Healthcare Innovation Drivers: Artificial Intelligence and Machine Learning

We live in an era of astounding healthcare innovation. Within living memory, we’ve seen the first cardiac pacemaker implanted and the first successful heart transplant. Measles, mumps, rubella, and chickenpox were, for all intents and purposes, eradicated in the US through public health vaccination campaigns. CT scans are commonplace. And, the human genome has been fully-sequenced.

The net result of all this innovation is longer life spans worldwide, going from an expected life span of 48 years in 1950 to 71.5 years in 2014.

Much of today’s healthcare innovation is focused on technological advances. With the advent of economical and ubiquitous computing power, it appears that the pace of healthcare innovation won’t slow down anytime soon.

Currently in the technological landscape artificial intelligence/machine learning (AI/ML) promise to have widespread impact on healthcare innovation.


Artificial Intelligence and Machine Learning Defined

It’s hard to clearly sum up just what artificial Intelligence (AI) or machine learning (ML) are. There’s no consensus on a single, clear definition for either term. Often these terms are used with an “I’ll know it when I see it” subtext.

The first thing to understand is that AI is not a single technology or product. AI, broadly speaking, is when a machine demonstrates some kind of intelligent ability that is normally attributed to a human. For example, when a computer plays chess, it’s demonstrating a form of AI; it has learned the rules, strategy, and tactics of chess and uses this knowledge to decide which move to make in the game.

In the context of healthcare innovation, AI is software that applies mathematical algorithms to data in order to come to a conclusion. One healthcare AI example is using software to read a scan and identify abnormalities. The computing power behind AI/ML enables this software to collect and analyze massive amounts of data quickly.

ML is when a machine, most often some kind of computer, has the ability to learn from the data it processes and does not have to be explicitly reprogrammed to improve or change its process. In the context of healthcare innovation, ML applies an algorithm to a set of data and extracts more and more exacting models of that data with each iteration. One example of this in healthcare is the use of ML to iterate the modeling of a potential drug molecule.

Part of what makes it so hard to settle on a definition for AI or ML is that these concepts are developing and evolving continually. The characteristics of the algorithms underpinning AI today are dramatically different from what was common in the 1980s and 1990s.

The algorithms behind AI continue to develop and evolve, making it hard to settle on a single definition for AI – Source


AI and ML are often spoken about together as the technology behind many of today’s healthcare innovations. AI/ML is expected to have an impact on healthcare research, care delivery, and administration.

Examples of AI/ML applications in healthcare include:

  • Disease identification and diagnosis
  • Personalized treatments
  • Drug discovery and development
  • Clinical trial research
  • Smart electronic health records (EHR)
  • Epidemic outbreak prediction and response


Investment in Artificial Intelligence and Machine Learning

The sizeable venture capital investment in digital health companies that are leveraging AI/ML provides hard evidence of the expected impact of AI/ML on healthcare. Between 2011 and 2017, Rock Health reports $2.7B was invested by venture capitalists in companies leveraging AI/ML. This represents just over 10% of all venture dollars invested in digital health during that time period.

Since 2011, just over 10% of all venture capital investment in digital health has gone toward AI/ML innovation – Source


Where AI/ML Will Deliver Value in Healthcare Innovation

In trying to characterize the potential impact of venture investment in AI/ML digital health, Rock Health categorized these investments by the value propositions offered by the companies. Nineteen categories were identified, ranging from acting as a catalyst for research and development (R&D) to providing a medical reference product.

Nineteen categories reflect the breadth of AI/ML’s potential impact on healthcare innovation – Source


The categories of AI/ML digital health companies receiving the most funding, listed by value proposition offered, are:

  • Research and development catalyst
  • Population health management
  • Clinical workflow
  • Health benefits administration
  • Diagnosis of disease

In this article, we’ll take a closer look at two of these: being a catalyst for research and development and health benefits administration.


Healthcare Innovation by Catalyzing R&D

When it takes the FDA a decade to approve a new drug, anything that shortens the process to screen molecules for drug discovery or improves the predictability of molecule efficacy early in the drug development process is welcomed by pharma. The computing power behind AI/ML is being applied directly to improving the likelihood of identifying viable molecules for development into new drugs early in the drug development process.

