Navigating towards AI success: A Comprehensive Roadmap for Implementation
Author(s): Ivo ten Voorde
Abstract
Artificial Intelligence (AI) is revolutionizing the pharmaceutical industry, enhancing efficiency, safety, and compliance. In this article, we delve into the steps necessary for successful AI implementation within pharmaceutical logistics. I’ll provide my own experience, real life cases, links to scientific articles, and adhere to guidelines from respected regulatory bodies. By following my approach ensures compliance with Good Manufacturing Practices (GMP), Good Distribution Practices (GDP), and various ISO standards and will aid in a successful implementation.
Introduction
Implementing a successful Artificial Intelligence (AI) within your
pharmaceutical manufacturing and/or logistics organisation is not
just about integrating cutting-edge technology-it’s about redefining
how quality management aligns with the rigorous standards
like Good Manufacturing Practices (GMP), Good Distribution
Practices (GDP),various ISO certifications and country specific
regulatory requirements [1-8]. Your approach must ensure that you
are embracing technological innovation, while in the meantime
ensuring compliance with regulatory requirements. In this article,
I’ll go into the outline of a comprehensive, step-by-step manual
to successfully implement AI within pharmaceutical organisation
and logistics and dive into my own experience and research from
others.
Initial Idea: Identifying Opportunities for AI Implementation
Your objective here is to identify areas within your pharmaceutical
manufacturing and/or logistics operations where AI can bring
significant benefits, such as enhancing efficiency, optimizing
supply chain management, improving predictive maintenance,
and ensuring regulatory compliance.
The steps below can be followed:
Establishing a Dedicated Cross-Functional AI Implementation
Team
Begin with assembling a dedicated team that includes a project
manager, quality manager(s), data scientists (QA/QC), regulatory
affairs specialists, and key operational staff. This diversity ensures
a well-rounded approach to integrating AI technologies, focusing
on operational excellence and regulatory compliance. If possible
add an AI expert as well, if you have this internally available.
Gap Analysis
Conduct a thorough assessment of current processes, challenges,
and opportunities within the organization’s manufacturing and
logistics operations. Then ensure you identify the areas AI can
or will be helpful to assist and ensure you have your scope and
limitations clear.
Tip from myside: use a patient journey map on your service/
product and this visualisation will help you to identify where
and in which stages AI could help (Research, testing trials,
manufacturing, diagnostic, quality control, administrative tasks,
analytics and so on) [9,10].
Data Strategy
Have a clear strategy what you want to improve and use your
scope and limitations as starting point. AI can help in many fields
but ensure you focus on key-areas identified in the GAP analyses
and I’ll mention here some areas where AI can give you quickwin like in:
- Data analytics analysing and cleaning up data (operations /
research / clinical trials / testing) [11,12]
- Linking datasets (anonymous patients data) and simulate
results to identify patterns and or predict (negative) side
effects (end user care / diagnostic aid / operational excellence)
[13-15]
- Quality control (imaging / anti-fraud / deviations avoidance)
[16,17]
- Predictive maintenance (operations) [18,19]
- Automation of inventory management (operations / logistics)
[20]
- Risk management (operations / logistics / end user care /
diagnostics) [21,22]
- Aiding in administrative tasks (scheduling meetings,
translations etc.) [23,24]
With your scope clear and you’ve identified which areas AI
could improve your organisation and can start consulting the
stakeholders.
Stakeholder Consultation
Engage key stakeholders and ensure you do include the (end)
users and impacted departments. This does mean you should not
forget to ask feedback from the workers, who’ll be impacted by
the change. Include your manufacturing- & logistics departments,
supply chain experts, regulatory compliance officers, end-users
(doctors/nurses/operators) and your IT professionals. Ensure you
have a good understanding of their pain points and potential areas
for AI intervention. With their feedback combined this in what
I call your Package of Demands (PoD). This PoD includes the
following important information and is the foundation of your
AI. It includes your AI vision, objectives and defences clearly
on the following topics:
- The quality and availability of the data the AI will be trained
on [25,26]
- Transparency and quality controls of the AI (deviations
avoidance and regulatory compliance, security issues) [27-31]
- Governance and monitoring (evaluation) controls [32]
- Scope and which legal frameworks to include (GMP/
GDP/21CRF-11/ISO)
- Budget & Cost effectiveness, is AI nice to have but too
expensive, cost saving and / or increase patient safety by
improving quality of product/service? [33]
- Training and increase the awareness of the users of the AI,
because if you enter “bad” prompts, you can’t expect a good
reply
- Avoiding biases and form a transparent ethical framework
[34-36]
- Patients approval and privacy concerns [37,38]
- Prioritize AI use cases based on their potential impact,
feasibility, and alignment with organizational goals and
applicable regulatory requirements.
