TRYING FOR THE TRIFECTA TELEHEALTH MEETS AI MEETS CYBERSECURITY By Thomas P. Keenan he social disruption of the COVID- and chest X-rays, after about five or six world—things like Wikipedia and so on— 19 pandemic has accelerated he’s getting tired, and after ten he’s “ready and used that data to learn how to answer Ttwo powerful trends in health to go the bar.” By contrast, the machine questions about the real world.”6 care—telemedicine (TM) and artificial treats each case with unbiased, fresh eyes, Today, we think nothing of saying, “Hey intelligence/machine learning (AI/ML). giving consistent results. “It [PUFF] also Google, turn off all the lights” or “Alexa, tell Advocates of digital transformation let out knows things I haven’t a clue about,” he me a joke about rabbits,” confident that our of collective “Yes!” as long-awaited changes added. One of the many experts who AI-enabled speakers will understand and happened almost overnight in early 2020. contributed knowledge to PUFF was a obey us. Because their knowledge base and Then, the ants of cybersecurity came along tropical medicine specialist. This contri- ML functionality reside in a remote com- to spoil the picnic! bution enabled the system to spot a rare puter, our virtual assistants will keep getting First, the good news. lung problem caused by breathing bat smarter and better informed. Perhaps • Suddenly, it became acceptable to guano in Central American caves. Alexa will have a joke about “one-legged “see” your doctor electronically; PUFF was an example of a rule-based rabbits” the next time I ask “her.”7 you could even send a picture of expert system. It sifted through care- AI is already embedded in many aspects that wart on your foot, though per- fully formatted symptoms and lab results, of today’s medicine, including surgery. haps not that rash in your groin.1 then spewed out a result like “THE LOW According to trade publication Robot- Health insurers, including Medi- DIFFUSING CAPACITY, IN COMBI- ics Online, “AI can determine patterns care, that had frowned upon NATION WITH OBSTRUCTION AND within surgical procedures to improve telemedicine suddenly decided it A HIGH TOTAL LUNG CAPACITY IS best practices and to improve a surgical was a great idea. CONSISTENT WITH A DIAGNOSIS robots’ control accuracy to submillimeter • The success of countries like South OF EMPHYSEMA.”4 Impressive for its precision.”8 Korea in “stopping the virus in time, but PUFF would feel like a dummy How about chest X-rays? In the almost its tracks”2 demonstrated that a (if programs could feel—a fascinating legal forty years since PUFF premiered, have technology-based, data-driven question5) beside one of today’s $40 smart we figured out how to automate that? approach to a major health prob- speakers. The answer is a qualified yes. In a study lem could actually work. It also Expert systems like PUFF had their day, published in 2019,9 Giovanni Montana raised troubling issues of pri- but real progress in AI required two more and colleagues used 470,388 adult chest vacy as some countries enforced things: massive computing power and new X-rays that had been analyzed by human quarantine with mandatory smart- approaches. radiologists to train a neural network. They phone apps, license plate readers, We got the first, in 1997, when IBM’s then gave the system an independent set even tracking of people through Deep Blue program stunned the world by of 15,887 X-rays to review. The program cell phone pings and credit card defeating human chess champion Garry called true positives “positive” seventy- transactions. Kasparov. It used specialized hardware and three percent of the time and identified true a brute force approach that could evalu- negatives ninety-four percent of the time. ARTIFICIAL INTELLIGENCE AND ate 200 million chess positions per second, While that isn’t perfect, it could certainly MACHINE LEARNING applying rules like working out the value of help to unclog the X-ray reading backlogs. High hopes for AI in medicine go back having your king in a safe position. The authors conclude that “we have dem- decades. In the 1980s I interviewed one Of even greater relevance is the 2011 onstrated the feasibility of AI for triaging of the early adopters of PUFF, an expert victory of the same company’s Watson chest radiographs.”10 system for the diagnosis of lung disease.3 over human Jeopardy! champs Brad Rut- One of the problems with AI in medi- I asked this doctor if PUFF was any good. ter and Ken Jennings. According to AI cine is that it can be too observant, finding “It’s a better diagnostician than I am,” he expert Murray Campbell, that system patterns that are not clinically meaning- laughed. The doctor further explained “used a machine-learning-based system ful. Speaking to Science, neurosurgeon