D42

67

theaiconf.com

Early adopters of applied AI have a unique opportunity to invent new business models, reshape industries, and build the impossible.

Put AI to work — right now.

AI is moving fast. Don’t fall behind.

Mike Barlow

AI and MedicineData-Driven Strategies for

Improving Healthcare and Saving Lives

Boston Farnham Sebastopol TokyoBeijing Boston Farnham Sebastopol TokyoBeijing

978-1-491-96145-2

AI and Medicineby Mike Barlow

Copyright © 2016 O’Reilly Media Inc. All rights reserved.

Printed in the United States of America.

Published by O’Reilly Media, Inc., 1005 Gravenstein Highway North, Sebastopol, CA95472.

O’Reilly books may be purchased for educational, business, or sales promotional use.Online editions are also available for most titles (http://safaribooksonline.com). Formore information, contact our corporate/institutional sales department:800-998-9938 or corporate@oreilly.com.

Editor: Nicole TacheProduction Editor: Colleen ColeCopyeditor: Rachel Monaghan

Interior Designer: David FutatoCover Designer: Karen Montgomery

July 2016: First Edition

Revision History for the First Edition2016-07-08: First Release2016-08-25: Second Release

See http://oreilly.com/catalog/errata.csp?isbn=9781491961452 for release details.

The O’Reilly logo is a registered trademark of O’Reilly Media, Inc. AI and Medicine,the cover image, and related trade dress are trademarks of O’Reilly Media, Inc.

While the publisher and the author have used good faith efforts to ensure that theinformation and instructions contained in this work are accurate, the publisher andthe author disclaim all responsibility for errors or omissions, including without limi‐tation responsibility for damages resulting from the use of or reliance on this work.Use of the information and instructions contained in this work is at your own risk. Ifany code samples or other technology this work contains or describes is subject toopen source licenses or the intellectual property rights of others, it is your responsi‐bility to ensure that your use thereof complies with such licenses and/or rights. Thisbook is not intended as medical advice. Please consult a qualified professional if yourequire medical advice.

Table of Contents

AI and Medicine:Data-Driven Strategies for Improving Healthcare and Saving Lives. . . . . 1A Wealth of Benefits for Millions of Patients 3Strength in Numbers 4Barriers to Entry 6Amplifying Intelligence with Patient Data 7Pursuing the Quest for Personalized Medicine 9Wearables and Other Helpful Gadgets 10Predicting Adverse Drug Interactions 10Machine Learning Is Key to Better, Faster Medical Research 11Insight from Yeast 12AI Is “Like a Small Child” 13It’s All About Sharing the Data 14Looking Ahead 15

iii

AI and Medicine:Data-Driven Strategies for

Improving Healthcare and SavingLives

For centuries, physicians and healers focused primarily on treatingacute problems such as broken bones, wounds, and infections. “Ifyou had an infectious disease, you went to the doctor, the doctortreated you, and then you went home,” says Balaji Krishnapuram,director and distinguished engineer at IBM Watson Health.

Today, the majority of healthcare revolves around treating chronicconditions such as heart disease, diabetes, and asthma. Treatingchronic ailments often requires multiple visits to healthcare provid‐ers, over extended periods of time. In modern societies, “the oldways of delivering care will not work,” says Krishnapuram. “We needto enable patients to take care of themselves to a far greater degreethan before, and we need to move more treatment from the doctor’soffice or hospital to an outpatient setting or to the patient’s home.”

Unlike traditional healthcare, which tends to be labor-intensive,emerging models of healthcare are knowledge-driven and data-intensive. Many of the newer healthcare delivery models will dependon a new generation of user-friendly, real-time big data analyticsand artificial intelligence/machine learning (AI/ML) tools.

1

Krishnapuram sees five related areas in which the application ofAI/ML tools and techniques will spur a beneficial revolution inhealthcare:

Population managementIdentifying risks, determining who is at risk, and identifyinginterventions that will reduce risk.

