In recent years, the pharmaceutical industry has faced challenges in improving operational efficiency due to various changes in the business
environment. In particular there is a strong need to reduce the burden of enormous paperwork.
Takeda Pharmaceutical Company Limited (hereinafter referred to as "Takeda Pharmaceutical Company") is promoting a number of advanced AI and DX-related initiatives, and as part of these initiatives, we have jointly developed an AI solution with Findability Sciences that reduces the work of document verification in quality control operations that require close data verification.
We interviewed Mr. Karashima of Takeda Pharmaceutical Company Limited, who led the project, about the details of the project and the future vision
of pharmaceutical
manufacturing using AI.
Takeda Pharmaceutical Company Limited Pharmaceutical
Science Sustainability & Technology Head, Innovation
In 2003, he completed a master's course at the University of Tokyo's Graduate School of Engineering. In the same year, he joined Takeda Pharmaceutical Company Limited, where he was engaged in pharmaceutical analysis research, physical properties, and pre-formulation research at the Development and Analysis Laboratory of the CMC Division. During that time, he obtained a Ph.D. in pharmacy from Meiji Pharmaceutical University in 2017. Since 2020, he has been in charge of promoting digital transformation in the CMC division, where he is currently in his current position.
Interviewee Sayaka Tanaka: In charge of the life science business at Findability Sciences Co., Ltd. Completed a master's program at the University of Tokyo's Graduate School of Pharmaceutical Sciences and a doctoral program at the Graduate School of Medicine. After working as an assistant professor at a national university, he was engaged in medical affairs work in the field of oncology at a foreign pharmaceutical company. Since 2020, he has been involved in business development in the healthcare field at SoftBank Corp., and in his current position, where he has been seconded since May 2023, he will be in charge of business promotion and business development in the life science business.
Table of contents
Tanaka:
In this project, CMC work for pharmaceuticals※１QC is one of the important tasks in QC※２The goal was to automate the document verification work
that occurs during checks. Please tell us about the general workflow of QC check work.
*1: Abbreviation for Chemistry, Manufacturing and Control. A general term for all operations related to the quality control of pharmaceuticals.
*2: Abbreviation for Quality Control.
Mr. Karashima:
We, the CMC department, are conducting clinical trials.※３It is responsible for the manufacture and quality check of drugs under development.
In the quality check of pharmaceuticals, we first set test items to confirm quality and standard values (standards) for each item, and then summarize
them as a standard document. Next, quality confirmation tests are conducted in-house or at contractors according to the method specified for each
test item, and the test results of each item obtained in the end are aggregated and CoA※４is created. It is the QC check work that ensures that the
contents of this CoA are indeed correct.
*3: A study conducted to confirm the efficacy and safety of a drug candidate on healthy adults and patients, as well as the treatment method
(appropriate dosage and method of administration).
*4: Abbreviation for Certificate of Analysis. A test report to ensure that the quality of the manufactured pharmaceutical product complies with
the standard.
Mr. Karashima:
Specifically, the procedure for creating a CoA is to first transcribe the test items and standards from the standard. Then, the results are transcribed
from the test report for each item obtained from the test provider. In other words, since the CoA is created by transcribing information from two
documents, the QC check must not only ensure that the test results conform to the standard, but also that the content posted in the CoA is consistent
with the standard and test results.
At Takeda Pharmaceutical Company at the time, the QC check procedure was first performed visually by the person who created the CoA, then
double-checked by employees other than the person in charge and finally checked by the quality assurance manager. In this process, we check from a
third party's point of view that there are no mistakes in the contents of the CoA, check them on a paper basis, and leave proof that they have actually
been confirmed. When a developed product is delivered to a medical institution that conducts clinical trials, this document is attached to the product
to ensure quality, and the current situation is that a lot of work is labor-intensive.
Tanaka:
Specifically, how long does it take?
Dr. Shinjima:
Medicines are actually taken by patients. Pharmaceutical companies have a very important responsibility to ensure the quality of their products, no
matter what the situation. The stance that all external materials must be accurate is common sense throughout the pharmaceutical industry, and each
company is screened with stricter standards than other industries. As a result, the workload is currently high, and DX is being promoted throughout the
pharmaceutical industry, regardless of department.
Especially in the field of CMC, due to the role of having the most important mission to ensure quality, mistakes cannot be allowed in documents that
guarantee quality. In order to always aim for 100% accuracy, it takes more than a certain amount of time to prepare the work because it requires
multiple checks after the document is completed, after carefully checking the validity of the content with the human eye, and then consolidating and
transcribing everything, and even after the document is completed, multiple checks are required.
