For many years, artificial intelligence has been touted as the next big thing to improve patient prognosis. But the health system has not yet seen this promise become a reality.

Although there are thousands of research papers introducing potential new AI models or verifying products already on the market, there are few documents on how hospitals can truly use these tools for a long time, and most of the system leaders can only solve the fragments themselves. And hope for the best.

“There is no centralized, coordinated fashion when it comes to reviewing products and putting them into practice,” said Dr. Maxendak, head of population health and data science at the Duke Health Innovation Institute. “Each health system is developing its own way to do this, and our assumption is that there may be a lot of truly valuable knowledge to share.”

All types of software AI have the risk of accidental injury. Mechanisms established by hospitals to monitor artificial intelligence can play an important role in patient safety.

“The question is,’Will the algorithm inadvertently cause harm and prejudice to patient care?’ Because the lack of patients will lead to delays in diagnosis,” said David Vidal, supervisory director of the Mayo Clinic Digital Health Center.

Vidal, Sendak, and a group of researchers from the University of Michigan, Duke University, and elsewhere are calling on the health system to publish reports on their experience in implementing artificial intelligence in healthcare settings, which others can use as best practice guidelines. Frontiers is an open access journal and plans to publish 10 such case studies next summer.

But when a hospital Artificial intelligence software may achieve great success, and another may fail. The specific details of the patient population, how to instruct clinicians to use the tool, and how to build it into the process ultimately determine its success.

UnityPoint Health, a multi-state non-profit system located in West Des Moines, Iowa, encountered technical limitations in establishing an AI early warning system for sepsis infection. This chain of hospitals enables triage nurses to identify potential cases of sepsis without using software — and humans are infected with more infections earlier than AI.

Ben Cleveland, chief data scientist at UnityPoint Health, said: “From an artificial intelligence perspective, this is unfortunate because I think the default idea is that these models will enter the healthcare field and completely reform them, which is not currently a Achievable goals.”

landscape For healthcare software, artificial intelligence is huge, and most products currently have not passed the FDA approval process as medical devices. As of the end of 2020, suppliers have only sought and obtained approval from the Food and Drug Administration for approximately 100 products.

“Practices and institutions need to be able to understand the software we are picking, and how it is trained and validated, so that they understand whether they think it will work in their American Institute of Radiological Data Science, Chief Medical Officer, Birmingham, Alabama Community Said Dr. Bibb Allen, a practicing diagnostic radiologist.

Nigam Shah, deputy chief information officer of Stanford Healthcare in Palo Alto, California, said that in the future, hospitals may establish governance committees with insurance companies to select artificial intelligence tools or create formulary or vetted products. “If the industry cannot self-regulate, then in 10 years, the government will eventually defeat its whip,” he said.

Suchi Saria, CEO and founder of software provider Bayesian Health, said that software companies themselves are responsible for ensuring that their products are properly designed and used. “I don’t think the health system will suddenly become an expert in building the correct monitoring infrastructure,” she said.

Nebraska Medicine in Omaha There is a team to evaluate artificial intelligence tools, but this non-profit system still relies mainly on word of mouth and other health system reports when choosing software. For each new product, clinical experts are responsible for reviewing the work process and information management of each unit. There are still obstacles when trying to get clinicians to actually use the information generated by the software.

“I hope this is part of this type of initiative being carried out in Michigan and elsewhere-how we can expand some of these developments and some of these success stories to smaller places,” the Nebraska Department of Medicine Provider Informatics Director Dr. Justin Birge said.