Stability Study Data Quality: Where Long-Horizon Programs Drift

Why a Single Pull Point Hides the Problem

Most quality controls in a pharmaceutical laboratory are built around a single event. A batch is released or it is not. A sample passes or it fails. An out-of-specification (OOS) result triggers an investigation. These controls work because the unit of decision is small and the evidence is all in one place at one time.

Stability data does not behave that way. ICH Q1A(R2), the guideline that has governed formal stability studies since its Step 4 adoption on 6 February 2003, recommends that long-term testing for a product with a proposed shelf life of at least 12 months take place every 3 months over the first year, every 6 months over the second year, and annually thereafter through the proposed shelf life, on at least three primary batches.¹ Multiply that by strengths, container-closure systems, and storage conditions, and a single product carries dozens of studies with hundreds of scheduled pulls spread across years.

The value of that data is not in any one result. It is in the series. Shelf life is established by evaluating the change pattern across time points, and ICH Q1E explains that whether extrapolation is appropriate depends on the extent of knowledge about the change pattern, the goodness of fit of any mathematical model, and the existence of relevant supporting data.² A trend evaluation, an out-of-trend (OOT) assessment, a shelf-life extension, and an inspector’s question about whether a product remains within specification all consume the series as a whole.

That is why single-point controls miss the drift. When an analyst tests the 36-month pull, the result is reviewed against the current specification, by the current method, in the current system. Nobody at that moment is asked whether the 36-month result is comparable to the 3-month result generated three years earlier by a different method version, against a different specification version, recorded in a system that has since been retired. Each pull passes its own review. The series accumulates inconsistencies that nobody owns.

Seven Ways a Long-Horizon Program Drifts

The seven patterns below come up repeatedly in programs that have run for more than a few years. They are listed roughly in order of how often they cause trouble at a submission or inspection, not how often they occur.

1. Analytical method changes without bridging data

Over a five-year study, the assay method will change. Columns get discontinued, a mobile phase is adjusted, an impurity method is replaced with one that resolves a new peak, a compendial monograph is revised, or the method is transferred to a contract lab. Each change may be perfectly justified and properly validated in isolation. The question the stability series asks is different: are the results before and after the change comparable enough to belong in the same regression?

ICH Q2(R2) and Q14, which FDA issued as final guidance in March 2024, describe a lifecycle approach to analytical procedures and explicitly aim to support more efficient, science-based, and risk-based post-approval change management.²¹ Q2(R2) introduces concepts such as revalidation, co-validation, and platform analytical procedures.²² None of that removes the need for a bridge when a stability-indicating method changes mid-study. If the new method reads assay 0.8 percent lower than the old one on the same sample, and nobody ran the two side by side and documented it, the stability trend now contains a step change that looks like degradation.

The failure mode is not that companies skip method validation. It is that the change control for the method lives in the analytical development or QC change system, while the stability protocol and its data live in the stability function, and nobody updates the study record to say “results from time point X onward were generated by method version Y, bridged by study Z.”

2. Chamber mapping and excursion records with gaps

ICH Q1A(R2) specifies the long-term condition as 25°C ± 2°C/60% RH ± 5% RH (or 30°C ± 2°C/65% RH ± 5% RH), and the tolerances are part of the condition, not decoration.¹ EU GMP Chapter 6 requires that equipment used for the ongoing stability program, stability chambers among others, be qualified and maintained under Chapter 3 and Annex 15.⁷

In practice, chamber records drift in three ways. Mapping studies are performed at qualification and then not repeated after a compressor replacement or a shelf reconfiguration. Excursion alarms are logged in a building management system that is not linked to the stability study, so the study record has no note that a chamber ran warm for 40 hours in year two. And the impact assessment for an excursion, when one is written, is filed in the deviation system rather than attached to the affected studies, so five years later nobody can tell which time points were exposed.

