AI-First Healthcare

Chapter 4. Digital Transformation and AI

Healthcare automation plays a key role in healthcare digital transformation by enabling stakeholders to focus on patient care and drive outcomes that make the healthcare system work for everyone. Complex healthcare processes and systems make digital transformation challenging. We define healthcare automation as the use of technology or machines to take on tasks so as to free people to focus on patient care and the health of populations. Digitization is defined as simply converting nondigital data (e.g., faxes or voice) into digital data, which allows opportunities for technology and machines to increase automation. The pandemic shows the value of healthcare automation where resources are limited and AI and other technologies can facilitate improving efficiencies and access to healthcare services. We define digital healthcare transformation as embracing both healthcare automation and digitization. It is a broad term that encompasses everything from implementing new technologies in healthcare operations to increasing digitization or new business models. It can also refer to modest activities such as launching a new website or a new mobile app.

Optimizing the impact of technology and improving the entire breadth of organization workflows and processes while improving access to data should be the goal for digital healthcare transformation. Implementing automation-based technology in silos across the organization may be a positive step forward, but reducing redundant applications or eliminating process inefficiencies often produces a far greater impact. Digital healthcare transformation entails increasing self-service technologies for patients, clinicians, and all stakeholders. The implementation of digital tools that make information and data more readily accessible should be a goal. Several studies suggest organizations with successful transformations deploy more technologies than other organizations do.¹ Increasingly, AI as a general purpose technology (GPT) will be a key enabling technology for healthcare digital transformation.

The opportunity presented by artificial intelligence, machine learning, natural language processing, and computer vision to make healthcare better means increasing and improving the automation and digitization of healthcare. AI solutions exist today that help reduce the time physicians spend taking notes, making patient-doctor interactions efficient and helping patients recall the key points of physician-patient interactions. These are examples of giving day-to-day tools a digital upgrade. AI enables clinical documentation that writes itself. Even the processing of patients’ voicemail messages for clinicians can be automated, enabling clinicians to take faster action on behalf of their patients.

The large incumbent healthcare companies will lead the way, with a big assist from technology companies. Many technology companies, start-ups, and healthcare companies are in a race to lower the cost of healthcare, improve experiences, provide better diagnostic tools, and reduce friction in the systems. Successful digital healthcare transformation is not all about implementing new technologies or implementing AI. However, digitization and AI are interlocked; the full value of digital healthcare transformation cannot be achieved without unlocking the value of AI.

Agreement on a definition of what digitization is remains as elusive as a definition of artificial intelligence in organizations. However, organizations must have a common definition of both to achieve digital healthcare transformation. In decades past, everyone agreed that moving from analog to binary, from paper to computer readable, or from vinyl records to CDs was digitization. Now we view streaming videos, not CDs, as digital. In the early 1990s, we viewed chess-playing computers as AI, but today we see them as a computational task. Narrow definitions of digital and AI don’t serve us anymore. Companies born in the AI era see the internet, mobile, and cloud as table stakes underlying their platforms, systems, and processes. They don’t debate what it means to be digital or AI because the two are interlocked and are a natural part of their ethos.

Healthcare digital transformation gets accomplished through the pursuit of two tracks, as illustrated in Figure 4-1. Digital healthcare transformation must better the way clinicians get their work done; this is people-centric digitization. Computers must fade into the background, and the applications or systems used by clinicians must become more intelligent and have less friction and easier interfaces; this is application-centric digitization.

Figure 4-1. People-centric and application-centric digital transformation

People-centric digitization focuses attention on the experiences, workflow, and outcomes of the stakeholders involved in the patient journey. For clinicians and patients, the nirvana state or goal is precision medicine, healthcare tailored to one’s specific needs. Enabling doctors to focus on prevention and early diagnosis changes the healthcare outcomes for patients. Patients will be able to get their most basic questions about benefits or insurance coverage answered on a moment’s notice. Your medical history being securely stored on your mobile device, your dermatologist connecting with you over video chat to review skin issues, and prescriptions being delivered to your doorstep are examples of people-centric digitization.

Detecting bed moves and falls and alerting caregivers to take immediate action can be achieved through applications or systems that sit silently in the background and detect movements using computer vision and machine learning. This is application-centric digitization.

People-centric digitization recognizes this shift and consumers’ expectations for novel, intuitive people-machine interactions. We saw this in banking, where consumers’ desire to exchange money with anyone in the world, at any place and at any time, gave rise to mobile digital payment services. Incumbent banks were dragged into mobile banking and mobile payment services because of the needs and demands of their consumers, who started using apps from start-ups providing services previously unavailable from their financial institutions. The same will most likely be the case with healthcare unless an incumbent seizes the opportunity for people- and application-centric digitization.

Digital Transformation of Healthcare

Digital transformation of healthcare is a broad area addressing a wide swath of domains, such as patient care, health and wellness, diagnostics, decision support systems, hospital and provider systems, electronic medical record (EMR) systems, triage, back-office administrative systems, home care, urgent care, emergency rooms, chronic care management, mental health, and more. Digital transformation means using technology to provide real-time, immediate outcomes, whether in approving an insurance claim, approving prescription coverage, or performing a diagnostic. Digital transformation eliminates the need for phone calls by making information readily and instantly accessible, whether you’re trying to determine your benefits, get your lab test results, or get approval for a prescription, a test, or surgery. It means substantially reducing misdiagnoses and medical errors thanks to highly interactive decision support systems at play in doctor’s offices or hospital settings. Digital transformation means price transparency, so that patients know the cost of a service at the time of inquiry and delivery. The pandemic shows telemedicine becoming a more important experience for patients and clinicians, taking our traditional in-person practice into the digital world. The number of use cases is limitless, but let’s explore a few through the lens of three paths.

Three distinct paths (see Figure 4-2) should be pursued by healthcare incumbents and by every organization tasked with making healthcare better, faster, and more accessible and with increasing its efficacy.

