By Allison Proffitt
March 5, 2020 | Data is the most valuable resource on the market today, Eric Topol told the crowd at the Molecular Medicine TriConference in San Francisco this week. At least according to the Dark Web, which, the founder and director of Scripps Research Translational Institute and cardiologist reported, holds medical data at five times the worth of personal financial data.
And there is a lot of it. We’ve long since passed data measurements in the terabyte, petabtye or even exabyte quantities, move now into zettabytes and yottabytes. Beyond that the nomenclature is up for debate. Topol recommends “helluvabyte.”
But unlike in Estonia, individuals don’t own and can’t access all of their health data, and it hasn’t improved healthcare for patients. In truth, we are individually so unique our own data are big—our genomes, phenomes, exposomes, and more. We are as bad at managing and making use of those data as we are at processing the data from the healthcare system as a whole.
And the situation is hard on doctors as well as patients. Doctors are the only profession that are managed by an entirely different type of professional, Topol said. Hospital and healthcare practice administrators are no physicians and their goals and priorities rarely match.
As a result, doctors are rushed, stressed, depressed. Physician burnout is a serious—and sometimes deadly—problem. Doctors have time only for diagnoses that are fast, intuitive, reflexive, automatic and shallow. Shallow medicine, Topol emphasized, is not our goal. The solution is neural networks. Not AI to replace doctors, but AI to empower doctors. Machines to bring humanity back to medicine.
Topol quickly ran through AI’s uses in various medical disciplines. Dermatologists and pathologists regularly use computer vision to diagnose skin cancers and medical images. AI can predict driver mutations for certain cancers, and for liver cancer, AI-assisted trainees performed as well as experts.
AI in these cases helps process the data, make predictions, and guide the physicians. And Topol argues that implementing this sort of assistance into our growing sensor data pool—particularly smart watch biosensors—and microbiome analysis are particularly promising. There is much opportunity for sensors to measure blood glucose and even depression and machine learning algorithms can help parse the data.
But Topol is a proponent of any technologies that remove the distance between patient and physician. Topol lauded the iPhone-connected ultrasound (his model is made by Phillips). “Within seconds of putting it on the chest you can see everything!” he marveled. “It is remarkable insight in seconds.”
He has imaged his entire body—from heart to gall bladder to his left foot—and his 6-year-old grandson can now perform an echocardiogram, he said, guided by voice instructions of where to place the sensor and which direction to tilt it. When the right image comes into view, software automatically captures it.
This isn’t a parlor trick. Topol employed his device to diagnose his own kidney stones. A finding he later paid $3,400 to confirm via CT. And this is the missed opportunity that AI can help facilitate—an intimate connection with the doctor, sharing findings with patients instantly, something an imaging tech can’t do.
Drug discovery, on the other hand, is still ripe for AI’s input. There are no AI-developed drugs just yet, though Topol mentioned compounds being tested by Deep Genomics and In Silico Medicine. There have been hundreds of retrospective studies using AI, but only six randomized controlled trials, and 11 prospective trials, he said.
AI will not “naturally” make medicine better, he said. There is ample opportunity to use the technologies badly. But there is great evidence, Topol says, that AI can give patients autonomy when desired, and reconnect physicians and patients when needed. AI assistance will improve both care and the doctor-patient relationship, he argued.
“Machines will not replace physicians,” Topol said, quoting Antonia Di leva, “but physicians using AI will soon replace those not using it.”
Editor’s note: This article also ran on Bio-IT World.