In collaboration

with ZS

Patient-First Health with

Generative AI: Reshaping
the Care Experience
WHITE PAPER
JANUARY 2024
Images: Getty Images, Unsplash

Contents
Foreword 3

Executive summary 4

1 Three broad categories of generative AI use cases  5

for patient engagement

2 How and why generative AI will revolutionize the way patients  6

engage with their health

3 Three barriers to safe, effective patient-facing generative AI 10

4 To make generative AI for patients a reality, healthcare stakeholders  12

must take these four steps

Conclusion  14

Contributors 15

Endnotes 16

Disclaimer
This document is published by the
World Economic Forum as a contribution
to a project, insight area or interaction.
The findings, interpretations and
conclusions expressed herein are a result
of a collaborative process facilitated and
endorsed by the World Economic Forum
but whose results do not necessarily
represent the views of the World Economic
Forum, nor the entirety of its Members,
Partners or other stakeholders.

© 2024 World Economic Forum. All rights

reserved. No part of this publication may
be reproduced or transmitted in any form
or by any means, including photocopying
and recording, or by any information
storage and retrieval system.

Patient-First Health with Generative AI: Reshaping the Care Experience 2

January 2024 Patient-First Health with Generative AI:
Reshaping the Care Experience

Foreword
Shyam Bishen
Head, Centre for Health Pratap Khedkar
and Healthcare, Chief Executive Officer, ZS
World Economic Forum

In 2022-23, the World Economic Forum and ZS Given that somewhere around 80% of all medical
collaborated to create a report outlining the most data is unstructured, these models hold strong
promising use cases for artificial intelligence (AI) promise to address global healthcare challenges
in healthcare, the stiffest barriers to adoption, a for which “traditional” predictive AI techniques are
framework for building trust and adoption for AI ill-suited.
solutions, and a call to action for stakeholders across
the healthcare ecosystem and beyond. Shortly This white paper looks specifically at generative
before the paper was published, OpenAI released AI’s promise to help empower patients. It draws on
ChatGPT, unleashing generative AI into the public interviews with digital and innovation leaders across
discourse and opening fresh new possibilities for using healthcare, technology and other sectors, who
AI to tackle healthcare’s most intractable problems. have shared their views on the most promising use
cases for patient engagement, their biggest barriers
For example, large language models (LLMs) trained to adoption and key enablers for using this new
on vast troves of unstructured data can create modality to improve global health outcomes.
original, human-like text based on minimal prompts.

Patient-First Health with Generative AI: Reshaping the Care Experience 3

Executive summary
Generative AI offers a powerful new
modality to address global healthcare
challenges, but it’s not a panacea.

The COVID-19 pandemic exacerbated numerous Many healthcare companies and other stakeholders
existing global healthcare problem areas – among have struggled to keep pace with changing patient
them a severe shortage of healthcare workers,1 expectations for how they engage with the health
widening health disparities and existential strain system, as described in the 2024 ZS Future of
on health system finances. The World Health Health Report. This paper focuses on the promise
Organization estimates the global shortage of of using generative AI to advance direct patient
healthcare workers today at 15 million (including engagement, a use case with high potential to
10 million doctors),2 a figure projected to decline alleviate health system burden, reduce healthcare
to 10 million by 2030. Around 95% of the shortfall provider burnout and improve patient experiences
is in low- and middle-income countries (LMICs).3 and outcomes.

This poses a significant threat to global health, as

there simply are not enough healthcare providers to Key takeaways include:
care for the volume of patients in need – let alone
to provide the type of preventive care required
Generative AI holds significant promise to
to extend the health span for global populations.
aid healthcare consumers across their full
Powered by rapid advances in machine- and deep-
health journeys – from dispensing reliable
learning techniques, artificial intelligence has been
health information across regions and
hailed as the answer to a broad array of endemic
cultural contexts to ensuring patients get
healthcare challenges.4 However, significant barriers
the appropriate level of care and helping
remain, as examined in the 2023 joint Forum and
them to manage their conditions.
ZS report Scaling Smart Solutions with AI in Health:
Unlocking Impact on High-Potential Use Cases.
The biggest barriers to the adoption of patient-
While generative artificial intelligence (AI) is not a facing generative AI solutions for healthcare are
healthcare panacea, it does offer stakeholders mistrust among doctors and the public, holes
a powerful new modality to address healthcare in the data foundation and scalability in low-
challenges that are difficult for existing predictive resource environments.
AI techniques to address. Yet, predictive AI’s
limitations are not the only factor spurring the
development and use of generative AI in healthcare. Encouraging generative AI adoption in
The consumerization of health – long under way healthcare depends on instilling models with
but hastened by the pandemic – is a powerful and empathy and domain-specific knowledge,
growing force catalysing healthcare transformation. mitigating bias by connecting the data
From the public’s growing reliance on the internet ecosystem and continuously fine tuning models,
to gather health information to US healthcare keeping humans in the loop and developing
pricing transparency laws and policies across more cost-effective ways to train and run multi-
Europe to make healthcare more accessible, modal foundation models.
patients everywhere demand more convenient and
accessible healthcare experiences.

