Prospective drug safety monitoring using the UK primary-care General Practice Research Database: theoretical framework, feasibility analysis and extrapolation to future scenarios.

BACKGROUND: Post-launch drug safety monitoring is essential for the detection of adverse drug signals that may be missed during preclinical trials. Traditional methods of postmarketing surveillance such as spontaneous reporting have intrinsic limitations, many of which can be overcome by the additional application of structured pharmacoepidemiological approaches. However, further improvement in drug safety monitoring requires a shift towards more proactive pharmacoepidemiological methods that can detect adverse drug signals as they occur in the population.
OBJECTIVE: To assess the feasibility of using proactive monitoring of an electronic medical record system, in combination with an independent endpoint adjudication committee, to detect adverse events among users of selected drugs.
METHODS: UK General Practice Research Database (GPRD) information was used to detect acute liver disorder associated with the use of amoxicillin/clavulanic acid (hepatotoxic) or low-dose aspirin (acetylsalicylic acid [non-hepatotoxic]). Individuals newly prescribed these drugs between 1 October 2005 and 31 March 2006 were identified. Acute liver disorder cases were assessed using GPRD computer records in combination with case validation by an independent endpoint adjudication committee. Signal generation thresholds were based on the background rate of acute liver disorder in the general population.
RESULTS: Over a 6-month period, 8148 patients newly prescribed amoxicillin/clavulanic acid and 5577 patients newly prescribed low-dose aspirin were identified. Within this cohort, searches identified 11 potential liver disorder cases from computerized records: six for amoxicillin/clavulanic acid and five for low-dose aspirin. The independent endpoint adjudication committee refined this to four potential acute liver disorder cases for whom paper-based information was requested for final case assessment. Final case assessments confirmed no cases of acute liver disorder. The time taken for this study was 18 months (6 months for recruitment and 12 months for data management and case validation). To reach the estimated target exposure necessary to raise or rule out a signal of concern to public health, we determined that a recruitment period 2-3 times longer than that used in this study would be required. Based on the real market uptake of six commonly used medicinal products launched between 2001 and 2006 in the UK (budesonide/eformoterol [fixed-dose combination], duloxetine, ezetimibe, metformin/rosiglitazone [fixed-dose combination], tiotropium bromide and tadalafil) the target exposure would not have been reached until the fifth year of marketing using a single database.
CONCLUSION: It is feasible to set up a system that actively monitors drug safety using a healthcare database and an independent endpoint adjudication committee. However, future successful implementation will require multiple databases to be queried so that larger study populations are included. This requires further development and harmonization of international healthcare databases.