Mean kinetic temperature (MKT) is the usual tool for assessing excursions. Jenkins, Cancel, and Layloff, writing in BMC Public Health in 2022, define MKT as a calculated single temperature at which the level of degradation over a period equals the total of the separate degradations that would result from a series of different temperatures over the same period, and they note that in the absence of manufacturer data a default activation energy is assumed, with consultation of the manufacturer being the preferred approach.²⁰ An MKT calculation is only as good as the temperature log it is built on. If the log has gaps, the MKT is a guess with a decimal point.

3. Pull-date slippage against the protocol schedule

Protocols specify time points. Reality delivers pulls a few days late because the analyst was out, a few weeks late because the instrument was down, and occasionally months late because the sample could not be found. 21 CFR 211.166(a)(1) requires test intervals based on statistical criteria for each attribute examined to assure valid estimates of stability.⁶ A pull tested 47 days after its nominal 12-month date is a 13.5-month data point, not a 12-month one.

Most programs allow a testing window. The drift is not the existence of late pulls; it is that the actual pull date and the actual test date are recorded separately from the nominal time point, and the trend analysis uses the nominal value. Over a long series, a pattern of consistently late pulls shifts the regression, and a shelf-life extrapolation built on nominal dates overstates what the data supports.

4. Specification version confusion

Specifications are revised over a product’s life: acceptance criteria tighten after more data, a new impurity is added, a dissolution criterion is changed at a regulator’s request. Each result in the stability series was evaluated against whichever specification version was current at the time of testing. Some LIMS configurations overwrite the specification reference when the specification is revised, so the historical result now displays against a criterion that did not exist when it was generated. Others keep the original reference but report the whole series against the current specification for trending purposes, which can turn a historical pass into an apparent fail.

Clarkston Consulting’s 2023 piece on LIMS data migrations gives a concrete example of how this becomes a migration problem: a legacy system stored stability specifications separately from release specifications, while the target system stores them together, so the migration had to transform the structure without losing which version governed which result.²⁵ Version history for static data, the same article notes, can produce a migrated version number that does not reflect the reality of the version.

5. LIMS or stability software migrations that lose or re-key data

A program that runs long enough will cross at least one system change. The migration is usually scoped around active studies and current master data. Historical time points get treated as archive. That is a defensible engineering choice right up to the day someone needs the complete series in the new system to support a shelf-life extension.

MHRA’s 2018 GxP data integrity guidance defines data migration as the process of moving stored data from one durable storage location to another, which may change the format of the data but not its content or meaning, and states plainly that the challenges of migrating data are often underestimated, particularly regarding maintaining the full meaning of the migrated records.²³ PIC/S PI 041-1, in force since 1 July 2021, expects that migration of data from one system to another is performed in a controlled manner, in accordance with documented protocols, with appropriate verification of the complete migration of the data.²⁴

Where migrations go wrong for stability specifically: results migrate but their audit trail entries do not; the link between a result and the method version or specification version is dropped because the target data model has no field for it; non-active studies are re-keyed from PDF reports rather than transferred as data, introducing transcription errors and losing raw data links; and the nominal time point migrates but the actual pull date does not.

6. Sample inventory that does not reconcile

A stability study starts with a defined number of containers placed in each chamber. Each pull removes some. Investigations, retests, and extra pulls for a post-approval change remove more. If the inventory is tracked on a chamber log sheet rather than in the study record, or if the two are tracked separately and never reconciled, the program eventually reaches a time point with no sample to test, or discovers samples in a chamber with no study to which they belong.

The regulatory expectation is direct. 21 CFR 211.166(a)(4) requires testing of the drug product in the same container-closure system as that in which the product is marketed, and 211.166(b) requires an adequate number of batches to be tested.⁶ A missing 48-month pull because the samples were consumed by an unplanned retest at 36 months is a gap in the shelf-life evidence, and it is one an inspector can find by counting.

