Selected use cases of AI in Health Care

Chapter 2 — Part 2: of “Transforming healthcare with AI”


EIT Health — European Institute of Innovation and Technology
March 3, 2020


This is an excerpt of the report “ Transforming healthcare with AI. Impact on the workforce and organisations”, published in 2020 by EIT Health. Survey of 175 healthcare professionals, health investors and AI start-up founders and executives


Executive Summary


by Joaquim Cardoso MSc.
Chief Editor of “The Health Revolution” Institute
AI Health Unit @ Digital Health Institute

March 25, 2022


Introduction

  • Despite the increasing levels of government attention on AI in healthcare and increased funding, we still need to define the critical use cases that can deliver the biggest impact in healthcare through AI (using our broader definition of AI as a spectrum) and what is actually being delivered on the ground or showing promise in terms of developments in the pipeline.

  • A summary of the key AI cases identified and profiled in the report are included in Exhibit 2.9 mapped onto the framework.

Exhibit 2.9 — AI use cases at each stage of the AI in healthcare framework


Table of Contents (TOC)

  • 2.4.1 Self-care, prevention and wellness
  • 2.4.2 Triage and diagnosis
  • 2.4.3 Diagnostics
  • 2.4.4 Clinical decision support
  • 2.4.5 Care delivery
  • 2.4.6 Chronic care management
  • 2.4.7 Improving population-health management
  • 2.4.8 Improving healthcare operations
  • 2.4.9 Strengthening healthcare innovation

2.4.1 Self-care, prevention and wellness

  • Patients are increasingly empowered to take care of their own health and wellness.

  • Countless wellness applications that aim to support people to live healthier lives (e.g., activity and sleep trackers) are part of the growing group of health applications that consumers can buy directly with no need for a healthcare consultation.

  • Many applications began as monitoring or tracking devices only, but there is a trend to enhance these solutions with AI models, for example to provide personalised plans and guidance based on individual health goals.

  • These apps collect large amounts of data, which can be used both to provide more personalised guidance to individuals, and to develop longitudinal views of broader population health and triggers for disease onset or deterioration, gradually adding a learning component.

  • The field also includes personal devices that cross over into diagnostics by monitoring vital signs among presumably healthy individuals and then flagging specific healthcare risks, such as higher than average blood pressure; and devices that cross over into remote disease monitoring tools that support the needs of patients with a chronic condition (e.g., atrial fibrillation).

  • In such examples, systems have evolved from generating large amounts of unprocessed and unorganised data available to clinicians for decision making, to assessing the patient’s condition and outlook, increasingly involving independent learning components.

  • Case study (see long version of the article): AliveCor — Personal Electrocardiogram (ECG) 

2.4.2 Triage and diagnosis

  • Symptom checkers such as Babylon, Mediktor, Ping An Good Doctor, Ada Health, K Health and others, can help triage patients and provide guidance if the symptoms require additional healthcare resources.

  • This e-triage is a promising way to provide direct and immediate access to care where otherwise there may have been delays …

  • It is also potentially a very effective way for hospital systems to ensure that only patients for whom hospital care is essential actually turn up in emergency rooms, while others are routed to more appropriate channels.

  • Case study (see long version of the article): Babylon, Mediktor, Ping An Good Doctor, Ada Health, K Health

2.4.3 Diagnostics

  • When further clinical work is needed to determine the underlying reason for symptoms, diagnostic AI solutions may help by improving accuracy or saving time.

  • Across the different diagnostic areas, we see a variety of applications for diagnostic tests, which started with the simple (e.g., moving from haemocytometers with counting grids, to automated live/dead cell counters) to more sophisticated diagnostic applications such as OLO, an AI-enhanced blood testing device that counts blood cells at the point of care.

  • Some medical specialties lend themselves naturally to AI applications, due to their large emphasis on pattern recognition, such as radiology, pathology, dermatology and ophthalmology.

  • In these areas, several AI applications have received regulatory approval by the FDA, such as Arterys, an AI-enhanced medical image-analysis platform with several applications including LungAI for lungnodule analysis allowing for the early detection of lung cancer.

  • Despite the rapid pace of development of diagnostic AI applications, they still generally focus on a specific, well-defined task.

  • Although several AI models have shown higher accuracy rates than board-certified medical experts, clinical trials are needed for further evaluation and to derive potential clinical implications from the findings.

  • Research also highlights that the relative accuracy of AI models compared to physicians can also be influenced by external context.

  • Case study (see long version of the article): Sight Diagnostic

2.4.4 Clinical decision support

  • With the rapid increase of medical knowledge, it is ever harder for physicians to keep up to date.

  • AI solutions that retrieve relevant medical knowledge for each patient and present it in a structured way can help the physicians decide on the best treatment option, saving time and leading to a more comprehensive evidence-based decision-making process.

  • In a routine clinical setting, AI models may also be able to detect patients at high risk of complications or early deterioration (e.g., DeepMind Health can predict acute kidney injury[10]51) and provide guidance for further clinical decision support, with the opportunity for prevention or early intervention.

  • Reducing complication rates by intervening early may result in improved health outcomes and reduced length of stay in hospital and related healthcare costs.

  • Despite the potential sizeable benefits, early applications have shown that clinical implementation may be less straightforward than other use cases.

  • CDS requires large comprehensive databases with high-quality data on which to build the decision tool (for example, it needs to cover all ages and ethnicities). This is as critical as the development process itself.

  • The careful codevelopment of CDS solutions in a multidisciplinary setting, with clinical and AI input, the continuous evaluation during all development stages (including after deployment) and thorough testing for validation will all be crucial to ensure a safe and efficient use in clinical practice.

  • Last, any solution perceived as a black box may face significant barriers to adoption and is therefore something developers need to preempt.

  • Case study (see long version of the article): DeepMind Health and Moorfields Eye Hospital NHS Foundation Trust

2.4.5 Care delivery

  • In care delivery, NLP-based solutions could support practitioners in various areas. For example, Moxi is a nurse-assistant robot that proactively completes tasks such as refilling stock, and there are AI solutions that can take notes or retrieve required information from medical records such as lab results or medical history.

  • Another potential use of AI in care delivery is in monitoring or treatment devices such as AI-powered artificial pancreas solutions for patients with type 1 diabetes.

  • A third area where AI can support care delivery is patient monitoring in an inpatient setting. Patient monitoring solutions such as EarlySense, which uses AI to provide actionable health insights, help nursing staff focus their attention where it is needed and provide immediate help in case of early signs of patient deterioration.

  • Case study (see long version of the article): Amelia — Virtual Health Agent Platform, and Bionic Pancreas by Beta Bionics


2.4.6 Chronic care management

  • AI solutions help patients (as well as relatives and caregivers) to manage their chronic disease on a day-to-day basis and potentially remain independent and stay at home longer.

  • For example, patients with congestive heart disease may be supported by virtual-nurse systems monitoring vital signs and symptoms, ensuring medication is taken and encouraging the adoption of healthy habits.

  • This could reduce the need for a 24/7 caregiver.

  • Personal monitoring and alert systems for use at home can help elderly people stay in their familiar environment for as long as possible.

