Healthcare is perhaps the most important sector in the U.S. economy. It is the largest: close to $4 trillion per year is spent on healthcare in the United States. It employs more people than any other industry, accounting for 11% of all American jobs. Nearly one quarter of all U.S. government spending is on healthcare.

At the same time, healthcare is the most broken sector in the U.S. economy. Healthcare costs have spiraled out of control in recent decades, from $355 per person in 1970 to $11,172 per person in 2018. Despite spending far more on healthcare per capita than any other country, the United States ranks 38th in the world in life expectancy, between Lebanon and Cuba. Access to healthcare remains worse in the U.S. than in any other developed country.

Artificial intelligence offers an unprecedented opportunity to cut this Gordian Knot and reshape the practice of healthcare. Of the many ways in which AI will transform our lives in the coming years, its impact may be more profound and far-reaching in healthcare than in any other field.

Machine learning and healthcare are in many respects uniquely well-suited for one another. At its core, much of healthcare is pattern recognition. A healthy human body and its various subsystems function in consistent, quantifiable ways. When a human organism is suffering from some affliction, it deviates from this homeostasis in ways that tend to be predictable across time and populations.

A constellation of data points—recent physical symptoms, blood pressure, genetic makeup, the bloodstream’s chemical composition, and so on—can be collected that, taken together and compared against population-level patterns, tells the definitive story of a person’s health. Similarly, the medicines that we create and prescribe consist of specifically defined substances that act upon the body’s internal systems in measurable ways.

If there is one activity at which machine learning excels, it is identifying patterns and extracting insights about complex systems given lots of data. Healthcare therefore represents an ideal challenge for AI.

In which specific areas of healthcare can we expect AI to have a significant impact? It is helpful to break down the sprawling field of healthcare into three main categories: clinical (the delivery of care to patients), administrative (the operational nuts and bolts that keep the healthcare system running), and pharma (the research and development of new medical drugs).

In each of these three areas, machine learning is already being applied in transformative ways. This will only accelerate in the years ahead.

This article will review the first two of these categories: Clinical and Administrative. The third category, Pharma, will be covered in a followup article.

Clinical

Imaging

Using computer vision to identify health conditions in medical images has become perhaps the most widely referenced use case for AI in healthcare. It is easy to understand why: examining a medical scan to determine whether a tumor, a skin lesion, a retinal disease, or some other indication is present is a clear-cut example of object classification, exactly what deep learning excels at.

As AI legend Geoff Hinton famously declared in 2016, “People should stop training radiologists now. It’s just completely obvious that within 5 years, deep learning is going to do better than radiologists.”

A number of startups have emerged over the past few years to automate analysis of medical images. Among the more notable are Caption Health, PathAI, Paige, and Zebra Medical Vision.

Yet despite the hundreds of millions of dollars of venture capital that has flowed into this category, the technology has not yet been widely adopted. It has proven challenging for AI companies to convince healthcare providers to alter their workflows to incorporate these solutions at scale—particularly given that this use case so directly threatens to render human practitioners obsolete.

Patient Intake and Engagement

Another area in which AI will improve care delivery is patient intake and engagement, a critical part of the healthcare journey.

Recent advances in natural language processing have made possible AI-based conversational interfaces that can automate patient screening and care navigation. For example, patients can share symptoms and questions via text message and receive automated clinical guidance in response. Similarly, AIs can be developed that communicate with patients on an ongoing basis to ensure that they remain engaged and compliant with their care regimen.

Using AI to automate these interactions can dramatically reduce costs and democratize access to healthcare by making expert health guidance available without the need for human physicians’ expensive time.

“Chatbots” have generated plenty of criticism and unfulfilled hype in recent years. But NLP technology is now advancing at a breathtaking rate, opening up new possibilities for conversational AI. Conversational platforms are most effective when they are purpose-built for a specific use case (e.g., patient engagement) and are designed to loop in a human when appropriate (e.g., a doctor). Expect asynchronous provider/patient communication to become increasingly automated in the years ahead.

The behemoth in this category is Babylon Health, which has raised an eye-popping $635 million, much of it from Saudi Arabia’s Public Investment Fund. Other startups building tools to automate patient intake and communication include Buoy, Gyant, Curai, and Memora.

“The entire computing revolution has unfolded without fundamentally changing how we deliver and access healthcare,” said Curai CEO Neal Khosla. “AI and NLP offer the potential to massively scale the availability of quality primary care, making it accessible to more people at lower cost. That’s our north star: a world where all 8 billion people in the world can access best-in-class primary healthcare.”

Remote Health

COVID-19 has greatly accelerated the adoption of remote health: the delivery of clinical services to patients over distance rather than in-person, using digital tools.

While the pandemic has served as a near-term catalyst, many experts believe that remote health (also called telehealth) is on its way to becoming a permanently important pillar of healthcare delivery. McKinsey estimates that up to $250 billion of healthcare spend will be virtualized in the coming years in the United States alone.

Today, telehealth often simply means a videochat with a clinician. Such remote sessions are worthwhile but rudimentary. Remote health will reach its full potential only when empowered by machine learning (and the right sensors). Several promising companies are tackling this challenge.