Market Research
Now with your Package of Demands you can sent out a tender
and do market research of what is available on the market of AI
solutions. Explore off the shelve AI applications and solutions
tailored made to the need of your organisation and related industry.
Even when third parties are sending in their offers, you yourself
should consider looking into case studies, whitepapers, and
industry reports from respectable peer review studies and sources.
It can save a lot of time and if known pitfalls can be avoided, you
should do so.
As an alternative solution you can also opt to build an AI yourself
- however this is a whole complete new undertaking. Your
organisation should have at least the personal available for this
(dedicated), and need to have (AI) experts inhouse already. If not,
my advice is strongly to NOT chose this option.
Selecting the Right AI Solution: Ensuring Effectiveness and
Compliance
Objective: Choose AI solutions that meet the specific needs of the
organization while ensuring regulatory compliance and mitigating
known pitfall like biases. The following steps can be followed:
Requirements Definition
Clearly define the functional requirements, technical specifications,
and regulatory constraints for the desired AI solution. Simply put,
link your (medical) context and limitations in which AI will aid
your organisation.
Vendor Evaluation
Evaluate AI vendors based on their experience in the pharmaceutical
or relevant industry, track record of successful implementations,
compliance with relevant regulations (e.g., FDA, EMA), and
approach to bias mitigation.
Be aware that regulators will demand you have your risk mitigation
and profiles in order no matter what solution you choose. The
“AI maturity” can be seen in these forms and amount of risk and
impact varies:
- Experimental: like the name suggest: it’s a new solution
and is excellent for testing and for research. Not feasible to
be implemented directly to wider public/patients
- Newly Introduced: new in your market (country), AI
solution where there is some clinical evidence and proof of
concept. The results obtained are from other countries, but
not introduced or used yet in the country of your organisation.
- AI - Version 2 or Higher: Used by many different countries
already or has substantial evidence and trails completed and
results are published and available. Even other organisations
and/or your competitors locally are using this AI solution or
are in process of doing so.
- Tender Execution: Let the market decide if they have the
right solution for your organisational wishes and tailor make
a solution.
- Own Creation: Alternatively as mentioned before, you
can invent your own AI model, but this is option is not my
personal recommendation.
Proof of Concept (PoC)
Conduct a PoC with your selected vendor(s) to assess the feasibility
and performance of their AI solutions in real-world scenarios.
Ensure that PoC includes rigorous testing for bias and fairness
and ensure compliance and mitigate your risks [39,40].
Compliance Review
Collaborate with legal and regulatory experts to ensure that the
selected AI solution complies with data privacy regulations (e.g.,
GDPR), industry standards (e.g., Good Distribution Practice), and
ethical guidelines (e.g., IEEE AI Ethics Initiative) [41].
Testing and Implementation: Deploying AI Solutions Safely
and Effectively
Objective: Deploy AI solutions in a controlled manner, ensuring
minimal disruption to operations while maximizing benefits, with
the following steps
Data Preparation
Cleanse and preprocess data to ensure accuracy, completeness,
and relevance for AI model training. Use techniques such as
data anonymization and de-identification to protect sensitive
information. You don’t want to have issue with this later on and
especially in markets like EU and US privacy is an important
matter [42].
Also, don’t forget that AI is only as smart as good as the data you
train/feed it, if you put“garbage in your output will be garbage as well” [43-46].
Model Development
Develop AI models using appropriate algorithms (e.g., machine
learning, deep learning) and techniques (e.g., supervised learning,
reinforcement learning) based on the nature of the problem and
available data.
Testing and Validation
Conduct thorough testing and validation of AI models using diverse
datasets and realistic scenarios. Evaluate model performance
metrics such as accuracy, precision, recall, and F1-score and
follow available guidelines from inspection institutes like the
FDA [47,48].
Pilot Deployment
Roll out the AI solution in a pilot environment or limited production
setting to assess its performance, scalability, and user acceptance.