  • Adhering to medication is another chronic-care challenge that AI could tackle.

  • Case study (see long version of the article): Sensely (Chronic Care Management) and Karantis360 (Elderly Care)

2.4.7 Improving population-health management

  • AI can be used on large datasets to predict health outcomes within a population, which helps health systems focus more heavily on prevention and early detection, improve population health outcomes and, over time, ensure the financial sustainability of the care system.

  • Using AI to analyse large datasets may prove useful both in healthcare settings and epidemiological studies. AI-powered models based on clinical data from a large population

  • In population health research, AI may be able to uncover previously unidentified correlations between factors ..

  • Case study (see long version of the article): Mount Sinai Health Systems — Risk prediction for hospital emergency admissions

2.4.8 Improving healthcare operations

  • “AI currently creates the most value in helping frontline clinicians be more productive and in making back-end processes more efficient […less so] in making clinical decisions.”[21]64

  • Potential areas for improving healthcare operations include: scheduling, hospital admissions, discharge and capacity management, optimising processes in the operating room and the emergency department, as well as moving patients between diagnostics and the ward.

  • Such applications can significantly and directly affect patients by reducing waiting times, and increasing transparency on process, times and outcomes — all of which lead to a better patient experience, as inefficiencies along the patient pathway are ironed out.

  • Case study (see long version of the article): Qventus (Improving Health Care Operations)

2.4.9 Strengthening healthcare innovation (pharma and medtech)

  • Overall, AI is now applied in different elements of the business system in the pharmaceutical and medtech industries in order to increase the speed to market of new products, reduce costs, enhance clinical outcomes and serve a variety of organisational goals.

  • Startups such as Recursion Pharmaceuticals and BenevolentAI are innovating, while big pharmaceutical and technology players are focused on realising opportunities from AI.

  • Big Pharma is also making major investments and partnerships to address opportunities ( Novartis and Microsoft; Bristol-Myers Squibb and Concerto HealthAI ; AstraZeneca with BenevolentAI and Schrödinger), but pharma companies have found it hard to realise the promise of AI.

  • One major challenge is data — although pharma companies have a lot of data, they are often poorly suited to AI due to quality issues, inconsistent formats or the challenges of linking data and obtaining the necessary consent to use in different use cases.

  • As a result, major investments are now taking a step back, focusing on developing distinctive data assets and consistent data formats that will enable future application (e.g. Sanofi’s DARWIN platform; Roche’s Navify Tumour Board solution)

EXCERPT OF THE REPORT (long version, including the case studies)

Introduction


Despite the increasing levels of government attention on AI in healthcare and increased funding, we still need to define the critical use cases that can deliver the biggest impact in healthcare through AI (using our broader definition of AI as a spectrum) and what is actually being delivered on the ground or showing promise in terms of developments in the pipeline
.

This section uses the framework below, that set out the key areas of impact of AI in healthcare, placing the patient at the centre, and across the healthcare value chain. Beyond the patient, we also highlight areas of impact of AI in terms of healthcare more broadly, by improving population health, healthcare operations and healthcare-related innovation.

A summary of the key AI cases identified and profiled in the report are included in Exhibit 2.9 mapped onto the framework.

These examples illustrate the range of areas where AI is having an impact in healthcare and are by no means exhaustive; nor does their inclusion constitute an endorsement of a specific solution or organisation.

Use cases listed were highlighted by our interviewees as prominent or promising, or reflect our collective experience in AI over recent years, and serve to demonstrate the diversity of applications available today.

Exhibit 2.9 — AI use cases at each stage of the AI in healthcare framework


Table of Contents (TOC)

  • 2.4.1 Self-care, prevention and wellness 
  • 2.4.2 Triage and diagnosis 
  • 2.4.3 Diagnostics 
  • 2.4.4 Clinical decision support 
  • 2.4.5 Care delivery
  • 2.4.6 Chronic care management
  • 2.4.7 Improving population-health management
  • 2.4.8 Improving healthcare operations 
  • 2.4.9 Strengthening healthcare innovation

2.4.1 Self-care, prevention and wellness


Patients are increasingly empowered to take care of their own health and wellness.

Countless wellness applications that aim to support people to live healthier lives (e.g., activity and sleep trackers) are part of the growing group of health applications that consumers can buy directly with no need for a healthcare consultation.

Many applications began as monitoring or tracking devices only, but there is a trend to enhance these solutions with AI models, for example to provide personalised plans and guidance based on individual health goals.

“Nearly all self-care options will have some type of AI component.”

Michal Rosen-Zvi, IBM Research and The Hebrew University

These apps collect large amounts of data, which can be used both to provide more personalised guidance to individuals, and to develop longitudinal views of broader population health and triggers for disease onset or deterioration, gradually adding a learning component.

The field also includes personal devices that cross over into diagnostics by monitoring vital signs among presumably healthy individuals and then flagging specific healthcare risks, such as higher than average blood pressure; and devices that cross over into remote disease monitoring tools that support the needs of patients with a chronic condition (e.g., atrial fibrillation).

In such examples, systems have evolved from generating large amounts of unprocessed and unorganised data available to clinicians for decision making, to assessing the patient’s condition and outlook, increasingly involving independent learning components.


ALIVECOR — PERSONAL ELECTROCARDIOGRAM (ECG)


What is it and what is its role in healthcare?

US-based company AliveCor developed KardiaMobile, a personal ECG device that can monitor heart rhythm and instantly detect and flag atrial fibrillation, bradycardia or tachycardia to clinical teams.

Atrial fibrillation affects 7.6 million people over 65 in Europe, a number expected to rise to 14.4 million by 2060.[1]42

How does it work?

Patients record their personal ECG remotely using KardiaMobile in combination with the Kardia app. Depending on the patient’s monitoring needs, a single or 6-lead ECG can be acquired with no need for additional leads and cables.

The Kardia app allows patients to track data over time and shares the ECG recordings directly with their physician.

What does this mean for healthcare practitioners and organisations?

Physicians can monitor patient data remotely on the KardiaPro platform. They can get additional datapoints from patients between visits and establish a more complete picture of the patient’s heart rhythm, to further inform clinical decision making.

ECGs may reveal intermittent atrial fibrillation not detected in periodic measurements at the physician’s office, and lead to change of treatment.

The existence of such applications in self-care, prevention or wellness more broadly, means that physicians need to be able to calibrate the point at which an intervention may be needed, as many technologies have a low threshold for alerts.

This can only be overcome by physicians collaborating in AI multidisciplinary teams to identify real trigger points for intervention versus noise in the system, taking a population view.

Some organisations have already started collaborating in this direction, for example Sheba Medical Center and two Health Maintenance Organisations in Israel are planning to combine their data to create predictive models to help proactively address the health of their population by identifying the right intervention trigger points.

For payors and insurers, the increased use of personal monitoring devices may require reimbursement policies, potentially through outcome-based models, for the time physicians spend analysing the recorded data, providing information and discussing individual findings.

