F Hoffmann1, M S Pfannkuche, G Glaeske. 1. Universität Bremen ZeS, AG Arzneimittelanwendungsforschung, Ausser der Schleifmühle 35 - 37, 28203 Bremen. hoffmann@zes.uni-bremen.de
Abstract
BACKGROUND: Claims data of German statutory health insurance companies have been used increasingly over the last years for various research purposes. However, little is known about the validity of these data. Aim of this study was to analyse whether information relevant for research purposes is accurately recorded in computerized out-patient medication claims data. METHODS: We used data of the Gmünder ErsatzKasse (GEK) over the years 2000 - 2006. A random sample of 300 prescriptions annually was drawn from a total of 6.3 to 7.5 million prescriptions. Data on the scanned original prescriptions ("gold standard") were validated against that information in claims data. RESULTS: Over the years 2000 - 2006 an increasing proportion of dates of writing was recorded accurately (77.3 % to 93.0 %; p for trend < 0.0001). Concomitantly the proportion of handwritten prescriptions declined and that of printed ones increased. Dates of dispensing were barely correct until 2003. Thereafter, the agreement ranged from 65.3 % in 2004 to 86.1 % in 2006 (p for trend < 0.0001). There were significant differences between pharmacy data processing centres (Apothekenrechenzentren). The accuracy of physician and pharmacy identifiers (97,7%, 99,75% and 100.0 %, respectively) was nearly constant over these 7 years. CONCLUSION: The accuracy of variables analysed in this study is very good for those which are immediately related to the flow of money in German out-patient medication claims data. Recording of dates of prescription and dispensing are fairly good quality.
BACKGROUND: Claims data of German statutory health insurance companies have been used increasingly over the last years for various research purposes. However, little is known about the validity of these data. Aim of this study was to analyse whether information relevant for research purposes is accurately recorded in computerized out-patient medication claims data. METHODS: We used data of the Gmünder ErsatzKasse (GEK) over the years 2000 - 2006. A random sample of 300 prescriptions annually was drawn from a total of 6.3 to 7.5 million prescriptions. Data on the scanned original prescriptions ("gold standard") were validated against that information in claims data. RESULTS: Over the years 2000 - 2006 an increasing proportion of dates of writing was recorded accurately (77.3 % to 93.0 %; p for trend < 0.0001). Concomitantly the proportion of handwritten prescriptions declined and that of printed ones increased. Dates of dispensing were barely correct until 2003. Thereafter, the agreement ranged from 65.3 % in 2004 to 86.1 % in 2006 (p for trend < 0.0001). There were significant differences between pharmacy data processing centres (Apothekenrechenzentren). The accuracy of physician and pharmacy identifiers (97,7%, 99,75% and 100.0 %, respectively) was nearly constant over these 7 years. CONCLUSION: The accuracy of variables analysed in this study is very good for those which are immediately related to the flow of money in German out-patient medication claims data. Recording of dates of prescription and dispensing are fairly good quality.