How to Avoid the Pinch of Pain Points in Clinical Care AI Tools

As health systems face urgency to adopt and implement AI tools in clinical care, several pain points are emerging.

Many of these pain points are highlighted in a new white paper generated by Qventus. The white paper is based on a survey and interviews with more than 60 CIOs, chief AI officers, CMIOs, and other senior IT leaders at medium and large health systems.

The white paper includes four key findings:

• The survey found that 94% of respondents said AI operationalization delays would put their organization at a competitive disadvantage compared to other health systems.
• Four out of five survey respondents said they struggle to determine or measure the ROI of AI tools.
• More than two-thirds of survey respondents cited limited IT resources from managing multiple AI vendors as a top execution obstacle.
• Three-quarters of survey respondents said reliance on their EHR vendor’s AI roadmap is an obstacle to executing their AI strategy.

Balancing urgency and risk avoidance

Despite the pressure to implement AI tools in clinical care, senior leaders at health systems must move deliberately to avoid unnecessary risk, a pair of experts say.

The No. 1 way health systems can feel the urgency to operationalize AI tools in clinical care while avoiding risk is to be intentional about the problems they are seeking to solve, according to Matthew Anderson, MD, MBA, CMIO for ambulatory care at HonorHealth.

“If you adopt an AI tool because you think it is going to be a catch-all solution; if you adopt an AI tool because a vendor sings its praises; or if you adopt an AI tool because you feel you have to keep up with the Joneses; that is not going to work out well,” Anderson says.

Senior leaders at health systems should feel the urgency to adopt AI tools but be disciplined in their actions, Anderson explains.

“You should not rush decisions, take a look at your risks, and review the AI tools that are available,” Anderson says.

Avoiding risk in AI tool adoption primarily involves understanding what AI tools are and what they are not, according to Joseph Sanford, MD, chief clinical informatics officer at the University of Arkansas for Medical Sciences (UAMS).

“There is a lot of rhetoric around the intelligence side and reasoning side of AI models, but AI tools are generally probabilistic data-generation engines,” Sanford says. “They are very good at several things, but they are not magic.”

To avoid risk, senior leaders must get educated about what an AI tool can and cannot do, Sanford explains.

“You need to understand there are many opportunities for AI tools to help your staff do their jobs in better and faster ways,” Sanford says. “You should focus on those opportunities rather than considering AI tools that vendors claim will solve everything.”

Assessing ROI of AI tools in clinical care

For AI tools in clinical care, there is “hard ROI” and “soft ROI,” according to Anderson.

“Examples of hard ROI include improving billing and coding, capturing Hierarchical Condition Category risk, increasing productivity and patient visits, and boosting the number of surgeries in operating rooms,” Anderson says. “Examples of soft ROI include improving patient experience and making the lives of clinicians easier.”

At HonorHealth, hard ROI is not the primary factor that drives adoption and implementation of AI tools in clinical care, with the focus more on clinician and patient metrics, Anderson explains.

“We want to know whether patient outcomes are better,” Anderson says. “With a deterioration index, we want to know whether we are intervening in patient care faster. With AI fall solutions, we want to know whether we are reducing patient falls. We want to see whether clinicians are spending less time in the electronic medical record or whether clinicians are reviewing charts faster with a chart summarization tool.”

Financial ROI is only one of several criteria that UAMS assesses in the adoption of an AI tool in clinical care, according to Sanford.

“We don’t put a hard dollar figure on a user’s personal time or frustration, but we know that there is a cost to overutilizing a provider’s time such as burnout and the expense of replacing clinicians who leave because of frustrations,” Sanford says.

“We look at how much time we can save for clinicians, whether we can reduce the time that clinicians spend working outside of normal working hours, administrative throughput efficiency such as scheduling patients at the right time and the right place, and metrics such as the time it takes to close a chart and time to complete a prior authorization,” Sanford says.

There are some relatively easy financial ROI measures for AI tools in clinical care, according to Sanford.

“When you are purchasing an AI tool that has per-user-per-month licensing costs, you want to look at two factors,” Sanford says. “First, is an AI tool going to increase the ability to generate revenue such as reducing the time it takes to see patients and increasing the ability to see more patients? Second, is an AI tool going to decrease expenses?”

Managing multiple AI vendors

While managing multiple AI vendors can be challenging, senior leaders should lean on their experience in dealing with multiple vendors in other areas of their operations, according to Anderson.

“Vendor management is relatively common at health systems,” Anderson says. “It is a little more complicated with AI vendors, but maintaining relationships with multiple AI vendors is mainly a matter of being intentional. In particular, you should know how your vendors approach co-development of AI tools.”

Data stewardship and transparency are primary challenges in working with multiple AI vendors. Leaders must be cautious about sharing data and be good stewards of patient data, Anderson explains.

“When you get in situations where you are working with multiple vendors trying to solve the same problem, you should be transparent with the vendors,” Anderson says. “You also want vendors to be transparent with you such as providing access to engineers as opposed to sales staff.”

The best way health systems can manage relationships with multiple AI vendors is to rely on a team of experts, according to Sanford. UAMS has independent governance committees that are operationally focused on clinical care, research, and education.

“Each of those committees is staffed by experts in their fields and chaired by a senior leader at UAMS who has organizational responsibility for one of our three core missions,” Sanford says. “My peers and I sit as voting members on our overarching AI governance committee, which sets policies and standards for the operational governance committees.”

The UAMS AI governance committee monitors the AI market and plays a key role in managing multiple AI vendors, Sanford explains.

“The AI governance committee works with our IT teams, EMR services, and information security office to vet contractual relationships with vendors,” Sanford says. “One of the big challenges is the developing environment of AI vendors and how they demonstrate data quality and trust.”

EHR-native AI tools versus point solutions

Senior leaders should carefully consider adoption of electronic health record-native AI tools and point solutions, according to Anderson and Sanford.

“AI tools in the EHR are great, but you need to be open to adopting point solutions, when necessary,” Anderson says. “When you consider adopting a point solution, it is important to have a partner that you trust.”

Senior leaders face an opportunity-cost question when assessing EHR-native AI tools and point solutions, Sanford explains.

“On the one hand, you must look at what you have already paid for and deployed because you are obligated to the organization to maximize utilization of that technology,” Sanford says. “On the other hand, you must look at technology that solves a problem but is only available as a point solution.”

When a vendor’s point solution is similar to the EHR-native offering, the EHR’s AI tool is usually going to be the most attractive option, according to Sanford.

“If a point solution is bringing something to the table that is similar to something that is EHR-native, then the competitive advantage of the point solution such as cost savings must be profound to justify its adoption,” Sanford says.