Figure 4-2. Paths to the digital transformation of healthcare

All three paths require employing transitional and game-changing technologies like artificial intelligence. AI enables digital capabilities previously unimagined. For example, we can employ AI to help people adhere to their medications, to use a person’s voice as a way to search for healthcare services, to evaluate a patient’s specific conditions to determine whether they would be better served by readmission to hospital A versus hospital B, or simply to determine that a patient’s best healthcare outcome will be achieved with a specific provider organization. Al allows us to use empirical evidence for where patients will obtain the best outcomes rather than relying on anecdotal comments on social media sites ranking providers. Many of these capabilities cannot be achieved without the application of AI, and this is only a small, partial list.

Organizations that fail to apply the right AI techniques in deep learning, machine learning, computer vision, and natural language processing (to name a few) will not achieve optimal digital healthcare transformation. Often AI alone will not be sufficient. For example, using technologies such as graph databases instead of other database technologies enables organizations to see relationships among data. Seeing the intersection of patients, providers, and clinicians allows organizations to create patient journeys, so that any authorized user can see the steps an individual patient makes as they interact with the healthcare system. Graph and other technologies are discussed in State of Healthcare Technology by Kerrie L. Holley et al. (O’Reilly); this ebook provides a summary of healthcare technologies required for optimizing healthcare and accelerating digital healthcare transformation.

Next, we’ll discuss the three different paths to the digital transformation of healthcare.

Path A: Creating Digital Operations and Processes

Path A creates and integrates digital operations and processes delivering on the customer value propositions. Customers include the ecosystem in healthcare, including patient consumers, clinicians, hospitals, providers, health services companies, and more. In this path, organizations focus on changing existing processes and their legacy systems supporting those processes to a digital state. Path A concentrates mainly on transitioning from the as-is state to a future state of current operations and processes. That is, you take existing products, systems used for ongoing operations, or legacy systems and focus on how to optimize, increase automation, or apply AI. Undertaking any or all of these three actions improves the path toward increased digitization.

Healthcare digitization is held hostage to several IT applications and systems. For example, electronic medical record (EMR) or electronic health record (EHR) systems provide digitized records of a patient’s healthcare encounters. Digitized records of a patient’s healthcare encounters were state of the art at one time, but not anymore. Today, time spent on EHR systems reduces the amount of time physicians spend with their patients, negatively affecting patient relationships. Storing data is not a clinical tool; it doesn’t provide healthcare. A 2018 Harris Poll conducted on behalf of Stanford Medicine delivered a treasure trove of AI use cases for EHRs: disease diagnosis, disease prevention, and population health management. In this study, 9 out of 10 physicians wanted their EHRs to be intuitive and responsive—a perfect storm for AI adoption.

Currently, physicians type up encounters or enter them into systems on tablets. Providers have to click on different sections of a record to access useful data such as previous medical notes or scanned documents, where a more complete clinical picture of the patient is found. The ICD-10 codes that are used for medical billing are also used in EHRs to identify diagnoses, but a patient’s coded diagnosis may not reveal the true issue with the patient. For example, if someone has chosen a code for congestive heart failure (CHF), then the doctor understands that this is a current condition of the patient, but the doctor does not know where the heart is failing, which is usually delineated by the terms diastolic or systolic CHF. That information also helps to determine the underlying cause of the CHF. These are all-important factors that are not always captured in one code.

A more intuitive and responsive system would simulate what paper charts did in the past. A clear description, often annotated by the physician, would be appended and carried through in the chart or would be immediately accessible for review to give the provider a clearer picture of the patient’s medical issue. Once the diagnosis and its implications are clear, then bigger issues such as disease prevention in populations can be addressed. So for diastolic CHF, a physician could look for predisposing sleep apnea or hypertension and address these conditions to prevent CHF from developing. Improving code accuracy has been addressed by ICD-10 but is not enough in itself to fully resolve the issue. Short-term improvements to date include voice as a modality acting as a scribe during patient visits, a highly viable AI engineering effort using deep learning and natural language processing.

In many cases, legacy systems must address poor system availability and stop the mindset that systems that read and write data cannot have the reliability and availability seen in the platforms of the large technology companies, such as Google’s search engine, Amazon’s commerce site, or Netflix’s streaming videos. Historically, incumbent companies spend the lion’s share of their budget on customer-visible capabilities, unlike digital-era companies, who make no such choice. Digital-era companies treat performance and availability as first-class constructs, the same as consumer-visible features and functions.

Path B: Building New Capabilities

While Path A focuses on improving existing products, services, systems, and capabilities, Path B is thinking up and deploying new capabilities. For example, clinical coding, as we discussed earlier, takes information about a specific patient case and assigns standard classification codes. Codes for heart disease, diabetes, and other conditions help with medical informatics and research in a variety of applications. Amazon created a new product, AWS Comprehend, to automate this activity using deep learning. This is classic Path B: building a new capability. Arguably, after creating enough new capabilities, Amazon becomes a competitor in healthcare, just as it moved from commerce to technology with the cloud. If an incumbent with a clinical coding product were to decide to improve its product using AI (i.e., deep learning), that would be pursuing Path A—making an existing product better using AI.

This example of clinical coding² highlights the variation in what is meant by digital healthcare transformation. Each improvement over time increases the automation of clinical coding and thereby improves digital healthcare transformation. That is, clinical coding can be automated or made digital in a variety of ways, leading to computer-assisted coding systems that increasingly make this an easier task for coders. Some clinical coding systems are superior to/more digital than others because they apply AI in the form of natural language processing. Clinical coding systems that use deep learning NLP are more accurate than those using NLP without deep learning. The application of deep learning NLP to coding systems would help solve the issue raised in the CHF example discussed earlier—hence the interlock of digital healthcare transformation and AI. Perhaps another way to approach this scenario is not to label a solution as digital or more digital but simply to tackle every problem with the intent to maximize the use of AI because both the old and the new reflect automation implementations.

