Stability Studies Q&A: Predictive Stability, AKM & Photostability
Your stability questions, answered: Expert Q&A from the 2026 Stability Studies Conference
Pharmaceutical stability studies raise complex questions, particularly as approaches such as predictive stability and accelerated kinetic modelling become more widely used alongside traditional long-term stability data.
On 21 October 2026, the stability community came together in Kildare, Ireland for the 5th Annual Stability Studies Conference, hosted by Q1 Scientific. It is Ireland’s only conference dedicated exclusively to stability studies, and a leading European forum for stability, quality, regulatory, analytical and development professionals from across the pharmaceutical and life sciences industry.
The day offered practical learning on topics from the ICH Q1/Q5C revision and OOS results to biologics, kinetic modelling and excipient-driven stability risks, alongside two “Stability Exchange” discussion sessions and plenty of opportunity to network and share experiences with peers facing similar challenges. In 2026 we also launched the first Stability Studies Conference USA in Durham, North Carolina, bringing the series to a new audience.
Thank you to everyone who joined us and took part in the live Q&A. There were more great questions than we could answer on the day, so our speakers have kindly followed up with written responses.
In this expert Q&A, Dr. Warren Roche (argenx), Mahmoud Tawfieq (Novo Nordisk), Janina Steeger (UCB Biosciences) and Aideen Glynn (Q1 Scientific) share their answers on topics including:
- Accelerated kinetic modelling and shelf-life prediction
- Photostability study design
- Accelerated predictive stability
- Excipient compatibility
- Predictive versus conventional stability data
- Making the most of existing stability evidence when new questions arise
What you’ll learn
This Q&A explores how pharmaceutical companies can approach some of the practical challenges that arise during stability programmes, from determining the limitations of predictive models to assessing whether existing stability evidence can support a packaging change or temperature excursion.
It also considers how accelerated and predictive approaches can complement conventional stability studies rather than simply replace them.
Accelerated kinetic modelling (AKM) and shelf-life prediction
Dr. Warren Roche, Senior Statistical Expert | Research Scientist at argenx
Warren specialises in pharmaceutical stability assessment and predictive modelling for shelf-life estimation, with 10 years of experience in drug product development. His work focuses on applying kinetic modelling and model-informed approaches to accelerate formulation development and strengthen regulatory submissions, in line with the updated ICH Q1 draft guidance. He holds a PhD from South East Technological University and has co-authored more than 20 peer-reviewed publications relating to statistics.
What are the limitations of AKM, especially as an empirical approach?
It’s really semi-empirical – you’re fitting a kinetic form rather than deriving mechanism from first principles – so it assumes the degradation pathway and its temperature dependence hold across the range you extrapolate into. If a new pathway or a physical/phase change intervenes, that breaks down. It’s also only as good as the data: short or sparse datasets invite over-fitted models that extrapolate poorly. The honest defence is to quantify uncertainty and resist over-fitting – which is the discipline the stability software Blue Sigma is built around, so you get a defensible prediction rather than a single optimistic number.
Can the software manage multiple batches, and how are they handled?
Yes – single-batch illustrations are just a teaching convenience. Blue Sigma allows you to compare batches and base the shelf life on the worst-case batch, so your answer reflects the batch that reaches the limit first rather than an optimistic average. The question of whether batches can legitimately be pooled follows the ICH Q1E approach, which is the right frame when presenting multi-batch results to an agency.
Reasonable extent of extrapolation from 6 months accelerated – and what might an agency accept?
There’s no real universal multiplier at the moment. Kinetic modelling from short, accelerated data is well established as supporting evidence for extrapolation of shelf-life, but agencies will still expect real-time data to be generated in parallel to confirm a registered shelf life once available.
How would you set pre-determined criteria for model quality and accuracy?