Patient-First Health with Generative AI: Reshaping the Care Experience 4

 1 Three broad categories of
generative AI use cases
for patient engagement
Generative AI can be applied in diverse
ways in healthcare to empower patients
and alleviate health system burden.

Generative After speaking with leaders responsible for vetting help create a more robust data foundation by
AI can augment and implementing generative artificial intelligence automating data management and extracting
providers by (AI) solutions across healthcare and technology, insights from unstructured data, a task that was
offering more three themes emerged. One, using AI and other extremely time-consuming, cumbersome and
emerging technologies to create value sits at inefficient using earlier AI techniques.
intuitive, human-
the top of the corporate agenda. Two, while the
like conversations healthcare field is rife with experiments, there is – Insights generators: Generative AI enables
with patients and limited agreement on which use cases are ready real-time analysis of both structured datasets
caregivers that for deployment and hold transformational potential. (such as health data) and unstructured datasets
can help inform Three, there is no consensus on how to group (such as physician notes), offering significant
care decisions generative AI use case types. promise as a “co-pilot” to help healthcare
and spur action. providers quickly analyse information and
Based on a review of the literature and qualitative interpret results. These tools also can help
interviews with digital and innovation leaders across accelerate data integration across datasets to
healthcare, technology and other sectors, use help doctors, nurses and health systems gain
cases fall into three general categories: faster, more accurate insights.

– Productivity boosters: With minimal – Action drivers: Traditional chatbots have

prompting, generative AI models can create frequently frustrated both patients and providers
images and text sufficiently human-like as to with incomplete, confusing or irrelevant
automate many of today’s manual tasks – information that fails to convert insights into
everything from transcribing doctor-patient action. Generative AI can augment providers by
visits and drafting emails to summarizing offering more intuitive, human-like conversations
clinical studies and dispensing general health with patients and caregivers that can help
information. Generative AI can also be used to inform care decisions and spur action.

Patient-First Health with Generative AI: Reshaping the Care Experience 5

 2 How and why generative
AI will revolutionize the
way patients engage with
their health
Generative AI is poised to upend how
patients access health information,
receive care and manage their conditions.

In the June 2023 report, Scaling Smart Solutions With the advent of generative AI, the question is:
with AI in Health: Unlocking Impact on High- Given what is known about the connection between
Potential Use Cases, more than 50 leaders from better patient engagement and improved health
across healthcare, government, tech, academia outcomes,5 can this technology be used to safely
and non-governmental organizations (NGOs) plug the dangerous shortage of healthcare workers
were interviewed. Many of them identified patient by communicating directly with patients across a
engagement as an area ripe for transformation range of healthcare interactions?
using AI and other frontier technologies; however,
they also pointed to the challenges, given the Traditionally, healthcare systems engage with
disconnected healthcare ecosystem, liability individual patients mainly at a few discrete points in
concerns inherent to engaging patients without a their health journey: when they see a doctor, when
“human in the loop” and the technical limitations of they start on therapy and when they discontinue
traditional chatbots. treatment. What healthcare systems need are
ways to engage with, influence and gather insights
from patients across their entire health journey.
Generative AI can assist with all three.

