AI Regulation Is Coming to Life Sciences: Three Steps to Take Now

Cognizant 20-20 Insights
September 2020
AI Regulation Is Coming to Life
Sciences: Three Steps to Take Now
To maximize the value of artificial intelligence and machine learning for patients,
healthcare providers together with life sciences enterprises must gear up to
meet the continually evolving regulatory landscape.
Executive Summary
Artificial intelligence and machine learning (AI/ML)
have enormous potential to improve patient outcomes
and drive new revenue streams for life sciences (LS)
companies. Whether understanding the complexities
of the human genome, assessing the effectiveness of
treatments or modeling the spread of disease, AI/ML could
help speed new treatments to market, make healthcare
more efficient and effective, and better track and control
pandemics. Insights from AI/ML solutions could be
monetized as chatbots that can help patients manage
their health, assist doctors as they diagnose and treat
conditions, and guide robotic surgery.
Predictive algorithms, as well as AI/ML, can also improve
and speed regulatory compliance by mining data from
clinical trials and caregivers and surfacing potential issues
before safety reports are submitted to regulators.
But the use of complex algorithms that improve
themselves over time raises complex new regulatory
issues. LS companies that understand, prepare for

2 / AI Regulation Is Coming to Life Sciences: Three Steps to Take Now
and help shape those regulations will be best
positioned to use AI/ML to drive the maximum
benefit for patients, healthcare providers and their
shareholders. Read on to learn how LS companies
should adapt their governance models, data
management, and reporting, tracking and validation
processes to meet the requirements of upcoming
AI/ML regulations.

The regulatory challenge
Because AI/ML involves such new and rapidly
changing computing technologies, and vast
amounts of training data, it poses significant new
regulatory challenges such as:
❙ Determining the accountability for providers of
AI/ML-based solutions and assigning liability
for harm caused by the “black box” of an AI/ML
process.
❙ Assuring the quality and safety of products or
therapies developed using AI/ML.
❙ Guaranteeing that the data and/or algorithms
used in AI/ML solutions are not biased against
underserved populations.
❙ Ensuring that the AI/ML solutions developed are
sustainable and environmentally friendly, and
that the recommendations are not in conflict with
societal priorities such as social justice.
❙ Protecting the privacy of patients and their data.
Existing patient privacy rules do not, for example,
protect patient data when it is shared and used by
technical or consumer marketing organizations
rather than healthcare providers.
In response, regulators are working to create adaptive,
real-time regulations, aided by next-gen-technology-
enabled interventions, pre-certification programs
and data-driven decision-making. The role of AI/ML
Current regulations governing AI/ML in life sciences
Figure 1
Regulator Program and Link Program Details Program Type Status
FDA Developing a software
pre-certification
program: A working
model (2018–2019).
Aims to provide tailored, pragmatic and least burdensome
oversight to ensure organizations have a culture of
quality and excellence. It includes streamlined pre-market
review and post-market verification for continued safety,
effectiveness and performance of software-as-a-medical-
device (SaMD).
Evolving
guidance
documents.
Accepting
applicants.
FDA Proposed regulatory
framework for
modifications of AI/ML-
based SaMD
(April 2019).
Proposed framework to regulate iterative algorithms within
AI/ML SaMDs to ensure patient safety.
Assures ongoing algorithm changes are implemented
according to prespecified performance objectives, follow
defined protocols and use a validation process committed
to improving the performance, safety and effectiveness of
AI/ML software, including performance monitoring.
Discussion
paper.
Currently
accepting
feedback.
UK Department
of Health &
Social Care
Code of conduct for
data-driven health and
care technology
(July 2019).
Provides 10 principles in a code of conduct to enable the
development and adoption of safe, ethical and effective
data-driven healthcare technologies.
Guidance
document/
code of
conduct.
Active.
National
Institute for
Health and
Care Excellence
(NICE),
Department of
Health, UK
Evidence standards
framework for digital
health technologies.
Describes the types and levels of evidence needed to show
the effectiveness and expected economic impact of a
digital health technology.
Evidence
standards
and
regulatory
sandbox.
Active.

