Health Catalyst points HIMSS20 attendees toward three AI trends

The artificial-intelligence-powered healthcare information technology vendor discusses delayed AI results, algorithmic bias and attracting data science talent.
By Bill Siwicki
01:09 PM
Health Catalyst points HIMSS20 attendees toward three AI trends

Update: HIMSS20 has been canceled due to the coronavirus. Read more here.

The rise of artificial intelligence into the mainstream of healthcare information technology is one of the biggest trends at HIMSS20, according to analytics vendor Health Catalyst, which will be in booth 2428.

Healthcare IT News asked Jason Jones, chief data scientist officer at Health Catalyst and a speaker at HIMSS20, about a few overarching trends surrounding AI that are important to HIMSS20 attendees. He says that a lack of results from healthcare AI implementations, algorithmic bias and difficulty attracting and retaining data science professionals are some key areas to watch.

A dearth of healthcare AI results

Jones said the industry is not seeing healthcare AI results in the timeframe and to the magnitude hoped for. On a related note, there is the question of how healthcare-provider organizations deal with the crush from AI-powered health IT vendors in the space.

"It is very easy for individuals or organizations to get excited about their first AI project," Jones said. "It is new, exciting and a bit magical. Out of dreams of doing good or pressure to perform, people would like to believe there is a solution. What is the problem? Building predictive models is very quick and easy."

Jones said the problems here are in four areas.

"First, ironically, the biggest obstacle toward solving a problem via leveraging AI can be that the problem to be solved is defined poorly or differently by different people," he explained. "Start with a great problem statement and common understanding of what 'awesome' looks like across stakeholders. Second, technically, the difficult part is getting high-quality data to train the model – commonly 50-100x more time and effort than building a predictive model."

"Focus on fundamentals, ask challenging questions, realize that AI typically fits into a workflow that requires multiple changes, and plan to monitor and improve over time."

Jason Jones, Health Catalyst

Evaluate whether the organization has the high-quality data it needs before starting an AI project, he advised; if not, acquire or improve available data or choose a different project, he cautioned.

"Third, most improvements in healthcare require behavior change on the part of physicians, nurses, administrators, members, patients, etc.," he said. "We do not need AI to tell us to eat and exercise well, it's just that it can be hard to do. When human behavior change is needed for success, we need tools and resources for change management."

And fourth, few AI efforts are set up for optimization or formal evaluation, Jones explained.

"If you fear you are being left behind in the AI race, consider the last time you felt left behind by an infomercial," he offered. "The claims of success for AI may not be much better founded. Focus on fundamentals, ask challenging questions, realize that AI typically fits into a workflow that requires multiple changes, and plan to monitor and improve over time."

Algorithmic bias

Then there is the artificial intelligence problem known as algorithmic bias. How do healthcare-provider organizations deploy AI in such a way that they do not exacerbate health disparities?

"There has been an increase in concern that the 'move fast and break things' approach may have done more harm than good in particular and in aggregate," Jones stated. "People are intolerant of breaking things in healthcare in ways they feel could have been anticipated. We are justifiably and particularly angry when the nature of the failure involves disparity based upon personal characteristics such as gender, ethnicity, geographic location and socioeconomic status."

But healthcare does want algorithms to discriminate – between people at greater or lesser risk for readmission or ready or not ready to quit smoking, for example.

"Remembering this helps us to think differently about AI," Jones said. "For algorithms to succeed, we should retain the right and accountability to define what we want the algorithm to do and not do and then measure against these desires. With that in mind, it is possible to go beyond fear of algorithmic bias to algorithms helping assure equity."

On whiteboards, healthcare-organization staff can convert equity from a balancing measure (possible harm) to an outcome (desired benefit) and then design and measure for that, he explained.

Data science talent

And Jones’ third healthcare AI trend surrounding HIMSS20 is how healthcare provider organizations attract and retain data science talent.

"It can feel as though it is very difficult and expensive to attract a data scientist," he said. "In healthcare, it can feel impossible to compete with the tech sector. If you feel this way, pause and consider your needs and assets. First, in healthcare, most of the technical time and effort is in gathering and preparing data – data engineering. You may not need as many data scientists as you think, or you may be able to 'rent' one when you have the need."

Second, think about what the organization needs a data scientist to do – for example, ask and answer questions better with data, and in a way staff can understand, he added.

"Test and evaluate for people who can do that," he advised. "Usually this means not using the 'Kaggle' (data competition) approaches. These are the aspects of data science that are both most technical and most easily automated."

And third, if a healthcare organization has a noble purpose, point this out and explain how the data scientist contributes, Jones advised.

"Give him or her opportunities to see that contribution firsthand – from call centers, to boardrooms, to nurses' stations," he concluded. "Taking these steps not only helps you attract and retain talent, but also helps you get better output through the data scientist better understanding the real problems and what solutions might look like."

Jones will be at HIMSS20 on a panel entitled "​Analytics to Algorithms: How to Maximize Impacts" on Monday March 9. He also will be presenting alongside Dr. Terri Steinberg during a presentation entitled "Machine Learning and Data Selection for Population Healthon Thursday, March 12.

Twitter: @SiwickiHealthIT
Email the writer: bill.siwicki@himssmedia.com
Healthcare IT News is a HIMSS Media publication.

Full coverage of HIMSS20

An inside look at the innovation, education, technology, networking and key events at the HIMSS20 global conference in Orlando.