FIGURE 1 Patient engagement and generative AI: addressing patient care gaps for better outcomes

Health education Co-pilots for Disease management

assistance patient triage interventions

Lifestyle support, early disease Patient care navigation in the Assist in monitoring, side
education and even prenatal first mile to help understand effect management,
care in LMICs*, with a focus where to direct patients for adherence and keeping
on consumers and caregivers diagnosis or treatment patients on treatment

*Low- and middle-income countries

Patient-First Health with Generative AI: Reshaping the Care Experience 6

  atient engagement use
P million of whom provide unpaid care for those 65
and older in the US11 – reliable, accessible health
case 1: health education information. (Globally, people – mostly women
assistance – log 16.4 billion hours doing unpaid care work for
people of all ages.12)
LLMs can According to a survey of US adults conducted
be trained to before ChatGPT became widely available, nearly However, health education is not just about
be polylingual, all US healthcare consumers (94%) say they use making information available, it is also about
the internet to educate themselves about their making it easy for healthcare consumers to
delivering answers
own health, self-diagnose, understand treatment understand (and therefore actionable). A recent
that are not only
protocols, manage side effects and address their survey estimated that gaps in health literacy cost
correct – the bare mental health.6 Respected entities such as Mayo the US economy as much as $238 billion per
minimum – but Clinic, a US-based health system and research year, particularly in underserved communities and
also sufficiently institution, already offer expansive, consumer- for the 8% of the population with limited English
understandable facing repositories of health information accessible proficiency.13 In the EU, that figure is between
to make them through traditional search engines like Google and €120 billion and €320 billion.14 Health literacy is
actionable. Bing. In the vast majority of cases, these resources even more limited in LMICs, where education
should not be used in lieu of seeing a doctor. If levels are lower, health systems are dramatically
used responsibly, however, platforms like Mayo’s under-resourced and access to health education
could significantly improve incidence of chronic is limited.15
illness like diabetes and cardiovascular disease
through upstream lifestyle improvements and better Large-language models (LLMs) trained exclusively
decision-making about preventive care. However, on health data that exceeds specified quality
studies have consistently found that very few health thresholds would eliminate the shortcomings that
websites meet established thresholds for quality7 plague traditional search. Moreover, they can be
and that healthcare consumers in the US,8 Europe9 trained to be polylingual, delivering answers that
and LMICs10 show limited health literacy. are not only correct – the bare minimum – but
also sufficiently understandable to make them
Generative AI may be able to provide healthcare actionable no matter the user’s language fluency,
consumers and their caregivers – more than 50 education level or cultural context.

BOX 1 Real-world vignette – Ada Health

Ada Health, a Germany-based medical AI updated. Its health assessments are deterministic,
company, has built a digital symptom checker meaning identical inputs will yield identical
based on natural language processing (NLP) outputs. In tests, Ada’s app has proven to be safe
enhanced by generative AI. Unlike when checking and accurate, correctly recognizing inflammatory
systems through a LLM like ChatGPT or Google’s rheumatic diseases more often than physicians
Bard, which are trained on the entire corpus of and besting competing apps in diagnosing a
the World Wide Web, Ada’s knowledge base range of conditions.
is curated, clinically vetted and continuously

Patient-First Health with Generative AI: Reshaping the Care Experience 7

  atient engagement
P away from non-urgent patients to those who require
timely, high-touch care. Platforms like these show
use case 2: co-pilots strong promise but have so far failed to scale due
for patient triage to a lack of open-source models for the 7,000-
plus languages spoken around the world and
The COVID-19 pandemic exacerbated an limits to predictive AI’s ability to understand and
existing global healthcare provider shortage, reflect nuances across varied cultural contexts and
with record levels of US doctors experiencing healthcare ecosystems. With domain-specific and
burnout,18 nurses leaving the field in droves19 culturally diverse training, generative AI can begin
and patient experiences plummeting.20 The addressing many of these gaps, offering a way
public is suffering the inevitable results in the for healthcare providers to safely stratify patients
form of increasing health disparities and rising entering the healthcare system and to route their
healthcare costs, decreasing life expectancy in care more effectively.
many parts of the world and increased burden of
care from ageing and chronic disease.21 In addition to stratifying patients and routing them to
the appropriate level of care, generative AI can help
Early patient-facing triage solutions, such as with the more prosaic – but just as critical – task of
MyndYou and others, have tried to solve care getting patients into the health system in the first
navigation gaps using chatbots as “virtual care place. By automating many aspects of patient intake,
assistants”. These solutions offer a way for AI can alleviate administrative burden and help
strapped healthcare providers to shift resources patients navigate the “first mile” of their care journeys.

