I’m not an expert on this, but have seen companies where I like what they are doing, and others where I thought there could be improvements.
Statistical Method - annual product review
In my experience there is an annual product review. This document looks back on the product release criteria, and tracks and trends the various parameters. This includes ALL lots of the prior year. At this time statistics can be done showing confidence levels, excursions from “Controls” and also excursions from acceptance levels (this should have been identified previously during release testing).
The excursions from controls (statistics, not pass/fail criteria) are explained or investigated.
As you can imagine, relying on annual product reviews to show statistical control means that the evaluation is often done months later, when any data to support an investigation might be old and less-helpful (how can you investigate something 8 months after it happened, when sometimes what happened might be in someones memory, not ideal, but it happens).
Statistical Method (real time)
Alternatively, having real time data anlytics is the idea, but is often expensive and for legacy/existing processes would require an investment of capitol.
Every so often (based on a schedule, or based on recommendation from QA, or due to upsets/trends) perform a re-validation. This is a mininimum of 3 lots, THis is the “old school” method and might be acceptable, but additional review would likely be requested.