in life sciences is so new that only a few proposed
regulations are industry-specific, with the rest
addressing the cross-industry ethical, governance
and social impacts of AI/ML.
Among the evolving regulatory frameworks are a
voluntary pilot software pre-certification program
developed by the U.S. Food and Drug Administration
(FDA) for makers of software-based medical
devices (see Figure 1, page 3), and an April 2019
discussion paper from the FDA seeking input
from pharmaceutical companies on a proposed
framework for regulating such devices.1
LS companies that work with regulators to help
create these regulations can gain a competitive
advantage by:
❙ Helping to shape the rules to provide the
maximum value to patients, healthcare providers
and the rest of the LS ecosystem.
❙ Understanding how AI/ML can add ongoing value
to both their clinical and commercial operations.
❙ Identifying the skills they need to leverage AI/ML.
❙ Leveraging partners skilled in AI/ML to
complement their own LS skills.
While the exact nature of forthcoming AI/ML
regulations is still evolving, we have identified three
areas – governance, data management and
reporting/tracking/monitoring – that LS companies
will likely need to address to stay ahead of the curve
(see Figure 2).
Governance
model
❙ Apply existing quality, compliance
and documentation processes to
AI/ML-based medical solutions.
❙ Involve staﬀ with regulatory
experience as well as legal, risk, IT,
security and analytics.
❙ Provide regulators with real-time
updates; iteratively test AI/ML solutions
for faster review and feedback.
Data
management
❙ Create adaptable processes for collecting,
organizing, maintaining, using,
preserving and managing the security
of data used in AI/Ml solutions.
❙ Use data testing/analysis models
and/or tools whose results can be
submitted as evidence to support
claims of safety assurance and provide
audit trails to verify testing and results.
Reporting, tracking
& validation
❙ Develop real-time reports, metrics
and KPIs that evolve to meet changing
AI/ML technological capabilities, as well
as regulatory and market needs.
❙ Assure mature change and risk
management, safety and audit, and
quality assurance processes.
The multiple dimensions of regulatory action
Figure 2

GOVERNANCE MODELS
LS companies have long experience assuring their
development, testing and manufacturing processes
comply with complex regulations in multiple
jurisdictions. But emerging regulatory, risk and
ethical considerations associated with AI/ML require
new governance models, implemented by staff with
the proper skills who are empowered to resolve
issues that could affect patient safety and privacy.
Compared to the traditional development process
for LS products, or the diagnostic process followed
by a provider, it is often difficult or impossible to
determine how an AI solution arrived at a conclusion.
Governance models must be adapted to account
for varying levels of risk tolerance and limits on the
explainability and interpretability of AI/ML solutions.
In addition, AI/ML governance frameworks must
allow for modifications in how the AI/ML
solutions are measured as the algorithms, and
their uses, change.
LS companies will need to demonstrate they already
have or are acquiring the appropriate AI/ML tools,
automation and test environments and have a
centralized and transparent process to implement
AI/ML solutions. Regulators will also expect evidence,
perhaps in the form of third-party audits, of a robust
and effective communication framework, reliable
feedback and learning mechanisms, and multiple
avenues of customer outreach. This information flow
is essential to ensuring AI/ML solutions not only meet
the needs of patients and caregivers but produce
ethical and unbiased recommendations.
Solution developers will need to identify specific KPIs
and targets for AI/ML solutions, explain outliers or
anomalies in the data used in those solutions, and
summarize actions planned or taken to address
those outliers or anomalies. Their governance
frameworks must also include mechanisms
for changing the measurement criteria as the
algorithms themselves evolve.
LS companies must also create and document
appropriate decision-making evaluation frameworks
for AI/ML solutions, backed by data analytics, as
well as data governance models to ensure that the
data used in AI/ML models is clean and reliable.
Evaluation criteria should include the level of risk
to patients, the criticality of the condition AI/ML is
being used to diagnose or treat, the type of changes
that have been made to the solution and the actual
impacts of the AI/ML treatment.
Other requirements include protocols for premarket
assurance, including change control mechanisms
and documentation, as well as for determining
acceptable risks and acting on those determinations.
As regulators introduce premarket reviews of AI/ML
solutions and changes to them, LS players can “self-
regulate” with agile methods to manage risk and
prove the safety and effectiveness of their solutions.
This can be done, for example, by using controlled
environments to iteratively test their products/
services/models.
Solution developers will need to identify specific KPIs and targets
for AI/ML solutions, explain outliers or anomalies in the data used in
those solutions, and summarize actions planned or taken to address
those outliers or anomalies.