Organisations also need to ensure interoperability and the ability to integrate readings from diverse sources (e.g., wearables) into a single view of the patient for the physician, integrating seamlessly into the typical clinical workflow.

What is its reach (and potential)?

KardiaMobile is available in 35 countries (13 in Europe) and has been used to record more than 50 million ECGs to date.

Devices can be used by patients, nurses or caregivers in situations when a full 12-lead ECG is not required or feasible.

As such, they could serve as mobile devices in home-care settings for the monitoring of patients with atrial fibrillation, bradycardia or tachycardia and help reduce a substantial rise across Europe of disabling strokes.


2.4.2 Triage and diagnosis


Symptom checkers such as Babylon, Mediktor, Ping An Good Doctor, Ada Health, K Health and others, can help triage patients and provide guidance
if the symptoms require additional healthcare resources.

This e-triage is a promising way to provide direct and immediate access to care where otherwise there may have been delays — for example, in healthcare systems with long waiting times, in primary care in different parts of Europe, in rural areas with shortage of primary care resources, or in some emerging markets.

It is also potentially a very effective way for hospital systems to ensure that only patients for whom hospital care is essential actually turn up in emergency rooms, while others are routed to more appropriate channels.

“The only way to know if you need to see a doctor, is to see a doctor. This maxim can and should change. Patients are now going to Google all the time. They prefer an imperfect answer now than a perfect answer in a week, and need to know whether they should be worried.”

Josep Carbó, Mediktor and Barcelona Health Hub


ONLINE SYMPTOM CHECKERS/E-TRIAGE TOOLS


What are they and what is their role in healthcare?

These tools aim to improve access to healthcare by letting patients check common pathologies typically addressed in primary care and providing information on related symptoms, potential treatments and outcomes.

Some offer follow-up via online chat or video consultations with physicians.

How do they work?

The symptom checkers in use today are broadly similar, but they do have some important differences (e.g., training of algorithm, access to data or scale). Key examples include:

· Babylon Health (UK), Mediktor (Spain) and K Health (Israel). Babylon’s AIpowered chatbot uses NLP to understand symptoms defined in the patient’s words and provides relevant health and triage information using algorithms trained on NHS data. It offers an initial diagnosis with possible scenarios and a percentage-based estimate of each being correct. Babylon is also working on a technique, inspired by quantum cryptography, that would allow medical databases to be tapped for causal links.[2]43

· Mediktor uses an approach also focused on e-triage and has been validated in a prospective observational study in a tertiary university hospital emergency department.[3]44

· K Health was developed in partnership with Maccabi Health Services in Israel, which granted the company access to the anonymised electronic medical record data of more than 2 million people from the past 20 years. Using NLP and advanced modelling, the algorithm was trained to understand symptoms and the likelihood of an underlying diagnosis being connected with a patient’s data. A built-in feedback loop lets the system learn with every case and improve continuously.[4]45

· Ada Health (Germany). Its system connects medical knowledge with AI and the app compares answers to questions about symptoms to similar cases drawing from extensive clinical literature.

· Ping An Good Doctor (China). The symptom checker is an integral part of a closed-loop ecosystem that connects patients with physicians, online or offline, after initial assessment. It is also deployed in a physical setting as an AI-enabled virtual doctor accessible in Ping An’s One-MinuteClinics. 
 These are found in eight provinces in China, with contracts signed for nearly 1,000 units in large and medium-sized enterprises, community centres, chain pharmacies and other high footfall areas.[5]46

What does this mean for healthcare practitioners and organisations?

Symptom checkers can boost productivity as practitioners spend less time collecting data and forming an early view of patients, and they can help reduce the risk of misdiagnoses.

They may also help relieve pressure on primary-care providers and organisations, lead to fewer people presenting in emergency departments and reduce overall caseloads.

This lets organisations dedicate more time to the patients with the highest needs (though patients who eventually present at hospital may have more complex needs on average).

Today, most symptom checkers are not reimbursed. Many grow through direct-to-consumer sales (e.g., in China), or collaborations with hospital organisations (e.g., in Europe).

What is their reach (and potential)?

Symptom checkers are widely available in the US, Europe and China. In 2019, Ping An Good Doctor registered users exceeded 300 million.

Babylon Health records 2.2 million AI consultations and almost 4 million users.

Ada Health claims it has made more than 15 million symptom assessments and has 8 million users.

Mediktor reports more than 3 million assessments, and K Health more than 2 million users and 80 million user questions answered in the app.[6]47


2.4.3 Diagnostics


When further clinical work is needed to determine the underlying reason for symptoms, diagnostic AI solutions may help by improving accuracy or saving time.

Across the different diagnostic areas, we see a variety of applications for diagnostic tests, which started with the simple (e.g., moving from haemocytometers with counting grids, to automated live/dead cell counters) to more sophisticated diagnostic applications such as OLO, an AI-enhanced blood testing device that counts blood cells at the point of care.

Some medical specialties lend themselves naturally to AI applications, due to their large emphasis on pattern recognition, such as radiology, pathology, dermatology and ophthalmology.

In these areas, several AI applications have received regulatory approval by the FDA, such as Arterys, an AI-enhanced medical image-analysis platform with several applications including LungAI for lungnodule analysis allowing for the early detection of lung cancer.

Despite the rapid pace of development of diagnostic AI applications, they still generally focus on a specific, well-defined task.

Although several AI models have shown higher accuracy rates than board-certified medical experts, clinical trials are needed for further evaluation and to derive potential clinical implications from the findings.

Research also highlights that the relative accuracy of AI models compared to physicians can also be influenced by external context.

For example, a recent Nature article highlights differences in the relative success of AI based on the frequency of mammography and number of reviewers.

In the US, where there are more frequent reviews but by a single practitioner, AI was more valuable than in the UK, where reviews are typically conducted by two practitioners, albeit less frequently.[7]48

Interviewees in our survey also highlighted the potential for significant added value in countries where resources to train the workforce are more limited or there are wider variations in clinical education and quality.

As Tencent Healthcare’s Dr. Alexander Ng says of China, “It would be impossible from a human-resources perspective to skill up everyone. The transition will have to be through digital tools to enable physicians to manage their patients to address their needs in a nascent primary care system.”

Another example shared during our interviews is a solution developed at Sheba Medical Center in Israel.

A new algorithm for image reconstruction can reduce the radiation exposure of a conventional chest CT to 4 percent and could be suitable for annual screening of smokers with a high risk of developing lung cancer.

A reduction of radiation exposure for CT scans could lead to broader benefits at the population level, as CT scans are high-volume diagnostic tools — in 2016, there were 36 million CT scans in France, Germany, the UK, Italy and Spain.[8]49


SIGHT DIAGNOSTICS


What is it and what is its role in healthcare?

Israeli company Sight Diagnostics has developed OLO, a point-of-care blood testing device that can perform a full blood count (FBC) using AI machine-vision technology.

Almost half a petabyte of anonymised blood image data was used to train the AI powering the blood diagnostics system.

The device allows a healthcare professional to carry out an accurate test within 10 minutes from a finger prick and requires only minimal training.

This makes it suitable for use in primary care settings, emergency departments or outpatients, and settings without a lab.

