(130) A comparison of the descriptive epidemiology derived from the GOLD and Aurum datasets in the Clinical Practice Research Datalink

Background: The Clinical Practice Research Datalink (CPRD) contains a wealth of clinical data sourced from UK general practice. It comprises two datasets: GOLD (from practices using the Vision patient administration system) and Aurum (from EMIS). As these datasets originate from different systems, we wished to compare estimates of incidence and prevalence derived from them to determine whether the results were consistent.

Objectives: To compare the incidence, annual prevalence, and lifetime (point) prevalence of over 1,000 diseases recorded in the Aurum and GOLD databases.

Methods: Medical codes used in Aurum and GOLD were mapped to three character codes of the ICD-10 classification and three metrics calculated: incidence, annual prevalence and lifetime prevalence. Only conditions with a minimum of 100 cases in each dataset were considered. Rate ratios were used to compare Aurum to GOLD. A rate ratio greater than 1.0 indicated that the recording of diagnoses was more frequent in Aurum. Additionally, exploratory analyses were conducted to consider factors which may partly explain observed findings.

Results: For incidence, 690 conditions were considered and 79% reported incidence rate ratios (Aurum:GOLD) greater than 1.0 (median: 1.32, interquartile range (IQR): 1.06-1.69). For annual prevalence, 718 conditions were considered and 91% of diseases reported annual prevalence rate ratios greater than 1.0 (median: 1.62, IQR: 1.31-2.06). For lifetime prevalence, 1,132 conditions and 57% reported lifetime prevalence rate ratios greater than 1.0 (median: 1.04, IQR: 0.89-1.25). An analysis of 207 practices that migrated from GOLD to Aurum revealed higher annual prevalence rates in the year following system migration than in the year prior.

Conclusions: There were significant differences in the recording of diseases between Aurum and GOLD, with Aurum reporting higher rates across most metrics. The analysis of migrated practices highlighted potential operational differences in diagnosis coding between the two patient administration systems. One consideration is whether the systems differ in their treatment of routine healthcare events, such as annual reviews. These findings underscore the importance of considering database-specific characteristics in epidemiological research and call for further investigation into the factors driving these differences