Care managementDesigning care plans for individual patients and closing gaps incare.

Patient self-managementSupporting and enabling customized self-care treatment plansfor individual patients, monitoring patient health in real time,adjusting doses of medication, and providing incentives forbehavioral changes leading to improved health.

System designOptimizing healthcare processes (everything from medicaltreatment itself to the various ways insurers reimburse provid‐ers) through rigorous data analysis to improve outcomes andquality of care while reducing costs.

Decision supportHelping doctors and patients choose proper dosage levels ofmedication based on most recent tests or monitoring, assistingradiologists in identifying tumors and other diseases, analyzingmedical literature, and showing which surgical options arelikely to yield the best outcomes.

Applying AI/ML strategies in each of those five areas will be essen‐tial for creating large-scale practical systems for providing personal‐ized and patient-centric healthcare at reasonable costs. In thisreport, I explore these areas and more through interviews conduc‐ted with leading experts in the field of AI and medicine.

2 | AI and Medicine: Data-Driven Strategies for Improving Healthcare and Saving Lives

A Wealth of Benefits for Millions of PatientsThe potential benefits of applying AI/ML to medicine and health‐care are enormous. In addition to improving treatment and diagno‐sis of various cancers, AI/ML can be used in a wide range ofimportant healthcare scenarios, including fetal monitoring, earlydetection of sepsis, identifying risky combinations of drugs, andpredicting hospital readmissions.

“Medicine and biology are very complicated and require humans tobe trained for a long time to be highly functional,” says Dr. Russ Alt‐man, director of Stanford University’s biomedical informatics train‐ing program. “It is intriguing that computers may be able to reachlevels of sophistication where they rival humans in the ability to rec‐ognize new knowledge and use it for discovery.”

ML and neural networks are especially useful, says Altman, for find‐ing patterns in large sets of biological data. Some of the most prom‐ising applications of ML in medical research are in the areas of“omics data” (e.g., genomics, transcriptomics, proteomics, metabo‐lomics); electronic medical records; and real-time personal health‐care monitoring via devices such as wearables and smartphones.

Real-time or near-real-time testing and analysis are particularly crit‐ical in self-management scenarios. For example, it’s essential forpeople with diabetes to monitor their blood sugar levels accurately.But waiting for a doctor or nurse to perform tests can impair theaccuracy of results and defeat attempts to manage the disease prop‐erly. “Let’s say a test shows your blood sugar is high,” says Krishna‐puram. “Maybe it was high because you ate too many carbs beforethe test, or didn’t sleep well the night before, or you were stressedout or didn’t get enough exercise that week. Each of those canimpact your blood sugar level.”

If your doctor relies on tests performed once every couple ofmonths at his or her office to set the proper dosage of your medica‐tion, it may be difficult to optimize your dosage and manage yourcondition effectively over time.

AI and ML tools can play a valuable role not only in analyzing testresults rapidly and optimizing dosages of medications, but also inprompting behavioral changes by communicating timely remindersto exercise, eat healthier foods, and get more sleep.

A Wealth of Benefits for Millions of Patients | 3

“People also need to change their behaviors,” says Krishnapuram. AIand ML can motivate and reinforce behavioral changes by “orches‐trating” multiple channels of communication between healthcareproviders and patients.

Strength in NumbersThe organized practice of medicine can be traced back to 3,000 BC.Although early physicians relied on supernatural phenomena toexplain the origin of many diseases, the idea of developing practicaltherapies for common ailments is not new. Even when the causes ofdisease were grossly misunderstood, physicians were expected tofind remedies or provide effective treatments for patients who weresick or injured.

Today, medicine is widely regarded as a science. New therapies areinvented. If they seem promising, they are scientifically tested. Thetests are carefully analyzed with rigorous statistical processes. If atherapy is shown to be safe and effective in a large enough numberof cases, it is approved and used to treat patients.