How does it work?

The AI technology interprets multiple images of a small blood sample producing an FBC test of similar quality to a traditional laboratory test.

Point-of-care blood testing means the healthcare professional can get the results much more quickly, which in turn means the patient can be diagnosed potentially immediately, rather than waiting hours or days.

It also reduces the need for samples to be transported, tracked and tested in a lab.

Performing the test from a finger prick reduces the need for a larger blood sample to be taken from a vein, which can require phlebotomy services.

The AI approach also removes the need for device calibration, which is typically done by lab technicians, or under their supervision.

What does this mean for healthcare practitioners and organisations?

OLO can save time by removing the need for follow-up appointments and even in the time needed to draw blood.

Time can also be saved for lab staff by reducing the amount of calibration required.

Training is required to use the device correctly and ensure accurate results, but new data and analytics skills are not required.

The main consideration for organisations is ensuring quality assurance processes are adapted to incorporate the use of new point-of-care diagnostics.

What is its reach (and potential)?

OLO has a CE Mark registration and has been deployed in several European healthcare systems, while undergoing clinical trials in the US.

The blood testing market globally is expected to exceed $62 billion by 2024, driven in part by the growing prevalence of chronic diseases such as diabetes and cardiovascular disease that may require more frequent testing.[9]50

Delivering an FBC through AI-augmented point-of-care blood testing regardless of setting, could revolutionise diagnostics and reduce complexity for millions of patients.


2.4.4 Clinical decision support


With the rapid increase of medical knowledge, it is ever harder for physicians to keep up to date.

AI solutions that retrieve relevant medical knowledge for each patient and present it in a structured way can help the physicians decide on the best treatment option, saving time and leading to a more comprehensive evidence-based decision-making process.

In a routine clinical setting, AI models may also be able to detect patients at high risk of complications or early deterioration (e.g., DeepMind Health can predict acute kidney injury[10]51) and provide guidance for further clinical decision support, with the opportunity for prevention or early intervention.

Reducing complication rates by intervening early may result in improved health outcomes and reduced length of stay in hospital and related healthcare costs.

“AI needs to serve as a decision-support tool. In the end, it is the decision of a doctor. AI is not going to substitute doctors in the foreseeable future.”

Dr. Eyal Zimlichman, Sheba Medical Center

Despite the potential sizeable benefits, early applications have shown that clinical implementation may be less straightforward than other use cases.

While many clinicians are excited about the potential use of AI in clinical decisions, others cast doubt on the readiness of the technology, or the relative value to focusing on CDS, compared to applying AI in areas that simplify routine processes and thus free up physician time.

For example, IBM Watson was one of the first companies eager to bring AI into clinical settings. It entered the field with clear strengths in NLP and machine learning but found it hard to deliver fully on the early promises of its Watson Health division and of Watson for Oncology, an AI system developed and trained in collaboration with the Memorial Sloan Kettering Cancer Center in New York City.[11]52

Such experiences show how complicated today’s healthcare system can be. Data can be incomplete or of poor quality and, more importantly, there can be a fundamental mismatch between the way machines learn and the way doctors work. More broadly, AI is only as good as the data used to generate it.

CDS requires large comprehensive databases with high-quality data on which to build the decision tool (for example, it needs to cover all ages and ethnicities). This is as critical as the development process itself.

The careful codevelopment of CDS solutions in a multidisciplinary setting, with clinical and AI input, the continuous evaluation during all development stages (including after deployment) and thorough testing for validation will all be crucial to ensure a safe and efficient use in clinical practice.

Last, any solution perceived as a black box may face significant barriers to adoption and is therefore something developers need to preempt. As Dr. Thomas Senderovitz, Director General of the Danish Medicines Agency, says, “There need to be transparency and ethics in AI. In healthcare, it is unacceptable that companies are asking us to trust that their algorithms work, without being able to examine it. 
 We need to push against that and develop mechanisms to assess such solutions, potentially developing centres that can specialise in assessing SaMDs [Software as a Medical Device].”


DEEPMIND HEALTH AND MOORFIELDS EYE HOSPITAL NHS FOUNDATION TRUST

What is it and what is its role in healthcare?

In 2016, Moorfields Eye Hospital (an NHS hospital in the UK), and DeepMind (a UK-based AI company owned by Google) started a partnership to explore how machine learning and AI can support medical research into eye disease, including age-related macular degeneration and sight loss as a result of diabetes.[12]53

How does it work?

The joint team used thousands of anonymised eye scans to train algorithms to identify signs of eye disease and recommend referral or treatment.

In 2018, results showed that AI could match world-leading experts in diagnosing a range of conditions, making the correct referral decision for more than 50 eye diseases with 94 percent accuracy.

The team uses a dual-layered neural network to avoid creating a black box. The first neural network analyses the scan to provide a mapping of types of eye tissue and features of disease, allowing professionals to reach a clinical decision independently. The second network analyses this map to present clinicians with a potential diagnosis and recommendation.

What does this mean for healthcare practitioners and organisations?

This type of algorithm can help reduce the time to diagnosis for conditions that threaten permanent loss of sight, allowing providers to treat patients or refer them to the right specialists for further treatment more quickly.

Today, eye-care professionals use optical coherence tomography scans to help diagnose eye conditions — 3D images offering a detailed map of the back of the eye — but these can be hard to read and need expert analysis to interpret.

At Moorfields, where practitioners may review more than a thousand scans a day, delays in triaging scans can make the difference between an urgent case being addressed properly or a patient losing his or her sight.

Such applications therefore help prioritise the most complex cases, increase productive time and potentially reduce clinical error.

To maximise their use, practitioners need to understand the technology behind these applications, their key features and limitations, and engage in their ongoing development.

In the absence of clinical trials or regulatory approval, the onus is higher on practitioners to integrate such solutions in their workflow, but the benefits to patient care and productive clinical time are becoming more pronounced.

For healthcare organisations, such applications must be well-governed and stewarded, and partnerships need to have a long-term perspective: Moorfields and DeepMind launched their collaboration in 2016 and development is still ongoing.

What is its reach (and potential)?

DeepMind is expanding CDS in breast-cancer detection and acute kidney injury, and Sheba’s collaborations in Israel focus on colorectal cancer — a few examples of the vast potential of CDS.


2.4.5 Care delivery

In care delivery, NLP-based solutions could support practitioners in various areas. For example, Moxi is a nurse-assistant robot that proactively completes tasks such as refilling stock, and there are AI solutions that can take notes or retrieve required information from medical records such as lab results or medical history.

In our interviews, AI solutions using NLP were mentioned several times as areas where people see potential for significant progress within the next couple of years.

As Antanas Montvila, a radiologist and Vice President of the European Junior Doctors’ Association says, “Once NLP gets started, every area of healthcare should be affected.”

“Voice tech, such as digital assistants, is changing the game. There is a huge opportunity with the shift to voice, breaking down the barrier of usability, in particular with older adults.”

Mary Lou Ackermann, SE Health

Another potential use of AI in care delivery is in monitoring or treatment devices such as AI-powered artificial pancreas solutions for patients with type 1 diabetes.

