Regulators no longer accept a process that "works" without an explanation of why it works and how far it can drift before quality fails. That expectation is codified in ICH Q8, and the practical engine behind it is design of experiments. This guide shows how ICH Q8 design of experiments process optimization actually happens: how DOE links critical process parameters to quality attributes, how those relationships become a design space, and where the Normal and Proven Acceptable Ranges sit inside it. If you want to build the screening or response-surface designs described here as you read, you can generate them free in a browser-based DOE generator.
The through-line is simple: quality by design asks for multivariate understanding, and DOE is the only efficient way to get it. One-factor-at-a-time work produces univariate ranges that ICH Q8 explicitly says do not, on their own, make a design space.
What ICH Q8 asks for (QbD & design space)
ICH Q8(R2) "Pharmaceutical Development" describes what goes into the development section of a regulatory submission, and it formalised Quality by Design, the design space, and the control strategy. It is guidance, not a checklist, and its stated aim is to enable flexible, science- and risk-based regulatory approaches built on genuine process understanding.
ICH Q8 defines quality by design as "a systematic approach to development that begins with predefined objectives and emphasizes product and process understanding and process control, based on sound science and quality risk management." Its building blocks form a chain:
- QTPP (Quality Target Product Profile) is a prospective summary of the quality characteristics the product should have. It is the design intent and the starting point.
- CQA (Critical Quality Attribute) is a property that must stay within a limit, range, or distribution to ensure quality, derived from the QTPP. For a biologic, think aggregation, charge variants, glycosylation, and potency.
- Critical process parameters (CPPs) are the process inputs whose variability affects a CQA, so they must be controlled.
- Design space and control strategy are what tie the CPPs to the CQAs and keep them there.
The design space definition is worth quoting verbatim, because the wording carries regulatory weight:
"The multidimensional combination and interaction of input variables (e.g., material attributes) and process parameters that have been demonstrated to provide assurance of quality."
Two consequences follow. Working within an approved design space is not considered a change, which is the regulatory flexibility QbD offers. Moving out of it normally triggers a post-approval change process. And crucially, the phrase "and interaction" means a design space must capture how parameters modify each other, which is exactly what design of experiments is built to measure.
ICH Q8 design of experiments process optimization
ICH Q8 design of experiments process optimization is the structured use of DOE to turn a list of candidate parameters into a validated, multivariate operating region. ICH Q8 states that a design space can be developed from prior knowledge, first principles, and/or experimentation, and DOE is the experimentation half done efficiently.
The reason OFAT cannot deliver a Q8 design space is structural: it holds every parameter but one constant, so it never observes an interaction. If temperature and pH jointly affect aggregation, an OFAT study optimising each separately will miss the combined effect and can land on a false optimum. A designed experiment varies them together and estimates the interaction term directly. For the full argument, see our comparison of DOE vs OFAT.
A Q8-aligned DOE campaign is usually two-stage. First a screening design (fractional factorial or Plackett-Burman) sifts many candidate parameters down to the critical few. Then a response-surface design (central composite or Box-Behnken, covered in our response surface methodology guide) maps curvature and interactions across the surviving parameters so the response can be predicted, not just ranked. The output is a model per CQA, and the intersection of the acceptable regions is the design space.
Build the screening and RSM designs for your Q8 study
The free DOE generator produces fractional factorial, Plackett-Burman, central composite, and Box-Behnken run tables with bioprocess presets, randomized and CSV-ready for a QbD campaign.
Mapping CPPs to CQAs with DOE
Mapping critical process parameters to critical quality attributes is a risk-then-experiment sequence: rank the risks, then use DOE to quantify the ones that matter. This is where ICH Q8 and ICH Q9 (Quality Risk Management) work together.
The sequence, from QTPP to a fitted model, runs like this:
- Translate QTPP into CQAs. Turn the target product profile into measurable quality attributes with acceptance limits (e.g. aggregate < 2%, main charge variant 55–70%, titer ≥ 3 g/L).
- Risk-rank candidate parameters (ICH Q9). Use an Ishikawa (fishbone) diagram and FMEA to list every parameter that could plausibly move a CQA, and score each by severity, occurrence, and detectability. High-RPN parameters become DOE factors; low-risk ones are fixed.
- Screen. Run a fractional factorial or Plackett-Burman design across the shortlisted parameters to confirm which truly are critical process parameters.
- Optimise. Run a response-surface design over the critical few to quantify interactions and curvature.