Collect feedback from end-users and stakeholders for iterative
improvement. Make sure your evaluate and do correction, before
the AI model goes full scale.
Full-Scale Deployment
Gradually scale up the deployment of AI solution across the
organization’s logistics and airfreight operations, ensuring proper
training, documentation, and support for end-users.
Don’t forget to train the staff on the new features and changes
and use the feedback and questions as learning opportunities to
tweak and improve your AI model.
Follow-up, Monitoring and Control Mechanism
Objective: Establish mechanisms for ongoing monitoring,
evaluation, and refinement of deployed AI solutions to ensure
effectiveness, compliance, and fairness. The following items
can be used:
Performance Monitoring
Implement monitoring tools and dashboards to track the
performance of AI models in real-time, including key metrics
such as accuracy, latency, and resource utilization [49].
Feedback Loop
Collect and encourage feedback from end-users, domain experts,
and stakeholders to identify possible issues, challenges, and
opportunities for improvement. Incorporating a feedback culture
into iterative model updates and refinements. Additionally ensure
a minimum of a yearly mock recalls or simulation deviations to
test if the AI is still doing what it supposed to do.
Bias Detection and Mitigation
Continuously monitor AI models for biases and unfair outcomes,
leveraging techniques such as fairness-aware algorithms, bias
detection frameworks, and diverse training data. Also to ensure the
AI links relevant criteria and avoid the trouble with early models,
where Wolfs were misidentified or the big disaster of Google’s
launch with Gemini image creator and Amazon recruitment
process fail [50-52].
Compliance Audits
Conduct periodic audits and reviews to ensure that deployed AI
solutions remain compliant with regulatory requirements, industry
standards, and ethical guidelines.
Create with AI as aid towards a culture of always doing it well,
also when no one is looking. Because patient safety must be of
utmost importance at all times.
Training and Education
Provide ongoing training and education to employees about AI
technologies, best practices, and ethical considerations. Foster a
culture of transparency, accountability, and responsible AI usage
within the organization.
Conclusion
Please go ahead and use my plan as a comprehensive framework
for a start of AI implementation.
I’ve covered some critical key stages starting from idea to
deployment and monitoring.
Emphasizing on the importance of key areas like bias mitigation,
and continuous improvement, which are crucial considerations in
highly regulated industries. The successful implementation of
AI within pharmaceutical and/or logistical organisations hinges
on a strategic approach that balances innovation with regulatory
compliance. By following the outlined steps and learning from
real-life cases, organizations can harness the power of AI to
enhance operational efficiency, improve product distribution and
improve the ultimate goal: Patient Safety
References
1. Yang Judy X, Li Lily D, Rasul Mohammad (2021) Warehouse
Management Models Using Artificial Intelligence Technology
with Application at Receiving Stage - A Review. International
Journal of Machine Learning and Computing 11: 242-249.
2. Ashley Welch (2023) Artificial intelligence is helping
revolutionize healthcare as we know it. Johnson & Johnson
https://www.jnj.com/innovation/artificial-intelligence-inhealthcare.
3. Iakovlev Igor, Kremleva Elmira, Guzanov Roman (2023)
Enterprise automation using artificial intelligence: Methods,
technologies and prospects. E3S Web of Conferences 460.
10.1051/e3sconf/202346004007.
4. Good manufacturing practice. European Medicines Agency
https://www.ema.europa.eu/en/human-regulatory-overview/
research-development/compliance-research-development/
good-manufacturing-practice.
5. Good distribution practice. European Medicines Agency
https://www.ema.europa.eu/en/human-regulatory-overview/
post-authorisation/compliance-post-authorisation/gooddistribution-practice.
6. ISO/IEC 42001:2023 Information technology-Artificial
Intelligence-Management system. ISO https://www.iso.org/
standard/81230.html.
7. ISO/IEC 23894:2023 Information technology-Artificial
intelligence-Guidance on risk management https://www.iso.
org/standard/77304.html.
8. ISO/IEC 25059:2023 Software engineering-Systems and
software Quality Requirements and Evaluation (SQuaRE)-
Quality model for AI systems https://www.iso.org/
standard/80655.html.
9. Amanda L Joseph, Helen M, Yuri Q (2023) Exploring Patient
Journey Mapping and the Learning Health System: Scoping
Review. JMIR Hum Factors 10: e43966.