Eric that, when given a stack of patient files that took a lot of data that existed in the Oermann noted that many of his hospital’s 10 TheSciTechLawyer FALL 2020 Published in The SciTech Lawyer, Volume 17, Number 1, Fall 2020. © 2020 American Bar Association. Reproduced with permission. All rights reserved. This information or any portion thereof may not be copied or disseminated in any form or by any means or stored in an electronic database or retrieval system without the express written consent of the American Bar Association. Published in The SciTech Lawyer, Volume 17, Number 1, Fall 2020. © 2020 American Bar Association. Reproduced with permission. All rights reserved. This information or any portion thereof may not be copied or disseminated in any form or by any means or stored in an electronic database or retrieval system without the express written consent of the American Bar Association. sickest patients had their X-rays done with by the appointment lasting over fifteen one example, it is now possible to buy portable machines. The hospital’s AI algo- minutes. There was some suggestion that a “medical-grade six lead EKG” from rithm started to incorrectly associate the doctors were being sloppy in tracking this, Amazon for $149.19 mere use of a portable unit with greater or even trying to “game” the system. With Millions of people wear fitness track- illness.11 telemedicine, data are collected automati- ers religiously, even to bed. This led to the No technology is perfect. Even Ama- cally and accurately. wonderful Gizmodo headline “Your Fuel- zon’s Alexa, which has been trained on TM can also make services available to band Knows When You’re Having Sex.”20 billions of interactions with millions of a wider population than traditional deliv- If you burn 100 calories in the middle of voices, can spew out some honkers. In a ery models. In the U.S., Congresswoman the night while taking zero steps, it knows YouTube video, an innocent toddler asks Robin Kelly (D-Illinois) has introduced what you’re doing. The only question is— Alexa to “Play Twinkle Twinkle.” The sys- a bill in the House of Representatives to who is it going to tell? tem hears this as “pussyanaldildo” and study the effects on the sudden move to Being ratted out by your wearable is begins to reply to that query, to the hor- doctors at a distance. She believes that not just a hypothetical risk. According to rified shouts of the adults in the room.12 technology can act as an “equalizer” in one news report, “a woman caught her Perhaps it was no accident. Alexa uses healthcare delivery.15 boyfriend cheating when his Fitbit activ- past interactions and machine learning It can also be a great compromiser of ity spiked at 4 a.m.”21 to tailor her responses to individual user personal privacy. While that data dump may not have preferences. One commentator on that When COVID-19 struck, Zoom rap- been in the boyfriend’s best interest, many video made this terse observation—“Dad’s idly became the most popular of the experts believe that collecting health data browser history!” videoconference platforms and was on a mass scale, and analyzing it with AI/ adopted by schools, businesses, and, of ML, may produce the greatest boon medi- TELEHEALTH AND course, healthcare providers. Part of its cine has ever seen. Projects like Columbia TELEMEDICINE appeal was simplicity. You could just send University–based OHDSI (ohdsi.org) are Canada, where I live, has a long history someone a link and they were in your pioneering a big data approach to medi- of using communications technology video meeting. You didn’t even need to cal research. for healthcare delivery. This is largely bother with a pesky password. A 2019 OHDSI-enabled paper pub- driven by geography, given the size of This quickly led to “Zoombombing,” lished in The Lancet “used insurance claim the country. It makes little economic where uninvited guests dropped into data and electronic health records from sense to fly a patient 1,000 km each way video sessions, disrupting them with 4.9 million patients across nine obser- from Churchill to Winnipeg, Manitoba, their comments or, often, their nudity. In vational databases, making it the most to have a toenail fungus checked. Also, addition, Zoom’s naming convention for comprehensive one ever on first-line Canada’s single-payer healthcare system stored recordings was easy to guess. The antihypertensives.”22 It also produced makes the decision to use technol- Washington Post reported “Thousands of a surprising result—“the most popu- ogy much simpler than in a world with Zoom Video Calls Left Exposed on Open lar hypertension drug isn’t the most HMOs and insurance companies and Web,” and said these included therapy ses- effective.”23 private and public providers.