Proper experience will be critical to compliance
success. The AI/ML governance task force should
include staff with regulatory experience who can
provide useful insights into the quality, format and
other characteristics of the data required to prove
compliance. It should also include C-level executives
and experts in areas ranging from legal and risk to
IT, security and analytics to evaluate and prioritize the
risks of AI/ML solutions.
Depending on the type of solution, development
staff should also include scientists and engineers
who can help shape data generation, collection and
maintenance techniques based on their domain and
business knowledge.
DATA MANAGEMENT
Data is central in developing and refining AI/ML
solutions, both to train the algorithms at the heart of
AI and for use in those models to make potentially
life-or-death decisions. The choice of data, its use
and how it is protected to maintain patient privacy
will be top of mind for regulators.
To manage risk, LS companies will need to create
data or information lifecycle management processes
to collect, organize, maintain, use and preserve
data. Data from clinical trials, pharmacovigilance,
customer feedback or commercial sources should
be gathered using validated or highly reliable
methods with appropriate controls implemented
by qualified experts. Those controls could include
national or international standards or previously
tested internal controls.
Producers of software-as-a-medical-device (or
SaMD; software used to diagnose or treat conditions
that is not part of a physical device) will need to
demonstrate how they can leverage usage data
to understand how their products are being used,
identify opportunities for improvements, and
respond proactively to safety or usability concerns.
Among the privacy and personalization
requirements we expect for data used in ML/AI are
the following:
❙ Any data used for training is accurate and has not
been tampered with.
❙ If a training dataset is created in real time, the data
within it should be anonymized in real time.
❙ Assurance that hidden biases or other defects
in the algorithms do not result in the wrong
diagnoses or treatment recommendations for the
intended patient.
❙ Adaptive systems should ensure that new data
and algorithms don’t contain malware or other
threats that could jeopardize patient data, the
security of the AI/ML system or the functioning
of other platforms with which the system
shares data.
Data is central in developing
and refining AI/ML solutions,
both to train the algorithms at
the heart of AI and for use in
those models to make potentially
life-or-death decisions. The
choice of data, its use and how
it is protected to maintain
patient privacy will be top of
mind for regulators.

LS organizations should also utilize data testing/
analysis models and/or tools whose results can
be submitted as evidence to support claims of
safety assurance. Capabilities to look for include
robust document management and reporting,
audit trails, content protection, integration with
other information management and workflow
systems, and the ability to present safety arguments
through diagrams. We also recommend that proven
processes to demonstrate the safety, effectiveness
and performance of medical devices be extended to
AI/ML solutions.
REPORTING, TRACKING AND
VALIDATION
LS companies will need to prove that their AI/ML
solutions give humans a substantial or even final say
in patient care, preserve privacy, provide “effective
redress” if they cause harm, and are auditable and
understandable. They may also, depending on the
political jurisdiction, need to prove their solutions do
not harm society or threaten the democratic process,
freedom of expression or freedom of identity.
Meeting these needs will require processes such
as change and risk management, safety and
audit, and quality assurance are well-established
and maintained and can be leveraged for AI/ML
solutions. Given the dynamic and unpredictable
nature of AI/ML algorithms, companies will need
to show they are investing in ensuring multistage
verification and validation of AI/ML solutions. This
could include engaging users in the verification
process at various stages, measuring the quality of
the output and ensuring the traceability of results.
Devices that rely on AI/ML must be validated using
analytical and clinical data, and follow good machine
learning practices (GMLP). These include the
relevance of available data to the clinical problem
and practice, that data is acquired in a consistent,
clinically relevant and generalizable manner, and
appropriate separation is maintained between
training, tuning and testing of datasets.
Explainability, i.e., documenting the logic, decision
methodology and data sources that influence the
output of the AI/ML solution, is not an easy task as
the algorithm(s) used are often highly complex. But
such explainability helps assure data privacy and
ethics as well as regulatory compliance. Explainable
AI (XAI), itself a form of ML, can help describe how
and why ML models made their predictions. For
relatively straightforward models, decision trees or
knowledge graphs used to make decisions can help
provide such understanding. For more complex
forms of ML, XAI uses methods such as highlighting
which data played the greatest role in a decision.
Defining ML uncertainty (describing questions the
system cannot answer) can help determine when to
trust it.
LS companies will need to prove that their AI/ML solutions give
humans a substantial or even final say in patient care, preserve
privacy, provide “effective redress” if they cause harm, and are
auditable and understandable.