10. Doris Wojcicka (2015) 10 Most interesting examples of
Customer Journey Maps. UXeria https://uxeria.com/en/10-
most-interesting-examples-of-customer-journey-maps/.
11. Chui KT, Lytras MD, Visvizi, A, Sarirete A (2021) An
overview of artificial intelligence and big data analytics for
smart healthcare: requirements, applications, and challenges.
12. Khan ZF, Alotaibi SR (2020) Applications of artificial
intelligence and big data analytics in m-health: a healthcare
system perspective. Journal of healthcare engineering 1-15.
13. Harmon SA, Sanford TH, Xu S, Turkbey EB, Roth H, et al.
(2020) Artificial intelligence for the detection of COVID-19
pneumonia on chest CT using multinational datasets. Nature
communications 11: 4080.
14. Bera K, Schalper KA, Rimm DL, Velcheti V, Madabhushi A
(2019) Artificial intelligence in digital pathology-new tools
for diagnosis and precision oncology. Nature reviews Clinical
oncology 16: 703-715.
15. Kuchana M, Srivastava A, Das R, Mathew J, Mishra A,
et al. (2021) AI aiding in diagnosing, tracking recovery
of COVID-19 using deep learning on Chest CT scans.
Multimedia tools and applications 80: 9161-9175.
16. Ferdosi BJ, Sakib MA, Islam MS, Dhar J (2021) Identifying
counterfeit medicine in bangladesh using deep learning. In
International Conference on Human-Centered Intelligent
Systems 46-55.
17. Moglianetti M, Pedone D, Morerio P, Scarsi A, Donati P,
et al. (2022) Nanocatalyst-Enabled Physically Unclonable
Functions as Smart Anticounterfeiting Tags with AI-Aided
Smartphone Authentication.
18. Bouabdallaoui Y, Lafhaj Z, Yim P, Ducoulombier L, Bennadji
B (2021) Predictive maintenance in building facilities: A
machine learning-based approach. Sensors 21: 1044.
19. Dalzochio J, Kunst R, Pignaton E, Binotto A, Sanyal S, et
al. (2020) Machine learning and reasoning for predictive
maintenance in Industry 4.0: Current status and challenges.
Computers in Industry 123: 103298.
20. Farchi F, Farchi C, Touzi B, Mabrouki C (2023) A
comparative study on AI-based algorithms for cost prediction
in pharmaceutical transport logistics. Acadlore Trans Mach
Learn 2: 129-141.
21. Hasani N, Morris MA, Rahmim A, Summers RM, Jones E,
et al. (2022) Trustworthy artificial intelligence in medical
imaging. PET clinics 17: 1-12.
22. Baryannis G, Validi S, Dani S, Antoniou G (2019) Supply
chain risk management and artificial intelligence: state of
the art and future research directions. International Journal
of Production Research 57: 2179-2202.
23. Alevizos V, Georgousis I, Simasiku A, Karypidou S, Messinis
A (2023) Evaluating the Inclusiveness of Artificial Intelligence
Software in Enhancing Project Management Efficiency--A
Review. arXiv preprint arXiv:2311.11159.
24. Tomasello L (2019) Neural machine translation and
artificial intelligence: what is left for the human translator?
Universita degli Studi di Padova https://core.ac.uk/download/
pdf/189853844.pdf.
25. Hacker P (2021) A legal framework for AI training data -
from first principles to the Artificial Intelligence Act. Law
Innovation and Technology 13: 257-301.
26. Whang SE, Roh Y, Song H, Lee JG (2023) Data collection
and quality challenges in deep learning: A data-centric ai
perspective. The VLDB Journal 32: 791-813.
27. Amann J, Blasimme A, Vayena E, Frey D, Madai VI, et al.
(2020) Explainability for artificial intelligence in healthcare:
a multidisciplinary perspective. BMC medical informatics
and decision making 20: 1-9.
28. Bodimani M (2024) Assessing the Impact of Transparent AI
Systems in Enhancing User Trust and Privacy. Journal of
Science & Technology 5: 50-67.
29. ISO/IEC 27013:20219(en), Information security,
cybersecurity and privacy protection - Guidance on the
integrated implementation of ISO/IEC 27001 and ISO/
IEC 20000-1.