7. Trend evaluations run on inconsistent data

The first six patterns converge here. EU GMP Chapter 6 states that the number of batches and frequency of testing should provide sufficient data to allow for trend analysis, requires at least one batch per year of product manufactured in every strength and every primary packaging type unless otherwise justified, and requires that out-of-specification results or significant atypical trends be investigated, with confirmed OOS results or significant negative trends affecting marketed batches reported to the competent authorities.⁷

Trend analysis assumes the points on the chart are comparable. ISPE’s Pharmaceutical Engineering, in a March/April 2025 article on identifying out-of-trend data in stability studies, describes the regression control chart, by-time-point, and slope control chart methods, and a modified approach that uses analysis of covariance to test whether historical batches can be pooled before applying confidence intervals.¹⁹ Every one of those methods is built on the assumption that a result at 24 months means the same thing as a result at 6 months. When the method changed at 12 months without a bridge, when the 24-month pull was actually tested at 26 months, and when the specification was tightened at 18 months, the OOT method is faithfully detecting the program’s own inconsistency and reporting it as product behavior.

What the Regulations and Guidelines Actually Require

The stability rules are older and more stable than most of the digital compliance topics leadership teams discuss. That is part of the problem: they are assumed rather than read. A short tour of what is actually in force, and what is changing, helps frame the audit approach that follows.

ICH Q1A(R2) and Q1E: the current standard

Q1A(R2) requires validated stability-indicating analytical procedures, formal studies on at least three primary batches, defined storage conditions with tolerances, and a data evaluation that covers assay, degradation products, and other appropriate attributes.¹ It also states that data from the accelerated storage condition, and where appropriate the intermediate condition, can be used to evaluate the effect of short-term excursions outside the label storage conditions, such as during shipping.¹ That sentence is the regulatory basis for most excursion impact assessments, and it presumes the accelerated data is itself sound.

Q1E, adopted the same day, sets the limits on extrapolation. Where long-term and accelerated data show little or no change over time and little or no variability, the proposed retest period or shelf life can be up to twice, but should not be more than 12 months beyond, the period covered by long-term data.² Where data show change or variability, statistical analysis of the long-term data becomes useful, and Q1E recommends testing batch poolability with analysis of covariance at a significance level of 0.25, chosen to compensate for the low power of a typical stability design.² Poolability tests are exactly where method changes, late pulls, and specification confusion show their effect: they inflate variability, pooling fails, and the shelf life is set by the worst individual batch.

The consolidated ICH Q1: status as of September 2026

ICH is replacing the Q1A through Q1F series and Q5C with a single guideline, ICH Q1 “Stability Testing of Drug Substances and Drug Products.” The draft was endorsed at Step 2 on 11 April 2025 and released for public consultation; the draft text states that ICH Q1 is a consolidated revision that supersedes the ICH Q1A-F and Q5C guidelines, and its scope covers synthetic and biological drug substances and products, with a dedicated annex for advanced therapy medicinal products.³ The draft includes a section on excursions outside a labeling claim, a section on ongoing stability studies, an annex on reduced protocol designs, and an annex on stability modeling. It also states that an extension of the approved shelf life based on acceptable stability data from a minimum of three production or primary batches may be submitted to allow a longer shelf life.³

It is not final. The ICH Q1 Expert Working Group work plan dated 11 February 2026 records the public consultation period as April to September 2025 and lists Step 3 sign-off and Step 4 adoption as an anticipated milestone for November 2026, with training material to follow in June 2027.⁴ The FDA rapporteur, presenting at CASSS WCBP 2026, stated that finalization of the Step 3 experts draft is expected by November 2026, with Step 4 adoption and Step 5 implementation to follow, and was explicit that the draft is expected to change based on consultation comments.⁵

21 CFR 211.166 and EU GMP Chapter 6

In the United States, 21 CFR 211.166 requires a written testing program designed to assess the stability characteristics of drug products, with sample size and test intervals based on statistical criteria, storage conditions for retained samples, reliable, meaningful, and specific test methods, and testing in the marketed container-closure system. Results are to be used in determining storage conditions and expiration dates, and an adequate number of batches must be tested.⁶ Note that “reliable, meaningful, and specific” methods and “statistical criteria” for intervals are the regulatory hooks for two of the seven drift patterns. A method that changed without a bridge is arguably no longer meaningful across the series. A test interval that slipped is no longer the statistically chosen one.

