Abstract

Background: Pharmacovigilance (PV) is the science and activities involved in monitoring and developing the safety profile of all marketed medicines. Adverse drug reactions (ADRs) for medicinal products can be identified through post-marketing studies by methods of signal detection. Traditional, qualitative methods involve clinical review of cases, and coupled with modern, quantitative methods which have evolved as PV has grown, may help surveillance of the large number of medicinal products on the market today. This research aimed to investigate combining traditional and modern methods of signal detection by adding statistical weighting to adverse event terms identified as requiring further monitoring pre-marketing, to improve identification and evaluation of ADRs post-marketing.

Methods: Four anti-diabetic drugs currently marketed were chosen to model the concept: gliclazide, pioglitazone, rosiglitazone and vildagliptin. Review of pre-marketing information for safety concerns highlighted two medical concepts: cardiac failure and acute pancreatitis. The Delphi method was adapted to identify and prioritise terms for these concepts to add statistical weighting to. The weightings were applied to two datasets, both from the UK Yellow Card Scheme (YCS): a two-year dataset (2005-2007) and a ten-year dataset (2000-2010). 

Key findings: Attribute agreement analysis (AAA) showed the Delphi method provided ‘slight’ to ‘fair’ overall agreement for terms ranked as very important (Fleiss’ kappa statistic (FKS) values between 0.1-0.4) on the first review by eleven physicians. There was good agreement for terms in the second review by four of the original reviewers (FKS 1.00, p>0.0001 for cardiac failure terms, and 0.983, p>0.0001 for acute pancreatitis terms). Weightings were established based on these ratings and applied. In the two-year dataset one term, oedema peripheral, became a potential signal with pioglitazone, one month earlier with the same number of reports as without weighting, and rosiglitazone, two months earlier with 2 less reports. In the ten-year dataset, oedema peripheral was reported 22 times with pioglitazone but was not considered a signal in the original data. With the weighting applied it became a potential signal for further review. All the drug/event term pairs that were identified as signals in the original data, remained signals with weighting applied. 

Conclusion: The weighting method did not result in missing any true signals. One term, oedema peripheral, was identified earlier in the two-year dataset, and is now an established side effect of anti-diabetic medication. The timing of signals was not available in the ten-year dataset. Monitoring timing of potential signals in a large spontaneous reporting database is necessary for validation of the proposed weighting method. Based on the results of this research, incorporating traditional, qualitative review of pre-marketing information with quantitative, signal detection methodology has the potential to improve PV, allowing quick, accurate identification of safety risks to monitor for marketed medicinal products. 

Additional Notes

Supervisor: Professor David Brown

Course: Master of Philosophy - MPhil

Date Deposited: 2017-05-12

URI/permalink: https://library.port.ac.uk/dissert/dis12883.html