Stability Bracketing and Matrixing: Which Cells to Test, Which to Infer

The short answer: what a reduced stability design does — and does not — permit

Under ICH Q1D, a stability protocol can be reduced before the study starts in two ways. Bracketing tests only the extremes of a design factor — for example, the lowest and highest strengths, or the smallest and largest container sizes or fills — at every time point, on the assumption that the stability of intermediate levels is represented by the extremes. Matrixing tests a selected subset of all factor combinations at one time point and a different subset at the next, on the assumption that each tested subset represents the stability of all samples at that time point.

Neither design is a discount on the baseline. ICH Q1A(R2) still expects formal stability data on at least three primary batches of the same formulation, packaged in the container-closure system proposed for marketing, and states that reduced designs can be applied only “if justified.” In a time-point matrix, every selected factor combination is still tested at the initial and final time points, and also at 12 months (or the last time point before submission) when full long-term data will not be available for review before approval. In the United States, 21 CFR 211.166 separately requires a written stability program whose sample sizes and test intervals rest on statistical criteria.

What a reduced design actually buys is inference. Untested strengths, packs and time points are assumed to behave like the tested cells. Q1D asks sponsors to weigh, before adopting a reduced design, the risk of establishing a shorter shelf life than a full design would have supported. When the data contradict the assumptions — extremes that degrade differently, batches or factors that cannot be pooled, a matrix too thin to estimate the missing combinations — the intermediates in a bracketing design are treated as no more stable than the least stable extreme, ICH Q1E permits no interpolation between extremes, and the smaller data set can support a shorter shelf life than a full design.

Full versus reduced designs: the baseline every reduction is measured against

Q1D defines a full study design as one in which samples for every combination of all design factors are tested at all time points; a reduced design is one in which they are not. Design factors are the variables evaluated for their effect on stability — typically strength, container size and/or fill, and batch. Q1D says a reduced design can be a suitable alternative when multiple design factors are involved, but “any reduced design should have the ability to adequately predict the retest period or shelf life,” and data from a reduced design are evaluated in the same manner as data from a full design.

The baseline schedule comes from Q1A(R2). For a product with a proposed shelf life of at least 12 months, long-term testing should normally occur every 3 months over the first year, every 6 months over the second year and annually thereafter — 0, 3, 6, 9, 12, 18, 24 and 36 months for a 36-month claim. For the general case, the long-term condition is 25°C ± 2°C / 60% RH ± 5% RH or 30°C ± 2°C / 65% RH ± 5% RH, and the accelerated condition is 40°C ± 2°C / 75% RH ± 5% RH, where Q1A(R2) recommends a minimum of three time points, including initial and final (for example 0, 3 and 6 months), from a six-month study.

Consider an illustrative solid oral product registered in three strengths and three bottle sizes, with three primary batches. A full design places 27 batch–strength–container combinations on long-term storage. On the eight-point, 36-month schedule that is 216 long-term test sets; the three accelerated time points add another 81, for 297 before any intermediate-condition testing. Each test set means the protocol’s full battery of stability-indicating tests, because Q1D says matrixing should not be performed across test attributes. These numbers are arithmetic, not product data; a real protocol is sized to its own factors, storage conditions and shelf-life claim.

Reduction risk is not uniform, and Q1D draws a distinction that much definitional content misses. A design that matrixes on time points only “would often have similar ability to that of a full design to detect differences in rates of change among factors and to establish a reliable shelf life,” because linearity is assumed and every factor combination is still fully tested at the initial time point and the last time point before submission. By contrast, matrixing on factors other than time points “generally has less precision in shelf life estimation and yields a shorter shelf life than the corresponding full design.” The shelf-life risk concentrates in designs that leave whole batch–strength–container combinations untested.

