For debate: In a clinical development program clinically meaningful safety signals CANNOT be reliably determined
The evidence is quite simple: the troglitazone signal for acute liver failure was not found by the company or regulators in its clinical development programme; the issue of fractures with PPAR-γ agonists was discovered 5 years after licensing in a study designed for completely separate purposes; concerns over ketoacidosis in people with both type 1 and type 2 diabetes using SGLT2 blockers appeared from clinical practice; and many other issues for DPP-4 inhibitors, SGLT2 blockers, GPR-40 agonists and PPAR-γ agonists have appeared only as a result of non-core monitoring of adverse events in studies designed to provide reassurance over cardiovascular (CV) side effects. And that is only diabetes!
Indeed without the CV studies, mandated only for CV safety, simple arithmetic shows the impossibility of showing less common adverse events in a typical phase 2/3 clinical development programme. A very large programme might expose as many as 2000 people to new drug, but usually only for 6 months for the most part. There is some requirement for a sub-population to be followed for longer, but in all we will typically not have more than 2000 people-years of exposure. It is immediately obvious that a serious side effect (say causing death) occurring at 1:1000 patient-years may not even occur in such a clinical development population, allowing for chance, while for the most part we would be concerned by much rarer events, as with troglitazone. In adverse event monitoring hundreds of serious events (SAEs) are dutifully recorded, but small statistical fluctuations in some are inevitable, so real increases in something like uterine cancer, pancreatitis, or suicide stand essentially no reliable chance of being identified.
Recently in diabetes CV outcome trials have been mandated. These have markedly increased exposure overall, and a positive side effect of this, with the usual careful SAE monitoring, has been detection of unexpected problems notably with the DPP-4 inhibitors and SGLT2-blockers. Curiously the regulators still see these studies as directed to CV safety, and indeed they are not even mandated in Europe and much of world. But even these studies cannot give us reassurance of long-term CV safety – for reasons of efficiency, time, and statistical power the study populations are atypical of those of ambulatory diabetes care, sometimes very atypical, and study durations are very short compared to life expectancy with type 2 diabetes. For example the EMPA-REG-OUTCOME results are in theory entirely compatible with increased death rate after 10 years exposure to empagliflozin, though there is only a signal for this for stroke.
It may be thought that post-marketing surveillance should fill the gap. Indeed the regulators, cognizant of the above problems, routinely demand it. But it is manifestly not fit for purpose as the problems over pancreatitis with GLP-1 receptor agonist emphasize, in particular due to problems over prescribing and reporting biases.
In conclusion safety signals cannot be reliably determined with the current clinical development programme. The reasons for this are fundamental and we have to live with them. It is for this reason that clinical benefit needs to be reliably established, such that the risk:benefit ratio is likely to remain positive when an undetermined side effect later appears, or indeed remains for even hidden.
Presented by: Professor Philip Home, Newcastle University, UK