A validated assay is only as trustworthy as its behaviour when the analyst, the column lot, or the lab temperature drifts a little. That is what ICH Q2 calls robustness, and proving it one parameter at a time is slow and blind to interactions. This guide covers ICH Q2 design of experiments assay optimization: how a designed experiment optimises an analytical method and demonstrates robustness before you enter formal analytical method validation. If you want to build the screening designs described here as you read, you can generate them free in a browser-based DOE generator.
The logic mirrors the one used for process development: an assay has factors (flow rate, pH, temperature), responses (resolution, recovery, %RSD), and interactions, so design of experiments is the efficient way to understand it. ICH Q2 lists robustness; DOE is how you deliver it.
What ICH Q2 covers (validation characteristics)
ICH Q2(R2) "Validation of Analytical Procedures" defines the performance characteristics an analytical method must demonstrate to be considered validated for its intended use. It is the harmonised reference that pharmacopoeial and regulatory expectations for analytical method validation are built on.
The characteristics depend on the type of procedure (identification, impurity test, or assay), but for a quantitative assay Q2 expects the full set:
| Characteristic | What it demonstrates | DOE useful? |
|---|---|---|
| Specificity | Response is for the analyte, free of interference | Indirect |
| Accuracy | Closeness of result to the true value (% recovery) | As a response |
| Precision | Repeatability, intermediate precision, reproducibility (%RSD) | As a response |
| Range | Interval over which accuracy and precision hold | Indirect |
| LOD / LOQ | Lowest detectable / quantifiable amount | Indirect |
| Linearity | Proportional response across the range | Indirect |
| Robustness | Result unchanged by small deliberate parameter shifts | Directly — multivariate by nature |
Robustness is the characteristic where design of experiments is not just helpful but the natural tool. ICH Q2 describes robustness as a measure of a procedure's capacity to remain unaffected by small but deliberate variations in method parameters, and notes it provides an indication of reliability during normal usage. Because "small variations in method parameters" almost always occur together in real labs, evaluating them jointly, with interactions, is precisely what a DOE does.
ICH Q2(R2), finalised in 2023, is paired with ICH Q14 on Analytical Procedure Development, which explicitly encourages a multivariate, risk-based approach. Q14 is where the DOE lives during development; Q2 is where the resulting performance is confirmed. Our full walk-through of the validation side is in the guide to analytical method validation for bioprocess (ICH Q2).
ICH Q2 design of experiments assay optimization
ICH Q2 design of experiments assay optimization is the structured use of DOE to develop and prove out an analytical method, so that robustness is demonstrated by data rather than assumed. It splits into two DOE roles: optimisation (finding the best method conditions) and robustness (confirming those conditions tolerate normal variation).
During development, an optimisation DOE explores a wide range of conditions to maximise a response such as peak resolution or signal-to-noise. Once the method conditions are chosen, a robustness DOE explores a narrow band around those nominal setpoints to confirm the method still passes when parameters drift. The two use different designs for different questions:
- Optimisation wants curvature and an optimum, so it uses a response-surface design (central composite or Box-Behnken) over a few key factors, or a definitive screening design when several factors are still in play.
- Robustness wants a yes/no on whether small shifts matter, so it uses a two-level screening design (fractional factorial or Plackett-Burman) that estimates main effects in the fewest runs.
The efficiency gain is the whole point. Checking 6 robustness factors one at a time, at two levels each with a centre reference, is slow and, worse, never reveals whether, say, pH and temperature interact. A designed screen does it in 8–12 runs and estimates every main effect independently. For the underlying argument on why univariate testing misses interactions, see DOE vs OFAT.
Build your robustness and optimisation designs
The free DOE generator produces Plackett-Burman, fractional factorial, central composite, and Box-Behnken run tables, randomised and CSV-ready, for an ICH Q2 assay study.
DOE for method robustness
Method robustness testing with DOE varies every method parameter by a small deliberate amount around its setpoint, all at once, and measures which parameters actually move the result beyond its acceptance limit. This is the single most common use of DOE in analytical work and the one ICH Q2 practically invites.
The classic reference on how to do it, Vander Heyden and colleagues' guidance, recommends a two-level screening design (Plackett-Burman or fractional factorial) with factor levels set to the extremes of expected variation, then interpreting the effects against the method's own acceptance criteria. The workflow:
- List the factors that could plausibly drift in routine use, and set a small ± range for each around the nominal setpoint (e.g. flow rate 1.0 ± 0.1 mL/min).
- Choose responses that are the method's own quality indicators: resolution, tailing factor, retention time, % recovery, %RSD.