- Model. Fit a regression or response-surface model per CQA, mapping parameter settings to each attribute.
The risk assessment is what keeps the DOE affordable. Rather than experiment on twenty parameters, Q9 tools justify fixing the fifteen that prior knowledge and mechanism say are non-critical, leaving a tractable five for the design. Our guide on CPP-CQA mapping and the design space walks through the risk-ranking step in detail.
Figure 1. A DOE effects Pareto on a mAb aggregation CQA. Temperature, the temperature×pH interaction, pH, and DO clear the significance threshold and are confirmed critical process parameters; feed rate does not and is fixed. The interaction term is invisible to OFAT.
Building the design space from a DOE
The design space is the multivariate region where every critical quality attribute simultaneously stays within its acceptance limit, read off the overlaid response-surface models. You do not design a design space directly; it emerges from the intersection of the CQA models.
Once each CQA has a fitted model, you overlay their acceptable regions. For a two-parameter slice (say temperature and pH at fixed DO), each CQA contributes a contour: aggregate below its limit carves out one region, titer above its limit carves out another. The area where all the acceptable regions overlap is the design space for that slice. Repeat across the parameters and you have the multidimensional space ICH Q8 defines. The boundary beyond which any CQA fails is the edge of failure.
Because the design space is defined by interactions, its true shape is usually not a rectangle. It is the irregular green region above, and a naive box drawn from univariate ranges can poke outside it, which is precisely the trap ICH Q8 warns against.
NOR vs PAR
A Proven Acceptable Range is univariate; a Normal Operating Range is the narrower band you actually run in, and neither is a design space by itself. Getting this hierarchy right is the single most commonly botched point in Q8 discussions.
ICH Q8 defines a Proven Acceptable Range (PAR) as "a characterised range of a process parameter for which operation within this range, while keeping other parameters constant, will result in producing a material meeting relevant quality criteria." The key clause is "while keeping other parameters constant" — a PAR is established one parameter at a time, so it captures no interactions. Q8 states plainly that a combination of PARs does not necessarily constitute a design space.
The Normal Operating Range (NOR) is not formally defined in Q8 but is established industry usage: the narrower range within which the process runs routinely. The hierarchy is:
| Term | What it is | Multivariate? | Regulatory status |
|---|---|---|---|
| NOR (Normal Operating Range) | Routine day-to-day operating band | No (setpoint ± margin) | Inside the design space |
| PAR (Proven Acceptable Range) | Univariate acceptable range, others fixed | No | Not a design space alone |
| Design space | Multivariate region all CQAs pass | Yes (interactions) | Approved; working inside = no change |
| Edge of failure | Boundary where a CQA fails | Yes | Do not operate at or beyond |
So NOR ⊆ PAR ⊆ design space ⊆ edge of failure. The NOR gives you operating margin: because it sits well inside the design space, normal run-to-run variability in a critical process parameter never pushes a CQA to its edge of failure. Setting the NOR is a risk decision about how much margin you want between routine operation and the boundary.
Worked example: a CHO mAb design space
Here is the whole chain on one process: a CHO monoclonal antibody bioreactor step, three candidate CPPs, and a design space for two CQAs. The numbers are illustrative but the structure mirrors a real QbD submission.
From QTPP to design space in six steps
1. QTPP → CQAs. The mAb must be safe and efficacious, giving two bioreactor-stage CQAs: aggregate < 2.0% and titer ≥ 3.0 g/L.
2. Risk assessment (ICH Q9). An FMEA over the bioreactor lists temperature, pH, DO, feed rate, and seeding density. Prior knowledge fixes seeding density and feed rate as non-critical, leaving three factors for the DOE: temperature (34–37 °C), pH (6.8–7.2), DO (30–60%).
3. Screen. A 2³ full factorial with centre points confirms temperature, pH, and their interaction dominate aggregate; DO mainly moves titer. Feed rate, checked in the screen, is confirmed non-critical.
4. Optimise. A central composite design (roughly 2³ + 6 axial + centre ≈ 20 runs) maps curvature. Fitted models: aggregate = 1.4 + 0.5·T + 0.3·pH + 0.4·T·pH and titer = 3.4 + 0.2·DO − 0.15·T (coded factors).
5. Design space. Overlay the two acceptable regions. Aggregate < 2.0% excludes the high-temperature, high-pH corner (where the +0.4 interaction bites); titer ≥ 3.0 g/L excludes the low-DO, high-temperature corner. The overlap is the design space, an irregular region, not a box.