But in reality, that’s where the science often grinds to a halt. Theoverwhelming majority of healthcare practitioners aren’t scientists.The term medical arts isn’t merely romantic—it’s an accuratedescription of how medicine is practiced in most of the world.

The application of AI, ML, and other statistical processes to medicalpractice—as opposed to just medical research—would be a leap for‐ward on the scale of the Industrial Revolution.

If the revolution fails, however, “we’ll look back at this century withthe same sense of horror we feel when we look at previous centu‐ries,” says Nate Sauder, chief scientist at Enlitic, a company thatdevelops ML technology for medicine. “Our feeling is that medicine—and in particular, medical diagnostics—is very much a data analy‐sis problem,” Sauder says. “Patients generate lots of data, everythingfrom genomic sequences to images from CT scans. It’s a natural fitfor machine learning techniques.”

For example, Sauder and his colleagues at Enlitic are helping medi‐cal radiologists improve the accuracy of their diagnoses. “We choseradiology because most of the reports and images are already in dig‐ital form, which makes it easier to manage the data. There’s also

4 | AI and Medicine: Data-Driven Strategies for Improving Healthcare and Saving Lives

been an explosion in the improvement of computer vision technol‐ogy.”

The combination of accessible data, high-quality computer vision,and ML techniques has the potential for improving the quality ofcare for millions of patients worldwide. “We started with a couple ofthe harder problems in radiology to validate our approach,” saysSauder. “For example, early discovery of lung nodules in a chest x-ray is incredibly important because there’s a huge difference in thesurvival rate between Stage 1 and Stage 4 cancer. We were able toidentify lung nodules 40–50 percent more accurately than a radiol‐ogist.”

One reason AI/ML processes can outperform humans is thathumans get tired after staring at screens for long periods of time.Another reason is that even in ideal conditions, it’s often difficult forhumans to spot small cancers on a lung scan. “What makes thisreally challenging is that your lungs have a bunch of tiny veins run‐ning through them. In a cross-sectional slice, a small mound of can‐cer and a tiny vein look very similar,” Sauder says.

It’s “easier” to see the difference between tumors and veins in three-dimensional scans, but human radiologists often find it difficult toread 3D images. Software, on the other hand, can be trained to read3D images as easily as 2D images. “As a result, a computer can lookat a three-dimensional scan and can spot tumors more accuratelythan a human,” says Sauder. “Additionally, a machine learning sys‐tem can look at 50,000 cases in the time it takes for a human to lookat one case. Those advantages can be translated into saving lives.”

Workflow integration, however, is a key ingredient in determiningthe success of an AI/ML product or service. “We really need toappreciate that many radiologists will view machine learning as areplacement for them or as a challenge to their established work‐flow,” says Sauder.

Like many of the experts interviewed for this report, Sauder sees AIand ML tools and techniques as aids, not replacements, for health‐care providers. He predicts AI and ML will become accepted com‐ponents of the medical diagnostic toolkit when their benefits aremore widely understood throughout the medical community.“Machine learning can improve diagnostics in two fundamentalways. First, it can help doctors perform diagnoses more quickly andmore accurately. Second, and perhaps more important in the long

Strength in Numbers | 5

term, is applying machine learning to screening. Screening is expen‐sive and churns out many false positives. But with machine learning,the computer can look at several hundred million screens and findthe smaller, weirder things that we humans tend to miss,” he says.

The long-range promise of machine learning is its ability to sortthrough very large numbers of screens and discover subtle or hid‐den patterns linking diseases with hundreds of variables, includingbehavior, geography, age, gender, nutrition, and genomics. “Thosehundreds of millions of screens create very rich data sets that can beculled by machine learning systems for medical insight,” says Sauder.

Barriers to EntryDespite the promise and potential of AI and ML to revolutionizemedicine, the majority of healthcare providers stick with traditionalprocesses to diagnose and treat patients. Part of the problem issemantics. For many people, “artificial intelligence” still evokesimages of sentient computers taking over the world, and very fewpeople understand the basic concept of “machine learning.”