In the EU, Chapter 6 of the GMP Guide (sections 6.26 through 6.36) establishes the ongoing stability program: a written protocol extending to the end of shelf life, qualified chambers, at least one batch per year per strength and primary packaging type, sufficient data for trend analysis, investigation of OOS results and significant atypical trends, reporting of confirmed OOS results or significant negative trends affecting marketed batches to the competent authorities, results available to the Qualified Person, and a periodically reviewed summary of all data generated.⁷ The phrase “summary of all the data generated” is worth pausing on. It presumes that all the data can be assembled.

Post-approval reporting: field alerts, annual reports, and variations

For an approved NDA or ANDA, 21 CFR 314.81 requires a field alert report to the FDA district office within 3 working days of receipt of information concerning any significant chemical, physical, or other change or deterioration in a distributed drug product, or any failure of a distributed batch to meet its specification.⁸ A confirmed stability failure on a marketed batch is a field alert event. The same regulation requires an annual report within 60 days of the anniversary of approval, covering manufacturing and controls changes and the status of postmarketing studies.⁸ FDA’s 1994 guidance on the CMC section of the annual report states that an extension of the expiration dating period should be accompanied by full shelf-life stability data on a minimum of three production lots, obtained using the stability protocol approved in the application.¹⁰

In the EU, the EMA guideline on stability testing for applications for variations to a marketing authorization was revised in Revision 3, which came into effect on 15 January 2026 and replaced Revision 2.¹¹ It carries forward the extrapolation rule: where long-term and accelerated data show little or no change over time and little or no variability, the retest period or shelf life can be extrapolated up to twice, but not more than 12 months beyond, the period covered by long-term data, with any extrapolation after a change showing an adverse effect on stability treated case by case.¹¹ The same rule appears in Annex II of the EMA guideline on stability testing of existing active substances and related finished products (CPMP/QWP/122/02 rev 1 corr).¹²

What Inspectors Keep Finding

Stability is among the most frequently cited areas in FDA drug inspections, and the pattern of citations is consistent. An ECA Academy analysis of FDA Form 483 data reported that, across 779 drug-related 483s in fiscal year 2019 and 349 in fiscal year 2020, the most common stability observations were the lack of a written stability program (67 and 18 citations respectively), a written program that was not followed (29 and 13), stability test methods that were not valid (22 and 5), and results not used to determine expiration dates or storage conditions (10 and 4).¹⁴ Deficiencies were found across nearly every subsection of 211.166.

Three warning letters from the past year show what those categories look like in practice, and how often they arrive bundled with data integrity findings.

No ongoing studies, no controlled humidity

In a warning letter dated 14 May 2026 following a November 2025 inspection of a facility in Alsip, Illinois, FDA cited a failure to establish and follow a written testing program under 211.166(a) and (b). The only stability data provided for one product consisted of a single batch tested at one temperature with no controlled humidity condition, and the firm had no ongoing stability studies for any drugs currently manufactured. The same letter documented an uncontrolled instrument password affixed to a laptop by a sticker and instrument data that was automatically deleted when the laptop shut down, leaving no records for the quality unit to review.¹⁵

Labeled storage conditions wider than the data

In a warning letter dated 9 April 2026 following an inspection from 29 September to 16 October 2025, FDA found that a firm’s stability study concluded an expiration period was supported within a defined temperature range, but the firm labeled a much wider storage condition range at the instruction of its customer, without adequate data to support the wider labeled conditions. FDA found the firm’s response inadequate because it did not include a comprehensive review of the stability program, did not provide the revised protocol, and did not describe a retrospective review of batches lacking adequate stability data to support labeled expiration dates and storage conditions.¹⁶

Failures at multiple time points, investigations closed early

In a warning letter dated 23 September 2025 following an inspection from 27 January to 6 February 2025, FDA stated that three OTC products failed to meet predefined quality attributes at multiple time points during long-term room temperature storage. Despite inconclusive Phase I investigations and documented recognition of trending, OOS investigations were repeatedly closed without a Phase II investigation. Formulation changes had been made without documented evidence that the reformulated products would consistently meet their quality attributes throughout shelf life. Laboratory personnel shared a common password, only a system administrator user type was attributable in the chromatography software, and audit trails were not independently reviewed after analysis or at an established frequency.¹⁷

Auditing a Running Program: What to Pull, Reconcile, and Fix First

An audit of a live stability program is a reconciliation exercise. The goal is not to re-review every result but to establish, study by study, whether the series is internally consistent and whether the records needed to defend it exist and connect. The following approach has five phases and is designed to be run by a small team in a matter of weeks, not as a multi-quarter remediation.

