Why Not Use Design of Experiments for QbD?

Confession

Today, I share an embarrassing story… I did not truly appreciate DOE. When I was teaching Six Sigma topics at Stanford University and Keio University, I only scratched the surface level. Main reason was that most project sponsors could not share their proprietary or confidential details (process parameters), hiding real numbers and details of  equipment set up etc. So I had to deal with “encrypted” data and assumed information for many projects. Now this took away the context of the data and how experiments were set up. So I was stuck in the “theory sandbox.”

What’s more surprising? I wasn’t the only one. Of many professors who teach Design of Experiments, very few have had the industry experience to understand the nuances of experiments – beyond mere theory.

This was frustrating. So I threw myself back into the industry. Then came many days of firefighting due to product recalls. My R&D team members were subject to layoff’s due to a launch delay (mainly from failed experiments). It was only after years of  wrestling ill-defined studies, inefficient experiments, and highly constrained setups – that I began to appreciate the power of a sound Design of Experiments (DOE) approach.

Afterwards, I had the fortunate opportunity to be involved in more than 50 experiments in the BioPharma and Medical Devices industry. I began noticing a behavioral pattern. The practicing scientists and engineers who have experienced the struggle appreciate the DOE approach. They are the ones who think, “If I had only known what I know now, I could’ve saved much time, resources, frustration and gotten better results!” So I wasn’t the only one. Fair enough.

Why is the adoption rate of DOE in BioPharmaceutical or Medical Devices lower than other industries?

Then I began to wonder, why the adoption rate of DOE in Biotech, Pharmaceutical or Medical Devices was lower than other industries.

Then came QbD. Quality by Design (QbD) has been a great regulatory initiative to push the multivariate analysis (MVA) or DOE. QbD provided the awareness but I don’t think this external motivation was enough. I was interested in what other factors were impeding this great tool. So I turned to my QbD community members.

Why not use Design of Experiments?

In a LinkedIn group, I conducted a survey with the intent to learn what mental blocks may exist for design of experiments (DOE) for QbD. Below are the results and comments. Read the comments as they are insightful.

DOE Survey

Inna Ben-Anat, Director, Global QbD Strategy and Product Robustness at Teva Pharmaceuticals

Hi Kim, I am really curious to see the results for this poll….as I truly believe that DOE is the ultimate way to experiment effectively and efficiently

Aled Finniear likes this

Yamak Mehta

Research Associate in Downstream Process Development (R&D)

Resource + Time for execution and analysis and ultimately = cost !

Sun Kim

Sr. Manager, Master Black Belt in Quality-by-Design, Agile Lean Six Sigma and Design Thinking

Top Contributor

Yamak, So more resources? Inna, yes, I hope we can get some insight.

Pushpa Tiwari

Assistant Manager at Dr. Reddy’s Laboratories

I sure have heard people complain it requires more experiment than it claims. The problem here is not DOE but the mind set with which people do development .they are so use to the usual OFAT way of doing experimentation that they still do it mostly that way and when the process is almost optimized use DOE because its a management mandate. i would like to add one more thing though it is not relevant and probably we should start another thread of discussion for that, and that is : where we go wrong in the usage of the DOE e.g variable selection ,trying to do DOE when the process is simple and the initial screening data shows that their is only one factor which impacts on the response. etc.

Sun Kim, Yamak Mehta like this

Sun Kim

Sr. Manager, Master Black Belt in Quality-by-Design, Agile Lean Six Sigma and Design Thinking

Top Contributor

Pushpa, excellent observation. This poll is an attempt to understand the mental blocks of doing DOE.

Yamak Mehta

Research Associate in Downstream Process Development (R&D)

Sun Kim , Actually I also want to mention about mind-frame of users and generation gap between old and new researches. I think new generation understand long term benefits of QbD and DOE based development but cant implement always.

Old generation ( Experienced ) researcher have prior experiences and also have shortcuts for development with limited resources , time and also cost-effective manner. And if this is tradition then also its not favorable to go with this time consuming , high cost and high resource utilizing methodologies as per their thinking.

