when public health decisions have to be made, if postmarketing safety signals are indeed found.
This last point is critical for RCTs in that one of their main disadvantages cited is that they take too long. An alternative perspective is that they are started too late. If an RCT is considered only when very strong suspicion of a safety signal arises, it may be too late to initiate one, particularly if the adverse events occur long after exposure. In addition, some of the ethical difficulties that arise in trying to conduct an RCT of a widely used drug are minimized if the RCT is initiated soon after market introduction. In 1999, when rosiglitazone and pioglitazone had just been approved, the TIDE trial would have been an important, timely, and well-designed study. Because at that time there would have been populations of patients who were not on the drug, individual or cluster randomization poses fewer logistical and ethical difficulties than would exist later. However, because of FDAAA restrictions, a study could only be required when there were premarketing signals of a potential for serious risk (for example, adverse lipid alterations, or low frequency serious adverse events) and observational studies were deemed inadequate. Early initiation of postmarketing investigations can dramatically change both the scientific value and the ethical calculus of such studies, making the optimal sequence dependent on when in the lifecycle of a drug the studies are being contemplated.
One key determinant of the kinds of designs that might be considered is whether the safety signal is a harm to be offset by a known benefit or the harm is a failure to provide expected benefit, either overall or in specific populations. FDAAA defines the latter as a safety concern, or more specifically “any failure of expected pharmacological action of the drug”.4 A failure to provide expected clinical benefit is most likely to be observed if that benefit was not directly tested in the approval process, for example, if surrogate endpoints were used, or if non-responsive subpopulations were not well represented in premarketing trials. If the safety concern focuses on the issue of no or reduced benefit, an RCT is more likely needed because confounding by indication in observational studies of drug benefit can be difficult to overcome. Conversely, if the harm is distinct from the mechanism of benefit, is unforeseen and not strongly linked to patient or disease characteristics, or is strongly linked to conditions of general practice (such as cotreatments or inconsistent monitoring), observational studies can often provide sufficiently reliable evidence related to risk. As previously noted, that degree of sufficiency also depends on the relative degree of increase in the risk that is deemed important to detect, weighed against the likely magnitude of confounding.
The dimensions of quality that must be judged for each combination of study design, data source, and analytic approach are the precision, bias, and transportability of the result. Each of those contributes to the observed effect’s potentially
421 USC § 355-1(b) (2010).