Consider the Mayo Clinic, the world’s largest comprehensive non-profit medical organization, which has created over 160 AI algorithms in cardiology, neurology, radiology, and other specialties. 40 of them have already been deployed in patient care.
To better understand how AI is being used in healthcare, I spoke with John Halamka, a medical informatics-trained physician and president of the Mayo Clinic Platform. As he explained to me, “AI is just a simulation of human intelligence through machines.”
Halamka made a distinction between predictive AI and generative AI. The former involves mathematical models that use past patterns to predict the future. The latter uses text or images to generate a kind of human-like interaction.
The first type is the most valuable for medicine today. As Halamka explained, predictive AI looks at the experiences of millions of patients and their ailments to ask, ‘How can we minimize holes in the way and guarantee the best possible trip? ?” helps answer the simple question.
For example, someone is diagnosed with type 2 diabetes. Instead of giving general recommendations to people with this disease, the algorithm will make the best care plan for that patient based on their age, geography, racial and ethnic background, pre-existing medical conditions, and nutritional habits. predict the
This kind of patient-centered treatment is not new. Doctors have long individualized their recommendations. In this sense, predictive AI is just another tool to aid clinical decision making.
Algorithm quality is determined by the amount and diversity of data. It was amazing to learn that the Mayo Clinic team has signed data partner agreements with clinical systems across the United States and around the world, including Canada, Brazil and Israel. By the end of 2023, Halamka expects the organization’s network to cover more than 100 million patients, whose depersonalized medical records will be used to improve care for others. there is
Predictive AI can also enhance diagnostics. For example, to detect colon cancer, gastroenterologists typically perform a colonoscopy to manually identify and remove precancerous polyps. However, some studies estimate that a quarter of cancerous lesions are missed during colonoscopy screening.
Predictive AI significantly improves detection. The software is “trained” to identify polyps by looking at many pictures of polyps, and if it detects a polyp during a colonoscopy, it alerts the doctor to take a closer look. In a randomized controlled trial conducted at his eight centers in the United States, United Kingdom, and Italy, using such an AI, he more than halved his rate of missing potentially cancerous lesions from 32.4 percent to 15.5 percent. found to decrease.
Halamka made provocative remarks that could be considered medical malpractice within the next five years. no Utilize AI for colorectal cancer diagnosis. But he also cautioned that “this is not AI.” Exchange Doctor but AI enhance Ask your doctor to provide further insight. ” Technology will not reduce the needs of healthcare providers because there are too many unmet needs. Instead, “we will be able to see more patients in more areas,” he argued.
Generative AI, on the other hand, is “an entirely different kind of animal,” says Halamka. Some tools, such as ChatGPT, are trained based on unselected material found on the Internet. The input itself contains inaccurate information, so the model can produce inappropriate and misleading text. Furthermore, while predictive AI quality can be measured, generative AI models produce different answers to the same question each time, making validation more difficult.
Right now, there are too many concerns about the quality and accuracy of generative AI for directing clinical care. Still, it has great potential as a way to reduce the administrative burden. Some clinics already use apps that automatically transcribe patient visits. Physicians save valuable time by compiling medical records rather than creating them from scratch.
Halamka is clearly a proponent of using AI in medicine, but he wants federal oversight. Much like the Food and Drug Administration scrutinizes new medicines, we need a process to independently validate algorithms and publicly share the results. Additionally, Halamka defends efforts to prevent the perpetuation of existing medical biases in AI applications.
This is a careful and thoughtful approach. Like any tool, AI must be rigorously researched and carefully implemented, keeping in mind the caveat of “do no harm first”. Nonetheless, AI has incredible potential to make healthcare safer, more accessible and more equitable.