At present, it takes several days from the time we receive the quality test results to the time we make and sign the CoA, and I think that the total work
time is less than 10 hours per CoA. This is required for all medicines. As a result, the burden on the site is very high, and internal surveys have conveyed
that "automation to reduce time and improve efficiency in document processing is a major need." Therefore, we decided to start with CoA, which has
a relatively simple structure.
The other day, when I presented this content at an academic conference, the response was very great, and many people agreed with it. As an issue
for the pharmaceutical industry as a whole, I feel that there is a strong demand for automatic creation of quality documents and automatic checks
from the field. Dr. Karashima presents at an academic conference on the AI solution developed in collaboration with Findability Sciences.
Tanaka:
I understand that you are currently in the process of preparing for full-scale implementation, but please tell us specifically what kind of procedure you
will use for the QC check system you developed this time.
Mr. Karashima:
Our current assumption is that we are thinking of using this system at the beginning of the workflow. Specifically, we envision a flow in which documents
compiled by the person in charge of creating the CoA are sent for check by the AI system, and the results are reliably transcribed and double-checked
to ensure that they conform to the standard and then sent to the secondary check.
On the other hand, quality testing work is often outsourced, and CoA may be prepared by the outsourced company. The CoA guaranteed by the person
in charge of quality assurance is delivered at the outsourced company, but the CoA created by the contractor must also be checked by the quality
assurance staff of Takeda Pharmaceutical Company Limited. Similarly in this flow, we are considering the use of the AI system in the first check by the
person in charge.
In addition, QC checks are performed many times, but the final quality assurance is stipulated as a standard in the pharmaceutical industry to be
"signed by a person with a human eye", so this will continue to be done even after the introduction of the AI system.
Tanaka:
Due to the nature of the work required to develop pharmaceuticals, the accuracy standards required are very high, but can this system meet those
needs?
Mr. Karashima:
In this AI system, there is a process of subjecting the document template to be analyzed to machine learning during development, but I think that
the document check using the learned document template is almost 100% accurate.
On the other hand, depending on the production lot, there may be a change in the document template, for example, depending on the experimental
results obtained, there may be one more line (of analysis results).
In order to prepare the system so that it can achieve sufficient accuracy even for such irregular documents, I think it is important to have machine
learning after predicting the variations of documents that may occur in the future to some extent. If we can do that, we believe that the accuracy will
be as close to 100% as possible.
Tanaka:
Please let us know if you have any reactions from people who have actually used the AI system.
Mr. Karashima:
In addition to the impression that the concept itself is very unique, the response from those who have already used the service is very good, saying that
they are very grateful for the realization of such an initiative in particular.
Tanaka:
From what we have said so far, we can clearly see the high need for automation of document verification work in QC checks. So, what are the important
requirements for a system development partner to tackle these issues?
Mr. Karashima:
As a partner in the development of AI systems, we believe that it is important to have the flexibility to flexibly respond to our needs. For example, by
flexibly proposing a system that can be developed at a scale according to the scale of the system and the needs of its use, it is possible to proceed with feasibility checks with a small investment, which has the advantage of shortening the time required to obtain a POC (Proof of Concept).
In addition, in the process of actually developing the system, the requirements of our side sometimes change, but I feel that it is very easy to work as
a partner if you can respond flexibly and quickly to them.
Next, I would like to talk about our technological capabilities. When selecting a partner, it is important to evaluate technical capabilities.
For document automation challenges such as this one, our partners have their own development teams for machine learning models,
natural language processing, and OCR ※５I think it is important to have a lot of experience and know-how for technical elements such as. More
importantly, we can expand globally. Since Takeda Pharmaceutical Company is currently a global company that is expanding its products around
the world, it is necessary to keep in mind the implementation of such a system on a global basis. In the implementation phase after system
development, it is necessary to go to the site and respond to on-site training and maintenance support, and I feel that a partner who can handle
this is very attractive.
As a final point, I would like to mention reliability. When selecting suppliers when introducing a large system and deploying it company-wide,
the reliability of the supplier company is very important as one of the grounds for internal explanations. However, in the case of major companies,
it tends to be a big project, and on the other hand, in the case of a start-up company, there are concerns about the sustainability of the business,
although there is flexibility. The balance between the two is a very troubling point, but in recent years, the number of partners who have the
advantages of both are increasing, and we feel very reassured.