By Ranging study and design space obtained from DOE experiment analysis, you can predict your end product quality and yield. And also avoid batch failures and also easy to convince quality assurance and regulatory authorities.

Because of this reasons , I think researchers are very well aware but it is partially (half) accepted. But in future it can be definitely possible, if regulatory authorities ask for the same.

Gawayne Mahboubian-Jones

Manager – Quality by Design at Philip Morris International

I also have similar experience … the gap is not so much the resource to undertake the DOE, but rather it’s two sets of understanding …. the understanding in the researchers of how to apply DOE effectively and efficiently, … and the understanding in the managers of the power and overall efficiency (including cost-saving) of applying DOE at an early stage in the process when it has most power to influence the outcome.

Sun Kim

Sr. Manager, Master Black Belt in Quality-by-Design, Agile Lean Six Sigma and Design Thinking

Top Contributor

Yamak, Gawayne, Thanks for sharing! I resonate with your observations. This poll is giving some insights on the “misunderstood” side of DOE’s. Please share more.

Jasmine , ASQ CQE

Senior Scientist- Quality by Design at Dr. Reddy’s Laboratories

I wish we find more participants to the survey!

🙂

Sun Kim

Sr. Manager, Master Black Belt in Quality-by-Design, Agile Lean Six Sigma and Design Thinking

Top Contributor

Yes. keep voting!

Inna

Inna Ben-Anat

Director, Global QbD Strategy and Product Robustness at Teva Pharmaceuticals

It is encouraging to see that zero think that DOE is not required (alternative definition of DOE is not needed, Do-it-Or-Else…:-))

Imre Molnár

President of Molnar-Institute

I think, there is no alternative to do solid science based on DoE. Some people think, DoE is generally needs too much time as there are a great numbers of different Designs, which looks like they would need a lot of time to be finished. However there are simple ways to reduce the number of experiments to the most important ones and have a modeling software, where you can generate f.e. out of 12 experiments over a million visual results, as it is done by revolutionary DryLab-software for HPLC. If the complexity of a DoE can be reduced, the outcome is a better understanding of the process with a higher flexibility of changes and getting less regulatory controls.

In summary: Use simple Designs with Modeling Software for great products.

Charles Lowery

Formulation/Scale-Up Chemist at Virbac Animal Health

I guess as one of the “old dogs” in the field, I haven’t really been shown the practicle application of DoE. And now as a formulator who is called in to “fix” a formulation that isn’t working properly I am limited in my responses of how to repair the formulation.

In many ways, having been in materials research for a number of years pior to becoming a formulator, my old approach of throwing number of possible solutions against a wall and seeing what stuck was a form of DoE. Once I had an idea of what could possibly answer the formulation question, I would design experiments around the answers that stuck to the wall.

Hans-Werner Bilke

Ihr Partner für multifaktorielle HPLC-Trennoptimierung und Robustheitsprüfung

Commercially available software for computer -assisted HPLC Method Development (CMD ), such as DryLab subject in its latest version, the limitations of a 3 -factor optimization and the use of only one target size (resolution Rs). The calculated optimum is due to this limitation is not the ” global optimum ” but mostly a possible ” local optimum “. This is most forcibly in the ” global optimum ” over optimized. The ” global optimum ” can only be found if all significant variables HPLC separation using the statistical design of experiments are varied simultaneously.

Sun Kim

Sr. Manager, Master Black Belt in Quality-by-Design, Agile Lean Six Sigma and Design Thinking

Top Contributor

These are great insights! Especially the honest opinions of those who are newer to DOE. Please share more personal stories.