Eko has built a platform of proprietary sensors and machine learning algorithms that can remotely monitor patients’ cardiopulmonary vital signs for early detection of heart and lung problems. Eko’s AI is significantly more accurate at detecting heart problems than are human physicians using a stethoscope. For instance, general practitioners detect atrial fibrillation with 70-80% accuracy, while Eko’s algorithms do so with 99% accuracy.

“We are able to augment the doctor’s judgment about cardiac and pulmonary diagnoses with data analyzed from tens of thousands of past patient exams in seconds,” said Eko CEO Connor Landgraf. “These algorithms can be accessed anywhere in the world, enabling better care for patients regardless of where they are.”

Along similar lines, Aluna offers a solution that enables patients to measure their lung health from the comfort of their home using a simple spirometer. Applying machine learning to the spirometry data, Aluna monitors asthma and cystic fibrosis in real-time and flags worrisome lung conditions.

Companies like Biofourmis, Current Health, and Myia have likewise developed AI and sensor solutions that enable practitioners to examine a patient’s health at a granular level, wherever that patient is in the world. Technologies like these will increasingly blur the line between “in-clinic” examinations and day-to-day health monitoring—making healthcare more affordable and accessible in the process.

In-Hospital Care

As promising as telehealth is, there will always be medical procedures that necessitate in-person visits. AI will augment the work of human clinicians in hospitals in various ways.

As one example, Gauss Surgical uses computer vision to monitor blood loss during childbirth. Human clinicians’ visual estimation of blood loss is notoriously inaccurate, and hemorrhage is the leading preventable cause of maternal mortality. At one hospital system, Gauss’ AI solution led to a 4x increase in hemmorhage recognition and a 34% reduction in delayed bleeding interventions.

As another example, Medical Informatics is a Houston-based company that uses machine learning to monitor the well-being of patients in hospital beds by ingesting and synthesizing data from bedside monitors, ventilators, a patient’s EMR, and various other data sources.

Even if AI solutions like these never replace human clinical decision-making and serve only as supplemental tools, they offer the potential to dramatically improve health outcomes and save lives.

Precision Medicine

In a way, precision medicine represents the pinnacle of AI’s promise to improve human health. The vision of precision medicine is more ambitious, the technical challenges more complex, and the potential impact greater than perhaps any other application discussed here.

In a nutshell, the field of precision medicine aspires to create treatments that are individualized for each patient based on his or her particular genetic, environmental and behavioral context.

Precision medicine is not a new concept, but the advent of “big data” (especially genetic data) and modern machine learning have brought its full realization within reach. Due to the proliferation of sensors, internet-connected devices, EHRs, mass-market gene sequencing, cloud computing, and other digital technologies, staggering amounts of highly detailed health data are now collected every day. Several trillion gigabytes of health data will be generated this year, a figure that would have been unimaginable a few short years ago.

The premise of precision medicine is that if a computational system knows your entire genome, your metabolic profile, your microbiome composition, what foods you eat, how often you exercise, how much you sleep, and a thousand other data points about you; and it also understands a disease’s particular pathway in your body down to the molecular level; then it can synthesize all this information and craft a pharmaceutical and/or behavioral regimen specifically tailored to optimize your body’s response.

No human could ever perform such herculean data crunching and latent pattern recognition. For the first time, AI makes it possible—at least in theory.

The most prominent company pursuing this lofty vision of AI-powered precision medicine is Tempus. Tempus has raised a whopping $620 million from investors including NEA and T. Rowe Price. The company is focused on cancer treatment, although it has recently devoted resources to the fight against COVID-19.

Other well-funded companies in this category include Syapse and GNS Healthcare.

Precision medicine has for decades stood as a tantalizing but unfulfilled possibility. Time will tell whether AI is the key that can unlock its vast potential.

Administrative

Compared to clinical or life sciences use cases, applying AI to the administrative side of healthcare may seem unglamorous. But an enormous opportunity for value creation exists here.

As anyone who has dealt with the healthcare system knows, it is plagued by waste and inefficiency. Over $600 billion per year is spent on healthcare administration and billing in the U.S. alone. Many billions of dollars of value will be unlocked in the years ahead by rationalizing and streamlining healthcare operations. AI can play a key role here.

Provider Operations

Every time a patient interacts with a healthcare provider, dozens of support processes take place in the background: patient check-in, benefit and verification discovery, claims processing, invoicing, prescription orders, supply chain management, and more. The way this work gets done today is manual and error-prone.

A promising set of companies is applying machine learning to automate many of these rote tasks. Perhaps the buzziest is Olive, which has raised $125 million from investors including General Catalyst and Khosla Ventures. Notable Health is a newer competitor with a similar mission. Leveraging software bots and computer vision, these companies’ solutions can be thought of as robotic process automation (RPA) built specifically for healthcare use cases.

One administrative function that is especially challenging and important for healthcare providers is revenue cycle management (RCM). RCM refers to the set of processes that providers use to track and collect payments for services rendered to patients.

