There certainly isn’t a regulatory rationale for having that policy, but I can think of a few business reasons. The first, if your shelf-life is 24-months or less, would be that until 18-months you might not have enough data for sound statistics, and if the 95% confidence interval gave disturbing projections they could be a false alarm simply because of the fewer data points. The other business rationale could simply be to reduce the resources needed since so many batches being analyzed could really balloon to a heavy statistical workload. The justification for waiting could simply be that the filed data shows that there’s really no danger of OOS so why not wait? Both of these business rationales might be worth some risk financially, but they’re based on the flawed assumption that statistics is only trying to justify shelf-life when, in fact, they are also trying to catch unexpected shifts in product performance long before shelf-life, if possible. You need to at least have some mechanism in place to assure the relevant agency that you’re checking lot performance routinely throughout the studies. If not statistics, then at least some sort of visual check of your graphical or tabular data.