Measuring drug use: Differences between medical records and healthcare utilisation data bases

Introduction: Drug use retrieved from healthcare databases has a key role in drug utilization studies and pharmacoepidemiology. Our aim is to compare and analyse the differences and level of agreement in prevalence of drug use between healthcare utilization (HCU) and medical records (MRs) databases, expressed in number of users. Methods: One-year period prevalence rates (PPRs) in 2008, expressed as number of users/1,000 people (MRs) and DDD/1,000 inhabitants/day (DIDs) transformed into number of apparent users (HCU) were calculated for the following medicines (Anatomic Therapeutic Chemical [ATC] code): calcium channel blockers (C08C, C08D), antiepileptic drugs (N03A), macrolides (J01FA), benzodiazepines (N05BA, N05CD, N05CF), and antidepressants (N06AA, N06AB). Data were available for the Netherlands (2 MRs, 1 HCU), Spain (1 MR, 1 HCU) and United Kingdom (2MRs, 1 HCU). Percentage differences, Pearson correlation coefficients and level of agreement with Bland Altman statistics were calculated. Data were aggregated by ATC level: ATC level 3 (N03A, C08C, C08D) and ATC level 4 (J01FA, N05BA, N05CD, N05CF, N06AA, N06AB). Results: Percentage differences between PPRs [median (interquartile range)] fluctuated from -34.4% (38.2%) at ATC level 3 to -65.2 (49.2%) at ATC level 4; and from -96.0% (1.6%) for macrolides to 28.2% (54.4%) for calcium channel blockers. The correlation coefficients were r = 0.88 (P <0.001) and r = 0.51 (P = 0.008) at ATC level 3 and 4, respectively. The level of agreement showed a deviation of -2.2/1000 users (limits of agreement [LA] -19.7 to 6.4) and -28.9/1000 users (LA -84.3 to 26.6) at ATC level 3 and 4, respectively. Conclusions: The study suggests that the level of agreement between users and DIDs were high the more aggregated data is presented. The wide LA might reflect not only the low number of groups of medicines studied, but also systematic bias in the measurement of DIDs and one-year PPRs.