Because of the U.S.’ third-party payer structure and labyrinthine reimbursement system, money flows in the healthcare system are complex. Numerous stakeholders are involved: providers, patients, private insurers, government agencies, employers. Clerical errors and costly delays are rampant. An estimated $21.3 billion was spent on RCM in 2017 in the U.S. alone.

There is a massive opportunity for AI to systematize and automate revenue cycle management, making it faster, cheaper, and more accurate. One promising startup focused on RCM is Alpha Health.

Another major administrative challenge for providers is hospital patient flows and resource allocation. Hospitals are complex systems, with patients and clinicians constantly moving through their units in dynamic ways. Hospitals’ razor-thin margins depend on orchestrating these flows efficiently. Yet at present, they are dramatically underoptimized. Up to 25% of ICU patient-days are unnecessary; an estimated 15% of total hospital occupancy is wasted due to ineffective flow management.

This is exactly the type of data-rich optimization problem at which AI excels.

Qventus is one company applying AI to drive better operational outcomes for hospitals. The company claims its technology has enabled hospitals to achieve a 30% reduction in excess days spent in hospital, a 20% decrease in ER door-to-doctor times, and a 0.8 day reduction in average length-of-stay. These operational efficiencies translate to better patient experiences and significant improvements to health systems’ bottom lines.

Finally, AI can unlock massive gains by automating regulatory compliance. Healthcare is, for good reason, one of the most heavily regulated of all industries. It is challenging and expensive for healthcare providers to keep track of and ensure adherence to the many legislative and regulatory requirements to which they are bound.

Two important areas are patient data privacy and controlled substances management. In both cases, machine learning can play a key role automating compliance activities like violation detection and auditing—thus protecting patients, bringing down costs, and allowing clinicians to focus their energy on care delivery. One company worth watching in this category is Protenus.

Data Infrastructure

One of the foundational challenges standing in the way of a better healthcare system is its deeply fragmented data landscape. Stringent regulatory restrictions, archaic software architectures, and misaligned stakeholder incentives all undermine valuable data sharing and collaboration. It is prohibitively difficult today to assemble a complete picture of a single patient’s health, a new treatment’s efficacy, or a population’s health patterns.

Data silos in healthcare are not just a bureaucratic burden. They hold back progress in medical research and impede delivery of the right care to the right patients at the right time, ultimately costing lives.

This is a sprawling, multi-dimensional challenge. A number of interesting companies are tackling different pieces of it: Komodo Health, Datavant, Abacus Insights, HealthVerity, Kyruus, Ribbon Health, and Redox, among others. These companies have differing product and go-to-market focuses but share the overarching vision of breaking down data barriers in order to drive better health outcomes.

Because machine learning thrives on large datasets, these solutions are also laying the groundwork for boundless future AI innovation. A more integrated data ecosystem will make possible countless new AI applications in healthcare, most of which have not yet even been imagined.

Medical Documentation

One last administrative area in which AI is poised to generate massive value is medical documentation. Recording notes from patient encounters consumes a significant portion of clinicians’ working lives. In the era of electronic health records (EHRs), this has become a real problem.

A recent AMA study found that the average on-duty physician spends 5.9 hours per day directly engaged with EHRs. There is widespread concern in the medical profession that, with computers in every examination room today, doctors must be so oriented toward their keyboards during patient visits that they are not able to connect fully with patients.

Machine learning can take over much of this administrative burden from physicians, allowing them to spend more time with patients and less time with screens.

The core technologies underlying natural language processing and speech recognition have improved dramatically over the past several years, as anyone who uses Alexa or Siri can attest. This has enabled the development of voice-based solutions (“AI scribes”) that physicians can verbally dictate to in lieu of manual EHR data entry. These solutions can be built to automatically integrate with existing workflows and with EHR software like Epic or Cerner.

The efficiency gains can be tremendous. Augmedix claims that its voice-based solution saves clinicians 2 to 3 hours per day. Suki, another competitor in this category, says that its AI generates 100% accurate notes and enables doctors to finish their notes 76% faster. Scaled across entire health systems, the cumulative impact of these technologies will be enormous.

Today, these solutions still require humans in the loop for quality control; NLP and speech technologies, while impressive, remain imperfect. As the underlying AI continues to improve, less and less human intermediation will be required—translating to even greater productivity gains and cost savings.

Conclusion

Healthcare is an intimate part of our personal and family lives in a way that no other sector of the economy is. It is therefore particularly troubling how dysfunctional the healthcare system is today.

No technology can be a silver bullet for a system as complex as modern healthcare. Yet artificial intelligence, perhaps more than any other force in the world, offers the potential to rewrite the rules of the game. When deployed thoughtfully, AI can upend long-accepted constraints and assumptions about how the healthcare system works. It can redefine the relationship between cost, accessibility and quality—one that today is badly broken.

There has never been a more exciting time to be an entrepreneur in healthcare.

Stay tuned for Part II of this article.

Note: The author is a Principal at Highland Capital Partners, which is an investor in Kyruus.