A reduced design also leaves less slack for operational failures. In a full design, an unusable sample at one intermediate pull usually leaves adjacent time points for the same combination; in a matrixed design, the same loss can remove one of only a few data points for that combination, and Q1D’s minimum of three time points (including initial) through the first 12 months for each selected combination leaves little room. Investigating and documenting that kind of gap is a separate question, covered in our analysis of how a missed stability pull affects the filing.

Bracketing eligibility: strengths, container-closure systems, and proving your extremes

Bracketing assumes the stability of any intermediate level is represented by the extremes tested. Q1D is direct about the precondition: a bracketing design “would not be considered appropriate if it cannot be demonstrated that the strengths or container sizes and/or fills selected for testing are indeed the extremes.” It also asks sponsors to assess the effect of a bracketing design on retest-period or shelf-life estimation before applying it.

For strength bracketing, Q1D allows multiple strengths of identical or closely related formulations. Its examples, which it says are not exhaustive, are:

• Capsules of different strengths made with different fill plug sizes from the same powder blend.

• Tablets of different strengths manufactured by compressing varying amounts of the same granulation.

• Oral solutions of different strengths whose formulations differ only in minor excipients, such as colourants or flavourings.

With justification, Q1D also allows strength bracketing where the relative amounts of drug substance and excipients change; that justification can include a demonstration of comparable stability profiles among clinical or development batches of the different strengths. Where different excipients are used among strengths, Q1D says bracketing “generally should not be applied.” The Egyptian Drug Authority’s September 2026 FAQ (NP.CAPP.113) applies the same line for Egyptian submissions and adds that its assessors look for a relatively constant active-to-excipient ratio across the strength range and unchanged excipient functionality, degradation pathways and critical quality attributes such as moisture sensitivity and dissolution.

Container-closure bracketing applies to the same container-closure system where either container size or fill varies while the other stays constant. When both vary, Q1D warns that “it should not be assumed that the largest and smallest containers represent the extremes of all packaging configurations.” Extremes should instead be selected by comparing the characteristics of the container-closure system that may affect stability:

• Container wall thickness.

• Closure geometry.

• Surface-area-to-volume ratio.

• Headspace-to-volume ratio.

• Water vapour permeation rate or oxygen permeation rate per dosage unit or unit fill volume, as appropriate.

These comparisons decide the design, not just the narrative. The EDA FAQ gives a concrete case: if the evaluation shows that an intermediate configuration — such as a mid-size container with a low fill volume — represents a higher risk, that configuration must be included in testing, and failure to demonstrate that the selected extremes represent all packaging configurations may lead to the bracketing design being rejected. Q1D allows bracketing for the same container when only the closure varies, with justification such as a discussion of relative permeation rates. Different container-closure systems — blisters versus bottles, for instance — fall outside Q1D’s bracketing conditions; Q1D contemplates matrixing, not bracketing, across different systems, and only with supporting data.

Q1D also addresses a common portfolio change. If, after the studies start, one of the extremes is no longer expected to be marketed, the bracketing design can be maintained to support the intermediates, with a commitment to carry out stability studies on the marketed extremes post-approval. For drug substances, by contrast, Q1D says bracketing is generally not applicable and matrixing is of limited utility.

Matrixing eligibility: factor rules, balance, and the time points you can never drop

Matrixing keeps every factor level in the study but rotates which combinations are tested at each intermediate time point. Q1D lists factors that can be matrixed: strengths with identical or closely related formulations (the same capsule, tablet and oral-solution examples used for bracketing), batches made with the same process and equipment, and container sizes and/or fills in the same container-closure system. With justification based on supporting data, matrixing can extend to strengths whose drug-to-excipient ratios change, strengths with different excipients, or different container-closure systems — for example, with data on relative moisture vapour transmission rates or light protection, or data showing the product is not affected by oxygen, moisture or light.

Two limits are fixed. First, “each storage condition should be treated separately under its own matrixing design”: long-term, intermediate and accelerated conditions each carry their own scheme rather than sharing one rotation. Second, “matrixing should not be performed across test attributes”: a combination pulled at a time point is tested against the protocol’s attributes, not a rotating subset of them. Q1D does allow alternative matrixing designs for different test attributes if justified, but that is a justified exception rather than a default.

