used, and their performance, so that other researchers considering using pharmacy databases may benefit from our experiences.
Data were collected from all nine pharmacy dispensing databases in an isolated New Zealand town for the period October 2005–September 2006. Probabilistic record matching was used to link individuals across pharmacies. Patient addresses from the pharmacy data were geo-located to small areas so an area measure of socioeconomic deprivation could be assigned. Medicines were coded according to the ATC-DDD drug classification system.
Data on 619,264 dispensings were collected. Record matching reduced an initial pool of individuals from 54,484 to 38,027. Socioeconomic deprivation ranks were assigned for 30,972 (93%) of the 33,375 unique addresses identified, or 36,048 (95%) of individuals. ATC codes were assigned to 613,490 (99%) of the dispensings, with DDDs assigned to 561,223 (91%). Overall, 93% of dispensing records had complete demographic and drug information.
The methods described in this paper generated a rich dataset for medicines use research. These methods, while initially resource-intensive, can to a great extent be automated and applied to other locations, and will hopefully prove useful to other researchers facing similar challenges with using pharmacy databases. However, it is difficult to envisage these methods being viable on a long-term or national scale.