Emil Ciurczak

Independent Pharmaceuticals Professional

Charles, without sounding cruel, a shoddy formulation for dogs seldom leads to lawsuits or 483’s. A poorly formulated tablet for humans is another matter. There are few ways (other than perhaps checking LD-50’s of clinical trials) CQAs and that is not done by throwing stuff against the wall…well, maybe in developing countries, but not in the EU and US.

amirhossein morovati

pharmacy student at islamic azad university of pharmaceutical sciences

I did not have any experience before but I am going to start with QBD. From what I have read as the other friends said one problem is the wrong insight to analyse the datas obtained from initial screening following wrong descisions . I am going to be familiar with QBD as I said so I will be happy if you Sun can provide some resources. Wish you best regards

Sun Kim

Sr. Manager, Master Black Belt in Quality-by-Design, Agile Lean Six Sigma and Design Thinking

Top Contributor

Amirhossein, I am working on a starter guide at QbDWorks.com. Please check in a few weeks. Emil, 483’s…it’s unfortunate I’m familiar with them too. It’s important for this discussion to be as open and honest as possible so we can understand the constraints.

Rajkumar Juturu

Method Development Scientist

In method development and validation, we follow a systematic approach considering critical parameters that affect the molecule. Tha approach may be univariate, one at a time, but it improvises target output and minimizes unnecessary experimentation. Also, everything is formulated per specific ICH guidelines. Validation parameters are systematically evaluated per ICH. Doesn’t implementing multivariate QBD approach waste time and resources, instead of using logical and robust approach to a solution.\

Emil Ciurczak

Independent Pharmaceuticals Professional

Wow! I can just imagine the Luddites saying, “We don’t need steam engines; we’ve done well with horses pulling plows for centuries.” Seriously, one-at-a-time is better at finding interactions among 2, 3, or 4 factors?

If you truly believe that 1) you are making very, very, very simple, high dose, immediate release tablets (or a solution of a rock-solid, never degrading drug) or, 2) you are one 483 from closing your doors.

A K M Hai

Chief Scientific Officer at MOLESCI & MATESCI INC

Top Contributor

Hi Emil,

you put a nice comments. Thanks

A K M Hai

Chief Scientific Officer at MOLESCI & MATESCI INC

Design Of Experiment required strong scientific knowledge on the resources involved like ingredients, processing equipments and analytical equipments and methods.

To do MFAT (Multiple-Factor-At A-time) for DOE rather go with traditional OFAT (one-factor – at-Time) really one needs sharp intelligence or shrewd guess on critical quality parameter

A K M Hai

Chief Scientific Officer at MOLESCI & MATESCI INC

OFAT (one-factor- at- a time) strategy is very applicable to understand reaction mechanism (reaction pathway) in chemical and biochemical synthesis.

In the pharmaceutical product formulation development

One can do DOE (Design of Experiment) by introducing key multiple parameter/factor at a time (MFAT) to save time instead of OFAT (one-factor- at- a time) to get quality product in the formulation development.

Charles Lowery

Formulation/Scale-Up Chemist at Virbac Animal Health

Hi Emil. If you had read my note correctly, you would have noticed that I said that “throwing stuff against the wall” was the way I use to start my formulation work. That was 30 years ago when I was in material research. Since moving into pharmaceuticals, I have had to make more scientific decisions in my formulations. QbD wasn’t the way I learned, so I have had to adapt. It just is that when QbD is mentioned, I am not always sure of what it means, and I think that different people here on the forum have different definitions.

And, just because I am currently in animal pharmaceuticals, do I get any slack on the quality of our work here. The FDA and EU authorities monitor us just like human pharma.

Emil Ciurczak

Independent Pharmaceuticals Professional

Charles, I have been, er, uh, “spitting” into the wind for years on MVA and am sensitive about another, “We don’ need no stinkin’ badges” remark. Even something as simple as pre-formulation would be infinitely better when you mix several potential excipients with the API would and check for “unexpected” interactions using a MVA approach. This would take an unusual amount of samples and take a LOT of work in the “conventional” manner.

One suggestion I have often made was to use NIR or Raman instead of traditional chromatography for screening purposes. Using a well-planned DoE to make the minimum number of samples to gain the maximum amount of information, the vials could be examined (through the containers) daily….non-destructively. When a change of any type is detected, the sample could be sacrificed and run via TLC or HPLC. This approach gives us potential interactions AND rates at which they occur. All for less work and, possibly, in less time than the “traditional” approach.