Whatever the degree of reduction, Q1D keeps certain time points fully tested in a time-point matrix:

• Initial and final: all selected factor combinations are tested at the initial and final time points.

• 12 months or the last pre-submission point: if full long-term data for the proposed shelf life will not be available for review before approval, all selected combinations are also tested at 12 months or at the last time point prior to submission.

• Three points in the first year: data from at least three time points, including initial, should be available for each selected combination through the first 12 months.

• Accelerated and intermediate conditions: testing should occur at a minimum of three time points, including initial and final, for each selected combination of factors.

Q1D asks for a matrix that is “balanced as far as possible so that each combination of factors is tested to the same extent” through the last time point before submission, while acknowledging that mandatory full testing can make complete balance difficult. Those anchors also explain why the labels overstate the savings. In Q1D’s Table 2 examples — two strengths, three batches and an eight-point schedule to 36 months, or 48 strength–batch–time-point cells — the “one-half reduction” ultimately removes 15 of the 48 cells (about 31%) rather than 24, and the “one-third reduction” removes 10 (about 21%) rather than 16, because the initial, 12-month and final points remain fully tested.

Applicability tracks variability. Q1D says matrixing is appropriate when supporting data show only small variability; where variability is moderate, the design should be statistically justified, for example by evaluating its power to detect differences in degradation rates among factors or its precision in shelf-life estimation; where variability is large, matrixing should not be applied. The more factors and levels a product has, the larger the reduction that can be considered — always subject to the design still predicting shelf life adequately. The EDA FAQ adds, for its own reviews, that all samples, including unscheduled ones, should still be placed on stability.

The coverage map: tested versus inferred cells across strength, package, and time point

To make the trade-offs concrete, apply Q1D’s designs to one illustrative product: three strengths (S1, S2, S3), three container sizes in the same HDPE container-closure system (A, B, C), three primary batches, and Q1A(R2)’s eight-point long-term schedule to 36 months (0, 3, 6, 9, 12, 18, 24, 36). The full design is 27 combinations × 8 time points = 216 long-term test sets. Rows 4 to 6 reproduce the structure of Q1D Table 1 (bracketing on strength and container size) and Tables 3a and 3b (matrixing on time points, and on time points and factors); rows 2 and 3 apply the same bracketing logic to one factor at a time. Accelerated and intermediate conditions would each need their own design and are excluded from the counts. All arithmetic is illustrative.

Three readings matter more than the percentages. First, the largest cuts come from never testing a combination at all — rows 4 and 6 — which is exactly where Q1D and Q1E locate the shelf-life risk. Second, time-point-only matrixing (row 5) saves the least but keeps every combination tested at the start and end, which is why Q1D says it often performs similarly to a full design. Third, Q1D gives no worked example that stacks bracketing and matrixing; it says that because the two rest on different principles, “careful consideration and scientific justification should precede” their combined use, and Q1E applies the same multi-factor statistical procedure to the combined data. A team that wants the deepest reduction should expect to defend every inferred cell individually.

When the inference breaks: diverging extremes, failed pooling, and unestimable combinations

Every reduced design rests on assumptions that the stability data can later contradict. Q1D and Q1E spell out what happens when they do, so the consequences can be planned for rather than discovered at review. The ledger below maps each reduction to the assumption it makes, the evidence expected before the study, and the documented consequence when the assumption fails.

The most familiar failure is extremes that age differently. Q1E’s own example is a bracketing design in which only the smallest and largest container sizes, P1 and P3, are tested: if P1 proves less stable than P3, the shelf life for the intermediate P2 “should not exceed that for P1,” and “no interpolation between P1 and P3 should be considered.” The same logic applies to strengths, so an untested middle strength cannot be given a shelf life halfway between its two brackets. The EDA FAQ adds a practical nuance for its own reviews: differing trends between extremes do not automatically invalidate the design, but a conservative shelf life based on the least stable trend is expected, and significant divergence may prompt a request for additional data or the inclusion of an intermediate configuration.

