Topol: AI in early stages, but has potential to revolutionize healthcare
Emerging technology has the potential to add efficiency and effectiveness to healthcare, according to Eric Topol, MD.
Emerging technology has the potential to add efficiency and effectiveness to healthcare, according to Eric Topol, MD.
Capabilities such as artificial intelligence, polygenic risk scores and digital health technology can equip physicians to improve and reduce the cost of care, while enabling them to better connect with patients, said Topol, the founder and director of The Scripps Research Institute.
Much of care today is not based on provable facts, Topol noted at Liberation 2019, the annual meeting of Medecision, in Frisco, Texas. He cited an article by Hannah Fry, MD, published last month in The New Yorker, which detailed research that found that, of every 1,000 people taking statins for heart conditions over five years, only 18 will avoid a major heart attack or stroke.
“Patients and clinicians exist in a world of insufficient data,” Topol said. “We need to stop shallow, error-laden, wasteful medicine. We can do this, but it takes the will to do this.”
The capabilities are there—for example, patients are now able to upload results from genetic testing services such as 23andMe to a free tool from the The Scripps Research Institute called MyGeneRank, which enables consumers to upload their data and then quickly get a genetic risk for various conditions. Results from MyGeneRank enable consumers to adopt behaviors that potentially can offset their genetic risks.
Clinicians are not universally supportive of using artificial intelligence, Topol contended, adding, “There’s some resistance in the medical community today.”
But certain specialties are advancing the use of AI, he noted. “Opthalmologists are into this—eye doctors are leading the charge,” he said. AI can discern problems in eye images, for example, by spotting issues with diabetics that may be able to prevent diabetic retinopathy, which can cause blindness. Symptoms of this condition are difficult for humans to spot in advance, but machine learning can discern the condition.
“Half of diabetic patients are never scanned for retinopathy,” he said.
Topol noted that almost 40 algorithms have been approved by the Food and Drug Administration, and he projects that the use of AI will result in about a 35 percent improvement in efficiency.
However, he admitted that the use of AI in healthcare is still in its early stages. “There are hundreds of retrospective studies underway, but few prospective efforts,” he said.
The use of AI does have to surmount issues, such as privacy and security, limited large datasets on which to train algorithms and concerns over the “black box” aspect of AI—demonstrating how machine learning reaches the conclusions it draws from data to alleviate the concerns of clinicians. However, he sees this fear is holding AI to a higher standard than current clinical practice.
“We’re expecting everything to be fully explainable with AI. However, with current clinical practice, there’s so much that we don’t understand, but we use (these treatments) every day. We are holding algorithms to a different standard, because we don’t have explainability for what we do today."
Capabilities such as artificial intelligence, polygenic risk scores and digital health technology can equip physicians to improve and reduce the cost of care, while enabling them to better connect with patients, said Topol, the founder and director of The Scripps Research Institute.
Much of care today is not based on provable facts, Topol noted at Liberation 2019, the annual meeting of Medecision, in Frisco, Texas. He cited an article by Hannah Fry, MD, published last month in The New Yorker, which detailed research that found that, of every 1,000 people taking statins for heart conditions over five years, only 18 will avoid a major heart attack or stroke.
“Patients and clinicians exist in a world of insufficient data,” Topol said. “We need to stop shallow, error-laden, wasteful medicine. We can do this, but it takes the will to do this.”
The capabilities are there—for example, patients are now able to upload results from genetic testing services such as 23andMe to a free tool from the The Scripps Research Institute called MyGeneRank, which enables consumers to upload their data and then quickly get a genetic risk for various conditions. Results from MyGeneRank enable consumers to adopt behaviors that potentially can offset their genetic risks.
Clinicians are not universally supportive of using artificial intelligence, Topol contended, adding, “There’s some resistance in the medical community today.”
But certain specialties are advancing the use of AI, he noted. “Opthalmologists are into this—eye doctors are leading the charge,” he said. AI can discern problems in eye images, for example, by spotting issues with diabetics that may be able to prevent diabetic retinopathy, which can cause blindness. Symptoms of this condition are difficult for humans to spot in advance, but machine learning can discern the condition.
“Half of diabetic patients are never scanned for retinopathy,” he said.
Topol noted that almost 40 algorithms have been approved by the Food and Drug Administration, and he projects that the use of AI will result in about a 35 percent improvement in efficiency.
However, he admitted that the use of AI in healthcare is still in its early stages. “There are hundreds of retrospective studies underway, but few prospective efforts,” he said.
The use of AI does have to surmount issues, such as privacy and security, limited large datasets on which to train algorithms and concerns over the “black box” aspect of AI—demonstrating how machine learning reaches the conclusions it draws from data to alleviate the concerns of clinicians. However, he sees this fear is holding AI to a higher standard than current clinical practice.
“We’re expecting everything to be fully explainable with AI. However, with current clinical practice, there’s so much that we don’t understand, but we use (these treatments) every day. We are holding algorithms to a different standard, because we don’t have explainability for what we do today."
More for you
Loading data for hdm_tax_topic #better-outcomes...