Similar to how we see video streaming and not CDs as the optimal digital experience today, using deep learning NLP for clinical coding rather than just NLP is a more optimal digital healthcare transformation service. The deep learning NLP service over time will get better and faster than an NLP without deep learning implementation. This is what is meant by maximizing AI. In the future, additional AI research may require a different choice beyond deep learning.

Path B builds a new set of capabilities around the desired customer end state and operating model. Path B means thinking about how to use a variety of technologies (e.g., AI, IoT, graph, ambient, augmented reality, and more) to create new products and services. Path B enables enhanced healthcare experiences by defining even better experiences and healthcare outcomes.

Path C: Transforming Business Processes

Path C transforms business processes and often requires taking an enterprise view and eliminating duplicative processing or else imagining a new process. This path is fraught with risk and high rewards. Path C includes moonshot projects, or as Safi Bahcall calls them, loonshots—novel and breakthrough ideas that can cure diseases or transform industries. Real-time healthcare is a moonshot because of the technical, structural, and cultural divisions preventing interoperability and data exchange among all stakeholders involved in patient care. Path C may mean acquiring new businesses, i.e., diversification. Path C focuses on the transformation of culture, systems, or business processes.

Paths to the Digital Transformation of Healthcare

Figure 4-2 highlights three pathways to digital healthcare transformation, but these pathways are not a digitization strategy. Organizations should develop and realize a digital healthcare transformation strategy that addresses one or more of the following aspects:

• Digital healthcare transformation requires organizations to provide a strategy for achieving their transformation that is focused on problem areas or pain points at which applying AI would make a difference.

• Cultural transformation means that organization models and talent must be aligned with the goals for digital healthcare transformation and must focus on data liberation versus silos.

• Technology adoption requires decisions on how and when business areas will apply AI and accompanying technologies.

• Business process transformation means thinking about what processes need to be reengineered or reimagined.

• Portfolio diversification suggests thinking about digital divestiture, which may extend digital transformation outside the walls of the organization.

• Business model innovation means that the organization must decide if it will change its underlying customer value propositions and/or the business operating model.

There is no single playbook for digital healthcare transformation. David Rogers asserts that digital transformation is about strategic thinking, not a better technology stack. Interestingly, the same can be said for AI—that is, organizations’ success with digitization and AI depends on how companies understand, define, and think about it. What works for one organization may not work for another. The next section explores what digital healthcare looks like through examples and use cases.

Digital Healthcare

In general, digital healthcare applies technology, using the data acquired through technology to improve our health and wellness. This technology includes apps, wearable devices, remote or ambient monitoring devices, telemedicine, health-related email, and electronic health records to incorporate a patient’s related data into their health management.

The benefits are apparent. For example, if a patient is diagnosed with hypertension, doctors and other caregivers will have real-time or near real-time data on how their blood pressure management is working. This scenario will allow more immediate adjustments or micro-adjustments to the patient’s care that would not be possible in our current healthcare model. Typically a patient would start treatment with medication or a lifestyle change and then return to their doctor in several weeks’ time to have their blood pressure rechecked. Figure 4-3 illustrates the effect of digitization where digital care is continuous: monitoring the patient’s blood pressure, doctor check-in, and suggesting lifestyle changes.

Figure 4-3. Hypertension patient’s timeline of care

Micro-adjustments result in improved care for the patient. The doctor will be able to tell if the prescribed treatment worked or if further adjustments are warranted. The less time the patient’s blood pressure is out of control, the safer their kidneys and other vital organs are from the damage resulting from uncontrolled hypertension. The doctor has more tools and information to help manage the patient’s condition, making the doctor’s job more comfortable and efficient.

Digital tools can help pinpoint new issues by identifying medical conditions that would not be apparent otherwise (i.e., would not be captured during a doctor visit). As in the example, digital tools can help monitor the effectiveness of treatment and identify the worsening of chronic conditions. As we discussed in the last chapter, identification of medical issues allows physicians to intervene faster, either treating immediately or preventing the development of medical disease. Better management of disease symptoms occurs, improving patients’ quality of life. Patients avoid long-term complications resulting from untreated disease. For hypertension, these complications include kidney and eye injury, as well as increasing the risk for stroke. Avoiding these complications of the disease improves the patient’s quality of life and life span. It avoids healthcare costs associated with the development and management of these complications. The sicker a patient is, the higher the healthcare cost, as treating conditions takes more resources.

To take this a step further, let’s say that the doctor and patient have the hypertension well managed through medication therapy, and the blood pressure readings are consistently within a healthy range. What else is left to do?

AI Applied to Digital Healthcare

Elevating individual ownership of health through AI helps in the digital monitoring portion of digital healthcare. Digital healthcare can include digital insurance cards, digital check-ins for appointments, digital/virtual visits, and so on. AI adds another dimension to your care. Using the same hypertension example, AI now points out to the patient and their doctor that a weight decrease of 20 pounds could control the patient’s blood pressure without medication. In the office, a clinician is usually harried and is potentially running behind schedule, with numerous demands on their time. For example, physicians see patients, supervise other clinicians (such as students and physician assistants), respond to patient emails and phone calls, respond to insurance coverage issues, and address billing complaints, all while covering these same responsibilities for other physicians who are out of the office. Unfortunately, it is easy sometimes for patient care to slip between the cracks. The potential for patient care to improve exists when AI use eliminates or streamlines nonclinical tasks, thus giving doctors more time for direct patient care and reducing their burnout and overload from the stimulus of addressing so many associated patient care items throughout their day. Also, AI could be used to fill the role of safety net for the clinician, to ensure that appropriate and timely care is not accidentally overlooked.

AI can serve many functions in the clinical space. AI augments the physician, reminding them that counseling and perhaps enrollment in a health and well-being program for weight loss might help with nonmedication management of hypertension. Also, AI can be used to provide doctors with reminders about best-practice management for chronic conditions.