I would fix them in the protocol before you see results, and separate how well the model fits from how well it predicts. Fit criteria centre on well-behaved residuals, defensible model selection, and parameters you can actually trust; predictive criteria come down to error against held-out data within a pre-set tolerance. The value of a tool like Blue Sigma here is that it gives you those quality measures in a form you can write straight into a validation protocol as pass/fail criteria.
Can AKM be used for products stored frozen long-term?
It depends on the data used to create the models. Arrhenius assumes a consistent-phase mechanism, and frozen storage breaks that – freeze-concentration and reduced mobility often mean a different mechanism than the one you accelerated. It has certainly been demonstrated – you just need to ensure you can verify the model quality using pre-defined criteria – like, for example, data not used to generate the models fall within the prediction bands. A huge application for this type of modelling for this material is in understanding handling and shipping temperature excursions.
How would you perform internal vs external validation (external using model-independent data)?
Internal validation tests the model against its own fit via for example residual plots or lack of fit tests. External validation tests it against data it never saw, and we can do this by holding back the long-term points, predict them from the model alone, and check the data land within tolerance. It gives you a clear, visible separation between the data that builds the model and the data that confirms it. This is one of the things Blue Sigma is designed to make easy.
Photostability study design
Mahmoud Tawfieq, Optical Science Specialist at Novo Nordisk A/S
Mahmoud is a physicist with a PhD in Optics who joined Novo in 2023. His interests include product photodegradation and germicidal UV light.
What are the considerations for conducting photostability studies for a drug product with refrigerated storage conditions?
It is important that the drug product is protected from light within the refrigerator, to avoid light degradation during storage.
Do you anticipate regulatory expectations around photostability assessment evolving as our understanding of light exposure improves?
Yes, I expect regulatory expectations to evolve as our understanding of real-world light exposure improves. The Q1 Scientific conference was one step in this direction, bringing the topic into discussion with peers and regulators. We are also contributing to this work through BioPhorum and upcoming pharmaceutical conferences and events.
How can companies improve photostability study design without making testing unnecessarily complex or expensive?
I would start with a simple question: Where is the product stored? From there, use representative lighting conditions, considering the spectrum, intensity and exposure time. A realistic worst-case scenario can help avoid unnecessary testing while still providing robust evidence that the product is adequately protected. Also, consider that fluorescent lamps are potentially the most harmful, then cold white light, then warm white light, and amber light would be the least harmful. I use this as a rule of thumb when evaluating the lighting conditions.
Predictive stability and excipient compatibility
Janina Steeger, Analytical Scientist in CMC Analytical Development at UCB Biosciences GmbH
Janina joined UCB in Germany four years ago and leads accelerated predictive stability (APS) activities while supporting various analytical development projects. She is pursuing a Ph.D. in Pharmaceutical Sciences at Ghent University, Belgium, focusing on the influence of excipients on the stability of moisture-sensitive solid dosage forms using APS, and holds a Master of Science in Applied Chemistry specialising in Instrumental Analysis and Laboratory Management.
How do you balance cost vs benefit? When what we find most stable brings less revenue to the company?
Ultimately, formulation and packaging decisions should be risk-based and patient-focused. The most stable option is not always the commercially preferred option, but stability data help quantify the potential risks associated with different development strategies. Predictive approaches can support these decisions by highlighting where relatively small formulation or packaging changes may achieve meaningful stability improvements without significantly increasing development or manufacturing costs.
How much weight would you give the results of binary excipient compatibility studies compared to results from prototype formulations?
I would give binary compatibility studies considerable weight during early screening, but much less weight when making the final formulation decision. They are intentionally simplified systems and are very useful for identifying potential risks and understanding API excipient interactions. However, they do not reproduce the composition of the final formulation or interactions among multiple excipients. Therefore, I would consider the prototype formulation more representative and give it greater weight for formulation selection. An excipient that appears problematic in a binary mixture is not necessarily unsuitable in the final product, as its concentration, the presence of other excipients, moisture distribution, and the overall microenvironment can modify the degradation behaviour.
If a predictive model and conventional stability data disagree, how would you recommend investigating and resolving that discrepancy?
