Though in practice it’s more complicated. Basically, make predictions based on accelerated testing and then validate them in real time tests. So, if things hold up at 3 and 6 months real time you can probably trust them over longer timeframes.
Exactly. Plus dating tends to get extended post-approval.
If Pfizer has a short-list of candidates back in May, they could have had 6 months of stability data at approval.
They can keep those studies going and by next May have 12 months of data for the FDA to review and potentially update the transportation and storage requirements.
Yea, prescription drug manufacturers have little incentives to seek multi year shelf lives. Combined with the slow approval process real time testing is likely considered sufficient.
I think you summarized it well. There is little incentive to test the limits of stability as manufacturers typically control channel inventory quite tightly. I know my company targeted maybe 14 days on hand? So going much beyond 2 years stability didn’t make much sense as most product was consumed well before it expired.
Not OP, but this is the same way lightbulb or hard drive manufacturers can claim "greater than 50,000 hours of operation". Instead of testing for over a decade, they test many samples over a shorter time duration, and then they extrapolate. Like any other extrapolation, these estimates have their caveats. But for biological reagents, something like an exponential decay is often considered reasonable.
Accelerated studies. Usually you test stability for shorter periods of time under harsher conditions. It’s not perfect, but if your product is fine at 50C for 2 months, it’ll probably be fine refrigerated for 12 months.
Assume that the reaction rate (shelf life) versus temperature relationship is described by the Arrhenius equation or some other model; perform experiments at elevated temperatures, and extrapolate to lower temperatures.
Though in practice it’s more complicated. Basically, make predictions based on accelerated testing and then validate them in real time tests. So, if things hold up at 3 and 6 months real time you can probably trust them over longer timeframes.