What to pull: the minimum evidence set

Auditors and inspectors both work from documents, so the evidence set should be defined in terms of documents that exist or do not. For each study the minimum set is: the approved protocol and every amendment; the study record with all time points; the chamber qualification, mapping, and continuous monitoring records for the chambers used, covering the study’s full duration; every excursion deviation that touched those chambers, with impact assessment; the method validation and any transfer, revalidation, or bridging reports for every method version used; the specification version history with effective dates; the sample inventory log; the OOS and OOT investigations opened against the study; and, where the study crossed a system migration, the migration verification record showing that this study’s data was migrated completely.

What to reconcile: the five checks in detail

What to fix first

Three things come first, in this order. First, any study supporting a submission that is in preparation or under review, because a question from the agency will arrive on the agency’s timeline, not the company’s. Second, any commercial study where a confirmed OOS or a significant negative trend exists on the record without a completed investigation, because that is a reporting obligation under both 314.81 and EU GMP 6.35, not a housekeeping item.⁸⁷ Third, sample inventory shortfalls, because a missing future pull cannot be recovered later; the only options are to source additional retained samples now or to document the gap before it occurs.

Stability Data as a Submission Asset

It is easy to treat the stability program as an obligation that produces reports nobody reads until something fails. That framing undersells it. A clean, reconciled stability series is one of the few regulatory assets a company can build after approval that directly reduces supply risk and enables change.

Shelf-life extensions

The rules on extension reward companies whose series is clean and penalize those whose series is not. In the United States, FDA’s guidance on changes to an approved NDA or ANDA places an extension of the expiration dating period based on full shelf-life data on production batches, obtained under a protocol approved in the application, in the annual report category. An extension based on data obtained under a new or revised protocol that has not been approved, or on full shelf-life data from pilot-scale batches, requires a prior approval supplement. A reduction of the expiration dating period, and an extension of a previously reduced expiration date, go in a changes-being-effected-in-30-days supplement even when the data came from an approved protocol.⁹ The 1994 CMC annual report guidance adds that the extension should be supported by full shelf-life data on at least three production lots, using the approved protocol.¹⁰

Read those categories against the seven drift patterns. If the method changed mid-study without a bridge, the data was arguably not obtained under the approved protocol as written. If pulls slipped, the “full shelf-life data” may not reach the proposed date on actual test dates. If the study crossed a migration that dropped the audit trail, the data’s integrity for a regulatory submission is questionable. A program that has drifted converts an annual-report change into a prior-approval supplement, or removes the option altogether.

In the EU, the extrapolation rule of up to twice but not more than 12 months beyond the long-term data applies only where the data show little or no change and little or no variability.¹¹¹² Inconsistent data reads as variability. The drift patterns, in other words, directly shrink the extrapolation a company is allowed to claim.

Post-approval changes

ICH Q12 draws a distinction that matters here. The formal stability studies in Q1A(R2) exist to establish a shelf life for a new product. Stability studies to support a post-approval CMC change exist to confirm the previously approved shelf life and storage conditions, and their scope and design are informed by the knowledge and experience acquired since authorization.¹³ Q12 also describes how established conditions for analytical procedures should include the elements that assure the procedure’s performance, with reporting categories for changes to those elements justified through risk management.¹³

“Knowledge and experience acquired since authorization” is the reconciled stability series. A company that can show ten years of consistent long-term data, with every method change bridged and every excursion assessed, can justify a targeted confirmatory study for a site transfer or a packaging change. A company that cannot assemble its own series has no basis for reduced testing and will default to a full protocol every time.