- Pick a screening design. For up to 7 factors a 12-run Plackett-Burman is the standard robustness design; a resolution IV fractional factorial is an alternative that also cleans up two-factor interactions.
- Run in randomised order to protect against drift and carry-over.
- Plot the effects. A Pareto or half-normal plot ranks the factors; anything whose effect pushes a response past its limit is a non-robust parameter that must be controlled.
Figure 1. A robustness DOE effects Pareto on an RP-HPLC assay. Mobile-phase pH, % organic, and column temperature clear the significance threshold and must be controlled tightly; buffer concentration and wavelength do not, so their tolerances can be relaxed.
The interpretation is directly actionable: a large, significant effect means the method is not robust to that parameter, so it must be specified tightly in the procedure. A negligible effect means the parameter can drift freely, which simplifies the method and its transfer to other labs. Robustness testing is therefore also a control-strategy exercise.
Factors in an assay DOE
The factors in an assay robustness DOE are the controllable method parameters most likely to vary between runs, analysts, instruments, or reagent lots. The right list depends on the technique, but the principle is the same: choose parameters that both matter mechanistically and realistically drift.
For the two most common bioprocess assay families:
| Assay type | Typical factors (± around setpoint) | Responses monitored |
|---|---|---|
| RP-HPLC / UHPLC | Flow rate, column temperature, mobile-phase pH, % organic, buffer conc., detection wavelength | Resolution, tailing, retention time, % recovery, %RSD |
| ELISA (ligand-binding) | Incubation time & temperature, coating conc., blocking conc., wash count, substrate time | Signal window, background, curve fit, %RSD |
| CE / icIEF | Voltage, capillary temperature, buffer pH & conc., injection time | Migration time, resolution, peak area % |
An SVG helps show the structure: controllable factors on the left feed into the assay, which produces the responses on the right, with the DOE estimating each factor-to-response relationship.
Worked example: an RP-HPLC robustness screen
Here is a full robustness screen on a reversed-phase HPLC assay: six factors, a 12-run Plackett-Burman design, and the decision it drives. The numbers are illustrative but the structure follows the Vander Heyden robustness workflow.
Six factors, twelve runs, one clear answer
1. Factors & ranges. Flow rate (1.0 ± 0.1 mL/min), column temperature (30 ± 5 °C), mobile-phase pH (3.0 ± 0.2), % organic (45 ± 2%), buffer concentration (20 ± 5 mM), detection wavelength (254 ± 2 nm). Each set to its ± extremes as the two levels.
2. Design. A 12-run Plackett-Burman design screens all six main effects (a 2⁶ full factorial would need 64 runs; one-at-a-time with a reference needs 13 runs but estimates no interactions and confounds drift). Runs executed in randomised order.
3. Responses. Critical-pair resolution (limit ≥ 1.5), % recovery (98–102%), and %RSD of six replicate injections (≤ 2.0%).
4. Effects. On resolution, the standardised effects rank: pH (6.8) > %organic (4.9) > temp (3.4) > flow (2.5) > buffer (1.3) > wavelength (0.8), against a significance threshold of ≈ 2.2. So pH, % organic, temperature, and flow are significant; buffer conc. and wavelength are not.
5. Check the limit, not just significance. At pH 2.8 (low end), resolution falls to 1.42, below the 1.5 limit. That is the decision-maker: the method is not robust to pH over ±0.2, so the pH tolerance is tightened to ±0.1 and a system-suitability resolution check is added.
6. Outcome. Buffer concentration and wavelength effects are negligible, so their tolerances stay wide, easing method transfer. The method proceeds to formal validation with a defensible, data-backed robustness section and a tightened pH control, exactly what an ICH Q2 dossier needs.
That is ICH Q2 design of experiments assay optimization in one study: six parameters screened in twelve runs, one non-robust parameter caught and controlled before validation rather than during a failed transfer. To size the design for your own factor count first, our guide on how many experiments a DOE needs gives the run counts, and you can build the Plackett-Burman table directly in the design of experiments calculator.
From DOE to validation
The DOE happens during method development and robustness testing; the ICH Q2 validation study then confirms the finalised method's performance characteristics against pre-set acceptance criteria. DOE and validation are sequential, not the same activity, and keeping them distinct matters for the regulatory story.
The clean sequence is:
- Develop & optimise (ICH Q14 mindset): risk-assess parameters, run an optimisation DOE to choose conditions, define the analytical target profile.