LS companies will also need to continually adapt
their reporting to meet evolving AI capabilities
and regulatory requirements. They will need to
tailor the frequency of their reports, and the type
of information in them, to the risk categorization of
the device, the number and types of modifications
made to it, and the maturity and reliability of the
algorithms the solution uses to make decisions. For
example, if an AI solution initially intended to assure
patients receive the proper treatment by tracking
their lifestyle habits evolves to the point it can also be
used to diagnose disease, the KPIs used to measure
its effectiveness must change as the potential risk
of failures rises. In such a case, reports on the safety
and effectiveness of the device might be done
semiannually or quarterly rather than annually.
This continually evolving reporting will be provided
by digital dashboards displaying data-driven and
quality-focused KPIs measuring the explainability
of the solution, how well it assures data quality and
privacy, and its risk level.
SaMD manufacturers will need to create reporting
frameworks, and choose reporting tools, that reflect
the SaMD pre-specifications (SPS) and the algorithm
change protocol (ACP) documents, which define the
changes a manufacturer plans in a medical device
and how it will implement those changes.2
Finally, LS companies will need to develop
frameworks to track and measure changes in the
intended use of their AI/Ml solutions, the algorithms
used in them, the data used to train them or the
output of the solution. For example, the creator of a
mobile app designed to monitor patient’s exercise
or eating habits for disease risk would need to
validate predictions made by the app if the intended
purpose evolved from monitoring to diagnosing by
comparing against real-world data from diagnosed
patients with similar lifestyles.
LS companies will need to develop frameworks to track and
measure changes in the intended use of their AI/ML solutions, the
algorithms used in them, the data used to train them or the output
of the solution.

Looking ahead: Help set the agenda
Collaboration and convergence are the keys to
success for the future of AI/ML regulations. In
order to gain advantage, the industry needs to
proactively increase its engagement with regulators
who will be converging at both national and global
levels, adopting approaches like co-regulation and
increased collaboration with patients, healthcare
professionals, industry and payers. Even before the
use of AI/ML becomes widespread in the LS field,
industry players can position themselves
to best meet emerging AI/ML regulations by:
❙ Preparing to “self-regulate” by proactively
creating pathways that use agile and iterative
methods to manage risk and ensure the safety
and effectiveness of AI/ML solutions. This can be
done in partnership with experts or in controlled
environments to iteratively test products, services
and models.
❙ Hiring and/or training staff who can work closely
with regulators in understanding and helping to
develop AI/ML regulations.
❙ Establishing, maintaining and refining quality
systems that can validate their analytics and
clinical outcomes, and adopting GMLP for data
management, feature extraction, training and
evaluation.
The power of AI/ML creates new challenges for
regulators and the LS companies that must adhere
to the emerging guidelines. Those who are “present
at the creation” of these regulations can do the most
to influence them and will be best prepared to meet
them efficiently and effectively.
Endnotes
1 Guo, Elizabeth, et al.,“FDA Outlines Proposed Framework for Regulating Artificial Intelligence Software,” Covington Digital
Health, April 10, 2019; www.covingtondigitalhealth.com/2019/04/fda-outlines-proposed-framework-for-regulating-artificial-
intelligence-software/.
2 Singh,Jagmeet,“Preparing for the Regulatory Challenges Wrought by Software as a Medical Device”; www.cognizant.com/
whitepapers/preparing-for-the-regulatory-challenges-wrought-by-software-as-a-medical-device-codex2513.pdf.=
Collaboration and convergence are the keys to success for the future
of AI/ML regulations. In order to gain advantage, the industry needs
to proactively increase its engagement with regulators who will be
converging at both national and global levels, adopting approaches
like co-regulation and increased collaboration with patients,
healthcare professionals, industry and payers.