In the hypertension example, the patient’s blood pressure is well controlled on medication therapy. The patient starts working on healthy weight loss and engaging in regular aerobic exercise. Their connected device provides timely input to their doctor that the regimen is working well, and their blood pressure is now not only controlled but borderline low. AI is used to identify abnormally low blood pressures, taking both the data from the patient’s biometric device and input from a tablet that indicates the patient may have been feeling dizzy intermittently. AI analyzes all this information and researches the clinical database and hypertension management guidelines to form a recommendation for intervention by the doctor.

The doctor takes this packaged and easily digestible information to make an immediate decrease in the patient’s medication dosage. Over time, with close connected monitoring and AI assistance, the patient is able to stop all medication therapy. The patient’s hypertension has resolved as long as the patient continues to adhere to a healthy lifestyle.

As the patient ages, blood vessels become less compliant, genetics starts to catch up with them, and they may require medication therapy again. Over their entire life cycle with this condition, from elevated blood pressure to fully diagnosed hypertension to resolution of hypertension to chronic hypertension, AI and digital connections have been used to provide personalized care with optimal outcomes.

Many of the largest technology companies, aka Big Tech, have invested in the healthcare industry. Whether through acquisitions, startup investments, or their own products, or by democratizing AI or providing cognitive AI and cloud services for the masses, Big Tech will change healthcare. The next section provides a small sampling of the impact on healthcare.

AI, Digitization, and Big Tech

Predictably, technology companies are jumping on the digital healthcare bandwagon—look at Google, for example, with its deployment into the wearables market with Google Wear and its launching of the health platform Fit in 2014. Moreover, Google recognizes the impact of AI on digital healthcare and has created ventures that include the DeepMind unit, such as in AI being used to identify eye disease from imaging scans. Google’s parent company, Alphabet, has a health sciences arm named Verily, which is working on such exciting projects as the Aurora study, in which AI is being used to identify physical biomarkers of mental trauma. Verily created a wearable specifically for this study.

While some companies are investing in future AI with digital solutions, Apple has focused on current-day applications. It has been expanding the Apple Watch’s healthcare uses with the addition of a falls detection monitor and an ECG monitor. The corporate goal is to create other offerings for providers as well, such as Apple’s HealthKit; the Apple Watch would be the relied-upon device to connect patients to their providers for care. Amazon, in partnership with others, has created its own healthcare company, Haven, in an attempt to reimagine the healthcare market and how technology is used.

Big Tech’s interest in healthcare is expanding for a variety of reasons. Healthcare has traditionally been recession-proof, with high spend and growing demands on healthcare resources leading to the need for cost-saving strategies and innovative solutions. To date, the market appears to be focused on preventive strategies. For example, knowing the correlation between being overweight and hypertension and identifying an overweight or obese patient before the development of hypertension is a preventive strategy. Other preventive strategies using AI and digital health are being applied to chronic diseases such as diabetes and hypertension. Preventing disease is much less costly than treating illness, and we’ll explore that topic further in the next section.

Big technology companies’ penetration into healthcare is relatively low compared to the incumbent healthcare companies. Consequently, their expected impact in making healthcare operations or processes more intelligent and measurably moving the needle of digital healthcare transformation is also low unless adoption by healthcare organizations occurs. The increasingly powerful AI and cognitive cloud services provided by many Big Tech companies create enormous upside in making healthcare operations and clinical care processes more intelligent.

Preventive and Chronic Disease Management

The Centers for Disease Control and Prevention (CDC) writes that 60% of Americans have at least one chronic disease, and these diseases are a leading cause of death and disability as well as of rising healthcare costs. But most chronic diseases could be prevented, says the CDC, if we simply moved around more, ate better, and got regular health checks. We can better accomplish those goals with the help of real-time feedback loops resulting from inputs (e.g., sensors, patches, or wearables) and output signals (e.g., atrial fibrillation, glucose, and others), where intelligent machines understand health signals and provide insights into our intent and behavior. Figure 4-4 illustrates this emerging scenario of turning passive inputs into informed action.

Figure 4-4. Turning passive inputs into informed action

Today, patients and consumers increasingly engage with things that take in more signals and information about their data. For example, a consumer might receive an alert from their Apple Watch of an episode of a heart arrhythmia such as atrial fibrillation and wonder whether it’s a false alarm or a real signal. This triggers the consumer to see their doctor and upon confirmation from their clinician, a care pathway for evaluation and management is established. As another example, someone with type 1 diabetes might use patches and wearables to manage their diabetes.

Figure 4-4 illustrates a real-time feedback loop in which AI, using machine learning, turns passive inputs into informed action. Sensors or intelligent objects in the home capture real-time data on falls, rising glucose, ear health, skin health, sleep, and more. Smartphone sensors are on the rise, making continuous monitoring of consumers’ health a reality and enabling informed action.³ Consumers or patients can use targeted alerts to take action before episodic conditions arise. Phones have several sensors allowing capture of data about usage patterns, motion, humidity, temperature, biometrics, and more. Using distributed AI, AI in the phone, the cloud, or edge devices, insights can be provided that will help maintain a consumer’s physical and mental health over time.

Wearables, internet-connected medical devices, and mobile devices can help us move more and eat better and can provide signals about our health. This can occur because AI is everywhere and distributed throughout our environments. Sensors in our bodies or our homes and wearables like intelligent watches provide signals. Insights and analytics will be part of a closed feedback loop in which computers will learn from our actions and behaviors and remind us when to exercise, take medicine, see our clinician, or modify a behavior.

All of this represents new data being turned into value. These closed data loops tighten the feedback loop among data, insight, and action. Machines (e.g., sensors, watches, or computers) powered with AI capture data in real time and use AI to process the data at scale. We are maturing from just capturing and storing data to producing insights and recommendations for action.

Wearable devices can track multiple health signals, but the large amount of data and its current incompatibility with most EHRs makes their use unwieldy. Apple’s Health Kit is one tool that integrates data from wearables into the EHR.

AI and Prevention

Preventive medicine, lifestyle modification, and AI monitoring solutions abound. They range from stress management solutions to monitoring data associated with our health trackers to detect a potential risk for disease development and offering counseling on management and prevention.