BOX 2 Real-world vignette – K Health and Cedars-Sinai

US-based healthcare innovator K Health partnered within the system. The virtual care platform uses
with Cedars-Sinai, a leading academic health AI technology developed by K Health, which
system in the US, to offer patients 24/7 virtual automates most aspects of patient intake and
access to Cedars-Sinai doctors through a data entry, reducing administrative burden and
new mobile app called Cedars-Sinai Connect. allowing providers to focus more on patient care.
Patients can access Cedars-Sinai healthcare Patients can share symptoms, health details and
professionals 24/7 for personalized acute, concerns through the app, which summarizes and
chronic and preventive care, including same-day presents them directly to clinicians, who then use
visits and evaluation for referrals to specialists their clinical expertise to treat each patient.

BOX 3 Real-world vignette – University of Rochester Medical Center

Doctors report spending as long as two hours natural language processing (NLP) techniques.
per day reading and responding to messages Powered by ChatGPT-4 and working in a secure
sent through patient portals, a number that Microsoft Azure environment, URMC has built a
continues to climb as patients increasingly use model to appropriately triage messages to either
these platforms. Leaders at the University of a doctor, nurse or staff member. The tool was
Rochester Medical Center (URMC) in Rochester, still being tested at publication time, but testing
New York, have been building and testing models showed reliability (repeating the same action when
for years to support clinical operations – with presented with an identical input) in the high-90s
limited success, thanks to the messiness of and accuracy (selecting the most appropriate
available data and the limitations of traditional recipient) in the mid-80s.22

Patient-First Health with Generative AI: Reshaping the Care Experience 8

 Patient engagement but they struggle to suggest (let alone execute)
effective interventions to keep at-risk patients
use case 3: disease on course. Generative AI can make predictive
management interventions algorithms significantly more powerful by taking
the output from a classical algorithm and using it
Once a patient has been diagnosed with a condition to create an intervention tailored to the individual
and has begun treatment, his or her health journey patient. This use case is relevant not only to
is only just beginning. Treatment adherence is a provider organizations but also to biopharma, which
key facet of managing illness, especially chronic spends hundreds of millions of dollars on patient-
illness, and yet studies have found that, in the first support call centres, apps and digital “nudges”
year of treatment, as many as 60% of patients to enable better disease management.
suffering from chronic illness either miss doses,
take the wrong dosage or abandon treatment Ongoing management of disease isn’t only about
altogether.23 Inadequate medication adherence keeping patients on their prescribed course of
(defined as taking less than 80% of the prescribed treatment. Generative AI also can shoulder some
dose) is estimated to cost the US healthcare system of the administrative burden inherent to treating
between $100 billion and $290 billion and health patients living with one or more chronic conditions.
systems in Europe €125 billion per year.24 It is also showing promise bringing together multiple
data streams, such as genetics and biometrics, to
Traditional algorithms can help predict when a tailor treatment plans and chart personalized care
patient is likely to miss a dose or drop treatment, paths for individual patients.

BOX 4 Real-world vignette – Mayo Clinic

Rheumatoid arthritis (RA) is a difficult disease generative AI models to quickly and accurately
to manage, with only about 15% of patients predict treatment effectiveness. This innovative
experiencing quick relief from its painful approach is akin to translating a patient’s unique
symptoms. RA’s complexity lies in a patient’s genetic code into a language these models
genetics, which play a big role in both the can understand. The result is a more tailored
disease’s progression and how well treatments and efficient way of finding the right treatment,
work. Mayo Clinic is pioneering the use of a potentially speeding up the process of symptom
patient’s genetic profile and developing multimodal relief for RA patients.

BOX 5 Real-world vignette – Amazon and Hurone AI

High-income countries earn the lion’s share of oncologists. Patients can track symptoms and
investment in cancer prevention, screening and side effects throughout treatment and remission,
diagnosis, yet cancer remains a major health with generative AI recommending personalized
problem in LMICs. Take Nigeria, where four out responses and interventions for doctors to send
of five cancer diagnoses is fatal.25 The problem to patients. The generative AI tools also provide
is one of resources: Rwanda, for instance, has doctors with personalized recommendations
a population of 14 million people but only 15 based on side effects, tolerability and efficacy.
oncologists.26 Amazon and Hurone AI, a US- A beta test saved doctors 75% of time spent
based healthtech start-up, have partnered with on specific care tasks – a potentially lifesaving
the governments of Rwanda, Nigeria and Kenya efficiency boost in regions where resources are
to develop a generative AI-powered patient- so limited.
facing application to communicate with local