These patients need their blood glucose levels checked frequently and insulin and glucagon have to be administered accordingly.

Having a solution that constantly monitors the blood glucose levels and autonomously administers the right dose can both free up time and potentially reduce the error rate when compared to manual calculations and administration.

A third area where AI can support care delivery is patient monitoring in an inpatient setting.

Healthcare organisations aim to provide timely delivery of care but emergencies or unexpected changes in schedule can lead to longer waiting times, a poor patient experience, and worse clinical outcomes.

For example, a nurse on an orthopaedic ward may receive an alert when a patient shows signs of cardiac arrythmia and thus might be at risk of suffering from an acute myocardial infarction.

Patient monitoring solutions such as EarlySense, which uses AI to provide actionable health insights, help nursing staff focus their attention where it is needed and provide immediate help in case of early signs of patient deterioration.

“The reason we become nurses is to care for patients and save lives, not to enter data.”

Rebecca Love, OptimizeRx


AMELIA — VIRTUAL HEALTH AGENT PLATFORM (Care Operations and Administration)


What is it and what is its role in healthcare?

Amelia, a cognitive virtual agent platform from IPSoft, demonstrates learning abilities and elements of emotional intelligence.

It can perform autonomic task management using conversational AI and manage some operational and administrative hospital processes.

How does it work?

Amelia can play the role of a care protocol “whisper agent” (e.g., reminding practitioners of steps that need to be followed), as well as a care operations agent, helping document a patient visit, admitting patients, retrieving medical history prior to a conversation, checking availability of hospital beds, retrieving lab results and scheduling specialist appointments.

The Amelia Health agents, enabled by AI technology, learn continuously with every completed task and can communicate through voice, mobile, web, and chat.

What does this mean for healthcare practitioners and organisations?

Nurses may save time using virtual agents to help them admit, discharge and transfer patients.

Practitioners preparing for patient consultations can save time by instantly retrieving the patient’s medical history rather than searching through several systems.

Virtual assistants can help physicians identify available colleagues from another specialty when a consultation is needed.

The ability to document patient visits can help practitioners focus on the direct interaction with the patient, without manually filing notes. Another impact area is lab test results. The time between taking samples and getting results can vary, depending on factors such as availability of transport and lab capacity. Amelia can provide that information directly, reducing the need for staff to repeatedly check manually.

Given the wide range of capabilities of virtual agents such as Amelia, organisations need to define where the technology will bring the most value in their specific context and guide their employees through the training and adoption process to interact with virtual agents.

What is its reach (and potential)?

Amelia is already available in the US and the UK, and IPSoft has also partnered with NHS Digital to build a Digital Virtual Data Assistant (ViDA), a chatbot to help extract information from the NHS’ healthcare data repository, e.g., on emergency system waiting times.[13]54

Such technologies may also support caregivers and patients in outpatient and home-care settings, reducing the time practitioners spend scheduling appointments or sharing with patients notes from their last visit and care plan details.


BIONIC PANCREAS BY BETA BIONICS (Care Delivery)


What is it and what is its role in healthcare?

The bionic pancreas (iLet, developed by US company Beta Bionics) uses machine learning to constantly monitor and independently manage blood sugar levels in insulin-dependent type 1 diabetes patients, mimicking the function of the pancreas.

This could provide vital support for patients, many of whom find adherence to strict monitoring and insulin-management regimes very restrictive, especially adolescent patients.

How does it work?

The iLet device is worn on the skin and connects wirelessly to a smartphone-sized portable unit that contains the hormone(s). It defines timing and dosage to administer through an algorithm.

What does this mean for healthcare practitioners and organisations?

Today, practitioners have to teach patients who use a conventional insulin pump or who manually inject insulin, how to count the intake of carbohydrates. With a bionic pancreas, this may no longer be required.

The time of caregivers, who used to measure blood sugar levels several times a day and then administer insulin and glucagon, will also be freed up.

Physicians and nurses may have to develop critical appraisal skills of the functionality of the bionic pancreas so they can identify patients who are suitable for this treatment and enable them to use it correctly.

The biggest impact could be on the patients, as they can care for themselves more effectively, improving their outcomes.

This type of support enables hospitals and other providers to focus on those interactions with diabetes patients that have the most value in terms of outcomes, and therefore prioritise the use of resources.

Despite the promise of such solutions, we are still a long way from scaling them.

Regulators and payors need to evaluate which devices are safe and effective, and payors would need to develop criteria for the reimbursement of using a bionic pancreas.

The workforce would need the right coaching and counselling and would have to be confident they were choosing the right patients for such applications in the first place.

What is its reach (and potential)?

Currently, iLet is used as an investigational device and has been tested for use in type 1 diabetes patients in clinical trials for outpatient and home use, with hormones from different pharmaceutical companies.

It could potentially be used to treat insulin-dependent type 2 diabetes patients, who make up more than 90 percent of the world’s diabetic population.

With the diabetic population (diagnosed and undiagnosed) estimated at 415 million — or 1 in 11 of the world’s adult population — rising to 642 million by 2040, such solutions could be one of the largest potential applications of AI healthcare devices in terms of global health outcomes and population impact.[14]55


2.4.6 Chronic care management


AI solutions help patients (as well as relatives and caregivers) to manage their chronic disease on a day-to-day basis and potentially remain independent and stay at home longer.

For example, patients with congestive heart disease may be supported by virtual-nurse systems monitoring vital signs and symptoms, ensuring medication is taken and encouraging the adoption of healthy habits.

This could reduce the need for a 24/7 caregiver. The patient may share data recorded in the application, providing relatives, caregivers and physicians with a comprehensive longitudinal dataset of personal health information, potentially reducing the need for some in-person visits to the physician.

According to Dr. Marco Inzitari, President of the Catalan Society of Geriatrics and Gerontology, “AI enables health professionals to better select patients who can be empowered to take action for their own care.”

“AI can be used to continue the care cycle without patients being hospitalised. With AI-based solutions, we can help people have a better quality of life outside of hospitals.”

Federico Menna, EIT Digital

Personal monitoring and alert systems for use at home can help elderly people stay in their familiar environment for as long as possible.

These solutions may be particularly helpful for people with increased frailty or with cognitive impairment or dementia, where a monitoring and alert system could provide enough oversight to allow them to live at home, while knowing that healthcare services could help in a timely manner if needed.

Adhering to medication is another chronic-care challenge that AI could tackle.

Many older patients have to take several prescription drugs at different times of the day. Remembering when to take which pill can be daunting, especially if the patient has (mild) cognitive impairment.

A personalised AI-enhanced pill-presorting delivery system could reduce the risk of patients making mistakes, or the need for caregivers to sort medication. AI applications may also be able to help patients by monitoring and encouraging treatment adherence.


SENSELY (Chronic Care Management)


What is it and what is its role in healthcare?

US company Sensely offers a virtual nurse assistant, with modules for chronic diseases that can be used for personalised monitoring and follow-up care.

Patients can use the virtual assistant on a tablet at home to support their daily routine of managing their care or connect with healthcare practitioners.

