ICH Q1D: Bracketing and Matrixing Rest on Different Logic

ICH Q1D lets a stability program test less than the full matrix of strengths, container sizes, batches, and time points that ICH Q1A(R2) otherwise calls for. Bracketing and matrixing are the two reduced designs it allows, and they are not interchangeable tools for the same problem. Each rests on a different assumption about what the untested combinations would have shown — and choosing the wrong one, or applying the right one past its limits, is what a reviewer flags, not the testing it saved.

The assumption bracketing actually makes

Bracketing assumes that if the extremes of a design factor hold up, everything between them will too. A product available in three strengths and three container sizes, for example, can bracket to testing only the smallest and largest combinations if the strengths share the same qualitative composition and manufacturing process, and the container sizes or fill volumes genuinely represent the extremes of the range. That is a claim about formulation and process similarity, not a convenience for the study design — and it is the first thing a reviewer checks. A stability program that brackets across strengths with materially different excipient ratios, or across container closures with different headspace or permeability, has not satisfied the assumption; it has just tested less. This is the same discipline a regulatory CMC strategy should already be applying to the underlying control strategy, not a separate exercise bolted onto the stability protocol.

The assumption matrixing makes instead

Matrixing does not reduce which combinations of factors get tested — it reduces which combinations get tested at each time point, on the assumption that the stability of each factor combination can be inferred from the subset actually pulled. A one-half or one-third reduction design, tested against every batch, strength, and container combination in rotation across the study's later time points, can support this — provided the design is balanced and specified in the protocol before data collection starts, not fitted to whichever subset looked cleanest afterward. What matrixing never touches is time zero and the last time point: every factor combination is tested in full on both occasions, because those are the two pulls a reviewer will use to sanity-check the whole reduced design against a complete data set.

• Long-term condition. 25°C ±2°C/60% RH ±5% RH, or 30°C ±2°C/65% RH ±5% RH — the manufacturer's choice for most climatic zones, and the condition matrixing applies to most comfortably.
• Intermediate condition. 30°C ±2°C/65% RH ±5% RH for 12 months, required once significant change occurs at accelerated conditions but the product remains stable long-term.
• Accelerated condition. 40°C ±2°C/75% RH ±5% RH for 6 months — short enough that Q1D does not recommend matrixing it, even though the guideline does not forbid it outright.
• Significant change. A 5% assay shift from the initial result, any degradation product over its acceptance criterion, or a failed appearance, pH, or 12-unit dissolution result — the trigger a reduced design must still be able to catch.

A reduced design does not lower the bar for what the study has to prove. It only changes how much testing is needed to prove it — and only when the assumption behind the reduction actually holds. Why the justification comes before the protocol

Where this connects to the rest of the submission

A stability program does not exist to satisfy Q1D in isolation. It exists to support the shelf-life claim and the specification a product has to hold to throughout it — the same specification ICH Q6A/Q6B's decision trees are built to justify. For solid oral products in particular, where oral solid dosage development work often spans multiple strengths built on a common blend, a well-justified bracketing design across strengths can be one of the more defensible reductions available — provided the formulation science backs it up before the protocol is written, not after a reviewer asks. Whatever design a program uses, the CTD stability section has to carry the statistical justification in full, not just the reduced data set; a table of results without the reasoning behind the reduction reads, to a reviewer, exactly like an unexplained gap.

None of this is exotic once the two designs are kept separate in the team's own thinking. Bracketing is a claim about the product family; matrixing is a claim about how much data is needed to characterize a single stable trend. Programs that reach for either one to save testing cost, without first confirming the underlying assumption, tend to discover the gap during review rather than before submission — at which point the fix costs considerably more than the original full design would have.