Accelerated stability models compared — Arrhenius, ASAP, AccelStab and probabilistic decisioning

1. Executive summary

Four families of model dominate accelerated and predictive stability work in pharma: the classic Arrhenius equation, the humidity-corrected Arrhenius used by the Accelerated Stability Assessment Program (ASAP), the Šesták–Berggren kinetic approach implemented in the open-source AccelStab R package, and the probabilistic Arrhenius + Monte-Carlo approach Synlogica Terminus Quality uses for excursion decisioning.

They are often discussed as competitors. They are not. Three of them (Arrhenius, ASAP, AccelStab) answer a development question — "what shelf-life can I claim from a short, stressed study?" — while Synlogica answers an operational question — "given this product's already-characterised kinetics and this actual logger trace, is this batch releasable, and with what confidence?" This paper states each model's equations and assumptions plainly, compares them, and shows where each belongs.

2. The shared foundation — Arrhenius

All four approaches inherit the Arrhenius temperature dependence of the degradation rate constant k:

What they add on top of this base equation — a humidity term, a non-linear degradation form, an uncertainty model — is exactly what distinguishes them.

3. Classic Arrhenius & MKT

The traditional approach fits k at several elevated temperatures, extrapolates to the storage condition, and reports a single shelf-life. For non-isothermal exposure (e.g. a distribution excursion) it is often collapsed to Mean Kinetic Temperature (MKT):

Assumptions: one dominant, rate-limiting degradation reaction; E_a constant across the whole temperature range; simple (zero- or first-order) kinetics; humidity either irrelevant or held constant; the sensor temperature equals the product temperature.

Limits: produces a point estimate that hides uncertainty in E_a, in sensor calibration and in spatial inhomogeneity; MKT discards the shape of the excursion; no humidity term; breaks down when the degradation mechanism changes with temperature.

4. ASAP — humidity-corrected Arrhenius

The Accelerated Stability Assessment Program (Waterman et al.) extends Arrhenius with an explicit, log-linear humidity term and uses an isoconversional design: hold products at several temperature/relative-humidity combinations for days to weeks, measure time-to-a-fixed-degradation-level, and fit:

Shelf-life and its confidence interval are then propagated by Monte-Carlo simulation over the fitted parameters. ASAP underpins commercial Accelerated Predictive Stability (APS) tools and is widely used in QbD submissions.

Assumptions: a single isoconversional degradation endpoint; log-linear humidity dependence; moisture equilibration with the (open or defined) package; Arrhenius behaviour across the stressed range; a known specification limit defining "failure".

Limits: needs a humidity-controlled study design; the log-linear RH term is an empirical approximation; isoconversional framing assumes the degradation pathway does not switch across conditions.

5. AccelStab — Šesták–Berggren kinetics

AccelStab is an open-source R package (CRAN; AccelStab on GitHub) that pairs the Arrhenius temperature term with the Šesták–Berggren reaction-model function, giving a flexible, closed-form non-linear fit to the whole degradation curve rather than a single isoconversional point:

It extrapolates degradation in both time and temperature, and quantifies uncertainty with a parametric bootstrap (kinetic parameters drawn from a multivariate t-distribution) and/or the analytical delta method — yielding confidence and prediction intervals on the predicted curve.

Assumptions: the Šesták–Berggren form captures the true degradation shape; Arrhenius temperature dependence; parameter estimates are approximately multivariate-t; a humidity-modified Arrhenius term can be added where humidity matters.

Limits: more parameters (m, n) need more/cleaner data to identify; still a development-study tool, not a per-shipment decision engine; open-source — the validated state is the user's responsibility.

6. Synlogica Terminus Quality — probabilistic excursion decisioning

Terminus Quality does not try to establish a shelf-life from a stressed study. It takes the product's already-characterised Arrhenius kinetics (E_a, A, and their uncertainty, from the ICH Q1A development study) and applies them to the actual, non-isothermal logger trace of a real distribution event to answer a release question:

Where the degradation is non-linear, Terminus fits the same Šesták–Berggren reaction model used by AccelStab (dα/dt = k·f(α)) by non-linear least squares, and reports confidence and prediction intervals by parametric bootstrap (kinetic parameters drawn from a multivariate-t distribution) and the delta method — in addition to Monte-Carlo propagation of E_a, A, sensor-calibration and spatial uncertainty. The engine therefore spans both the simple integrated-Arrhenius case and the full Šesták–Berggren kinetic case, choosing per product.

The output is a probability-weighted recommendation — RELEASE / CONDITIONAL / HOLD with a stated confidence interval — packaged with every input, model version and assumption for audit. The human (QA / QP) makes the binding call.

Assumptions: the development-stage Arrhenius parameters are valid and version-pinned; the logger trace, after calibration and gap screening, represents the product within stated spatial bounds; cumulative exposure (the integral) is the relevant quantity. Humidity is included where the product's stability data support a humidity term.

What it is not: it is not a replacement for ASAP/AccelStab in development, and it is advisory by design — it does not sign the batch disposition.

7. Side-by-side comparison

8. Assumptions & failure modes they share

Every model here will mislead if its assumptions are violated. The common failure modes:

• Mechanism switching — if the rate-limiting degradation pathway changes between stressed and storage conditions, Arrhenius extrapolation (and everything built on it) is invalid.
• E_a uncertainty ignored — a ±15% error in activation energy can move a shelf-life estimate by months; point estimates hide this.
• Sensor ≠ product — logger calibration error and spatial inhomogeneity inside a shipment mean the measured profile is not exactly the product's profile.
• Humidity neglected for moisture-sensitive products — classic Arrhenius and a humidity-blind integration both overstate confidence.

9. Which model, when

• Setting a registered shelf-life fast, humidity matters: ASAP / APS.
• Modelling the full degradation curve, open and reproducible: AccelStab (Šesták–Berggren).
• Quick internal sanity check: classic Arrhenius / MKT — knowing its blind spots.
• Deciding release on a specific shipment after an excursion: Synlogica Terminus Quality — it consumes the development model and the real trace and returns an auditable, probability-weighted recommendation.

10. References

1. Waterman K.C. et al. — The Application of the Accelerated Stability Assessment Program (ASAP) to Quality by Design (QbD) for Drug Product Stability. AAPS PharmSciTech (PMC3167263).
2. Humidity-corrected Arrhenius equation: the reference condition approach. Int. J. Pharmaceutics, 2016 (S0378517316300473).
3. AccelStab: Accelerated Stability Kinetic Modelling — CRAN package & vignette; source at github.com/AccelStab/AccelStab (Šesták–Berggren + Arrhenius; parametric bootstrap / delta-method intervals).
4. ICH Q1A(R2) Stability Testing; ICH Q1E Evaluation of Stability Data.

11. Takeaways

• All four methods share an Arrhenius core; they differ in humidity handling, degradation form, uncertainty method, and purpose.
• ASAP and AccelStab are development tools for shelf-life; Synlogica is an operational tool for per-batch release decisioning — they are complementary.
• Whatever the model, the defensible output is a distribution with stated assumptions, not a single number — and, for release, an auditable decision package with the human as the decision authority.