In its 2007 renewal of PDUFA, Congress deemed it wise to earmark funds specifically to safety surveillance. The underlying concern that prompted this action is that prior user fee legislation has resulted in substantial increases in FDA resources applied to review and approval of new medicines but insubstantial increases in resources applied to review of drug safety once medicines have been marketed. It would be difficult to argue against this view; all evidence indicates it’s accurate. Technology advances—large-scale, real-time electronic capture of prescription fills and health outcomes at the patient level—offer the promise of better post-marketing drug safety surveillance, and the public rightly wants to see its government avail itself of these advances to better define risk and provide early warnings of serious safety issues.
Of course, surveillance is an imperfect method. Most surveillance relies on passive observation to identify health outcomes, establish relationships to drug use, and define the magnitude of risk. That is, the observers collect information but don’t intervene in the methods used to generate that information beyond, perhaps, enforcement of coding standards (e.g. ICD, CPT, ATC). A problem with this sort of surveillance is that substantial noise (adverse events) relative to signal (adverse drug reactions) is generated, leading to significant problems interpreting data. This is one of the major problems with FDA’s current and proposed surveillance efforts. More resources can increase the amount of data entering the surveillance funnel (by providing more sources of passive adverse event information), but it won’t necessarily improve the information flowing from the other end.
In the experimental setting such as exists during pre-marketing testing, safety signals may be distinguishable from noise by techniques such as third-party review of adverse events; active intervention to determine causality (e.g. withdraw and careful reintroduction of drug for non-serious events); capture of ancillary history, physical exam and lab tests; and, most usefully, a comparison of event incidence between two experimental treatment groups (ideally including a placebo group). In the “real world,” where much of the proposed heightened surveillance efforts live, such interventions by observers are impractical to implement. Nevertheless, there are opportunities for experimental and quasi-experimental observational safety studies to be conducted in the post-marketing setting, and these types of studies may be used by FDA to supplement traditional surveillance techniques. Examples include nested studies in ongoing, large-scale observational cohorts (e.g. Nurses and Physicians Health Studies) and de novo studies in cohorts whose healthcare is administered by a single institution or network with central administration (e.g. certain large HMOs, VA Health System). By automatically enrolling participants and assigning contemporaneously matched non-drug users to a control group, some of the noise inherent in observational studies may be reduced. It’s smarter than passive capture of data with post-hoc sorting of signal from noise, but it’s still imperfect.
Ideally, sponsors and FDA should be able to identify all serious risks of an investigational drug before bringing it to market, while the drug is still in the experimental (i.e. interventional) setting. Identifying all serious risks before marketing would minimize ‘at-risk’ exposure, establishing those risks at the earliest time practical (assuming that postmarketing surveillance takes much longer than premarketing experimentation to generate unequivocal risk definition, despite much larger exposure to drug in the former). Sponsors would also reduce their exposure to product liability risk and the potential tangible and intangible costs of a recall.
Despite these benefits, however, both sponsors and the FDA have suggested that identifying all serious adverse effects of a drug prior to market approval is impractical, as the numbers of subjects required is prohibitive.
The most pressing safety issue FDA and sponsors grapple with, at least for chronic-use drugs, isn’t what it once was. It’s not identifying rare drug reactions. Although in the past rare, serious drug reactions eluded detection until after approval, sponsors with guidance from FDA have improved their ability to detect the most frequent types of reactions before marketing approval (probably the best example of this is enhanced premarketing surveillance for torsades de pointe due to drug-induced QT interval prolongation).
These days the stickiest safety problems are adverse reactions that may be disguised as serious, unrelated conditions. The voluntary recall of Vioxx is particularly instructive, because it’s a recent example of a serious adverse drug reaction—thrombogenic coronary artery disease—masquerading as a common but serious disease that was missed during premarketing review (some have argued that it wasn’t missed as much as ignored). Much larger populations are required to detect relative increases in the incidence rate (assessed as the hazard ratio) of such relatively common conditions than are required to detect single occurrences of rare conditions.
Understanding this, regulators have occasionally relied on the power of large numbers to help sort out risks premarketing. Notable examples include the large, simple safety studies of Vanlev and Ketek. However, such studies are not demanded routinely. In these two examples, specific risks that had been suggested by smaller studies were investigated further. They were not used to detect risks de novo.
Sponsors are unsurprisingly reluctant to add to the cost and time of premarketing safety evaluations beyond what is already required, and FDA, taking increasingly large chunks of funding from the industry, is unsurprisingly reluctant to ask for such sacrifices. Is there a way out of this conundrum with today’s technology? I believe that large, simple safety studies are an interim answer for chronic-use drugs aimed at large treatment populations, particularly for conditions where suitable alternative therapies exist. In such cases, a population risk threshold should be established for common, serious conditions (essentially serving as the excess allowable hazard for study power/sample size calculations). There is no simple formula for this, as it requires expert assessment of potential population attributable benefits and risks. Different drugs will require different sized populations. Such studies can be run in parallel with routine phase 3 studies and should be started when sufficient preclinical information is available, usually after carcinogenicity studies are complete.