The second failure is statistical. Q1E tests whether batches share a common slope and intercept before their data are combined. If equality of slopes is rejected, the data from all batches are not combined; shelf lives are estimated for individual batches using individual slopes and intercepts with the pooled mean square error from all batches, and the shortest estimate becomes the shelf life for all batches. Where differences are readily attributed to a particular factor, such as strength, Q1E allows different shelf lives to be assigned to different levels of that factor instead of one shortest value for the whole product. Either way, a reduced design that relied on pooling to reach a target claim can lose that claim when pooling fails — and a factor-matrixed design has fewer observations per combination with which to demonstrate poolability in the first place.

The third failure is structural. Q1D warns that if a matrix excessively reduces the number of factor combinations tested and the tested data cannot be pooled into a single shelf life, “it may be impossible to estimate the shelf lives for the missing factor combinations.” Q1E adds that an unbalanced design can leave some factors or factor interactions non-estimable. Finally, an out-of-specification result at a tested extreme is treated as a signal about the whole bracket: the EDA FAQ calls it indicative of a potential worst-case failure, considers all intermediate configurations supported by the design to be at risk, and expects a thorough investigation whose corrective actions may include shelf-life reduction or additional stability testing. Separating a genuine trend from an isolated result is its own discipline, covered in our analysis of how stability out-of-trend signals derail launch supply.

The shelf-life price of reduction: what Q1E statistics do with a smaller data set

Q1E explains why a reduced design can cost shelf life even when every observed result is within specification. For an attribute known to decrease with time, the shelf life is estimated as the time at which the lower one-sided 95% confidence limit for the mean intersects the acceptance criterion; for an attribute known to increase, the upper one-sided 95% limit is used; two-sided limits apply where both upper and lower criteria matter or the direction of change is unknown. The estimate depends on the width of that confidence band, not only on the mean trend.

Q1E’s poolability rules then decide how much data each estimate can draw on:

• Batch-related terms at 0.25: Q1E recommends testing batch poolability at a significance level of 0.25, to compensate for the expected low power of designs with the relatively limited sample sizes typical of formal stability studies.

• Non-batch terms at 0.05: for multi-factor designs, Q1E recommends 0.25 for batch-related terms and 0.05 for non-batch terms such as strength or container size.

• Slopes before intercepts: equality of slopes is tested first. If slopes are common but intercepts differ, the data can be combined to estimate a common slope, individual-batch shelf lives are estimated with individual intercepts, and the shortest estimate applies to all batches. If neither slopes nor intercepts differ, the data from all batches can be combined into a single estimate.

• When pooling fails across factors: Q1E allows either a separate shelf life for each level of the factors remaining in the model, or a single shelf life based on the shortest estimate among them.

• A stated model for matrixing: the analysis should identify the procedure and assumptions — including any assumption that interaction terms are negligible — and state the final model on which the shelf-life estimate rests.

Fewer observations widen the confidence band in two ways. With fewer data points, the residual degrees of freedom fall and the critical t-value used for the one-sided 95% limit rises. And when time points are removed, the remaining observations carry less information about the slope, so the uncertainty of the predicted mean at later times grows. Both effects bring the confidence limit to the acceptance criterion earlier, which is why Q1E cautions that “the use of a matrixing design can result in an estimated shelf life shorter than that resulting from a full design.”

The size of that effect is product-specific, and this article does not quantify it: doing so requires the product’s own degradation rate, variability and acceptance criteria. Q1D’s practical answer is to assess it before the study starts — a statistical justification for a matrix can rest on its power to detect differences in degradation rates among factors or on its precision in shelf-life estimation. A CMC team targeting a specific commercial shelf life should run that evaluation against the claim it needs, not against the testing budget alone. Q1E also limits how far a shelf life may be extrapolated beyond the period covered by long-term data, so a thinner data set leaves less room to propose a claim longer than the real-time results.