6. NOR & control. Pick a setpoint safely inside (T = 35.5 °C, pH = 7.0, DO = 45%) and set a NOR of ±0.5 °C, ±0.1 pH, ±10% DO around it. That NOR sits comfortably inside the design space, so routine variability never reaches the edge of failure.
The payoff. Because the interaction was quantified, the submission can claim a multivariate design space, not a set of PARs, and gain the regulatory flexibility ICH Q8 attaches to it: moves within the space are not changes.
That is the entire value of ICH Q8 design of experiments process optimization in one process: DOE turned three parameters and two CQAs into a defensible, interaction-aware operating region, in about twenty runs. To scope the run count for your own factors before you start, our guide on how many experiments a DOE needs sizes each design, and you can generate the screening and response-surface tables directly in the design of experiments generator.
Frequently Asked Questions
What does ICH Q8 say about design of experiments?
ICH Q8(R2) does not mandate design of experiments, but it identifies DOE as a key tool for gaining the enhanced product and process understanding that Quality by Design requires. Q8 states that a design space can be developed from a combination of prior knowledge, first principles, and experimentation. In practice, DOE is how developers quantify the main effects and interactions of process parameters on quality attributes, and that multivariate understanding is what lets them define a design space rather than a set of one-factor-at-a-time ranges.
How is a design space defined under ICH Q8?
ICH Q8(R2) defines the design space as the multidimensional combination and interaction of input variables and process parameters that have been demonstrated to provide assurance of quality. Working within the approved design space is not considered a change; moving out of it normally initiates a post-approval regulatory change process. Because the definition explicitly includes interactions, a design space cannot be built from univariate proven acceptable ranges alone. DOE, which estimates interactions directly, is the standard route to it.
What is the difference between NOR and PAR in QbD?
A Proven Acceptable Range (PAR) is a characterised range of a single process parameter that yields acceptable quality when the other parameters are held constant, so it is a univariate concept. A Normal Operating Range (NOR) is the narrower band within which the process is routinely run day to day. The NOR sits inside the PAR, which sits inside the edge of failure. ICH Q8 warns that a combination of PARs does not necessarily constitute a design space because PARs do not capture interactions between parameters.
How does DOE link critical process parameters to critical quality attributes?
DOE varies several critical process parameters together in a structured pattern and measures the resulting critical quality attributes, then fits a statistical model that maps parameter settings to each attribute. Screening designs first identify which parameters are critical; response-surface designs then quantify the curvature and interactions needed to predict the response. The multivariate region where every critical quality attribute stays within its acceptance limits is the design space.
Which ICH guidelines support Q8 and QbD?
ICH Q8(R2) is one of a family. ICH Q9 (Quality Risk Management) supplies the risk-assessment tools, such as FMEA, used to decide which parameters warrant a DOE. ICH Q10 (Pharmaceutical Quality System) houses the control strategy and change management that make a design space usable. ICH Q11 extends QbD to the drug substance, and ICH Q14 extends it to analytical procedure development. Q8, Q9, and Q10 are usually cited together as the QbD triad.
Related Tools
- DOE Experiment Generator — Build the screening and response-surface designs a Q8 design-space study needs, free in the browser with bioprocess presets.
- CHO Process Troubleshooter — Diagnose the CQA excursions (aggregation, lactate, titer) your design space is meant to keep in check.
- Clone Scorecard — Rank cell lines on the quality attributes that become your CQAs.
References
- ICH (2009). ICH Q8(R2) Pharmaceutical Development. International Council for Harmonisation. Official guideline PDF
- Rathore, A.S. & Winkle, H. (2009). Quality by design for biopharmaceuticals. Nature Biotechnology, 27(1), 26–34. DOI: 10.1038/nbt0109-26
- Horvath, B., Mun, M. & Laird, M.W. (2010). Characterization of a Monoclonal Antibody Cell Culture Production Process Using a Quality by Design Approach. Molecular Biotechnology, 45(3), 203–206. DOI: 10.1007/s12033-010-9267-4
- Suksaeree, J., Monton, C. & Maneewattanapinyo, P. (2026). Quality by Design Integration of Design of Experiments for Tablet Formulation Optimization and Process Validation. AAPS PharmSciTech, 27. DOI: 10.1208/s12249-026-03335-4