Implementation best practices: Dealing with the complexity of AI

By Bill Siwicki
04:35 PM
Share
Four experts in artificial intelligence technology offer advice to healthcare CIOs on how to best begin implementing an AI system.
Implementation best practices: Dealing with the complexity of AI

Artificial intelligence is just as complex as it sounds. Successful deployment of the various technologies that are necessary for AI to work requires planning and strategy..

To help chief information officers and other IT professionals better understand these best practices for implementing AI at their health systems, hospitals, group practices and other provider organizations, we spoke with four experts in AI technologies who offered their advice for effective rollouts.

Identifying use-cases

Know and understand artificial intelligence use-cases and successes from various vendors, advised Ryan Pretnik, research strategy director at KLAS Research.

“Most healthcare providers should look to find and understand use-cases in the industry and how their peers have adopted AI technology,” he said. “This is a good first step, since there are many AI companies in the market that claim they can help healthcare organizations with AI. Being able to understand use-cases and their successes in the healthcare market, while consulting with clinicians internally on the use-cases, typically helps drive healthcare organizations in the right direction.”

If an IT decision-maker buys an AI system without consulting clinicians on the use-cases and benefits, then the tool may end up being a low-adoption system and looked at simply as another piece of technology the organization purchased, Pretnik said.

“If IT consults and works with clinicians from the start and shares the use-cases and successes and engages them on how to use the technology, the clinicians are more likely to use and adopt the product, which in turn helps drive outcomes,” he added.

“By identifying use-cases and successes, you can help categorize vendors who have a proven track record of success while reducing the financial risk your healthcare organization takes on by purchasing an AI tool," said Pretnik.

The expected value

First, the focus for CIOs implementing artificial intelligence in hospitals and health systems must be on the expected value, said Jean-Claude Saghbini, chief technology officer at Wolters Kluwer Health.

“AI is a once-in-a-generation transformative technology,” he said. “As such, expect its impact to be on the scale of the advent of electricity or the internet. Understandably, AI has been accompanied with a certain amount of hype, particularly in healthcare. Right now, an astute CIO has to navigate a narrow path between what is hype and what is reality.”

Those who watch from the sidelines will risk being left behind, Saghbini contended. Meanwhile, aggressive early adopters face the risks of implementing technologies that are premature or simply don’t deliver on promises, he added. Healthcare users expect technologies to be vetted to make sure they work to reduce uncertainty and resistance to change, he said.

“When it comes to implementing AI, there is no substitute for sound business principles,” he explained. “CIOs should apply the same rigor in the adoption of AI that they apply in the adoption of any other new technology, from cloud storage to secure messaging. AI has the potential of impacting virtually every process or domain in the hospital, both on the clinical side and the administrative.”

CIOs have to train their focus on problems that need solving, where considerable value can be extracted, and where the value statement is evident and can be clearly articulated, Saghbini insisted, noting that this is especially true right now, as hospital executives are trying to combat revenue and margin compressions while providing increasingly better care for patients.

"AI is a once-in-a-generation transformative technology. As such, expect its impact to be on the scale of the advent of electricity or the internet."

Jean-Claude Saghbini, Wolters Kluwer Health

“Where should CIOs start? Initial efforts should be geared toward quick wins for key stakeholders,” he said. “Moonshots can be very tempting, and often, this is what makes headlines in the media. Starting with these initiatives however, is inherently fraught with risk given the combined uncertainty in technology performance and in healthcare user adoption.”

Early investments in AI initiatives and systems should be aimed at yielding rapid yet sizeable results while building up internal knowledge in the AI domain, Saghbini said. More important, from a cultural perspective, these early wins can start to acclimate people in the rest of the organization to the art of the possible, he added.

“Finally, I would add that as CIOs pursue a portfolio of initiatives, it’s critical to work with partners who can introduce solutions to a variety of areas in the hospital or network,” Saghbini said. “As we move into an era where AI becomes more acceptable, and eventually expected, first-hand experience with the technology is going to be key. CIOs need to be the catalysts who ensure that clinicians are getting first-hand exposure, early on, to the transformative capabilities of AI and to its potential impact.”

The operational state

A key must-have when implementing an AI system is a clear vision of an organization's operational state and business goals, said Gurjeet Singh, CEO and co-founder of Ayasdi, a vendor of an AI-powered platform and enterprise-grade, intelligent applications. This may sound obvious, but experience says otherwise, he said.

“Very often, there is a desire to see what the AI tool can do versus thinking about the operational state and business outcomes,” he said. “This requires thinking about a number of things upfront: How often do you anticipate interacting with the system? Daily, weekly, monthly? What will be the ongoing inputs to the system? What will be the ongoing outputs to the system? What is the expected audience of users? What is the plan to expand usage over time?”

If one begins the project with the intent to have it be operational versus some proof of concept, one will make better, more informed decisions and increase the chances of success, Singh added.

“The second thing to understand is how the organization intends to validate the findings of the system,” he said. “There is no operationalizing a black box. There has to be a deep, clinician-understandable explanation of what the machine is recommending.”

Every lab, drug, order and test: This radical transparency, called justification, is what is required to build trust, Singh said. Trust is what is required to be successful in any system, particularly one as transformational as AI, he added.

Focus on outcomes

When implementing AI technology, the focus should be on outcomes, said Lois Krotz, research strategy director at KLAS Research, a healthcare IT research and consulting firm.

“From numerous conversations with provider CIOs and vice presidents of technology: They like the idea of using AI but are unsure how to, and what results could be driven from the solution,” she said. “Set goals and make sure you have ways to benchmark the success of the AI solution – know how long it will take to see an outcome.”

"There is no operationalizing a black box. There has to be a deep, clinician-understandable explanation of what the machine is recommending."

Gurjeet Singh, Ayasdi

Because AI systems can take smaller subsets of data (structured) and/or larger more holistic data sets (structured and unstructured), there tends to be a question of what results one’s data can produce.

“If internal teams – IT and physicians – can work together to determine a couple use-cases, they would like to start with something like ‘Diagnosing sepsis or diabetes for a specific ethnicity,’ ‘Patient readmittance rates for a specific use-case,’ or ‘Clinical variation management for hip replacements,’ they can then start researching specific AI vendors, their use-cases, and outcomes peer healthcare organizations are deriving from the solution.”

Furthermore, when one works with AI vendors, or consulting firms before one engages with an AI vendor, ask them to help define specific measurables that can be seen within a specific timeframe with the readily available data at hand, Krotz explained.

“For example, in care management, we might want to help decrease the onset of a heart attack – for patients with congestive heart failure – by 25 percent and also help decrease readmittance rates of these patients by 20 percent,” she said. “In addition, it is important to understand a vendor’s implementation approach.”

Will the vendor conduct a pre-implementation evaluation in helping determine ROI/outcome areas and their related measurement and target? Will they help provide EHR integration? Will they provide best practices or recommendations for processes/resources to support the new workflow and technology?

“In general, considering how new the market is, clients typically would like more hand-holding and partnerships from AI vendors during the implementation period,” said Krotz.

Twitter: @SiwickiHealthIT
Email the writer: bill.siwicki@himssmedia.com

Health IT implementation best practices

This 20-feature series examines in-depth what it takes to deploy today's most necessary technology and tools.