BOX 6 Real-world vignette – Microsoft and Epic Systems

The annual volume of messages flowing to patient inquiries. Generative AI, which at
through electronic patient portals doubled least one study has found to provide answers
between 2019 and 2022 to more than 1 billion. displaying higher levels of empathy than those
On average, physicians spend more than 15 drafted by human clinicians, drafts responses
hours per week on administrative tasks, with based on the patient’s inquiry and medical
almost two-thirds of that time taken up by history. A doctor then has the choice to accept
electronic health record documentation – work the responses unedited, to revise as they see
that can be largely done by generative AI. fit or to draft their own response from scratch.
This use case holds significant potential both
Microsoft and Epic Systems, the US’s largest to reduce provider burden and improve patient
provider of electronic health records, are partnering experience, which studies have consistently
to pilot the use of Azure OpenAI to respond linked with improved patient outcomes.

Patient-First Health with Generative AI: Reshaping the Care Experience 9

3 Three barriers to safe,
effective patient-facing
generative AI
Adoption barriers include data issues,
mistrust of outputs and barriers to scale
outside wealthy countries.

In the previous joint briefing paper on AI in healthcare, Using generative AI to engage patients faces many
technology and healthcare leaders identified of the same challenges, plus some new ones. These
numerous barriers slowing the widespread adoption have been ranked from highest to lowest priority.
of AI in healthcare, including holes in the data
foundation, insufficient trust in algorithms, inadequate
technological infrastructure and scalability issues.

Data and trust remain the two foundational barriers for adopting generative
AI in healthcare.
Mustaqhusain Kazi, Global Head, Informatics Strategy and Digital Innovation, Roche

Mistrust  H
 oles in the

data foundation
Interviewees singled out trust as the While many organizations have made progress
most formidable barrier to widespread adoption in building their own data management
of generative AI for patients, with one blanket infrastructure and augmenting it with data
issue rising above all others: inaccuracy. consortia and even federated learning, additional
work is needed to ensure models are trained on
sufficiently robust data to produce outputs that
Inaccuracy: For reasons that are often unclear are fit enough for safe and reliable patient use.
to users, LLMs occasionally generate incorrect or Two data issues predominate:
misleading answers or responses in the absence of
sufficient data. These hallucinations can be hard to
discern, leading to confusion and misinformation. Biased outputs: As with predictive AI techniques,
Since the risk of faulty health information is so high, generative AI outputs reflect underlying bias in the
it is paramount that providers integrate generative data on which they’ve been trained. This creates
AI into workflows such that AI augments, rather than intense urgency to implement governance that
automates, clinical decision-making. In addition, enhances transparency across these models’ entire
careful prompt engineering by developers in initial life cycle – especially for those models, such as
releases, and continued refinement on new and those from OpenAI, Google and others, that have
domain-specific datasets for future releases, is been trained on the broad corpus of the internet.
needed to ensure LLMs avoid misinformation and can
explain why a model responded in a specific way. Patient privacy: Patient health data is
incredibly sensitive, so organizations must build
Companies like Perplexity AI are seeking to increase security, privacy and ongoing data management
trust in outputs through greater transparency by protection into governance models. Organizations
building in functionality that explains how a model like Johnson & Johnson and Syntegra use
arrived at a given output. This can help users generative AI to create “synthetic data”, allowing
evaluate the degree of trust they should place in it. them to use the technology while bypassing
security and privacy issues altogether.

Patient-First Health with Generative AI: Reshaping the Care Experience 10

 S
 calability across regions and contexts

As with many novel technologies, there is a risk that generative AI exacerbates existing disparities.
Multilateral, cross-border partnerships will be key to accelerating generative AI’s impact outside high-
income countries. Two barriers stand out:

Language: A dearth of high-quality training data Resources: Generative AI models must be

in less common languages has limited the impact trained on large volumes of data, which can be
of text-based predictive AI techniques based on expensive to collect and store, and they require
NLP. While generative AI may help address these heavy computational resources to run. In addition,
limitations to a degree, barriers remain – most cloud computing services are needed to deploy
prominently, the need for context-specific prompt and especially scale generative AI applications. All
engineering to ensure outputs are safe and reliable of these will prove onerous for resource-strapped
across the spectrum of global populations. LMICs, even if the technology’s deployment will be
a long-term resource (and life) saver.