Weight management, stress management, sleep, exercise, and financial support are all the major categories of AI that are in place today. Consumers/patients use this technology to assess and take ownership of their health. The majority of consumers today are comfortable with and even embrace this technology for wellness.

Not only can hypertension possibly be averted through exercise and maintaining a healthy weight and diet, but a broad range of other conditions can be prevented as well. We can use AI to monitor a patient’s sleep—and rather than the patient being required to spend the night attached to monitors at a medical office, at-home sleep studies are now the standard. AI has the capability of identifying abnormal sleep patterns or restless sleep. AI can enhance data gathered from the sleep study and help manage the sleep disorder by using other signals like weight and alcohol intake, which, if the patient is overweight and alcohol intake is elevated, may cause worsening of the sleep disorder and prevent current treatment from being effective. An app takes in manual input on alcohol consumption, and a signal from a connected scale provides information on patient weight. Each of these signals, as reflected in Figure 4-4, is part of a closed feedback loop that provides the patient and their doctor with insight into other factors that may be contributing to the sleep disorder and impacting effective management.

Of course, in this same example, a sleep disorder can also be identified early on, providing the consumer an opportunity to change their lifestyle, modify their diet with weight loss, and have immediate data available to them and their provider on the management of their possible sleep disorder. The Apple Watch ECG app that can identify potential heart rhythm disorders that predispose to stroke is the start of numerous uses of wearables using AI in preventive medicine. This naturally extends to chronic disease management once it develops.

AI and Chronic Disease

Type 2 diabetes and hypertension are two of the most common chronic medical conditions today. The CDC reports that 1 in 9 Americans have type 2 diabetes, and 1 in 3 Americans have hypertension. Each of these conditions is a significant risk factor for heart disease, which remains the leading cause of death in America. 7 out of 10 people having their first heart attack have hypertension, and people with diabetes are twice as likely to have heart disease or a stroke as compared to people without diabetes, and at an earlier age. Sixty percent of patients with hypertension also have diabetes, and the health risk increases exponentially, with the potential for long-term nerve, eye, and kidney damage.

Interestingly, diabetes and hypertension can be “silent” for a prolonged period, until a crisis is reached. People with diabetes may slowly develop increased thirst and urination and not know that this could indicate an underlying disease state. Similarly, hypertension is known as the “silent killer,” as patients frequently will present with stroke or other neurological symptoms that are later tied back to undiagnosed and untreated hypertension. Both are managed through a healthy diet and lifestyle and medication therapies.

Patients with diabetes or hypertension self-monitor their blood sugar levels and blood pressure. The data input to healthcare providers and systems through the use of connected devices can be enormous. AI health solutions can track, understand, and report results from this patient-generated healthcare data (PGHD). For example, AI has been developed in which abnormal results are sent to healthcare management systems that can then reach out and provide telephonic counseling to improve disease management.

Alternative interventions through AI include the use of apps and virtual consultants to address abnormal results. One example is teledermatology, an application of telemedicine in which photos of skin conditions are sent remotely to dermatologists for diagnostics. Teledermatology will soon evolve to where a mobile phone can be used to take the pictures. AI embedded in the phone will classify the disease state from images taken over the course of a year. A dermatologist may have told a patient that they would see them again in one year, but the AI on the phone, in examining the photos regularly taken by the patient, sees a suspect mole turning into melanoma.

Besides the use of AI to enhance chronic disease management, AI can also promote and improve self-education for the patient. For example, a patient using a mobile device or laptop can enter a query or question regarding their diabetic diet and what their allowed intake is. Patients can use a voice speaker in their house, an app on their phone, or a virtual assistant with a web application on their computer that uses AI/NLP to provide advice on a diabetic diet. Doctors can access near real-time data and perform micro-adjustments to their patients’ care, which will lead to the avoidance of complications of the disease and associated costs.

AI provides benefits to healthcare payers as well. As mentioned earlier in discussing the telephonic case management program, patient members are given customized guidance and counseling. At the same time, clinical decision support is used to guide the patient to the best next step in care action. Should they meet with their provider to discuss starting medication therapy, or should they try shifting to a low-salt diet first? Should they start a weight-loss program? Providers may be contacted for care integration. All of this leads to enhanced disease management and improved quality of life without requiring the patient to self-manage, identify, or necessarily understand the underlying connections that lead to integrated care and improvement in their disease management.

Through these AI modalities come several other benefits: healthcare costs are decreased, and quality of life is increased for patients; there is increased accessibility and application of data to improve disease outcomes, as well as decreased cost from complications associated with the unmanaged or mismanaged disease; permanent end-organ damage is mitigated by improved management; and financial loss from hospital or emergency department visits is avoided because AI allowed intervention and care before a crisis.

Our physical well-being often is easier to detect than our mental health, i.e., our psychological, emotional, and social well-being. We are increasingly able to detect mental illness using artificial intelligence.⁴

AI and Mental Health

The use of AI in disease management and prevention is not limited to diseases like hypertension, or to what are typically bucketed as clinical medical conditions. Mental illness is common in the US, where nearly one in five adults lives with a mental illness—that was 46.6 million people in 2017, per National Institutes of Health (NIH) statistics. There are apps that allow AI to check a person’s behavioral health by monitoring their smartphone and recording how many social interactions have occurred that day, including social network site visits, texts, and calls, as well as physical activity level and general smartphone use. That data is then analyzed by AI to determine whether depression or other behavioral health disorders are on the increase or are being managed. In this way, AI allows self-assessment and connection to one’s provider for near real-time monitoring of one’s mental health. Some chatbots provide mental health counseling, and there are apps for cognitive behavioral therapy. The development and use of these tools is essential, as there is a growing gap between the amount of mental healthcare needed and the number of providers who can provide this mental/behavioral healthcare.