Annual reports and inspection readiness

The annual report status of ongoing stability studies, the EU GMP periodic summary of all data generated, and the product quality review all draw from the same series. When the series is reconciled, those documents are generated, not authored. When it is not, each one is a separate reconstruction, and the reconstructions disagree with each other in ways an inspector comparing them can see.

What Stability Software Solves and What It Only Relocates

Every LIMS vendor and several specialist vendors offer a stability module. The pitch is that the software schedules pulls from the protocol, generates sample identifiers and labels, manages chamber assignments, records results against the study, and produces trend reports. Most of that is true, and a well-configured stability module is a large improvement over a spreadsheet and a chamber log book. It is worth being precise about which of the seven drift patterns software addresses and which it simply moves.

What it solves

Pull-date slippage is the clearest win. A system that generates the pull schedule from the protocol, records the actual pull date and actual test date as separate fields, and flags pulls outside the protocol window removes most of the slippage problem, provided the trend reports use actual dates. Sample inventory is the second: if the study record decrements inventory at each pull and at each investigation withdrawal, and the chamber log is the system rather than a separate sheet, the inventory reconciles by construction. Specification version is solvable if the system stores the specification version identifier on each result at the time of entry and does not overwrite it on revision; this is a configuration decision, and it is frequently made wrong.

What it relocates

Method changes are not solved by stability software, because the method change happens in a different system and the bridging study is a separate piece of analytical work. The software can carry a method version field on each result, which is valuable, but it cannot know that version 4 and version 5 were bridged unless someone links the bridging report. The drift pattern moves from “we did not know the method changed” to “the system shows the method changed and nobody attached the bridge.”

Chamber excursions are relocated similarly. The building management system logs the excursion. Unless there is an interface that maps chamber identifier to the studies resident in that chamber on that date and creates a study-level flag, the excursion still lives outside the study record. Interfaces like this are possible and are increasingly common, but they are an integration project, not a module feature.

Migration losses are the pattern that stability software can make worse. Adopting a new stability module is itself a migration, and the choices Clarkston describes, from whether to migrate historical results or only active studies to whether to recreate data in the target environment for ongoing studies, decide whether the new system has a complete series or a fresh start with an archive nobody can query.²⁵ Astrix’s structured approach to LIMS decommissioning stresses a data audit of the legacy system before migration, reconciliation and data integrity testing before and after transfer, and parallel runs before the legacy system is retired.²⁶ The MHRA and PIC/S expectations quoted earlier apply in full.²³²⁴

Trend evaluation is relocated rather than solved. Most stability modules produce regression plots and OOT flags. Those flags are exactly as reliable as the consistency of the underlying series. Software that trends an unbridged method change will report a degradation step. Software that trends nominal dates will overstate shelf life. The statistics are correct; the inputs are the problem, and the inputs are a process discipline, not a feature.

Conclusion

Stability programs drift because they are long, because they cross organizational boundaries that nothing else in the quality system crosses in the same way, and because the controls that exist are built around single events rather than series. The seven patterns described here are ordinary. Every mature program has some of them. What separates a program that can defend its shelf lives, support its changes, and answer an inspector from one that cannot is whether anyone has assembled the series and reconciled it against the protocol, the method history, the specification history, the chamber records, and the physical inventory. The regulations already require the pieces. ICH Q1A(R2) and Q1E, 21 CFR 211.166, EU GMP Chapter 6, and the reporting rules in 314.81 and the EMA variations guideline all assume the series can be produced on demand. The consolidated ICH Q1, when it reaches Step 4, will assume it more strongly.

Sakara Digital works with pharma and biotech organizations on exactly this kind of data quality reconciliation: establishing what a long-running program’s records actually support, fixing the links that exist, documenting the gaps that do not, and configuring systems so the drift stops recurring. If you are preparing a shelf-life extension, planning a LIMS migration that will carry stability data, or simply unsure whether your program’s series could be assembled in a day, we are happy to have that conversation.

For Further Reading