Q1D also allows a design to be strengthened midstream. During a reduced-design study, a change to full testing or to a less reduced design can be considered if justified, with proper adjustments to the statistical analysis for the increased sample size; once changed, the fuller testing should continue through the remaining time points. That option is a mitigation, not a substitute for sizing the design correctly at the start.

What to document: justification content and where assessors look for it

Q1D says the use of any reduced design should be justified, and that certain assumptions should be assessed and justified before a reduced design is considered. The type and level of justification depend on the available supporting data; in some cases the conditions Q1D describes are sufficient justification, while in others additional justification is needed. A practical checklist drawn from Q1D and Q1E:

• Design description: which factors are bracketed or matrixed, which combinations are tested at which time points, and a table that distinguishes tested from untested configurations.

• Sameness evidence: the formulation relationship among strengths and the identity of the container-closure system across sizes and fills.

• Extremes evidence: the container-closure comparison of wall thickness, closure geometry, surface-area and headspace ratios, and permeation per dosage unit or fill volume.

• Variability and stability evidence: supporting data on variability and expected stability, and, where variability is moderate, a statistical evaluation of power or precision.

• Statistical plan: the poolability approach and significance levels, the model and its assumptions, and how shelf life will be assigned if pooling fails or extremes differ.

Where that justification sits in the dossier depends on each authority’s expectations. The EDA FAQ is unusually explicit for Egyptian submissions, distributing bracketing content across Module 3:

• 3.2.P.2.4 (container closure system in pharmaceutical development): the packaging rationale, including material equivalence, surface-area-to-volume comparisons and moisture/oxygen permeation rates showing the intermediates are represented by the extremes.

• 3.2.P.7 (container closure system): descriptions, engineering schematics, analytical methods and material specifications for the commercial packaging components.

• 3.2.P.8.1 (stability summary and conclusion): a concise summary of the bracketing design, a matrix table identifying tested and untested configurations, and the proposed shelf life and storage conditions for the product line.

• 3.2.P.8.2 (post-approval stability protocol and commitment): the protocol showing how post-approval batches will continue to be monitored using the reduced design.

• 3.2.P.8.3 (stability data): data tables, trend lines and regression or poolability analyses showing that the extremes support the intermediate configurations.

Two EDA expectations should not be generalized. The FAQ requires a bracketing protocol to be formally endorsed through preassessment before it is applied to primary stability batches, and it says bracketing must not be applied to clinical batches during investigational stages. These are Egyptian procedural requirements. ICH Q1D itself requires justification, not prior agency approval, and FDA and the EU implement Q1D as guidance — FDA as a final guidance issued in January 2003 (docket FDA-2001-D-0164) and the EU as a Step 5 scientific guideline (CPMP/ICH/4104/00) with a legal effective date of 1 August 2002.

Underneath the guidance sits binding US regulation. 21 CFR 211.166 requires a written testing program with sample sizes and test intervals based on statistical criteria for each attribute, testing of the drug product in the same container-closure system in which it is marketed, and testing of an adequate number of batches to determine an appropriate expiration date. Accelerated data may support tentative expiration dates only while full shelf-life studies are under way. A reduced design does not relax any of these requirements; it is one way of meeting the statistical-criteria requirement, and its justification is what shows that it does. When CMC questions surface late in review, the commercial consequences can be significant, as our analysis of how commercial teams should read CMC-only complete response letters describes.

On the horizon: the 2025 ICH Q1 draft and reduced-design strategy

On 11 April 2025, ICH endorsed a Step 2 draft of a revised ICH Q1, “Stability Studies for Drug Substances and Drug Products,” for public consultation. The draft describes itself as a consolidated revision that supersedes ICH Q1A–F and Q5C. Reduced designs move into Annex 1, “Reduced Stability Protocol Design,” and a new Annex 2 covers stability modelling.