Technology usage for mental healthcare is on the rise, and people’s comfort level is growing. The meditation app Headspace told CNBC that, due in large part to COVID-19 and the resulting pandemic in mental and behavioral healthcare, it has seen a greater than 500% increase in inbound interest from companies seeking mental healthcare. AI for mental health makes sense, as more consumers are taking responsibility for their mental healthcare. Again, the COVID-19 pandemic brought digital technology to the forefront, with people in isolation and their states shut down. Estimates show that the mental health burden rose in proportion to the prolongation of isolation associated with social distancing during the pandemic. Because the population was encouraged to stay at home and avoid social contact, digital healthcare grew. COVID-19 may inadvertently drive the increased utilization of digital technologies based on necessity. Time will tell if the trend continues, but we believe the pandemic has changed the healthcare paradigm to digital healthcare made possible with AI.

The coronavirus pandemic shows what is possible with telemedicine. Improvements in the state of “conversational” AI and everything described in Figure 4-4 will make telemedicine soar in efficacy.

AI and Telemedicine

In an ideal digitized healthcare environment, AI takes in all the data from connected devices, and that data is then processed by AI to provide quick and timely interventions in healthcare, all of which is supported by AI analysis of the most up-to-date and relevant treatment options. Connected devices facilitate remote delivery of care, or telemedicine. Telemedicine, as evidenced during the COVID-19 pandemic, has broad reach and has made billions of users aware of a new modality of healthcare. AI has a special and significant role in telemedicine.

The transformation of our healthcare system is already happening. COVID-19 was the impetus behind a rapid and exponential increase in the utilization of telemedicine. AI is being studied and has shown success in decreasing hospital readmissions.⁵ In one hospital readmission reduction program, each patient, upon discharge from the hospital, is given a WiFi-enabled device that transmits vital signs and other important patient data (heart rate, blood pressure, temperature, and so on) to their provider for ongoing outpatient management. AI is constantly monitoring this data, and at any sign of abnormality, the patient, their provider, and any other caregiver can be notified to address the finding and hopefully avert a hospital readmission or emergency department visit.

It’s all about timely intervention. An example of just such a partnership now in the market is a pilot program between NHS (National Health Systems) hospitals at Dartford and Gravesham and Current Health (formerly snap40) that is aimed at remote monitoring of patient vital signs with AI analysis. The patients in this pilot are fitted with WiFi-enabled armbands and given a chatbot-equipped tablet for medication reminders and remote communication with their providers. The patients receive all their tools prior to discharge from the hospital. In this manner, provider and care teams are able to keep a remote watch on patients, with the ability to perform micro-adjustments to care that keep their patients healthier and out of hospitals.

Telemedicine has now been widely accepted and can provide clinical support to the world of data that AI has opened. This integration provides numerous benefits:

• Early diagnosis and timely intervention

• Personalized care

• Remote patient monitoring on a real-time basis

We’ve discussed early diagnosis and treatment several times throughout the book. Again, early diagnosis allows for interventions either through changes in lifestyle, habits, or diet or via medications prior to complications of disease development and prior to development of long-term damage to the body related to uncontrolled disease. Traditionally, patients would have symptoms and no other associated information leading them to seek care at a doctor’s office. With AI, symptoms are paired with personalized patient data generated by the patient’s connected devices. Therefore, telemedicine providers have more information at the time of patient evaluation. These teledoctors have the patient to provide a description of their symptoms, data from connected devices (with AI analysis pointing out any abnormal findings), and AI-assisted analysis of clinical treatment guidelines, ensuring the most up-to-date and best care or best treatment or best next evaluation step for each patient based on their own individual data.

Personalized care via AI is possible, as all data generated from connected devices or alternate data input from other devices, such as an iPad with symptom generators and checklists, is derived from an individual patient. Thus doctors can make interventions or treatment decisions based specifically on that patient’s data. Information on the impact of each micro-adjustment is immediately available and allows for finer management based on individual patient data. In addition, AI can enhance the treatment algorithms to ensure the best quality of care and the best next step in management for each individual patient. These treatment plans are then monitored through AI to ensure adherence and control of the disease, and if any change occurs, treatments are changed accordingly.

Remote monitoring is a relatively new concept. Traditionally in a hospital, a set of vital signs is obtained on a regularly scheduled basis. If the patient is sick enough, then monitors may be placed such that real-time monitoring occurs, and alarms go off for abnormal readings, leading to clinical evaluation and possible adjustments in treatment. In the outpatient setting, there was no routine monitoring of any vital signs except in rare cases, and this was managed by the patient, who either had further appointments to have monitors placed on them for home evaluation or had to stop in at clinics or other healthcare sites for follow-up readings. AI has transformed the process in the outpatient setting, where devices are now common and use of AI has increased significantly. Today a constant stream of near real-time data is available to both the clinician and the patient. Any abnormalities are analyzed by AI so that early intervention and adjustments can be made. Timely intervention is key, as timeliness prevents long-term complications and allows for fewer intolerable side effects of treatment (since monitoring would identify, say, too low a blood pressure).

Telemedicine is also quickly evolving almost in symbiosis with digital modalities. Telemedicine (or telehealth) has emerged as an essential component of healthcare during the COVID-19 crisis. Although telemedicine administration is now widely accepted, certain related complexities must be managed for its continued success. “Having regulatory experience is imperative, in particular during this time when things are very fast-moving and fast-paced and we’re seeing changes in CMS rules and regulations almost biweekly,” says Chevon Rariy, director of the Telehealth Program at Cancer Treatment Centers of America.⁶ The Centers for Medicare & Medicaid Services (CMS) has instituted new billing and coding recommendations for telemedicine, which allows for the alignment of incentives for the provider and the ongoing growth of telehealth. Telemedicine was once called a “one-and-done” visit. Previously used as an alternative to urgent care medicine, telemedicine is now evolving into a primary care system. Several telemedicine providers are developing or have already deployed primary care provider telemedicine care platforms. Digitization provides support for telemedicine in a patient-provider relationship that is ongoing and not just a one-time visit. Through enabled connections with intelligent devices, AI has the capacity to strengthen this type of meaningful patient-doctor interaction by augmenting the information available to providers and other caregivers so that it extends beyond the telemedicine visit itself.

