VIDEO: How hospital IT teams should manage implementation of AI algorithms

"Hospital IT departments are woefully unprepared for this onslaught of AI," Bogdan explained. "They don't have the resources to be able to understand what implementation entails and what is needed for lifecycle management."

When a department wants to install an AI algorithm, the biggest need is a clinical liaison from the department to interface with the IT department so there is a clear understanding of what the AI does, why it is needed and how it needs to fit into the clinical workflow. 

Bogdan said there are two things the IT department needs to do when adopting an AI algorithm. First, make sure the AI is a good fit for the problem at hand. Second, validate it from both an IT and business perspective.

"Make sure you have a clear case and a value proposition for what you are doing," Bogdan said. "And then make sure you have the IT resources to implement it. You also need to ask what you are doing from a data policy prospective and what are you doing from an integration prospective, ask where that data is going and who owns that data and the outcomes that come out of that data?"

He said there are implications from a security prospective, data integration prospective and also knowing how to get the AI data back into the workflow.

"Just throwing in AI is not going to solve the problem for you, there are still multiple steps involved," Bogdan said. "You need to think about how to implement a clinical pathway to address what to do with the data or risk scores created by the AI." 

AI integration into medical IT systems is key for adoption

Numerous AI vendors have created AI apps stores where various market-cleared algorithms can be purchased. In the past couple years, that has morphed further with several larger health IT vendors patterning with numerous AI start-up companies to fully integrate seamlessly their AI into the large vendor's systems. This is being done because vendors say unless AI is fully embedded in daily workflows, no one uses it.

"Absolutely, that is the way things are going to go, because without that integration it has limited clinical capability. It has to be a part of the clinician's workflow, it has to be a part of the administrator's workflow, because they are unlikely to switch between systems to get the information they need. That is the future and the way I see it happening," Bogdan explained. "And the vendors are seeing the need to beef up their own systems by introducing the AI algorithms natively inside their technology. That integration is key for driving adoption." 

Partnerships in AI are key because no one company can do everything

A few years ago, IBM Watson was seen as a rising start in the AI world. It purchased health IT vendor Merge Healthcare to get access to the patient datasets needed to train its health AI applications. IBM Watson Health then laid out plans to create a myriad of algorithms to diagnose radiology images, auto calculate risk factor assessments and offer clinical decision support to clinicians. It had a major partnership with MD Anderson to develop AI to diagnose and guide therapy for various cancers. However, it became apparent IBM was having issues after its partnership agreement with MD Anderson was cancelled in 2017. One of the biggest issues was the vast amount of deep integration needed with data in electronic medical records and information from reports or notes that were not structured, which caused interface issues.

The vendor also greatly scaled back its AI messaging and admitted at conferences it could not do everything and was starting to partner with several promising start-up AI vendors. 

"Unlike other industries, in healthcare there are so many variables that make it hard," Bogdan said. "That is what happened to Watson, because taking cancer is a pretty big and broad space and they did not take into account all of the variables. And you cannot model effectively a lot of those scenarios, because it introduces a lot of variability."

And it's not just Watson. Other AI vendors have also fallen into a view that IT teams can program algorithms to solve problems, but find that what works in controlled bench testing environments does not always translate in real-world patient care. 

"The ones that are more successful are the ones that incorporate that clinical input and feedback from the beginning of the AI development, with the action oriented piece in mind for what will you be doing with the output from this algorithm and how are you going to use it to move the needle on patient care," Bogdan said. 

The model of partnering with start-up AI vendors quickly spread across heath IT. While these start-ups are seeing their names attached to some big-name IT vendors in the market, Bogdan said there will likely be a lot of consolidation in the coming years as these small AI vendors get bought up by larger ones that hope to make integration of AI easier and with fewer contracts to sign.  

This is part of a 5-part series of interviews with Bogdan on various aspects of AI in healthcare. Here are the other videos in the series:

VIDEO: 9 key areas where AI is being implemented in healthcare

VIDEO: AI can help prevent clinician burnout

VIDEO: Use of AI to address health equity and health consumerization

VIDEO: Understanding biases in healthcare AI

Find more AI news and video