There has been much talk about the potential of AI in health, but most of the previous research has been representative of practical medicine practices. It is a simulated scenario that predicts what the impact of AI is in the healthcare environment.
However, in one of the first real-world tests of AI tools working alongside Kenya clinicians, researchers have shown that AI can reduce medical errors by up to 16%.
A study available on Openai.com submitted to the scientific journal, researchers from Openai and Penda Health, a network of primary care clinics operating in Nairobi, have found that AI tools can provide powerful support to busy clinicians who can’t know everything about all medical conditions. Penda Health employs clinicians who have been trained for four years in basic health care. It is equivalent to a US physician assistant. The health group, which operates 16 primary care clinics in Nairobi, Kenya, has unique guidelines to help clinicians navigate symptoms, diagnosis and treatment. However, the range of knowledge required is difficult for any practitioner.
Dr. Robert Korom, Penda’s Chief Medical Officer, said: “So one of the biggest things is the breadth of the tool.”
Read more: Psychiatrist casting teenage figures with therapy chatbots. The conversation was amazing
Previously, Korom said he and his colleague, Dr. Sarah Kiptinness, Head of Health Services, had to create guidelines for guides for each scenario that clinicians may encounter in general (e.g., complex malaria cases, or adult malaria cases, or patients with low platelet count situations. AI is best at accumulating all of this knowledge and distributing it under appropriately matched conditions.
Korom and his team built the first version of the AI tool as a fundamental shadow of clinicians. If a clinician had questions about which diagnosis to provide or which treatment protocol to follow, he could press the button to pull a block of related text that was matched by the AI system to aid in decision making. However, clinicians used the function for about half of their visits, Korom said.
So Penda has improved a tool called AI Consult. This was carried out quietly in the background of the visit, essentially overshadowing clinician decisions and encouraged only when suspicious or inappropriate behavior such as prescribing antibiotics.
“It’s like having experts there,” Korom says. This is similar to how a doctor reviews a health care resident’s care plan. “In some ways, that’s how (this AI tool) works. It’s a safety net. It’s not about determining what care is, it just gives corrective nasty and feedback when needed.”
Read more: The world’s wealthiest woman opens medical school
Penda worked with Openai to conduct AI consulting research, documenting how it had impacted both on making diagnoses and prescribing treatments on helping around 20,000 doctors reduce errors. A group of clinicians using AI consulting tools reduced diagnostic errors by 16% and treatment errors by 13%, with a 13% reduction compared to 20,000 PENDA providers who did not use it.
The fact that the study involved thousands of patients in a real-world setting is that Dr. Isaac Kohane, professor of biomedical and informatics at Harvard Medical School, sets a strong precedent for how AI can be effectively used in healthcare delivery and improvement, according to Dr. Isaac Kohane, professor of Harvard Medical School who saw the study. “In contrast to retrospective studies, we need more of these types of prospective research. (Researchers) use AI to look at large observational data sets (health outcomes), which is what I’ve been waiting for.”
This study not only showed that AI can reduce medical errors and thus improve the quality of care patients receive, but the involved clinicians viewed this tool as a useful partner in medical education. That was a surprise to Openai’s Karan Singhal, Health AI lead who led the research. “It was a learning tool for (the people who used it), and helped them to educate themselves and understand the wide range of care practices they need to know,” says Sinhal. “It was a bit surprising because it wasn’t something we were trying to study.”
Kiptinness says that AI Consult will act as a key confidence builder and help clinicians gain experience in an efficient way. “Many of our clinicians now feel that they have to stay to help AI consultants feel confident in their care and improve the quality of care.”
Clinicians receive instant feedback in the form of green, yellow and red light systems that assess clinical action, and the company gets an automatic assessment of their strengths and weaknesses. “We want to provide more personalized feedback in the future, such as, ‘You’re good at managing obstetric cases, but in pediatrics, these are areas to be considered,” says Kiptinnes. “There are many ideas about customized training guides based on AI feedback.”
Read more: Amazing reasons why rural hospitals are closed
Such co-operation can be a practical and powerful way to incorporate AI into healthcare delivery, especially in most areas of high needs and healthcare professionals. The findings “changed what we expect as standard care within the penda,” Korom said. “Perhaps you don’t want a clinician to be completely present without this.”
The results set a more meaningful stage of research on AI in healthcare that shifts practice from theory to reality. Dr. Ethan Goh, executive director of the Stanford AI Research and Science Assessment Network and associate editor for the journal BMJ Digital Health & AI, predicts the study will stimulate similar things in other settings, including the US. “Maybe we’re just catching mistakes today, but what happens if we can surpass tomorrow?
Tools like AI Consult may further expand access to healthcare either by putting it in the hands of non-health people, such as social workers, or by providing more specialized care in areas where such expertise is not available. “How far can I push this?” Korom says.
The key, he says, is to develop highly customized models that accurately incorporate provider and patient workflows in specific settings, as Penda did. For example, Penda’s AI consulting focuses on the types of diseases most likely to occur in Kenya, and is the most likely condition that clinicians will be seen. If such factors are considered, he says, “I think there are a lot of possibilities out there.”