As a result, discussions about applying AI/ML techniques in health‐care scenarios tend to be one-sided and uncomfortable. On theother hand, most people agree that healthcare is expensive, incon‐venient, and often ineffective. There is a genuine hunger for afforda‐ble solutions to modern healthcare problems, but it’s difficult formost people to understand how AI and ML can help.

Another roadblock to more widespread usage of AI and ML in med‐icine is extensive government regulation, which often puts a damperon innovation and creativity. “You can’t just drop new software intoa medical monitor device,” says Josh Patterson, director of field engi‐neering at Skymind, an open source, enterprise deep-learning pro‐vider. “There are many regulations that create barriers to entry,making it difficult for smaller companies to compete.”

Long integration cycles also slow the adoption of new approachesbased on AI, ML, deep learning, and neural networks. “Hospitals arenotoriously hard to sell into unless you are an already establishedvendor, and established vendors are less inclined to aggressivelyoffer new features once they have the contract,” says Patterson. “Ifthe established vendor does want to offer a new ML or AI feature,then they have to figure out how to integrate it into their product.”

6 | AI and Medicine: Data-Driven Strategies for Improving Healthcare and Saving Lives

There are four broad obstacles to wider adoption of AI/ML techni‐ques in healthcare, according to Krishnapuram:

• Confusion around data ownership and privacy. AI/ML pro‐cesses are fueled by data. But which set of stakeholders ownsmedical data? Is the data owned by patients, doctors, hospitals,research centers, or technology vendors? Can medical data bemined for clinical insights without compromising privacy orviolating existing regulations?

• Dysfunctional incentives. In its current form, the healthcarepayment system revolves around volume of care. Shifting to asystem that rewards quality of care and improved outcomes willrequire a fundamental overhaul of most healthcare models.

• Liability and responsibility. It’s not clear which parties wouldbe held accountable when something goes wrong with an AI orML system. Who bears the risk? Who is responsible and whopays for damages? Can an AI system be sued for malpractice?

• The traditional research paradigm doesn’t support personal‐ized medicine. How do you conduct statistically meaningfulclinical trials when each patient is treated individually and everycare plan is customized for an individual patient? How do youestablish baselines, set standards, and develop common proce‐dures when each patient is a “market of one”?

“Those aren’t trivial questions,” says Krishnapuram. Resolving themwill require study, public debate, legal reform, and the emergence ofa new social consensus around the value of data analytics.

Amplifying Intelligence with Patient DataGiven the obstacles, it’s easy to see why healthcare organizationshave been slow to adopt big data and AI/ML solutions. That said, itis imperative for society to find practical ways for solving wide‐spread healthcare issues. AI/ML techniques offer the best and fastestpath to achieving the goals of personalized, outcome-based medi‐cine.

“Compared to other domains, such as retail and finance, healthcareis the least developed field in terms of AI and ML,” says Eric Xing, aprofessor in the School of Computer Science at Carnegie MellonUniversity. Xing has two PhD degrees, one in molecular biology

Amplifying Intelligence with Patient Data | 7

from Rutgers University and another in computer science from UCBerkeley.

Lack of medical data isn’t a problem, he says. “There’s a lot of data…from patients, from doctors, and from scientific studies. But the datais underused. It just sits in databases.”

For example, healthcare providers collect clinical data from patientsevery day. But most of that information is seldom used. “It is first‐hand information, collected directly from patients. It’s incrediblyvaluable, but it’s rarely looked at again unless the same patientcomes back for a visit,” says Xing. “So we’re not making effective useof the data.”

Xing’s team at Carnegie Mellon is developing an AI program to inte‐grate patient data from multiple sources such as x-rays, blood tests,tissue samples, demographics, and freehand notes from caregivers.“Once we have highly integrated data from patients, we can deploy itin a machine learning algorithm and generate predictive models,” hesays.