*5: Abbreviation for Optical Character Recognition (or Reader).
A processing technology that converts characters recorded as image data into digital data as text.
Tanaka:
The target of this project is QC checks of CoA in Japan, but please tell us if you have any plans for applying it to other operations.
Karashima:
Currently, we are implementing it in Japan, but of course we would like to expand it in the United States in the future. In addition, there are many
documents other than CoA that are relatively easy to structure, and we believe that a similar platform can be used for stability testing, for example,
to regularly verify the quality of products stored under certain conditions and to ensure long-term quality. Therefore, as the next step in the use of AI,
we would like to expand the automation of paperwork in stability testing.
Tanaka:
Do you have any specific expectations for improving efficiency as you expand the operation widely within the company?
Mr. Karashima:
In terms of development items, Takeda Pharmaceutical Company Limited issues about 100 CoAs a year, but the frequency of operations itself is
not high. On the other hand, many factories that handle post-market items that have been developed and approved for manufacturing and marketing
by regulatory authorities handle 10 times or more than we do, which means that the frequency of quality tests and accompanying QC checks is high,
and document verification occurs in proportion to the production volume.
Therefore, if we introduce a similar system in the factory department that manufactures more products and increase the number of situations were
it is used throughout the company; we will be able to achieve considerable time savings and efficiency. Post-market products are actually used by
patients, and some of them rely on the power of pharmaceuticals to sustain their lives for the rest of the day. Such products must be reliably kept in
supply. In doing so, we must also ensure quality, so we believe that the most important thing for us in the end is to handle far more items than we
develop and to develop them in factories that have higher responsibilities.
Tanaka:
In recent years, I think that supply stability has become an issue for the pharmaceutical industry as a whole. In addition to the automation of document processing that you used this time; do you expect the use of AI in pharmaceutical manufacturing to improve supply stability?
Mr. Shinjima:
I think this is an area where we can expect a lot of promise. Takeda Pharmaceutical Company Limited has also set up a new data science and digital implementation department in the factory division and is implementing a platform that monitors various parameters of each process of the product during production to predict the final quality of the finished product through the entire manufacturing process before quality testing is performed.
Specifically, we are trying to detect anomalies and predict the actual final quality in addition to constantly monitoring the analysis results by placing an analyzer called a probe in the manufacturing equipment. Currently, all quality test items must be actually tested before shipping, but we are aiming to introduce "real-time release," a method that guarantees the quality of the final product based on the analysis results in the process.
Tanaka:
So far, you have talked about the expectations for the introduction of AI in pharmaceutical quality control from various perspectives as a
pharmaceutical company, but I recognize that the government and regulatory authorities also expect the use of AI in drug development
and manufacturing. On the other hand, I think that there is naturally a cautious idea from the perspective of final guarantee of quality and safety.
To what extent do you think it is regulated to allow AI-powered automation to skip processes?
Mr. Karashima:
In the past, regulatory authorities have commented that they do not deny the use of AI through academic conferences and other means, but in May
of this year, the FDA announced that it would not deny the use of AI in drug development, especially in the field of CMC, and points to keep in mind.※６
A discussion paper has been issued that formally summarizes the way of thinking.
*6: Abbreviation for Food and Drug Administration. U.S. Drug Regulatory Authority.
Mr. Karashima:
In this report, the FDA states that the use of AI is unavoidable in pharmaceutical manufacturing in modern times, and that development companies are
clearly required to fulfill sufficient accountability when AI is used in drug development in the future.
As a pharmaceutical developer, we need to take responsibility and understand the characteristics of AI, thoroughly discuss which operations AI can be
applied to and prove that the system we have introduced is functioning reliably when we actually start operation.
PMDA, the regulator of Japan※７It is expected that the discussion will accelerate in the future. In the future, we would like to go beyond the company
and involve the entire pharmaceutical industry to spread the use of AI in all kinds of ways, including document automation like this one.
*7 PMDA: Abbreviation for Pharmaceuticals and Medical Devices Agency. Pharmaceuticals and Medical Devices Agency.
Tanaka:
Thank you very much. I sincerely look forward to a world in which the use of AI will be promoted throughout the pharmaceutical industry, and medicines
will be delivered efficiently to as many patients as possible.
.png){kind=small}
%20(1)%20(1).jpg)