Specialty telemedicine providers are also emerging. Specifically, diabetes specialists, or endocrinologists, are now being utilized in several different programs to help with ongoing care and the timely micro-adjustment of insulin regimens to provide the best possible care for diabetics. AI again empowers these doctors by notifying them of abnormalities and adherence, as well as supporting doctor treatment/management decisions based on best care practices identified by AI.

Just as digitization has the capacity to enhance telehealth, so too does it have the ability to impact further aspects of the provider-patient relationship, such as medication management. Digitized AI not only can provide meaningful insights on the patient to their doctor but also can be used to understand behavior and drive medication adherence once treatment is prescribed. Instead of issuing blind “take your medication” alerts, AI can be used to find the best time and place to send an alert to take your medication, thus optimizing medication management and adherence.

Medication Management and AI

AI continues to expand its reach and use cases and is taking a leading role in improving medication adherence. Medication adherence is the likelihood that a patient will take the medication prescribed for them as directed, or at all. Why is medication adherence important? Over the past two decades, healthcare costs have been rising exponentially, and prescription medications are a large component of these costs. For example, of the $101 billion spent on diabetes in 2013, half was spent on medications. Meanwhile, studies suggest that one-third to one-half of all patients do not take their medication or do not take it properly, according to a 2017 report by the US Agency for Healthcare Research and Quality. This results in nearly 125,000 premature deaths each year and costs the nation about $290 billion in associated hospitalizations and other complications from disease events, per NEHI (Network for Excellence in Health Innovation) and the American College of Preventive Medicine.⁷

The technology response to this problem is digital therapeutics, which encourage patients to take their medications as directed. Studies are using mobile devices and sensors to capture data and provide real-time alerts on medication adherence through browsers, apps, and medical devices. Examples include smart packing and pill dispensers, wearables that provide reminders to take medications and can track medication use, tablet apps providing reminders on refills and taking medications, and even virtual pillboxes with images displaying the size and shape of pills for identification so as to avoid confusion. When connected with sensors and combined with AI, the possibilities for medication and patient management are tremendous.⁸

Medication Adherence

The Medication Event Monitoring System (MEMS) captures when a patient takes their medicine. MEMS uses a bottle cap that fits on standard prescription pill bottles and includes a tiny microprocessor that records the occurrence and time of day when the pill bottle is opened and closed.

A study supported by the NIH used a reinforcement-learning-based medication health program to focus on medication adherence. Patients were randomized using two scenarios:

1. MEMS cap on pill bottle plus text messaging reminding the patient to take the medication and AI management to determine the type and frequency of messaging

2. MEMS cap only, with no AI

The results showed improved medication adherence at three months for both study groups. Of note, medication adherence rose from a baseline of 69–80% to 84–92% with the use of AI.

Management of chronic conditions involves complex behaviors, and patients vary in their medication adherence based on these behaviors. One study of patients who had had a heart attack episode noted that one month after hospital discharge, fewer than 50% of the patients were continuing to take their daily low-dose aspirin as prescribed during the heart attack. Another study showed that of 5,000 hypertensive patients, most patients took their medications only intermittently, and 50% had stopped taking their medications without informing their provider.⁹ In the NIH study, there was greater medication adherence when the MEMS cap was accompanied by AI-augmented text reminders. Mobile health services such as text message reminders have been shown to produce up to a twofold increase in medication adherence.

Of note, the NIH study used AI to determine whether patients were nonadherent with medication and to identify what type of messaging and what frequency of texting would spur increased adherence. Within one month, AI had decreased text messages by 46%, due to it learning that more frequent notifications did not lead to increased adherence. AI was also able to tailor notifications to those patients with the greatest need for reminders, and it decreased or stopped notifications to patients who were adherent as indicated by MEMS.

Digital Medication

AI use in medication adherence is not limited to apps, sensors, and wearables facilitating patient and provider/caregiver interactions. It extends into digital medications. In 2017 Proteus Digital Health announced that the US Food and Drug Administration (FDA) had approved the world’s first digital medication, Abilify MyCite. Abilify MyCite contains a sensor that records that the patient has ingested the pill.

The way this ingestible sensor works is illustrated in Figure 4-5. In this example, a silicon sensor that is the size of a grain of salt is attached to each of the patient’s pills. The patient swallows a pill with the sensor, which gets activated by the gastric juices in the patient’s stomach.

Figure 4-5. How ingestible sensors work

The sensor sends signals identifying the medicine and the date and time that the person took the pill. Like any high-fiber food, the sensor passes through the gastrointestinal tract in several days. Signals are transmitted from the sensor to a disposable adhesive patch on the patient’s body. Several metrics are compiled from the sensor and the patient’s body, and the analytics and insights are readily available to patients, clinicians, and caregivers.

Abilify MyCite is specific to the medication Abilify, which is used in the treatment of schizophrenia and certain other forms of mental illness. This type of technology is very important, as it is well known that schizophrenics who are not medication compliant have more psychotic episodes and may require more hospitalizations and intensive management, as well as being a potential harm to themselves and others.

The application of AI to digital pills shows even more potential. Not only will medication adherence be tracked, but providers and other caregivers will be able to apply AI to data intakes from wearables and other sensors to provide additional information on whether the medication is working and whether real-time micro-adjustments to treatment regimens must be made. The potential benefits of AI are obvious.

Other ways in which AI is transforming medication management include:

• Improving medication safety (digital data in conjunction with utilization review with AI can detect medication errors)

• Predicting health risks and outcomes (Michigan is working on using patients’ medication histories along with EHRs and prescription drug monitoring programs to calculate drug overdose risk and predict the risk of overdosing from a prescribed opioid¹⁰)

• Improving the medication prior authorization process (regardless of improvements, many medications still entail duplicate data entry, delays, and rework after criteria for authorization are not met; AI can be used to extract the relevant data for the prior authorization along with digital information/data that can help augment and streamline the process of authorization by providing and importing the relevant data that may be required for approval of therapy)

Medication management shows what might be developing with digitization. As both AI and IoT advance, the art of what’s possible evolves. At the heart of these advancements are the real-time insights into how to possibly change behavior that are gained from previously discarded signals. So many opportunities to digitize are present, and perhaps no bigger opportunity exists than in administrative healthcare tasks.