That’s all I was saying…

Charles Lowery

Formulation/Scale-Up Chemist at Virbac Animal Health

No problems with QbD and/or MVA. I guess that the point I was making is that QbD, as a recognized means of research, wasn’t the way I was taught, and was frowned upon all those years ago. Now I do things differently, and still don’t have a firm grip on how to properly se up a study.

And I have sensitivitiy about people thinking that animal pharma is different from human pharma as far as viewed by regulatory agencies We have the same hoops to jump, just that our target users can’t tell us what is happening to them. And often, our targets are many species, not just the human one.

Stefano Selva

Senior Scientist Formulation – Pharmaceutical Sciences at Aptuit

Emil, great idea the application if NIR/Raman also for screening purposes! It seems like to use a ‘PAT-like’ approach during early phases.. thanks for the suggestion (non destructive and allow to check rates). What kind of MVA model do you use for that (PCA to monitor trend from t0 samples? others? You know, the calibration efforts should be ‘limited’ on this phase..).

I use DoE approaches (and also Multivariate data mining on existing data/processes) for all the new developments (early phases, starting from the first excipients compatibility..) as well as for Process Validation and CPV (Multivariate process Monitoring).. I am sure there are no other ways to really understand your formulations and processes. Consider also the application of MVA on existing products which could be ‘similar’ (in terms of QTPP, CQAs) to the ones you have to start developing.. this Risk Assessment can represent the ‘start’ of your work..

Emil Ciurczak

Independent Pharmaceuticals Professional

Stefano, Since there is no precedent for following “changes” in preformulation studies, I have used the same software as we use for raw material qualification. The mixture (in question) is shaken up, tapped down, and scanned. Three scans of the mix are then considered the “material.” Depending on your software, this will establish a mean or cluster, with each individual “sample” (shaken sample) of that one blend having a distance from the mean/center.

After a period, during which the sample has been subjected to heat/humidity, it is reshaken and tapped down again and scanned. The distance from the should be in the same magnitude, if there is no change. [E.g., if the distance was ~ 1.1, then 1.2 or 0.9 could be expected. If you get 2.1, 2.3, etc., there is clearly “something” different and the sample could be sacrificed and assayed.]

In time, you will have enough data to populate a MVA system, such as a PLS equation which will show which mixture had the largest effect(s). This is the backbone of most commercial DoE software, which would also suggest which mixtures to make for the experiment.

Does that help?

Stefano Selva

Senior Scientist Formulation – Pharmaceutical Sciences at Aptuit

Thanks Emil! Yes, really interesting! It could be useful to plan the mixtures using a Mixture Design approach (so as you can model the further results on this). The NIRS/Raman with this ‘dynamic distance’ to the time zero clusters can be used as a response to check the modifications if compared with the original spectra (also considering that NIRS could detect physical modification over the time). The Mixture Des can then define the potential excipients (or combination of) impacting your formulations. Really nice!

Mark Zagorski

Umetrics – Regional Sales Manager

The first challenge it to prevent people from doing too many mistakes in the DOE process, guiding procedures and robust functions,

The second part is the risk analysis (our MODDE) -Probability Contour Plots, The classical contour plot without understanding the prediction variability have created missinterpretations of the qualtiy in the results.

Another thing to focus on is the graphicl GUI, many DOE softwares are still very statistical in the communication on top of the multidimentional problem it can be to much to grasp and interprete.

What about Your Experience?

Do the survey results represent your case?

Please participate in the living survey.

Just click on your vote and you can see the new results.

[yop_poll id=”3″ tr_id=”1″]

Do you have any experiences to share?

Get New QbD Tips!
I agree to have my personal information transfered to AWeber ( more information )
Want More Tips from QbD Practitioners? Then join our newsletter!
We hate spam. Your email address will not be sold or shared with anyone else.