How does it work?

Sensely’s avatar-based chronic-care platform provides personalised conversational content using text-to-speech and speech-recognition technologies. It helps guide the patient through daily monitoring needs and can assess symptoms to determine whether to contact a healthcare professional.

The assistant guides the patient step-by-step through the process (e.g., “Now it’s time to take your blood pressure. Please make sure the cuff is on by pressing the orange button”) and provides instant feedback (“Your blood pressure is a little high today”). Using speech recognition, the solution also allows the patient to speak to the virtual assistant, for example to report symptoms.

What does this mean for healthcare practitioners and organisations?

The practitioners who receive the recorded measurements from their patients need to understand the functionality of the platform (including potential customisation options for patients with comorbidities and complex diseases) and determine which patients may benefit from using the solution.

To ensure the self-monitoring results are of value, the caregiver needs to ensure the patient can follow the assistant’s instructions and take the measurements correctly.

The healthcare practitioner may also receive alerts when measurements are above a certain threshold, allowing him or her to monitor patients and identify situations that require further action.

This enables them and their organisations to have an ongoing but targeted interaction with patients, focusing the service where it adds most value. Such solutions can bring significant value to payors, as they may increase patient adherence and may allow for early interventions to prevent risk and reduce complications and avoidable hospitalisations.

What is its reach (and potential)?

Sensely provides content for chronic-care management in 32 languages across 14 conditions, including congestive heart failure, chronic obstructive pulmonary disease and diabetes, and is increasingly adopted not just by healthcare providers but also by insurers.

Virtual-assistant applications more broadly may also be of value for policyholders (e.g., providing wellness information, e-triage and online customer service) or for pharmaceutical companies to support clinical-trial monitoring or patient education, and to improve pharmaceutical adherence and engagement.


KARANTIS360 (Elderly Care)


What is it and what is its role in healthcare?

UK company Karantis360 has developed an automated, personal monitoring and alerting system that enables elderly people to live independently, ensuring caregivers and families can stay informed.

How does it work?

Karantis360 has partnered with IBM Watson and EnOcean to provide a comprehensive solution using AI and Internet of Things capabilities combined with intelligent sensors linked to a mobile device.

The device shares information via a web and mobile dashboard, and can send reports and alerts to caregivers and family members.

The sensors can provide information about the patient’s daily routine, e.g., when individuals get up or go to bed, use the bathroom or leave the house.

Using AI, the system identifies deviations from typical behaviour that could indicate that something is wrong — such as a fall — and informs caregivers and families about these abnormalities in order to inform the plan of action.

What does this mean for healthcare practitioners and organisations?

Professional caregivers can stay informed about their patients and react to urgent situations or intervene to help prevent deterioration.

The real-time speech-recognition feature allows caregivers to fully concentrate on providing care, while patient records are updated in the background.

Being able to immediately connect with the patient as required may provide the assistance needed for the patient to be able to live at home rather than move into a nursing home or assistedliving facility.

Caregivers may need to undergo training to calibrate alerts and prioritise when to intervene.

Such applications have important implications for healthcare providers.

They not only help prioritise when and how to deploy outpatient, community-based support, but also enable patients to return home earlier from hospital, freeing up hospital capacity to be used by others in need and reducing waiting times and costs.

Providers may also find such systems offer them a competitive advantage, as families and caregivers increasingly look for peace of mind when it comes to their loved ones.

Over time, AI-generated insights can be used for population-health management, enabling the system to prioritise use of resources and proactively identify, on a population-basis, drivers and triggers of deterioration.

What is its reach (and potential)?

The percentage of the population who are 65 years or over is growing, rising to 19.2 percent in the EU in 2016, and almost one third of those people lived alone at home in 2015.[15]57

The annual growth in demand for care and housing for elderly people in the EU is expected to grow by 3.5 percent and 5.5 percent respectively.[16]58

The increasing unsustainability of European healthcare systems when factoring in the increasing care needs of ageing populations, indicates the impact such systems could have if scaled across Europe.


2.4.7 Improving population-health management


AI can be used on large datasets to predict health outcomes within a population, which helps health systems focus more heavily on prevention and early detection, improve population health outcomes and, over time, ensure the financial sustainability of the care system.

Using AI to analyse large datasets may prove useful both in healthcare settings and epidemiological studies. AI-powered models based on clinical data from a large population (e.g., patients within a health region, or an integrated provider system) may help identify early risk factors that can trigger preventative actions or early interventions at a system level.

They may also be useful in determining what to prioritise during times of staff shortages. Similarly, identifying an increased risk of unplanned hospital admissions could help practitioners intervene preemptively to avoid them.

“We have changed how people use the healthcare system. We have algorithms to predict future health risks and, based on them, we have developed interventions.”

Ossi Laukkanen, Mehiläinen

In population health research, AI may be able to uncover previously unidentified correlations between factors, for example combining data collected by wearables and health outcomes, which can be investigated together in hypothesis-driven studies with the goal of a better understanding of underlying factors that cause diseases.

Such algorithms are already used in systems that are at least partially integrated across provider settings and on a population-health region basis, or that have shared incentives to deliver population-based outcomes (clinical, operational or financial).

At Sheba Medical Center in Israel, an AI solution aims to predict which of the colorectal cancer patients undergoing surgery will suffer from leakage as a complication.

By identifying the at-risk patients early, it aims to minimise or remove the risk through intervention. AI solutions may also help in prevention as part of national screening programmes, improving accuracy rates and enabling earlier detection of problems.[17]59

While these applications generate a high degree of enthusiasm, there can be a concern that they are of limited use beyond the populations on which they are trained (their “generalisability”).

In practice, this means further development, external validation and testing in clinical practice and across settings and geographies will be crucial to allow for a more wide-spread use of AI-based models.


MOUNT SINAI HEALTH SYSTEMS — RISK PREDICTION FOR HOSPITAL EMERGENCY ADMISSIONS


What is it and what is its role in healthcare?

Mount Sinai Health Systems, a hospital network in New York City, has developed a model to identify patients from their population health programme that are at risk of unplanned admission.

This is part of Mount Sinai’s effort to transition to a delivery model focused on value and risk-based population health.[18]60

How does it work?

Mount Sinai’s Department of Population Health has been using machine-learning algorithms to mine data that identifies patients who are at risk of an unplanned admission among the system’s 500,000 patient population health programme and develop predictive modelling features.

What does this mean for healthcare practitioners and organisations?

The system can enable clinical pathways and protocols to be redesigned towards intervening proactively in the highest risk cases. This shifts the working patterns of practitioners from reactive care to proactive care.

To adequately address the risk, practitioners and social workers need to understand how the model identified the patient and which factors may need to be addressed to mitigate the risk. They may, therefore, become more alert to the risk factors that the model identifies, which can help reduce unnecessary admissions.

In turn, to increase the clinical validity of the model, they need to feed back into the model which interventions were made and whether patients were admitted to the hospital as predicted.

Organisations that can implement risk-prediction models and intervene accordingly may also deliver better health outcomes for their patients, as well as reduce overall avoidable hospitalisation costs.