First, I would determine whether the difference is within the expected model uncertainty or represents a true systematic deviation. If it is systematic, I would go back to the assumptions of the predictive model and to the experimental conditions used to generate it. For an APS model, I would particularly look at whether the same degradation mechanism is operating under accelerated and long-term conditions, whether temperature and humidity dependencies are adequately described, and whether factors such as moisture equilibration, physical changes in the formulation, or packaging have been represented appropriately. At the same time, I would also critically assess the conventional stability dataset, including analytical variability, batch-to-batch differences and storage conditions. Additional targeted conditions can then help determine the source of the discrepancy. Ultimately, I would use the conventional long-term data as the reference for the actual shelf-life behaviour and refine or restrict the predictive model if necessary. (But maybe an interesting note – I have no known example within our company where predictive and real stability data did not match).
How well can predictive approaches distinguish between an excipient that is inherently problematic and one that becomes problematic only when combined with a particular API or formulation?
A predictive model alone cannot automatically make that distinction as it describes the behaviour of the specific system used to generate the model. But a predictive stability study combined with an appropriate experimental design may help distinguish these effects well. For example, if several formulations are investigated systematically by changing the type or amount of one excipient, we can assess whether degradation consistently follows that excipient or whether the effect depends on the surrounding formulation. Binary compatibility studies can provide an additional mechanistic level. So I would avoid describing an excipient as inherently ‘good’ or ‘bad’ based on one APS experiment. Stability is fundamentally system-dependent: the API, excipient grade and amount, other formulation components, environmental conditions, and potentially processing all contribute.
Using existing stability data for packaging changes and temperature excursions
Aideen Glynn, Quality Manager at Q1 Scientific
Aideen has over 18 years of pharmaceutical industry experience and oversees quality assurance and regulatory compliance for stability storage operations. Previously she held senior Quality Control and Quality Assurance roles at EirGen Pharma, Nypro Healthcare and Newport Synthesis. She helps clients prepare for and respond to stability-related disruptions, such as chamber failures, natural disasters or supply chain interruptions, through secure, validated backup storage.
The same product can need several different stability studies. When a new question arises, such as a packaging change or a temperature excursion, how do teams establish whether the evidence they already hold is suitable?
The first step is to clearly define what has changed, what question we are trying to answer and what evidence is required to support the decision being made. For example, with a packaging change we would consider whether the new packaging provides equivalent protection against moisture, oxygen and light and whether there is any impact on product stability. Stability data generated in the original package may not fully address the risks associated with the new packaging. For a temperature excursion, we would look at the temperature reached, the duration of exposure and whether existing accelerated, intermediate or excursion studies provide sufficient evidence to support product quality. We would also consider known degradation pathways and any observed trends in critical quality attributes. Any decision should be supported by a documented risk assessment and scientific justification. If existing data and available evidence adequately addresses the identified risks, then additional testing may not be necessary, but where significant knowledge gaps remain, a targeted stability study may be more appropriate than repeating an entire programme. The goal should not be to generate new stability data every time a new question arises. The goal should be to determine if the available evidence is sufficient to support the decision being made.
Continue the stability conversation
Stability questions rarely have one-size-fits-all answers. Product characteristics, formulation, packaging, storage conditions, available data and regulatory expectations all influence the most appropriate approach.
What these answers highlight is the value of combining robust stability data with scientific judgement, appropriate risk assessment and a clear understanding of the question the study needs to answer.
We would like to thank Dr. Warren Roche, Mahmoud Tawfieq, Janina Steeger and Aideen Glynn for continuing the discussion and sharing their expertise following the conference.
Join us in 2027
The conversations that start in the Q&A are exactly why we bring the stability community together.
In 2027 we will be hosting events in both the United States and Ireland, giving even more people the chance to hear from leading experts, share ideas and network with peers.
Dates and venues will be announced soon on LinkedIn and our conference website: conference.q1scientific.com