MJA
MJA

Evaluating medicines: let’s use all the evidence

Chris W Kelman, Sallie-Anne Pearson, Richard O Day, C D'Arcy J Holman, Erich V Kliewer and David A Henry
Med J Aust 2007; 186 (5): 249-252. || doi: 10.5694/j.1326-5377.2007.tb00883.x
Published online: 5 March 2007
Evolution of pharmacovigilance

Historically, procedures for establishing the effectiveness of medical treatments have been suboptimal, with many commonly used treatments still awaiting rigorous evaluation.4 The safety of many medicines has largely been taken on faith, on the assumption that claimed benefits outweigh (un)known risks. In the early 1960s, the connection between thalidomide and severe birth defects was observed;5 this was a potent catalyst for clinicians and regulators to consider systems for monitoring drug safety.

Limitations of the current system
Premarketing evaluation

There are a number of reasons why premarketing evaluation cannot fully explore the effects of new medicines.7

Postmarketing evaluation

The reporting systems for adverse drug reactions in most countries are voluntary and depend on individuals becoming suspicious that an event is related to a treatment, and then actually reporting it. Identifying medicine-related adverse events is straightforward for unusual reactions such as severe skin rashes or liver failure, and those with a clear temporal relationship, but difficult for common events with well known causes, such as heart attacks and stroke. In either case, risk quantification (rather than mere signal generation) requires measurement of incidence, as well as strength of association between exposure and outcome — such information is not available from adverse event reporting databases.

A recent example of the need for formal ongoing evaluation has been reported by McEwen et al.10 The disease-modifying antirheumatic drug leflunomide was approved in Australia in 1999 for use as monotherapy for rheumatoid arthritis. Actual usage has proven to be different, and the incidence of pancytopenia after use of leflunomide in combination with methotrexate now appears to be more than 1/1000 per person-year. Preregistration examination of such concomitant use was restricted to only 30 patients. While the voluntary reporting scheme has now alerted regulators to this problem, this example emphasises that neither the manner of clinical use nor real-world safety can be assumed when approving new medicines. A routine system for the ongoing examination of actual postmarketing experience seems a wise precaution.

While useful postmarketing information about medicines is sometimes generated by formal studies (a highly desirable but costly approach), observational data about these medicines are already available in Australia as a by-product of administrative processes. These data describe large patient populations and provide a promising new opportunity for monitoring safety of medicines. However, a formal system for examining these data is yet to be implemented, for a number of reasons, including concerns about patient privacy, lack of political will, and restrictions in access to and linkage of the various data collections.

A new approach

The urgent need for increased medicines surveillance has been recently argued by international bodies, such as the International Society for Pharmacoepidemiology, and considered as a component of the new requirements for postmarketing epidemiological studies by the US Food and Drug Administration and the European Agency for the Evaluation of Medicinal Products. Similar discussions are underway locally as legislation is developed for the upcoming Australia New Zealand Therapeutic Products Authority.11

An optimal postmarketing surveillance system should assess all available information about the performance of medicines, drawing on worldwide pre- and postmarketing trials and voluntary reporting, with the addition of findings from regular, systematic and ongoing evaluation using linked observational data.1,7,8,12-14 A number of centres in several countries are already performing high-class pharmacoepidemiological studies of this type.15-19

The utility of linked observational data is well illustrated by the aftermath of the worldwide withdrawal of the cyclooxygenase-2 inhibitor rofecoxib (Vioxx) in response to concerns that the medicine increased heart attack risk. The international literature has since carried a series of over 20 controlled observational and cohort studies evaluating the excess risk associated with use of rofecoxib and related drugs in the United States, Canada and Europe.20 All but two studies employed linked observational data (or linked electronic medical records), permitting rapid evaluation of potential harms. Had these data been analysed cumulatively as soon as they became available (as would occur in a routine surveillance system), rofecoxib might have been withdrawn earlier. A recent meta-analysis of controlled observational studies has indicated that diclofenac has a similar risk profile.21

Australia, as a fortuitous consequence of a complex multi-jurisdictional health care funding system, is replete with observational health care data in linkable databases. They describe the vast majority of health care encounters across the nation, including doctor visits, medicine dispensing, hospital admissions, deaths, and various disease registries. Recent work has shown the practicality of bringing these data together for active monitoring of the outcomes of new drugs while they are being used in the “real world”.22-24 Specific methodologies have been developed to support pharmacoepidemiological analysis of these large datasets.25,26 Encouraging progress has been made over the past few years to improve access to these health datasets. A major linked data resource is now available in Western Australia and a similar resource is being developed in New South Wales.27 To ensure ethical and appropriate use of these data, a nationally accepted protocol has been designed to guarantee both strong privacy protection and accurate linkage.28 While these data have some known limitations and biases, they portray actual practice in use of medicines — and their long-term consequences. With the adoption of appropriate case-inclusion criteria, most of the biases in these data can be controlled.29 Although there are as yet no overseas examples of “routine” medicines monitoring, there have been useful studies based on ad-hoc linkages of medicines and health-outcomes data.30,31

Our preferred approach would support the stated regulatory goals of Australia’s National Medicines Policy.32,33 We propose a staged introduction, incorporating the following features:

Online responses are no longer available. Please refer to our instructions for authors page for more information.