Clinical epidemiological studies of drug safety and disease risk factors using large primary care databases

• Main Author: Vinogradova, Yana
• Published: University of Nottingham 2017
• Subjects: WA Public health
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.728458

BACKGROUND: Observational studies of drug treatments complement pre-marketing drug trials and provide real-world outcomes of effectiveness and safety. Large UK primary care databases offer cost-effective access to clinical information for long-term studies requiring great statistical power and deliver findings representative of the general population. However, such data are not collected primarily for research, so all share weaknesses that must be offset by sophisticated use of statistical methodologies. This paper clarifies the current strengths and limitations of these data sources and discusses their potential. In the context of routinely collected primary care data sources, studies focusing on drug safety are used to show appropriate application of statistical techniques, and present a contribution to existing methodological practice based on multi-database use. METHODS: Methodological approaches to address potential data-related biases and drug safety study-related issues are discussed. These include coding differences, analyses of exposure, confounding factors to be included in models, missing data, misclassification bias from outcome uncertainty and prescription-only information, and use of sensitivity analyses to estimate the impact of information gaps and verify the validity of findings. A novel application of triangulation between the findings of separate identical analyses of two databases is introduced. RESULTS: The submitted papers are used to exemplify the delivery of more accurate estimates of risks than previous studies, with further comment on how the methodologies were used to address potential issues. The results of triangulation between the findings of two separate identical analyses based on different databases show confidence intervals for the combined results on average 30 per cent narrower than those of the original analyses. CONCLUSIONS: The application of emerging/developing methodologies enables large UK primary care databases with national coverage to deliver robust findings applicable to the general population and derived from long-term studies with great statistical power. The potential for future development is also shown, including use of multi-databases to further increase statistical power.