About the authors
Mini Nair
Senior Manager, Cognizant Consulting India Practice
Mini Nair is a Senior Manager within the Cognizant Consulting India Practice. She has about 18 years
of LS consulting experience managing various strategic and digital transformation projects through
her domain acumen and program management competencies. Prior to Cognizant, Mini worked in the
brand management space, driving pharmaceutical marketing strategies. She holds a master’s degree in
pharmaceutical management from SIESCOMS, Mumbai. Mini can be reached at Mini.Nair@cognizant.com |
www.linkedin.com/in/mini2008/.
Rohan Desai
Senior Manager, Cognizant Consulting India Practice
Rohan Desai is a Senior Manager within the Cognizant Consulting India Practice. He has 12 years of
consulting, program management and service delivery experience in life sciences, in the areas of digital
strategy, commercial operations, supply chain and IT operations transformation. He received a master’s
degree in business administration at the Indian Institute of Management, Indore. Rohan can be reached at
Rohan.Desai@cognizant.com | www.linkedin.com/in/rohancdesai/.
Brian Williams
Chief Digital Officer for Life Sciences, Cognizant
Brian Williams is Cognizant’s Chief Digital Officer for Life Sciences and is responsible for designing digitally
enabled solutions to facilitate care access and delivery. He is also the Global Life Sciences Consulting Leader
for Life Sciences, inspiring teams in North America, Europe and Asia that provide management consulting
services to the leading biopharmaceutical and medical device companies. Brian earned his undergraduate
degree from Indiana University. He can be reached at Brians.Williams@cognizant.com.
Shirali Desai
Life Sciences Senior Consultant, Cognizant Consulting India Practice
Shirali Desai is a Life Sciences Senior Consultant within the Cognizant Consulting India Practice. She has
over six years of consulting experience in the areas of digital strategy, process optimization, ERP integration
and digital health. Shirali received a master’s degree in business administration at the Indian Institute of
Management, Udaipur. She can be reached at ShiraliJitesh.Desai@cognizant.com | www.linkedin.com/in/
shirali-desai-28750343/.
Lakshay Bhambri
Life Sciences Consultant, Cognizant Consulting India Practice
Lakshay Bhambri is a Life Sciences Consultant within the Cognizant Consulting India Practice. He has
five years of consulting experience in the areas of digital analysis, big data, business process improvement
and commercial life sciences. Lakshay received a master’s degree in business administration at the Institute
of Management Technology, Ghaziabad. He can be reached at Lakshay.Bhambri@cognizant.com |
http://linkedin.com/in/lakshay-bhambri-ba588993.

© Copyright 2020, Cognizant. All rights reserved. No part of this document may be reproduced, stored in a retrieval system, transmitted in any form or by any means, electronic, mechanical,
photocopying, recording, or otherwise, without the express written permission from Cognizant. The information contained herein is subject to change without notice. All other trademarks mentioned
herein are the property of their respective owners.
Codex 5980
About Cognizant Life Sciences
Cognizant’s Life Sciences business unit is committed to helping improve millions of lives by partnering with clients to build solutions to healthcare
challenges, continually upgrading the way companies business, setting the pace in clinical development, strengthening the regulatory infrastructure and
increasing competitiveness. Cognizant serves the top 30 global pharmaceutical companies, nine of the top 10 biotech companies and 12 of the top 15
medical device companies. With a large team of dedicated professionals including doctors, pharmacologists, physicians, biomedical engineers, pharmacists,
biostatisticians, medical writers and GxP consultants, the practice provides domain-aligned consulting, IT, business process, software and as-a-service
solutions globally. Visit us at http://cognizant.com/life-sciences.
About Cognizant
Cognizant (Nasdaq-100: CTSH) is one of the world’s leading professional services companies, transforming clients’ business, operating and technology
models for the digital era. Our unique industry-based, consultative approach helps clients envision, build and run more innovative and efficient businesses.
Headquartered in the U.S., Cognizant is ranked 194 on the Fortune 500 and is consistently listed among the most admired companies in the world. Learn
how Cognizant helps clients lead with digital at www.cognizant.com or follow us @Cognizant.
World Headquarters
500 Frank W. Burr Blvd.
Teaneck, NJ 07666 USA
Phone: +1 201 801 0233
Fax: +1 201 801 0243
Toll Free: +1 888 937 3277
European Headquarters
1 Kingdom Street
Paddington Central
London W2 6BD England
Phone: +44 (0) 20 7297 7600
Fax: +44 (0) 20 7121 0102
India Operations Headquarters
#5/535 Old Mahabalipuram Road
Okkiyam Pettai, Thoraipakkam
Chennai, 600 096 India
Phone: +91 (0) 44 4209 6000
Fax: +91 (0) 44 4209 6060
APAC Headquarters
1 Changi Business Park Crescent,
Plaza 8@CBP # 07-04/05/06,
Tower A, Singapore 486025
Phone: + 65 6812 4051
Fax: + 65 6324 4051

AI Regulation Is Coming to Life Sciences: Three Steps to Take Now

More Related Content

What's hot (20)

Similar to AI Regulation Is Coming to Life Sciences: Three Steps to Take Now (20)

More from Cognizant (20)

AI Regulation Is Coming to Life Sciences: Three Steps to Take Now