For protocol planners, three changes stand out. First, Annex 1 keeps the bracketing and matrixing principles — including the least-stable-extreme rule and the warning that matrixing can produce a shorter estimated shelf life — and says reduced designs may be proposed for any formal stability protocol: primary, commitment, ongoing (annual) and product lifecycle. Second, it adds knowledge- and risk-based protocol reductions beyond bracketing and matrixing, in which attributes, time points, samples and/or storage conditions may be reduced where a risk assessment supported by data and product knowledge justifies it, along with worst-case strategies in which a predictable worst-case presentation is tested on behalf of the others. Third, for drug substances it keeps the position that matrixing is usually of limited utility and bracketing generally not applicable, while allowing reduced time points and/or attribute testing to be justified where little or no degradation occurs.

The draft’s general principles stay close to Q1D’s logic: any reduced design should meet the study objective “with a defined and acceptable risk as compared to a full design,” the use of any reduced design should be justified, and combining more than one reduction principle needs careful consideration and scientific justification. If a reduced design is introduced after the original marketing authorisation, change-management procedures apply in accordance with regional requirements.

Until the revised guideline reaches Step 4 and is implemented regionally, it is a consultation text, not a compliance basis. ICH Q1D, as implemented by FDA and in the EU, remains the operative guidance for reduced designs — guidance rather than binding law, sitting alongside binding requirements such as 21 CFR 211.166 in the United States. A protocol written today should be justified under Q1D and Q1E; the draft is mainly useful as a signal of where knowledge- and risk-based justification is heading.

Sources

• International Council for Harmonisation (ICH). ICH Q1D Bracketing and Matrixing Designs for Stability Testing of New Drug Substances and Products. Current Step 4 version, 7 February 2002. https://database.ich.org/sites/default/files/Q1D%20Guideline.pdf

• U.S. Food and Drug Administration (FDA). Q1D Bracketing and Matrixing Designs for Stability Testing of New Drug Substances and Products — FDA guidance page. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/q1d-bracketing-and-matrixing-designs-stability-testing-new-drug-substances-and-products

• International Council for Harmonisation (ICH). ICH Q1A(R2) Stability Testing of New Drug Substances and Products. Current Step 4 version, 6 February 2003. https://database.ich.org/sites/default/files/Q1A%28R2%29%20Guideline.pdf

• International Council for Harmonisation (ICH). ICH Q1E Evaluation of Stability Data. Current Step 4 version, 6 February 2003. https://database.ich.org/sites/default/files/Q1E%20Guideline.pdf

• U.S. National Archives and Records Administration. 21 CFR 211.166 — Stability testing. Electronic Code of Federal Regulations (eCFR). https://www.ecfr.gov/current/title-21/section-211.166

• International Council for Harmonisation (ICH). ICH Q1 Stability Studies for Drug Substances and Drug Products. Step 2 draft endorsed 11 April 2025. https://database.ich.org/sites/default/files/ICH_Q1EWG_Step2_Draft_Guideline_2025_0411.pdf

• European Medicines Agency (EMA). ICH Q1D Bracketing and matrixing designs for stability testing of drug substances and drug products — Scientific guideline (Step 5). Reference CPMP/ICH/4104/00. https://www.ema.europa.eu/en/scientific-guidelines/ich-q1d-bracketing-matrixing-designs-stability-testing-drug-substances-drug-products

• Egyptian Drug Authority (EDA). Notice to Applicant: FAQ on Bracketing and Matrixing in Stability Studies. Central Administration of Pharmaceutical Products, General Administration for Stability, NP.CAPP.113, Version 1, issued 14 September 2026. https://www.edaegypt.gov.eg/media/gy2gnxu4/notice-to-applicant-faq-bracketing-and-matrix_en-ver1.pdf