Data without action is useless, and generative AI has vast potential to empower
decisions, especially in middle- and low-income countries, where healthcare
resources are extremely scarce.
John Sargent, Founding Partner, BroadReach Group

Patient-First Health with Generative AI: Reshaping the Care Experience 11

 4 To make generative AI
for patients a reality,
healthcare stakeholders
must take these four steps
Overcoming adoption barriers for healthcare
generative AI demands action and collaboration
from players across the healthcare ecosystem.

The release of ChatGPT set off a frenzy of public interest. The barriers slowing the widespread
interest, private investment and experimentation. adoption of generative AI and the corresponding
Moving past the razzle-dazzle phase will require enablers for overcoming them mirror those for
sustained public-private engagement to create predictive AI. These are covered in depth in the
urgency, enable change and protect the common June 2023 Forum-ZS insight report.

FIGURE 2 Ecosystem and principles for generative AI acceleration

Patients Providers

Payers Life
sciences

Connectors1 Government

Usable and Trust in AI and Partnerships and Actions by individual

representative data stakeholder education scaling mindset organizations

Note: 1. NGOs, advocacy and patient groups, distributors, group purchasing organizations, investors and other organizations in healthcare.

Patient-First Health with Generative AI: Reshaping the Care Experience 12

 All stakeholders The recommendations below are intended as a 3  eep humans in the loop. Generative
K
must help generative AI-specific action plan for health and AI’s uncanny ability to mimic human-like
educate patients healthcare players to begin addressing barriers text creates a false sense of security. Yet
and providers to the widespread adoption of patient-facing the technology is fallible. All stakeholders
on generative generative AI. must help educate patients and providers
on generative AI’s limitations, especially its
AI’s limitations,
1 B
 uild trust through empathy and potential for hallucination. At the same time,
especially its
interactivity. To instil trust in generative AI providers must work to integrate patient-
potential for models, providers, payers, biopharma and facing tools into existing clinical workflows
hallucination. tech companies must prioritize fine-tuning and implement processes to review
models to make them more empathetic – recommendations and intervene accordingly,
including by having doctors “test” responses especially in high-risk discussions and with
to improve outputs. When these models have patients with severe disease.
been trained on healthcare domain-specific
data, and users can engage them in a dialogue 4  lan to scale across contexts. Running large
P
to evaluate their outputs, then patient trust and models is extremely energy-consumptive, so
engagement should follow. technology companies must work to develop
and deploy more cost-efficient means to run
2  itigate against bias. Providers, payers,
M them. At the same time, governments, NGOs
government and biopharma all possess rich and connectors (defined here as advocacy
consumer and patient datasets, but they sit and patient groups, distributors, group
in disconnected data layers. Given that it’s purchasing organizations, investors and other
impossible to fully correct for bias inherent organizations in healthcare) must also develop
in models’ training data, mitigating output flexible deployment models that recognize the
bias requires creating a richer data layer by varying needs and exigencies across regions,
connecting the data ecosystem and then cultures and contexts. With generative AI and
measuring response bias on an ongoing basis healthcare, one size does not (and cannot) fit all.
– especially in underserved populations.

Patient-First Health with Generative AI: Reshaping the Care Experience 13

Conclusion
Patient-facing generative AI poses
risks – but the risk of maintaining
the status quo is graver.

A person’s health is inherently precious. Any This is true at the patient level and even more
stakeholder developing, deploying or vouching so at the population level. Global patient demand
for a generative AI-powered tool that could put for medical treatment outstrips supply by many
even one patient’s health at risk through faulty, orders of magnitude, leading to great loss of life,
confusing or ill-timed information should be reduced lifespan for millions and growing fractures
cautious. However, the risk of medical or ethical to global health and healthcare systems. Generative
malpractice stemming from the responsible use of AI, especially when paired with predictive AI
these tools pales compared with the more general techniques, offers the greatest current hope
malpractice of choosing to ignore their boundless of translating today’s golden age of scientific
capacity for patient impact. innovation into equitable improvements to global
health and healthcare.