For instance, the data can be used in analytics that would help a doc‐tor assess the risks of subdiseases associated with a patient’s primaryailment or help the doctor predict the symptoms a patient is likely toexperience before a follow-up examination.

AI programs can also help doctors devise safe, effective, and practi‐cal treatment plans for individual patients. “It’s usually very difficultfor a doctor to come up with a treatment plan unless the doctor haslots of experience treating similar patients,” says Xing. “With an AIsystem, you can look at all the potential dangers and get a better ideaof what can go wrong. The system’s knowledge base includes mil‐lions of patients, and an algorithm would allow the doctor to searchfor similar patients in a matter of seconds.”

In a very real sense, AI and ML systems enable individual doctors toexpand their medical knowledge and experience far beyond whatwould be possible under traditional circumstances. “You can con‐nect one patient with a database of patients, making it easier to gaindeeper insights into disease mechanisms,” says Xing. From thepatient’s perspective, “it’s like assembling a large team of doctorswith vast experience.”

8 | AI and Medicine: Data-Driven Strategies for Improving Healthcare and Saving Lives

Like Sauder, Xing does not envision AI as a replacement for doctorsand other healthcare providers. “The goal of an AI system is notreplacing the doctor in a clinical setting. The doctor is still centerstage, making decisions and delivering care.”

Xing used the analogy of autopilot systems designed for airplanes.“They aren’t a substitute for human pilots. The humans in the cock‐pit still make the important decisions when they see a problem. Theautopilot just helps them by making it easier to fly the plane,” hesays.

Pursuing the Quest for Personalized MedicineThe idea of personalized precision medicine has been around formore than two decades, but AI and ML have the potential for bring‐ing it closer to reality. “Personalized medicine is built on the uniquegenetic characteristics of a patient,” Xing explains. “But it’s very dif‐ficult to practice because we don’t really understand the entiremechanisms and genetic underpinnings of many diseases. Also, thedata is hard to understand. You have a million polymorphic sites,and you don’t know which one of them is actually causing the dis‐ease or just along for the ride.”

Typically, doctors check a handful of key mutations that are gener‐ally believed to be associated with a particular patient’s disease—buttaking that shortcut effectively circumvents the value of personal‐ized medicine. “When you look at a mutation that is common tomany patients with a specific disease, you lose the power of person‐alization because everyone will be treated the same way,” Xing says.

Xing and his colleagues are building machine learning programsthat analyze an individual patient’s genomic, proteomic, and meta‐bolic data—including incremental risk factors—to generate a highlypersonalized profile for the patient. “You can use machine learningmodels for deriving the unique patterns underlying specific diseasesand symptoms, as well as for identifying potential targets for drugs,”he says.

Pursuing the Quest for Personalized Medicine | 9

Wearables and Other Helpful GadgetsAI will also be integral to the development of genuinely useful wear‐able and mobile devices for improving health. In addition to moni‐toring vital signs such as pulse, blood pressure, and respiration, thenext generation of mobile health tech would also provide personal‐ized real-time alerts and recommendations for modifying behaviorto achieve specific health goals.

“The mobile health domain will increasingly become part of every‐day life,” says Xing, who is working with his team on a mobile app tohelp patients with Parkinson’s disease. “It’s not just a passive timerreminding you when to take your medications. It will actually moni‐tor your past, present, and future activity. It will monitor your envi‐ronment and your risk levels. Then it will provide active suggestionsfor dosage, timing, and frequency of medication, and offer precau‐tions and advice for lowering your risk.”

Xing says mobile platforms are the best way to provide patients withreal-time feedback and advice. “But generating those servicesrequires AI, because the platform must learn from the patient’s dataand from existing medical data. It must be able to detect patterns inbehavior and then make helpful recommendations based on thosepatterns,” he says.