The promise is that AI/ML will shorten drug development cycles and increase the likelihood of successful drug development. If AI/ML can deliver on these promises, the cost of developing new drugs will be greatly reduced and new needed drugs will be brought to market more quickly.


Healthcare Innovation by Administering Health Benefits

As the population ages and the incidence of chronic illness grows, insurers and providers are looking for ways to streamline their administrative processes. By implementing AI/ML-driven algorithms, insurers and providers are coming to rely on computerized systems to diagnose patients and determine their level of care.

Insurers and providers expect that significant, sustainable cost reductions will result from applying AI/ML to a wide range of administrative processes-everything from patient intake and diagnosis to billing. There is also hope that AI/ML-driven algorithms will support personalization of care by removing bias and favoritism from the decision-making process.


Healthcare Innovation Needs People to Succeed

The press is full of headlines trumpeting a healthcare future made up of robots replacing doctors and algorithms making diagnoses.

The press would have us believe the future of healthcare is filled with robots and algorithms taking over – Source


Even with all the remarkable things AI/ML can do, they cannot completely replace the contribution of people, including patients, caregivers, and providers, in healthcare innovation. At least, not yet.

As with any data processing system, AI/ML is subject to the concept of garbage in, garbage out. No matter how complex or innovative the algorithm, the results are only as useful as the information it is built on and the data fed into it.


Drug discovery and development requires comprehensive and diverse patient groups

When discovering and developing a new drug, data from the patient group needs to be comprehensive and diverse. Otherwise, the resulting drug may have limited efficacy.

A recent study from UCSF found that albuterol, a bronchodilator widely used to treat asthma, is less effective for African American and Puerto Rican children. In examining how the drug was developed, the study found that 95% of the population engaged in lung research is of European descent. Because of this, the research did not register the effect of a genetic variation among African Americans and Puerto Ricans that resulted in significant segments of these populations not responding to albuterol.

This study points to the dramatic effect that drugs developed on restricted data sets can have on patients. No amount of AI/ML power can correct for data that simply isn’t present.


Healthcare algorithms require transparency and human oversight

Algorithms have the potential to automate the complex processes that are required for medical record review and treatment plan development. Because of the complexities of these processes, it’s critically important that healthcare algorithms are based on comprehensive information and thoroughly tested before they are implemented.

In a What Happens When an Algorithm Cuts Your Healthcare article, The Verge reported on the dire consequences for patients when poorly-vetted healthcare algorithms were implemented by Medicaid programs across several states.

In Arkansas, after some patients’ allotments of home care hours were reduced and the state was unable to explain how these decisions were being made, the issue of whether the patients’ rights had been violated ended up in federal court. Part of the hearing unveiled the inner workings of the healthcare algorithm being used.

Several aspects of the system were found to have a significant, possibly outsized, impact on the calculated results for the patient assessments. It was discovered that the weight given to some responses varied dramatically between the scores. This difference between a score of three versus four could translate into significant changes in the number of care hours authorized. Also, they found that the scores given for a particular patient’s condition could vary significantly from one assessor to another. The variation that comes with human judgement was not eliminated.

But the most damning discovery was that an incorrect calculation was being used by the algorithm. A mistake by a third-party software vendor implemented a version of the system that didn’t account for health issues related to diabetes. It was later reported that 19% of Medicaid patients in the state experienced negative impacts due to this omission. Additionally, the software had issues with the code related to cerebral palsy.

Ultimately, the federal judge ruled that the state had insufficiently implemented the program. Since then, the state has made changes to help people understand the system.

Better testing and more proactive human oversight could have avoided the potentially life-threatening impacts Medicaid patients in Arkansas experienced as a result of the mistakes made in implementing and administering a healthcare algorithm.


AI/ML-Driven Healthcare Innovation Still Promises Much

The promise of dramatic healthcare innovation based on AI/ML can’t be dismissed.

A future filled with AI/ML-driven healthcare that delivers better health outcomes at lower cost for more people seems within reach. However, as with any new and developing technology, we must remain mindful of its effect on the humans involved: patients, caregivers, and providers in particular.

What kinds of AI/ML-driven innovations in healthcare do you anticipate for your organization?

Share via
Send this to a friend