- Prove robustness (screening DOE): confirm the chosen conditions tolerate normal variation, tighten any non-robust parameters, and set the method's stated ranges inside the acceptable region, the method operable design region.
- Validate (ICH Q2): run the formal study, specificity, accuracy, precision, range, linearity, LOD/LOQ, and a confirmatory robustness section, against acceptance criteria and generate the report.
- Transfer & monitor: because robustness was proven multivariately, transfer to QC labs carries far less risk of surprise failures.
Done this way, robustness is not a nervous end-of-validation check but a designed result. The DOE evidence also strengthens the submission: reviewers can see the method's parameter sensitivities quantified, which is the analytical analogue of the process design space in ICH Q8. For the acceptance criteria and worked calculations of the validation study itself, see our analytical method validation guide.
Frequently Asked Questions
Does ICH Q2 require design of experiments?
ICH Q2(R2) does not mandate design of experiments, but it lists robustness as a validation characteristic and describes it as an evaluation of the effect of deliberate variations in method parameters. Assessing several parameters at once, with their interactions, is exactly what a DOE does efficiently, so DOE has become the standard way to demonstrate robustness. ICH Q14, the companion analytical procedure development guideline, goes further and explicitly frames a multivariate, DOE-based approach as the enhanced route to method understanding.
What is the difference between ICH Q2 and ICH Q14?
ICH Q2(R2) covers validation: it defines the characteristics an analytical procedure must demonstrate, such as specificity, accuracy, precision, range, and robustness, and their acceptance expectations. ICH Q14 covers development: how you arrive at a good method, including risk assessment, defining an analytical target profile, and using design of experiments to build a method operable design region. In practice you use Q14-style DOE during development to optimise and prove robustness, then confirm the final performance characteristics under Q2 during validation.
How is a DOE used for method robustness testing?
Robustness testing asks whether small, deliberate changes in method parameters, such as flow rate, column temperature, mobile-phase pH, and percent organic, meaningfully change the result. A screening DOE, typically a fractional factorial or Plackett-Burman design, varies all these factors together across a narrow band around the setpoint in as few as 8 to 12 runs. The fitted effects show which parameters move responses like resolution or percent RSD beyond their limits. Those become the parameters that must be controlled tightly, and the method's stated ranges are set inside the region where every response stays acceptable.
Which factors go into an assay robustness DOE?
For a chromatographic assay the usual factors are flow rate, column oven temperature, mobile-phase pH, percent organic modifier, buffer concentration, and detection wavelength, each varied by a small amount around its nominal setpoint. For a ligand-binding assay such as an ELISA, typical factors are incubation time and temperature, coating and blocking concentrations, and wash count. The responses are the method's quality indicators: peak resolution, tailing, retention time, percent recovery, or percent RSD for a chromatographic method, and signal window, background, and curve fit for an ELISA.
Do I need statistics software to run an ICH Q2 robustness DOE?
No. The screening designs used for robustness, fractional factorial and Plackett-Burman, are standard tabulated designs that a free browser-based DOE generator can produce, complete with a randomised run order. You then plot the effects to see which factors are significant. Dedicated packages like JMP, Minitab, or MODDE add automated modelling and reporting, but the core robustness screen, a small orthogonal design plus an effects plot, does not require paid software to build or interpret.
Related Tools
- DOE Experiment Generator — Build the Plackett-Burman and fractional factorial robustness screens an ICH Q2 study needs, free in the browser with a randomised run order.
- ELISA Analyzer — Fit 4PL/5PL curves and read %RSD for the ligand-binding assays a robustness DOE optimises.
- Osmolality Calculator — A quick QC utility for the buffer and reagent prep that feeds assay setup.
References
- ICH (2023). ICH Q2(R2) Validation of Analytical Procedures. International Council for Harmonisation. Official guideline PDF
- ICH (2023). ICH Q14 Analytical Procedure Development. International Council for Harmonisation. Official guideline PDF
- Vander Heyden, Y., Nijhuis, A., Smeyers-Verbeke, J., Vandeginste, B.G.M. & Massart, D.L. (2001). Guidance for robustness/ruggedness tests in method validation. Journal of Pharmaceutical and Biomedical Analysis, 24(5–6), 723–753. DOI: 10.1016/S0731-7085(00)00529-X
- Ganorkar, S.B., Dhumal, D.M. & Shirkhedkar, A.A. (2017). Development and validation of simple RP-HPLC-PDA analytical protocol for zileuton assisted with Design of Experiments for robustness determination. Arabian Journal of Chemistry, 10(2), 273–282. DOI: 10.1016/j.arabjc.2014.03.009