Although artificial intelligence in hospitals is still in the early stages of regulation, Akron-area hospitals are implementing it in a variety of ways, along with training and education.
Christopher Congeni, a partner in the Cleveland office of Amundsen Davis LLP, said there are many concerns that hospitals, physician groups and practitioners should consider when deciding to use AI.
These include maintaining bias, transparency, and the use of AI as a tool rather than a replacement for humans.
“Healthcare is very highly regulated and we’re still figuring out how to regulate AI, so that’s a challenge,” he said.
Hospitals are deploying AI algorithms to help with everything from interpreting radiology reports to summarizing findings to quickly identifying stroke patients in the emergency room.
Congeni said the use of AI in hospitals is still at the stage of risk assessment, and as regulations and laws develop, it is important to minimize risks through comprehensive compliance planning.
“We need to see how far we have to go,” Kongeni said. “Is it part of the actual diagnosis and treatment, or is it just a basic part of making things easier for the health care provider?”
Potential data bias
Naomi Scheinerman, assistant professor of bioethics at Ohio State University, said doctors need to think about the data used to train the AI models they use.
“We don’t have a perfect picture of our knowledge of the situation in society and how it affects different groups and populations,” she says. “Majority-dominant groups are disproportionately represented in our data.”
Congeni said AI algorithms could inadvertently amplify existing biases, leading to discriminatory outcomes in patient care.
Steve Worrell, CEO of Riverain Technologies in Miamisburg, developed the algorithm used by both University Hospitals and the Cleveland Clinic. He said his company takes this into account when training its AI.
As part of the company’s development process, employees ensure that the data captured is diverse and captures variation among different patient populations.
“When training these systems, it’s really important to be able to properly represent different patient populations,” Worrell says. “Generally speaking, the more data you have with these algorithms, the better.”
Without proper training and vetting, AI could potentially be harmful, said Deborah Shapiro, an associate professor of medical ethics at Ohio University.
“As ethicists, some may be concerned that we are potentially harming patients without a clear understanding of the risk-benefit profile,” she asked.
Doctors may rely too much on AI
Shapiro said there are concerns that doctors will become dependent and overly trusting of AI.
“There’s a question that if you actually use artificial intelligence over a long period of time, potentially both in the medical profession and other fields and other professions, your critical thinking skills and your accuracy and your attentiveness will become a little bit duller,” she said.
Shapiro said evidence is needed to ensure that doctors still have a job of critical thinking and analysis.
Dr. Pohao Chen, deputy director of artificial intelligence at the Cleveland Clinic’s Diagnostic Laboratory, said the AI will never make a diagnosis on its own within the lab, but instead humans will oversee the process and make the final decision.
Requires careful implementation
Shapiro said the process of using AI in hospitals needs to be planned and methodical.
“I worry that we are not being as careful as we should be in integrating and deploying artificial intelligence tools in hospitals and healthcare settings,” she says.
Dr. Leonard Kayat-Bittencourt, vice chair for innovation at University Hospitals, said UH has developed a deliberate approach to implementing AI.
“We do a few times what the industry calls ‘shadow mode,’ which launches an AI tool in the background for a very small number of selected users and monitors the situation for weeks or months, depending on the complexity of the AI tool,” said Bittencourt, who also works in abdominal imaging.
Experts then reconvene periodically to assess how the tool is performing before a formal implementation decision is made. The results will be collated with the data collected and, if satisfactory, deployed to clinical practice in parallel with education.
Bittencourt said the university takes training seriously and offers multiple educational opportunities for staff.
“We take very deeply our mission and our work to continually educate people,” Bittencourt said.
AI in healthcare is still not fully regulated
Congeni said he was concerned that some groups would take the regulation of AI more seriously than others.
“Where it ends and where it begins matters a lot,” he says. “It is important to define clear boundaries in your compliance plan.”
He emphasized the importance of having written policies, procedures and compliance plans.
“The underlying concern people have these days is that the use of AI is becoming more pervasive. [medical] I didn’t practice as much as I should have,” Shapiro said.
Scheinerman said the potential of AI could be promising if used correctly.
“If we can make this technology very well trained and effective, we can speed it up, be more accurate, faster, and save lives,” she said.
Lauren Cohen is a community reporting intern at Akron Beacon Journal and Signal Akron. This position is funded by a grant from the Knight Foundation, a financial supporter of Signal Akron.