Predicting Adverse Drug InteractionsBartenders will tell you, “Never mix, never worry.” But manypatients take more than one medication, and not everyone reacts thesame way to various combinations of drugs.

“Humans aren’t very good at predicting when two drugs will interactand cause problems,” says Nicholas Tatonetti, an assistant professorof biomedical informatics at Columbia University and a member ofthe Data Science Institute. Two drugs that are harmless when usedseparately might cause adverse reactions when used together by thesame patient. Predicting “drug/drug interactions,” however, is noto‐riously difficult.

In one of their recent projects, Tatonetti and his lab colleagueslooked for pairs of drugs that might cause cardiac arrhythmia. “Wegathered 20 years of medical record data from Columbia and

10 | AI and Medicine: Data-Driven Strategies for Improving Healthcare and Saving Lives

trained a machine learning algorithm to look for drug interactionswith a high probability of causing a heart arrhythmia,” Tatonettiexplains. “The algorithm initially came up with about 1,000 drug/drug interaction hypotheses. Then we investigated those interactionsand evaluated them for causality. We narrowed the field down toabout 20 interactions—using data analysis only. There was nohuman intuition involved at all.”

Eventually, the machine learning algorithm identified a combinationof two drugs, ceftriaxone and lansopravole, which can generate theconditions leading to heart arrhythmia. “That is a hypothesis thatnobody would have ever explored before, since those two drugs arenot suspected of causing this problem,” says Tatonetti. “Because thealgorithm we trained to look for arrhythmia found a pattern, it wasable to identify this new and potentially dangerous drug interac‐tion.”

Machine Learning Is Key to Better, FasterMedical ResearchHuman beings are great at seeing “the big picture.” We assemble auniverse around ourselves by sampling bits and scraps of informa‐tion, and then creating stories and narratives on the fly. We’re alwaystaking mental shortcuts—the “fast thinking” heuristics described sowell by Daniel Kahneman and Amos Tversky.

But when it comes to understanding and managing complex phe‐nomena—like cancer and Alzheimer’s disease—our innate humanability to rapidly leap from a handful of facts to a sweeping conclu‐sion is our Achilles heel.

Machine learning is a potential antidote to our highly evolved, butnot always useful, talent for manufacturing reality from informationgathered by our senses. Machine learning excels at identifying latentpatterns and connections that we are too highly evolved to perceive.We create myths by ignoring or skipping over details. Machinelearning, on the other hand, happily feeds on minutia.

Most diseases, it turns out, are made up of smaller subdiseases,which themselves are caused by even smaller subdiseases. Multiplelayers of interrelated biological processes are involved, making ithighly difficult to apply simplistic “rule of thumb” approaches.

Machine Learning Is Key to Better, Faster Medical Research | 11

Machine learning’s ability to find patterns and to uncover hiddenrelationships among subdiseases is what makes it especially attrac‐tive to medical researchers. For example, one of the harder problemsin medical research is bridging the gap between genetics and diseasephenotypes. Here’s a quick and useful definition of the genotype/phenotype distinction from the Stanford Encyclopedia of Philosophy:

The genotype is the descriptor of the genome which is the set ofphysical DNA molecules inherited from the organism’s parents. Thephenotype is the descriptor of the phenome, the manifest physicalproperties of the organism, its physiology, morphology, and behav‐ior.

Despite an abundance of genomic and phenotype data, bridging thegap between the genome and disease phenotypes requires a shift tocomputational models that incorporate the causal complexity inher‐ent in our biology, says David Beyer, a principal at Amplify Partnersand author of The Future of Machine Intelligence: Perspectives fromLeading Practitioners.

“In the last decade, researchers have transitioned from the applica‐tion of shallower machine learning techniques (primarily linear innature) to a new class of approaches, including deep learning, a sub‐class of ML broadly defined around the idea of multilayered neuralnetworks,” says Beyer. “And just as deep learning has shown break‐through performance in categories such as vision, the hope is toextend that success to biology and medicine.”

