EU GMP 1.10 says the review exists to verify consistency and 'to highlight any trends'. Paragraph 1.11 exists to decide whether corrective and preventive action or revalidation is needed. EU GMP Annex 15, section 5.28 onwards, requires manufacturers to monitor product quality under ongoing process verification 'to ensure that a state of control is maintained throughout the product lifecycle with the relevant process trends evaluated', and 5.30 says statistical tools 'should be used, where appropriate, to support any conclusions with regard to the variability and capability of a given process'. None of these can be satisfied by a table of results with a column saying 'pass'. This guide covers the minimum statistical evaluation that satisfies them, and how to write about it.
What 'trend' means to an inspector
An inspector reading a PQR is not looking for a p-value. They are looking for evidence that someone plotted the data over time, compared it with the specification and with the previous period, applied a defined rule for what counts as a signal, and wrote down what they saw. A trend, in this sense, is any of the following: a shift in the mean, a change in spread, a drift in one direction, a run of results on one side of the centre line, a cyclical pattern, or a result that is within specification but outside what the process has previously produced (out of trend). The procedure should define which of these the site looks for and how.
Which chart
For batch release results, where each batch gives one value per test, the individuals and moving range chart (I-MR) is the standard choice. It plots each result in batch order with a centre line at the mean and control limits at plus and minus three standard deviations estimated from the moving range. For in-process data with multiple measurements per batch (tablet weights, fill volumes), an X-bar and R or X-bar and S chart is appropriate. For attribute data such as defect counts, a p or c chart. For most PQRs the I-MR chart on every critical finished product and in-process parameter is the backbone.
Control limits are not specification limits. Specification limits are the registered acceptance criteria; control limits describe what the process actually does. A process whose control limits sit well inside the specification is capable; one whose control limits approach or cross the specification is producing conforming batches by luck. Plot both on the chart and the reader sees the relationship at once.
How many batches
Control limits calculated from fewer than about twenty to twenty-five points are unstable, and a review period with eight batches will not support them. Options: carry limits forward from a longer baseline (previous periods, stated on the chart), pool periods for the chart while reporting the current period's statistics separately, or, for very low-volume products, present the data against the specification with a plain statement that the number of batches does not support control limits and that each result was compared with the historical range. What is not acceptable is calculating limits from six batches and reporting the process as 'in control' because none of the six fell outside limits derived from themselves.
Capability
Capability indices summarise how much room the process has inside the specification. Cpk uses the within-subgroup or moving-range estimate of variation and describes short-term potential; Ppk uses the overall standard deviation of all the results and describes what was actually delivered over the period. For batch release data Ppk is usually the more honest number, and the report should say which was calculated and how. A commonly used threshold is 1.33, below which the process is generally regarded as needing attention, and below 1.0 the process is expected to produce out-of-specification results. The threshold is not regulatory; it is a site decision that belongs in the procedure with its rationale.
Capability is only meaningful if the process is stable and the data is roughly normal. A Ppk of 1.8 computed across a step change in the mean is meaningless. Check the chart before quoting the index. And do not compute capability for one-sided or non-numeric tests, or for results reported as 'complies' or 'less than the limit of quantitation'; say why instead.
Run rules and out-of-trend
Define the rules your site applies. The Western Electric and Nelson rule sets are the usual references: one point beyond three sigma; two of three beyond two sigma on the same side; four of five beyond one sigma on the same side; eight or nine successive points on one side of the centre line; six successive points increasing or decreasing. Pick a subset, write it into the procedure, apply it consistently, and report every rule triggered. An out-of-trend result, one that is within specification but outside the process's demonstrated range or breaks a rule, should have been handled at the time under the OOT procedure; the PQR checks that it was and looks for OOTs that were not recognised.
Stability trending
Item (vii) asks for adverse trends in stability. For a degradant or assay result over time, the question is whether the slope, extrapolated to the end of shelf life, stays inside the limit with margin. Plot each study's results by time point on the same axes as previous batches of the same product; a batch whose line is steeper than the rest is an adverse trend even if every point so far complies. Regression across batches, as used for shelf-life estimation under ICH Q1E, is the formal tool; for the PQR, an overlaid plot with a stated judgement is usually enough, provided the judgement is made.
Linking trending to the other sections
Trending is where the twelve sections connect. A shift in the mean should be datable, and the date should match a change control in item (v), a material lot change in item (i), or a deviation in item (iv). If the chart shows a shift and none of those sections explains it, the review has found something, and the conclusion should say so. If the sections do explain it, the review has demonstrated that it understands its own process, which is what verifying consistency means.
Module 4 of the course provides the trending workbook, walks through I-MR and capability calculations on a real data set, and drills the evaluation sentences until they are habit.