Protocol: Design of Experiments (DoE) Study

This is a ready-to-use Design of Experiments (DoE) study protocol. It applies to any DoE phase, screening, characterization/optimization, or mixture, and to any design family, factorial, response surface, or mixture-constrained. Replace every <<FILL: ...>> placeholder with your own specifics, set your document numbers and dates, and route it through your normal protocol review and approval before execution. A worked filled specimen follows the blank template. Confirm each cited regulation against the current source before you rely on it. This content is general educational reference, not regulatory or professional advice; adapt it to your product, process, and quality system.

Approval page

1. Objective

To quantify, using a designed experiment, the effect of <<FILL: factor(s)>> on <<FILL: response(s), normally one or more CQAs>>, so that the effects and interactions are known with statistical confidence and can support Critical Process Parameter (CPP) or Critical Material Attribute (CMA) designation and, where applicable, construction of a design space. This study supports the process design work under ICH Q8(R2) and Stage 1 of the FDA process validation lifecycle.

2. Scope

This protocol covers the design, execution, and analysis of the DoE described in section 6, for the unit operation and factors named above. It does not itself designate CPPs or CMAs; that decision is made in the criticality risk assessment (<<FILL: RA-ID>>) using the effect sizes this study produces. It does not cover confirmatory process performance qualification (PPQ) batches, which are executed under a separate protocol once the design space and control strategy are finalized.

3. Definitions

• Factor: an input deliberately varied in the study (a process parameter or material attribute), continuous or categorical.
• Response: the measured output, normally a CQA or an attribute that predicts one.
• Center point: a run at the mid-level of all continuous factors, used to detect curvature and estimate pure error.
• Design space: per ICH Q8(R2), the multidimensional combination of inputs shown to deliver assured quality; this study may feed one but does not itself constitute one.
• Mixture design: a design used when factors are proportions of a fixed total (they sum to 100 percent), requiring a simplex-based or D-optimal design rather than an independent-factor design.
• Confirmation run: an execution at a chosen setpoint, not used to fit the model, used solely to verify the model’s prediction against an independent result.

4. Responsibilities

5. Prerequisites (confirm before starting)

1. The criticality risk assessment (<<FILL: RA-ID>>) has identified the factors in section 6 as warranting study.
2. Every analytical method used to measure a response is qualified and validated for the range expected in this study. See method validation essentials.
3. Equipment and instrumentation used to set and hold factor levels are within qualification and calibration for the execution period.
4. Raw material lots used in the study are representative of the range the process will see commercially, or the lot-to-lot variation is explicitly blocked.
5. Acceptance criteria in section 9 are approved before any run is executed.

6. Factors, levels, and responses

If any factor is a mixture component (a proportion that must sum to 100 percent with other components), state the constraint set explicitly rather than listing independent low/high levels:

7. Design selection and rationale

8. Randomization and blocking plan

State the actual run order (not the standard order) and any blocking variable. Randomize the execution sequence and record the realized order here before execution begins; a design run in standard order without randomization is a common and preventable deviation.

9. Acceptance criteria

10. Run log and results: design matrix

Record the actual realized factor settings (not just the coded levels) alongside every measured response, and the model’s predicted value once fitted, so the residual (actual minus predicted) is visible for every run.

11. Model diagnostics checklist

12. Confirmation-run results

13. Deviation handling

Any departure from the protocol (a factor level missed, an out-of-window measurement, an equipment fault, a run executed out of the randomized sequence) is recorded as a protocol deviation with an impact assessment on the fitted model. A run that cannot be corrected is documented as excluded, with the statistical justification and the effect on remaining degrees of freedom, rather than silently dropped. See validation test failure management.

14. Summary and conclusion

Summarize which factors showed significant main effects and interactions, the direction and magnitude of each, whether curvature was detected, the final model form, and whether confirmation runs matched the prediction. State which factors are candidates for CPP or CMA designation based on effect size, to be finalized in the criticality risk assessment. State explicitly whether the study met every acceptance criterion in section 9.

15. Attachments

• Raw data export from the DoE software, including the fitted model equation and ANOVA table.
• Residual plots (normal probability, residuals vs predicted, residuals vs run order).
• Analytical method validation summaries for every response measured.
• Design generation record (candidate set and optimality criterion, if D-optimal).

16. References

ICH Q8(R2), Pharmaceutical Development (DoE as the basis for a credible design space). ICH Q9(R1), Quality Risk Management (prioritizing which factors enter the DoE program). ICH Q11, Development and Manufacture of Drug Substances (applying DoE to drug substance and CMA-linked attributes). FDA Guidance, Process Validation: General Principles and Practices, 2011 (Stage 1, Process Design).

Confirm the current version of each reference before issue.

17. Revision history

Filled specimen (excerpt)

The following shows an illustrative central composite design (CCD) for a monoclonal antibody cell culture step, studying the effect of culture pH and temperature shift timing on titer and a charge-variant CQA. Numbers are illustrative; replace them with your own.

Design: CCD, 2 factors, 5 center points, alpha rotatable, 13 runs total. Responses: titer (g/L, target maximized, not less than 3.0), acidic charge variants (percent, target not more than 12 percent).

Model diagnostics: overall model p less than 0.001; lack-of-fit p = 0.31 (not significant); adjusted R-squared 0.94, predicted R-squared 0.88 (gap within the pre-set 0.20 tolerance); residuals showed no pattern against predicted value or run order.

Confirmation runs: three runs at the model-predicted optimum (pH +0.8 coded, temperature shift +1.0 coded) produced titers of 3.8, 3.9, and 3.7 g/L against a predicted 3.85 g/L, and acidic variants of 10.1, 10.4, and 9.8 percent against a predicted 10.2 percent, all within the model’s 95 percent prediction interval.

Conclusion (specimen): culture pH and temperature shift timing both show significant main effects on titer with no significant interaction; pH shows a significant effect on acidic charge variants. The model met every acceptance criterion in section 9. Both factors are recommended for CPP evaluation in the criticality risk assessment, with the optimum region confirmed by three independent runs.

Common inspection findings this protocol prevents

• A design space or CPP designation asserted with no DoE report behind it, or a report with no fitted model, only a table of raw results.
• Runs executed in standard (unrandomized) order with no documented rationale, exposing the study to time-related confounding.
• No center points in a factorial study, so curvature cannot be assessed and a linear model is assumed by default.
• A high adjusted R-squared reported with no predicted R-squared shown, masking overfitting.
• No confirmation runs, or confirmation runs only at the center of the studied region, never near a proposed boundary.
• A mixture study analyzed with an ordinary factorial model rather than the correct mixture model form.
• An analytical method used to measure a response with no validation record for the range studied.

How to adapt this protocol

1. Set your document number, unit operation, product, and DoE phase on the approval page.
2. Populate section 6 with your actual factors, levels or mixture constraints, and responses, sourced from the criticality risk assessment.
3. State your real design-selection rationale in section 7; if corner conditions are unsafe or physically infeasible, document that as the reason a Box-Behnken or mixture-constrained design was chosen over a central composite design.
4. Set the acceptance-criteria thresholds in section 9 to match your organization’s statistical practice and the criticality of the responses studied.
5. If the study is a mixture design, replace the factor-level table in section 6 with the mixture-constraint table and confirm section 11 includes the Scheffe-form model check.
6. Confirm every regulation in section 16 against the current published version before issue.