Patient-First Health with Generative AI: Reshaping the Care Experience 14

Contributors
Lead authors

Dan Reiss
Principal, ZS

Antonio Spina
Lead, Digital and AI Health,
World Economic Forum

Acknowledgements

Thanks are extended to the following individuals Erica Chain

for their invaluable contributions to this paper. Digital Innovation Lead (Architecture, Strategy
& Innovation Ecosystem Catalyst), Roche

ZS Mustaqhusain Kazi
Global Head, Informatics Strategy
and Digital Innovation, Roche
Alex Anokhin
Lead, Health and Healthcare Innovations Pankaj Paul
Adviser to the Chairman and Founder,
Megan Wolfe Burjeel Holdings
Senior Director
John Sargent
Nicolas Zimmerman Managing Director, Founding Partner,
Content Strategy Lead BroadReach Group

Vivian Tan
World Economic Forum Vice-President, Strategic Information Management
& Global Relationships, Kaiser Permanente

Jitka Kolarova Ruan Viljoen

Lead, Health and Healthcare Innovations Chief Technology Officer, BroadReach Group

Andy Moose
Head, Health and Wellness Production
The authors would like to thank the following
interviewees for their time and input, which were Laurence Denmark
crucial to the production of this paper. Creative Director, Studio Miko

Allon Bloch Martha Howlett

Co-Founder and Chief Executive Officer, K Health Editor, Studio Miko

Bernard Katz Oliver Turner

Director, Product Management, BroadReach Group Designer, Studio Miko

Patient-First Health with Generative AI: Reshaping the Care Experience 15

 Endnotes
1. “Measuring and addressing physician burnout”, American Medical Association (AMA), 3 May 2023,
https://www.ama-assn.org/practice-management/physician-health/measuring-and-addressing-physician-burnout.
2. Boniol, Mathieu, et al., “The global health workforce stock and distribution in 2020 and 2030: a threat to equity and
‘universal’ health coverage?”, BMJ Global Health, vol. 7, no. 6, 27 June 2022, https://www.ncbi.nlm.nih.gov/pmc/articles/
PMC9237893/.
3. World Health Organization (WHO), Global strategy on human resources for health: Workforce 2030, 2020,
https://iris.who.int/bitstream/handle/10665/250368/9789241511131-eng.pdf?sequence=1.
4. World Economic Forum, Global Health and Healthcare Strategic Outlook: Shaping the Future of Health and Healthcare,
2023, https://www3.weforum.org/docs/WEF_Global_Health_and_Healthcare_Strategic_Outlook_2023.pdf.
5. “Patient Engagement”, Centers for Disease Control and Prevention (CDC), n.d., https://www.cdc.gov/healthliteracy/
researchevaluate/patient-engage.html.
6. AHIMA Foundation, Understanding Access and Use of Health Information in America, 2021, https://ahimafoundation.org/
media/ngfbggsk/oct2021_understanding_access_use_health_information_america_ahima_foundation.pdf.
7. Devine, Theresa, et al., “Making Quality Health Websites a National Public Health Priority: Toward Quality Standards”,
Journal of Medical Internet Research, vol. 18, no. 8, 2 August 2016, https://www.ncbi.nlm.nih.gov/pmc/articles/
PMC4987491/.
8. “Talking Points About Health Literacy”, CDC, n.d., https://www.cdc.gov/healthliteracy/shareinteract/TellOthers.html.
9. “Why improving health literacy is important”, MSD, 3 March 2022, https://www.msd.com/stories/why-improving-health-
literacy-is-important/.
10. Meherali, Salima, et al., “Health Literacy Interventions to Improve Health Outcomes in Low- and Middle-Income
Countries”, Health Literacy Research and Practice, vol. 4, no. 4, 2020, pp. 251-266, https://journals.healio.com/
doi/10.3928/24748307-20201118-01.
11. National Alliance for Caregiving (NAC) and AARP, Caregiving in the U.S. 2020, 2020, https://www.caregiving.org/research/
caregiving-in-the-us/caregiving-in-the-us-2020/.
12. International Labour Organization, Care Work and Car Jobs for the Future of Decent Work, 2018, https://www.ilo.org/
wcmsp5/groups/public/---dgreports/---dcomm/---publ/documents/publication/wcms_633135.pdf.