30. ISO/IEC 27005:2022(en), Information security, cybersecurity
and privacy protection -Guidance on managing information
security risks. ISO https://www.iso.org/obp/ui/#iso:std:isoiec:27005:en.
31. ISO/IEC 27001:2022(en), Information security, cybersecurity
and privacy protection -Information security management
systems - Requirements IS0.
32. Niazi SK (2023) The coming of age of ai/ml in drug discovery,
development, clinical testing, and manufacturing: The FDA
perspectives. Drug Design Development and Therapy 2691-
2725.
33. Farchi F, Farchi C, Touzi B, Mabrouki C (2023) A Comparative
Study on AI-Based Algorithms for Cost Prediction in
Pharmaceutical Transport Logistics. Acadlore Trans Mach
Learn 2: 129-141.
34. Norori N, Hu Q, Aellen FM, Faraci FD, Tzovara A (2021)
Addressing bias in big data and AI for health care: A call for
open science. Patterns 2.
35. Challen R, Denny J, Pitt M, Gompels L, Edwards T, et al.
(2019) Artificial intelligence, bias and clinical safety. BMJ
quality & safety 28: 231-237.
36. Noseworthy PA, Attia ZI, Brewer LC, Hayes SN, Yao X, et
al. (2020) Assessing and mitigating bias in medical artificial
intelligence: the effects of race and ethnicity on a deep
learning model for ECG analysis.
37. Price WN, Cohen IG (2019) Privacy in the age of medical
big data. Nature medicine 25: 37-43.
38. Vollmer S, Mateen BA, Bohner G, Kiraly FJ, Ghani R, et al.
(2020) Machine learning and artificial intelligence research
for patient benefit: 20 critical questions on transparency,
replicability, ethics, and effectiveness. BMJ 368.
39. Tangi L, van Noordt C, Rodriguez Muller AP (2023) The
challenges of AI implementation in the public sector.
40. (2023) IEEE Introduces New Program for Free Access to AI
Ethics and Governance Standards. IEEE SA https://standards.
ieee.org/news/get-program-ai-ethics/.
41. Khalid N, Qayyum A, Bilal M, Al-Fuqaha A, Qadir J (2023)
Privacy-preserving artificial intelligence in healthcare:
Techniques and applications. Computers in Biology and
Medicine 106848.
42. Eckhardt S, Knaeble M, Bucher A, Staehelin D, Dolata M,
et al. (2023) Garbage In, Garbage Out: Mitigating Human
Biases in Data Entry by Means of Artificial Intelligence.
43. Hanson B, Stall S, Cutcher-Gershenfeld J, Vrouwenvelder K,
Wirz C, et al. (2023) Garbage in, garbage out: mitigating risks
and maximizing benefits of AI in research. Nature 623: 28-31.
44. Teno JM (2023) Garbage in, garbage out-words of caution on
big data and machine learning in medical practice. In JAMA
Health Forum 4: e230397-e230397.
45. Hyde SJ, Bachura E, Harrison JS (2023) Garbage in, garbage
out: A theory-driven approach to improve data handling in
supervised machine learning. In Methods to Improve Our
Field 101-132.
46. Natasha Sharma (2023) Understanding and Applying F1
Score: AI Evaluation Essentials with Hands-On Coding Example.
47. (2002) General Principles of Software Validation. FDA
https://www.fda.gov/regulatory-information/search-fdaguidance-documents/general-principles-software-validation.
48. Rath KC, Khang A, Rath SK, Satapathy N, Satapathy SK, et
al. (2024) Artificial Intelligence (AI)-Enabled Technology
in Medicine-Advancing Holistic Healthcare Monitoring and
Control Systems.
49. Dave Pearson (2021) AI biopsy dilemma: Wolf or husky,
equity or bias? Healthexec.com https://healthexec.com/topics/
precision-medicine/ai-biopsy-dilemma-wolf-or-husky-equityor-bias.
50. Mark Humphery-Jenner (2024) Google’s $90 Billion AI
disaster. substack.com https://financemark.substack.com/p/
googles-90-billion-ai-disaster.
51. Dan Milmo, Alex Hern (2024) We definitely messed up’: why
did Google AI tool make offensive historical images? The
Guardian
52. Jeffrey Dastin (2018) Insight - Amazon scraps secret AI
recruiting tool that showed bias against women | Reuters
https://www.reuters.com/article/idUSKCN1MK0AG/.
View PDF
