Can AI Save Healthcare? (Part 1)

Artificial intelligence (AI) is a hot investment topic in healthcare. A recent report identified 106 AI healthcare startups, half with their initial funding in the last two years.¹ In this frothy environment, offering “artificial intelligence” may have as much to do with marketing as with the mathematical algorithms that can effectively assist in making better decisions.²

The above picture of robots at their computers comes from Larry Birnbaum’s recent article, “Understanding the hype vs. reality around artificial intelligence.”³ He pragmatically states, “Today’s AI may be just ‘mere’ computer programs – lacking the sentience, volition, and self-awareness – but that does not negate their ability to serve as intelligent assistants for humans”.

Following Birnbaum’s pragmatism, I will limit my focus to how artificial intelligence (more intelligent compilation, interpretation, and presentation of data) can improve the effectiveness of healthcare. In this Conversation I’ll begin with my personal experience with AI as a physician and administrator. In the next Conversation, I will examine the big picture of healthcare and the pitfalls of expecting that artificial intelligence will necessarily improve the effectiveness of the health industry.

I was exposed to artificial intelligence in the late 1970’s when commercial EKG machines began to interpret electrocardiograms. Despite significant skepticism, by the early 1980’s the machine algorithms, followed by human expert overview, had made EKG interpretation both faster and more accurate – a clear increase in effectiveness.

The digitalization of pathology specimens over the last two decades is slowly replicating the experience with EKG’s. Although adoption has been irregular because of early technological challenges, set up expense, and regulations, the AI capacity to enhance the quality of final pathology reports is inevitable. AI is also revolutionizing the world of diagnostic imaging, again because, on average, it clearly enhances the reading speed and accuracy of radiologists.

In 2012-2014, I experienced a “total immersion” with AI during an electronic health record (EHR) installation in the 400-bed hospital where I had just started as Chief Medical Officer. Despite the positive aspects of EHR, living through this project was a nightmare. Yet after six months of 24/7 repair of initial snafus, processes such as ordering, discharge instructions, and the need to interpret physicians’ illegible handwriting had improved.

However, the challenges for physicians have been complex. One frustrating example (and yes, it was fixed) was the pop-up warning screens that would flash homilies such as, “Morphine is a respiratory depressant” each time a doctor ordered a drug with severe potential side effects. Such frequent reminders had the cumulative effect of Chinese water torture on our medical staff’s mental status.

The term for this phenomenon is “warning fatigue.” Although these admonitions were designed to improve safety, they were so poorly designed that for six months, our hospital was less safe than it had been.

To get a better flavor for the “warning effect”, imagine the following stop sign improvement: “Stop! Don’t forget to put your foot on the brake! You might kill someone!” Doctors were not amused.

The huge nationwide push for Electronic Health Records was part of the 2009 Obama stimulus package. The consensus of economists on the stimulus has been positive for both shortening and limiting the depth of the 2008 recession.⁴ However, the assessment of the effectiveness of the law’s investment to EHR installation is more controversial.

The $200,000 per bed investment left our hospital with what is already an antiquated mainframe solution (total installation, training, and customization costs – $80 million) with inflexible software requiring frequent upgrades. However, as noted above, there were also advantages. From a workflow perspective, the improvements were a mixed bag with some specialists experiencing improved processes while others experienced obstacles in effectiveness.

From an outpatient clinic perspective, EHR increased the productivity of non-physician staff. For physicians, the time with paperwork has noticeably increased, resulting in less face-to-face time with patients. Over the years, time has healed some of the initial angst, but the longer-term increase in physician job dissatisfaction stems at least in part from physicians’ unsatisfactory interface with EHR.⁵

Several years ago a UCSF professor wryly noted to a medical resident the good and bad of electronic records, “The good news is that now I always know where to find a resident; they are at the computer station. The bad news is that before, when I couldn’t find them, they were with their patients.”

We can learn from these AI experiences to create better designs for the future. However, as will be discussed in Part 3, achieving transformational healthcare improvements will require a major upgrade in the use of our own intelligence in conjunction with further improvements in AI.

Breakthrough To Better,
Carl

¹Funding for healthcare AI is on the rise
²Algorithms to Live By
³Understanding the hype vs. reality around AI
⁴American Recovery and Reinvestment Act of 2009
⁵Biennial Physician Survey

____________________________________________________________________________________

Switch Conversations is a bi-weekly blog exclusively for designated key employers.

____________________________________________________________________________________