What is its reach (and potential)?

AI-based models could help reduce the significant numbers of avoidable admissions and subsequent lengthy hospital stays, a feature of many EU health systems. Ambulatory care sensitive conditions, for example, account for up to 14 percent of emergency admissions in the UK.[19]61

This would imply systems aggregating larger population datasets and aligning resources and incentives to adequately address identified risks while continuing to improve AI models.

AI-based models can also help predict the progression of chronic disease, a leading cause of death and disability within the EU, slowing down disease progression and significantly improving population outcomes.[20]62


2.4.8 Improving healthcare operations


The use of AI in healthcare may be more readily accepted when it helps free up practitioners from routine, low value-add administrative tasks, to increase direct time with patients.

According to one study, “AI currently creates the most value in helping frontline clinicians be more productive and in making back-end processes more efficient […less so] in making clinical decisions.”[21]64

Potential areas for improving healthcare operations include scheduling, hospital admissions, discharge and capacity management, optimising processes in the operating room and the emergency department, as well as moving patients between diagnostics and the ward.

Such applications can significantly and directly affect patients by reducing waiting times, and increasing transparency on process, times and outcomes — all of which lead to a better patient experience, as inefficiencies along the patient pathway are ironed out.

“Most of the potential value in healthcare via AI will be realised through the cognitive augmentation of routine interactions — rather than the more headline-grabbing applications in clinical decision support that tend to get much of the attention. By assuming delegated routine tasks, these applications free up human professional carers for those high-touch patient interactions that only humans can provide and that allow them to take care of more people.”

David Champeaux, Cherish Health and HIMSS Advisory Council Member[22]63


QVENTUS


What is it and what is its role in healthcare?

Qventus is an AI-based software platform that solves operational challenges that occur in the hospital.

Delays or cancellations of surgeries plague hospitals and can result in worse clinical outcomes, ineffective use of healthcare resources (e.g., theatres, anaesthetist time) and higher costs per patient.

How does it work?

Qventus’s operating-room solution helps optimise the different steps of the perioperative flow and proactively addresses potential bottlenecks using machine learning.

During patient preparation, the platform identifies missing requirements to avoid last-minute cancellations. The software detects unexpected orders and late-start risks during the preoperative phase and optimises the block schedules in real-time.

It helps hospital teams prioritise, for example, by identifying high-priority actions and nudging the teams to resolve issues.

The influx of patients to peri-acute care units, intensive care units or standard in-patient wards can be predicted to locate suitable hospital beds, minimise postop holds and speed up inpatient recovery.

What does this mean for healthcare practitioners and organisations?

The platform can help staff complete all the presurgery requirements, e.g., reminding the physician to complete consent forms or issuing reminders for missing diagnostics.

Ward nurses benefit from having reliable start times and estimates for the length of surgery, which helps them better plan daily work and prioritise which patients need to be prepared, when a staff member needs to provide postoperative care or more time-consuming monitoring, or plan the admission of a new patient who has a predicted need for intensive care after surgery.

For physicians, reliable operating times reduce waiting times for the operating room (e.g., taking care of patients on the ward instead of waiting in the operating room for a delayed surgery to start).

For surgery coordinators, the software helps schedule operations and day-to-day staffing decisions through instant AI-enabled recommendations.

Such solutions can have significant impact on hospital flows. According to Qventus, its operating room solution has helped clients reduce same-day cancellations by 25 percent, and peri-acute care unit transfer delays by 23 percent, while seeing a 20 percent increase in patient satisfaction score.

To be set up for success, hospitals have to develop an adoption plan that allows for an initial transition period during which staff members can be trained on the software to reap its full potential.

What is its reach (and potential)?

There is significant need for operational improvements in hospital operations — in October 2019, 84 percent of NHS patients had to wait more than four hours in A&E to be seen, and waiting times are increasing.[23]66

AI solutions that solve operational challenges have huge potential across healthcare — from hospitals to outpatient clinics, assisted living and nursing facilities, as well as home care.


2.4.9 Strengthening healthcare innovation


AI is being applied to many pharmaceutical R&D activities although, as in clinical practice, the opportunities identified are often far ahead of the impact on the ground.

Early applications include disease state and target understanding, lead selection and optimisation, clinical dose and endpoint selection, therapeutic tailoring and portfolio management.

Applications in development, regulatory and safety support include protocol optimisation, adaptive development plans, trial planning and execution, portfolio management and active safety surveillance.[24]68

Startups such as Recursion Pharmaceuticals and BenevolentAI are innovating, while big pharmaceutical and technology players are focused on realising opportunities from AI.

Big Pharma is also making major investments and partnerships to address opportunities.

In 2019, Novartis and Microsoft announced a partnership to apply AI to developing personalised therapies for macular degeneration, cell and gene therapy and drug design; Bristol-Myers Squibb entered a multiyear strategic agreement with Concerto HealthAI to use machine learning to help design protocols for precision treatment; and AstraZeneca announced partnerships with BenevolentAI and Schrödinger to accelerate drug discovery using machine learning.[25]69

However, even as capital-rich global organisations invest heavily in data and capabilities to address priority applications, pharma companies have found it hard to realise the promise of AI.

Novartis CEO, Vas Narasimhan, has reflected that beyond applications in clinical trial operations and finance there is “a lot of talk and very little in terms of actual delivery of impact” from machine learning and AI.[26]70

One major challenge is data — although pharma companies have a lot of data, they are often poorly suited to AI due to quality issues, inconsistent formats or the challenges of linking data and obtaining the necessary consent to use in different use cases.

As a result, major investments are now taking a step back, focusing on developing distinctive data assets and consistent data formats that will enable future applications.

Sanofi’s DARWIN platform, for example, applies AI to anonymised data from the records of 450 million patients to accelerate and deepen insights on treatment effectiveness, safety and value.[27]71

Roche’s Navify Tumour Board solution, which curates and presents data to promote collaboration and accelerate workflow in tumour boards, also highlights the potential for structuring data in a way that can improve the efficiency of healthcare delivery to develop datasets with consistent formats of high value for research.[28]72

Overall, AI is now applied in different elements of the business system in the pharmaceutical and medtech industries in order to increase the speed to market of new products, reduce costs, enhance clinical outcomes and serve a variety of organisational goals.

“The digital revolution in healthcare provides new ways to both collect high-quality data from each patient and connect it to data from large pools of patients for analysis with artificial intelligence-based algorithms. 
 
 This enables us to arrive at a deeper understanding of how to treat an individual… Real-world evidence, molecular information generated from next-generation sequencing, data from wearable devices and mobile apps, and novel clinical trials are transforming the future of care.”
[29]57


This is an excerpt of the report “ Transforming healthcare with AI. Impact on the workforce and organisations; EIT Health and McKinsey & Company, (2020). Survey of 175 healthcare professionals, health investors and AI start-up founders and executives

References


[1]
European Society of Cardiology (ESC), Press release, 6 June 2019, https://www.escardio.org/The-ESC/Press-Office/ Press-releases/Atrial-fibrillation-set-to-affect-more-than-14-million-over-65s-in-the-EU-by-2060.