AI and Digitization Applied to Administrative Tasks

In November 2016 the Harvard Business Review published an article by Vegard Kolbjørnsrud, Richard Amico, and Robert J. Thomas entitled “How Artificial Intelligence Will Redefine Management”. In the article, the authors explained that AI is expected to automate many of the administrative coordination and control tasks performed by humans and that the transitioning of labor-intensive tasks to AI just makes sense.

In previous sections we’ve described how administrative tasks continue to comprise a significant proportion of providers’ time and prevent clinicians from having the time to spend on face-to-face interactions with patients. AI can simplify humans’ lives by efficiently performing administrative tasks and giving providers and other caregivers time to focus on human-based interactions.

Already, AI has made great strides in imaging tasks, such as analyzing X-ray images, detecting cancer, and assisting doctors with clinical decision support and management of patients. Supporting clinicians with some of the “easier” tasks allows time for more human-focused interactions with patients. You often hear patients complain that their doctor saw them for five minutes and charged them X amount. With AI supporting the doctor, they will have more time to spend with their patients.

Similarly, AI can be a facilitator in population health management, where physicians look for trends in cohorts of patients of similar age and with similar disease conditions, risk factors, and so on. The advent of EHRs has propelled the digitization of healthcare data on broad populations, which AI can then be applied to. AI provides the analysis on those cohorts of patients and parses out the details on which patients need management and/or intervention in care and at what time. Patterns and predictive trends used to rely on data scientists querying healthcare datasets and on clinician review along with clinician experience to determine population health targets. Now AI can either perform this task for us or augment the current process.

Why is this important for doctors? Doctors have long relied on experience and evidence-based medicine guidelines to help facilitate management of large populations. Today AI can augment that management through the above noted analysis. From this broad overview of patient populations, AI algorithms can then analyze millions of data points and comb through the latest published research databases to quickly find relevant patterns and determine next best steps for the doctor and patient. This approach makes a big difference in costs and in patient health. Dr. Robert Pearl, a Stanford University professor, used consensus algorithms, along with oncology data entered in EHRs, and cross-walked them with the hundreds of established treatment regimens to recommend the most appropriate combination of chemotherapy medications for a patient. Furthermore, this same research and use of AI allowed the creation of a predictive model that could identify which cancer patients would end up in the ICU in the near future. AI facilitates doctors in making judgments on management that save in costs and, more importantly, impact people’s health and lives.

AI also has the potential to go beyond the human experience through the use of unsupervised learning. Unsupervised learning is a subset of machine learning that analyzes data and discovers patterns and anomalies with minimal human involvement. Unsupervised learning has the potential to uncover patterns and trends missed by clinicians based on potential clinician bias.

Last, AI can augment claims and processing as well as prior authorizations (as mentioned previously). All of this results in efficiency and time savings as well as cost savings. AI can empower transactions and data management such that patients should be able to pay for doctor visits in real time, with the claim processed and completed and the needed prior authorizations fulfilled by the time the appointment is over. This all leads to improved quality of life and care for patients and clinicians.

Summary

A lesson of digitization is that organizations cannot apply their old way of thinking to disruptive technologies like artificial intelligence. Digital natives speak the language of digital; they embody the culture of digital in platform, architecture, process, and organization. The mythical digital playbook is just that—mythical.

Digital natives see agility as having a measurable return on investment and high availability of their platforms and products as table stakes. They automate everything. They embrace failure. They constantly evolve their application code. They push new features daily. Experimentation is a way of life. This is the culture of digitization, the ethos of digitization.

Digitization is needed for healthcare, and to have the greatest impact in helping people live healthier lives, digitization must focus on each area shown in Figure 4-6.

Figure 4-6. Digitization focus areas for improving healthcare

The more digitization can be realized in diagnostics, patient care, prevention and wellness, triage and diagnosis, chronic care management, and clinical decision support, the more we raise the efficiency and effectiveness of clinicians. We make healthcare more ubiquitous and available to wider populations. This is the power of digitization and AI.

In the next chapter we address the use of AI to eliminate or minimize waste, whether that be in terms of clinicians’ time or in terms of fraud. So we spend less time on the business of healthcare and more on the back-office and operational aspects of healthcare.

¹ See “Unlocking success in digital transformations”, McKinsey & Company (survey), October 29, 2018.

² Thomas H. Davenport and Steven Miller, “The Future of Work Now—Medical Coding with AI”, Forbes, January 3, 2020.

³ See Sumit Majumder and M. Jamal Deen, “Smartphone Sensors for Health Monitoring and Diagnosis”, Sensors 19, no. 9 (2019): 2164.

⁴ “AI in psychiatry: detecting mental illness with artificial intelligence”, Health Europa, November 19, 2019.

⁵ Wenshuo Liu, “Predicting 30-day hospital readmissions using artificial neural networks with medical code embedding”, PLOS ONE 15, no. 4 (2020).

⁶ “Impact of COVID-19 on Telehealth”, American Health and Drug Benefits 13, no. 3 (2020): 125–126.

⁷ Jennifer Kim et al., “Medication Adherence: The Elephant in the Room”, U.S. Pharmacist, January 19, 2018.

⁸ Jae-Yong Chung, “Digital therapeutics and clinical pharmacology”, Translational and Clinical Pharmacology 27, no. 1 (2019): 6–11.

⁹ Steven Baroletti and Heather Dell’Orfano, “Medication Adherence in Cardiovascular Disease”, Circulation 121 (2010): 1455–1458.

¹⁰ Jesse Adam Markos, “Michigan’s Enhanced Prescription Monitoring Program and New Analytic Tools for Controlled Substances Help Protect Both Patients and and Providers”, Wachler Associates, n.d.