Insight from YeastThe genotype-phenotype divide has limited the practical value ofgenomic science in treating disease, since people with the samegenetic mutations can experience different symptoms of the samedisease, or in some instances, experience no symptoms at all.

Genomic medicine is also an area in which machine learning tech‐niques can generate highly valuable insights. At Tatonetti’s lab,researchers studied yeast genetics to understand why some humangene mutations are harmless by themselves, but deadly when com‐bined with other mutations. The phenomenon is called syntheticlethality, and it’s a hurdle that makes it difficult to use genetic infor‐mation for curing human diseases.

Understanding synthetic lethality in humans is critical to developingtargeted and personalized cancer therapies that spare healthy cells

12 | AI and Medicine: Data-Driven Strategies for Improving Healthcare and Saving Lives

while killing cancer cells. “We know a lot about synthetic lethality inyeast, but it’s hard translating that knowledge from yeast tohumans,” Tatonetti explains. “Humans have 5 to 10 times more pro‐teins than yeast. So the number of potential interactions is exponen‐tially higher in humans than in yeast.”

Many of the previous attempts to use yeast for understanding moreabout human disease had been unsuccessful because they focusedon the mechanism of the proteins themselves. “We took a differentapproach,” says Tatonetti. “We set up a supervised machine learningalgorithm and told it which pairs of genes were synthetic lethal toyeast. Then we applied the algorithm we had trained on the yeast tomaking predictions for human genes. The algorithm didn’t know itwas looking at human genes; it just ran. And it predicted about amillion lethal pairs of human genes.”

The team then compared the algorithm’s output to a previous highlydetailed investigation of lethality in human cancer cells. “We valida‐ted our findings against the smaller ‘gold standard’ set and foundthat we had achieved practically the same performance.”

Tatonetti and his colleagues successfully deployed a machine learn‐ing algorithm for translating knowledge from unicellular organismsto multicellular organisms. “Instead of trying to understand thefunctions of every protein in the human body, we let the machineidentify the important patterns for us,” he says.

AI Is “Like a Small Child”If there’s a rock star of AI in medicine, it’s Dr. Lynda Chin, associatevice chancellor and chief innovation officer at the University ofTexas System. “The human brains are limited in their capacity,” shesays. “Medicine is becoming more and more complex, as more andmore data are collected about the patients and the medical knowl‐edge base grows exponentially. No single human being can possiblykeep up, especially if their job is taking care of patients. We needhelp.”

Chin sees AI as a helpful tool for augmenting human cognitive capa‐bilities. AI would serve doctors in much the same way that para‐legals or law clerks serve trial lawyers and judges. Paralegals andclerks aren’t substitutes for lawyers and judges, yet they are neces‐sary for an effective legal system. For example, AI can help doctors

AI Is “Like a Small Child” | 13

organize and synthesize the ever-increasing amount of data—aboutthe patients, the disease, and the treatment options—into intelli‐gence that is actionable.

“Imagine if a doctor can get all the information she needs about apatient in 2 minutes and then spend the next 13 minutes of a 15-minute office visit talking with the patient, instead of spending 13minutes looking for information and 2 minutes talking with thepatient,” says Chin.

However, she describes AI in medicine as maturing, still a smallchild who is growing up fast. “Training AI systems to be useful inmedicine is like parenting—no easy task! Not only does the underly‐ing AI analytics need to mature, application of AI in medicine itselfis a brand new challenge that requires an iterative learning process.”

It’s All About Sharing the DataFrom Chin’s perspective, one of the biggest barriers to developingand applying AI in medicine is access to longitudinal medical andother health-related data that truly represent the diversity of thepatient population and the heterogeneity of the diseases. “These dataare all over the place, not shared, and worse yet, not standardized,with each silo being too small and too narrow...which means they’renot good for training an AI system,” she says.