13. Lambert, Veronica and Deborah Keogh, “Health literacy and its importance for effective communication. Part 1”,
Nursing Children and Young People, vol. 26, no. 3, 2014, pp. 31-37, https://pubmed.ncbi.nlm.nih.gov/24708337/.
14. Baccolini, V., et al., “What is the Prevalence of Low Health Literacy in European Union Member States? A Systematic
Review and Meta-analysis”, Journal of General Internal Medicine, vol. 36, 2021, pp. 753-761, https://link.springer.com/
article/10.1007/s11606-020-06407-8.
15. Meherali, Salima, et al., “Health Literacy Interventions to Improve Health Outcomes in Low- and Middle-Income
Countries”, Health Literacy Research and Practice, vol. 4, no. 4, 2020, pp. 251-266, https://journals.healio.com/
doi/10.3928/24748307-20201118-01.
16. Balasubramanian, Sai, “Google Cloud and Mayo Clinic Partner To Improve Heathcare With Generative AI”, Forbes, 7 June
2023, https://www.forbes.com/sites/saibala/2023/06/07/google-cloud-and-mayo-clinic-partner-to-improve-healthcare-
with-generative-ai/.
17. Gupta, Aashima and Amy Waldron, “A responsible path to generative AI in healthcare”, Google Cloud, 13 April 2023,
https://cloud.google.com/blog/topics/healthcare-life-sciences/sharing-google-med-palm-2-medical-large-language-model.
18. “Measuring and addressing physician burnout”, AMA, 3 May 2023, https://www.ama-assn.org/practice-management/
physician-health/measuring-and-addressing-physician-burnout.
19. Mandowara, Khushi and Leroy Leo, “One-third of US nurses plan to quit profession, survey shows”, Reuters, 1 May 2023,
https://www.reuters.com/world/us/one-third-us-nurses-plan-quit-profession-report-2023-05-01/.
20. Muoio, Dave, “Pandemic ushered ‘unprecedented’ declines in hospital patient experience scores, study finds”, Fierce
Healthcare, 25 August 2023, https://www.fiercehealthcare.com/providers/pandemic-ushered-unprecedented-declines-
hospital-patient-experience-scores-study-finds#:~:text=Compared%20to%20expected%20trends%20based,of%20
2021%2C%20per%20the%20study.
21. Heuveline, Patrick, “Global and National Declines in Life Expectancy: An End-of-2021 Assessment”,
Population and Development Review, vol. 48, issue 1, 2022, pp. 31-50, https://www.medrxiv.org/
content/10.1101/2022.01.14.22269109v1.
22. Hagland, Mark, “At the University of Rochester Medical Center, a Breakthrough With Generative AI”, Healthcare
Innovation, 14 November 2023. https://www.hcinnovationgroup.com/analytics-ai/article/53078074/at-the-university-of-
rochester-medical-center-a-breakthrough-with-generative-ai.
23. Wood, Brenda, “Medication Adherence: The Real Problem When Treating Chronic Conditions”, U.S. Pharmacist, vol. 37, issue
4, 2021, https://www.uspharmacist.com/article/medication-adherence-the-real-problem-when-treating-chronic-conditions.
24. Kripalani, Christopher P. et al., “Medication Non-Adherence - A $290 Billion Unnecessary Expenditure”, Annals of Internal
Medicine, 2015; “€125 billion lost each year across Europe due to non-adherence to medication”, International Longevity
Centre UK, 7 April 2022, https://ilcuk.org.uk/125-billion-lost-each-year-due-to-non-adherence/.
25. WHO, Nigeria Fact Sheet, 2020, https://gco.iarc.fr/today/data/factsheets/populations/566-nigeria-fact-sheets.pdf.
26. “This AWS-powered startup gives Rwandan cancer patients better access to life-saving treatments”, 31 January 2023,
https://www.aboutamazon.com/news/aws/this-aws-powered-startup-gives-rwandan-cancer-patients-better-access-to-
life-saving-treatments.

Patient-First Health with Generative AI: Reshaping the Care Experience 16

The World Economic Forum,
committed to improving
the state of the world, is the
International Organization for
Public-Private Cooperation.

The Forum engages the

foremost political, business
and other leaders of society
to shape global, regional
and industry agendas.

World Economic Forum

91–93 route de la Capite
CH-1223 Cologny/Geneva
Switzerland
Tel.: +41 (0) 22 869 1212
Fax: +41 (0) 22 786 2744
contact@weforum.org
www.weforum.org