[2] Heaven, D. “An algorithm that can spot cause and effect could supercharge medical AI”, MIT Technology Review, February 5, 2020, https://www.technologyreview.com/s/615141/an-algorithm-that-can-spot-cause-and-effect-couldsupercharge-medical-ai/.

[3] Smith, N., et al., “AI-Driven Automation in a Human-Centered Cyber World”, In 2018 IEEE International Conference on Systems, Man, and Cybernetics, October 2018, (SMC) (pp. 3255–3260). IEEE.

[4] https://www.khealth.ai/post/how-k-delivers-free-personalized-healthcare-information.

[5] Ping An Good Doctor “Ping An Good Doctor Launches Commercial Operation Of One-Minute Clinics And Signs Contracts For Nearly 1,000 Units, Serving More Than 3 Million Users”, January 4, 2019, http://www.pahtg.com/en/news/ in-the-news/ping-an-good-doctor-launches-commercial-operation-of-one-minute-clinics-and-signs-contracts-for-nearly-1- 000-units-serving-more-than-3-million-users/.

[6] “Ping An Good Doctor has become the first online healthcare platform with more than 300 million registered users”, PR newswire, September 2019, https://www.prnewswire.com/news-releases/ping-an-good-doctor-has-become-the-firstonline-healthcare-platform-with-more-than-300-million-registered-users-300923026.html; https://www.babylonhealth. com/blog/business/10-things-we-did-in-2019/; https://ada.com/milestones/; https://www.mediktor.com/en-us; https:// www.khealth.ai/post/k-health-two-million-users-and-just-getting-started.

[7] McKinney, S.M., Sieniek, M., Godbole, V. et al. “International evaluation of an AI system for breast cancer screening”. Nature, 2020, 577, 89–94.

[8] See: Healthcare resource statistics — technical resources and medical technology”, Eurostat, Statistics Explained https:// ec.europa.eu/eurostat/statistics-explained/pdfscache/37388.pdf, https://ec.europa.eu/eurostat/statistics-explained/ pdfscache/37388.pdf.

[9] https://www.marketresearchhighlights.org/industry-reports/blood-testing-market/.

[10] DeepMind blog “Using AI to give doctors a 48-hour head start on life-threatening illness”, July 31, 2019, https://deepmind.com/blog/article/predicting-patient-deterioration; Tomašev N., et al., “A clinically applicable approach to continuous prediction of future acute kidney injury”. Nature, 2019 572(7767) pp116–9.

[11] “How IBM Watson overpromised and underdelivered on AI health care”, IEEE Spectrum April 2, 2019, https://spectrum. ieee.org/biomedical/diagnostics/how-ibm-watson-overpromised-and-underdelivered-on-ai-health-care.

[12] “Moorfields and DeepMind Health research partnership latest update”, November 2018, https://www.moorfields.nhs.uk/ node/2558; “Google-backed DeepMind Creates Deep Learning CDS for Eye Diseases”, Health Analytics, 14 August 2018, https://healthitanalytics.com/news/google-backed-deepmind-creates-deep-learning-cds-for-eye-diseases.

[13] https://digital.nhs.uk/data-and-information/virtual-data-assistant.

[14] https://www.diabetes.co.uk/diabetes-prevalence.html (accessed January 8, 2020).

[15] https://ec.europa.eu/eurostat/cache/infographs/elderly/index.html (accessed January 13, 2020).

[16] “Elderly care and housing demand in the EU — Golden opportunities, but mind the cultural gap“, ING Economics Department, May 2019, https://think.ing.com/uploads/reports/ING_-_Elderly_care_and_housing_demand_in_the_EU_-_ May_2019_1.pdf.

[17] Yamada, M., et al, “Development of a real-time endoscopic image diagnosis support system using deep learning technology in colonoscopy”, Scientific reports, 2019, 9(1), pp.1–9; https://www.bowelcanceruk.org.uk/news-and-blogs/ campaigns-and-policy-blog/the-future-of-artificial-intelligence-(ai)-in-the-nhs/.

[18] “Icahn School of Medicine at Mount Sinai to establish world-class center for artificial intelligence”, Newswise, June 11, 2019, https://www.newswise.com/articles/icahn-school-of-medicine-at-mount-sinai-to-establish-world-class-center-forartificial-intelligence-hamilton-and-amabel-james-center-for-artificial-intelligence-and-human-health.

[19] “Emergency hospital admissions in England”, The Health Foundation, May 2018, https://www.health.org.uk/ publications/emergency-hospital-admissions-in-england-which-may-be-avoidable-and-how.

[20] Causes of death-standardised death rate, EU-28, 2016, Eurostat, http://ec.europa.eu/eurostat/product?code=hlth_cd_ asdr2&language=en&mode=view; https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Causes_of_death_ statistics (Accessed January 13, 2020).

[21] Kalis, B. et al., “10 promising applications of AI in healthcare”, Harvard Business Review, May 10, 2018, https://hbr. org/2018/05/10-promising-ai-applications-in-health-care.

[22] Published in zu Putlitz, Jasper (ed.) et al., The future of Medicine: disruptive innovations revolutionise medicine and health, (in German), Medizinisch Wissenschaftliche Verlagsgesellschaft, May 2019.

[23] “NHS performance and waiting times”, The Health Foundation, 22 November 2019, https://www.health.org.uk/newsand-comment/blogs/nhs-performance-and-waiting-times.

[24] Darino, L., Knepp, A., Mills, N., and Tinkoff, D., “How pharma can accelerate business impact from advanced analytics”, McKinsey & Company, January, 2018, https://www.mckinsey.com/industries/pharmaceuticals-and-medical-products/ourinsights/how-pharma-can-accelerate-business-impact-from-advanced-analytics.

[25] https://www.novartis.com/news/novartis-and-microsoft-announce-collaboration-transform-medicine-artificialintelligence; “BMS signs deal with Concerto Health AI”, Pharma Times, March 29, 2019, http://www.pharmatimes. com/news/bms_signs_deal_with_concerto_healthai_1282903; “AZ to accelerate drug discovery with BenevolentAI partnership”, Pharma Times, April 30, 2019, http://www.pharmatimes.com/news/az_to_accelerate_drug_discovery_with_ benevolentai_partnership_1286156.

[26] David Shaywitz, “Novartis CEO who wanted to bring tech into pharma now explains why it’s so hard”, Forbes, January 16, 2019, https://www.forbes.com/sites/davidshaywitz/2019/01/16/novartis-ceo-who-wanted-to-bring-tech-into-pharmanow-explains-why-its-so-hard/#402f02957fc4.

[27] https://www.sanofi.com/en/about-us/our-stories/real-world-evidence-turns-experience-into-knowledge.

[28] https://www.navify.com/tumorboard/.

[29] https://ec.europa.eu/eurostat/cache/infographs/elderly/index.html (accessed January 13, 2020).

Total
0
Shares
Deixe um comentário

O seu endereço de e-mail não será publicado. Campos obrigatórios são marcados com *

Related Posts

Subscribe

PortugueseSpanishEnglish
Total
0
Share