Learning from her earlier work in partnership with IBM Watson todevelop MD Anderson Oncology Expert Advisor©, a virtual cogni‐tive expert system designed to democratize cancer care knowledgeand expertise, Chin believes that the promise of AI in medicine willremain elusive until the barrier to data is removed. “The sharingneeds to go beyond individual hospitals or hospital systems,” shesays, “because no single entity has enough data.”

In her effort to remedy the data challenge, Chin began working withPricewaterhouseCoopers to develop a “super-compliant” cloud plat‐form for sharing medical data and analytic insights safely andsecurely across disparate institutions and organizations.

A necessary component of the platform is a governance frameworkto assure all stakeholders that the data will be used only for the spe‐cific purpose, with no unspecified secondary uses. “Rarely anyoneobjects to the stated use of data,” she says. “It’s the unintended orunexpected use of private data that worries most people. We need to

14 | AI and Medicine: Data-Driven Strategies for Improving Healthcare and Saving Lives

acknowledge the importance of keeping data private, even when it isshared.”

To achieve that goal, her group is also partnering with AT&T todevelop secure dedicated networks for transmitting healthcare data.“We need security and privacy when data is transmitted, not justwhen it’s at rest,” she says.

In addition to a cloud-based infrastructure that can securely connectto disparate data sources across health and other related industries,she hopes to see “more data from more patients and more institu‐tions, more networking and more aggregating of data from moresources.”

Chin also envisions application of these novel technologies andcapabilities in the battle against access and affordability of health‐care for the disadvantaged in medical desert areas. “We need tothink about providing care for the people who can’t afford or don’thave access to healthcare,” she says. “AI, along with wearables andmobile devices, can potentially extend more affordable and qualitycare to these people.”

Looking AheadAs a sexagenarian Baby Boomer, I now interact with the healthcaresystem more often than ever before. From my perspective as some‐one who writes frequently about data science, I am regularly aston‐ished and dismayed at how poorly my medical information iscollected, stored, analyzed, and shared. No retailer would ever coun‐tenance the indifference to data that is routinely demonstrated byhealthcare workers at practically every level.

As a child, I was taught that medicine is both a science and an art. Ihave seen the art part in action. Watching skilled medical practition‐ers set broken bones or slice away diseased tissue is like watchingmiracles being performed.

Like many people, I’m still waiting for the scientific part of medicineto really kick into high gear. Clearly, AI and its various componentshave the potential to play enormous roles in improving manyaspects of healthcare. But the full potential of AI in medicine won’tbe realized until there’s a new social consensus on healthcare data.

Looking Ahead | 15

It’s time to begin a national dialogue about how we treat healthcaredata. Will we treat it as private property that is owned and sold, orwill we treat it as common property that is shared freely? Theanswer to that question will largely determine the eventual impact ofAI on medicine.

16 | AI and Medicine: Data-Driven Strategies for Improving Healthcare and Saving Lives

About the AuthorMike Barlow is an award-winning journalist, author, and communi‐cations strategy consultant. Since launching his own firm, CumulusPartners, he has worked with various organizations in numerousindustries.

Barlow is the author of Learning to Love Data Science (O’Reilly,2015). He is the coauthor of The Executive’s Guide to EnterpriseSocial Media Strategy (Wiley, 2011), and Partnering with the CIO:The Future of IT Sales Seen Through the Eyes of Key Decision Makers(Wiley, 2007). He is also the writer of many articles, reports, andwhite papers on numerous topics such as smart cities, ambient com‐puting, IT infrastructure, predictive maintenance, data analytics,and data visualization.

Over the course of a long career, Barlow was a reporter and editor atseveral respected suburban daily newspapers, including the JournalNews and the Stamford Advocate. His feature stories and columnsappeared regularly in the Los Angeles Times, Chicago Tribune, MiamiHerald, Newsday, and other major US dailies. He has also writtenextensively for O’Reilly Media.

A graduate of Hamilton College, he is a licensed private pilot, avidreader, and enthusiastic ice hockey fan.