You are on page 1of 3

The NEW ENGLA ND JOURNAL of MEDICINE

Perspective September 28, 2017

Lost in Thought The Limits of the Human Mind


and the Future of Medicine
Ziad Obermeyer, M.D., and ThomasH. Lee, M.D.

I
n the good old days, clinicians thought in come maddeningly complex. Pa-
Lost in Thought

groups; rounding, whether on the wards or tients and clinicians want simple
answers, but we know little about
in the radiology reading room, was a chance whom to refer for BRCA testing
for colleagues to work together on problems too or whom to treat with PCSK9 in-
hibitors. Common processes that
difficult for any single mind to Its ironic that just when clini- were once straightforward
solve. cians feel that theres no time in ruling out pulmonary embolism or
Today, thinking looks very dif- their daily routines for thinking, managing new atrial fibrillation
ferent: we do it alone, bathed in the need for deep thinking is now require numerous decisions.
the blue light of computer screens. more urgent than ever. Medical So, its not surprising that we
Our knee-jerk reaction is to knowledge is expanding rapidly, get many of these decisions wrong.
blame the computer, but the with a widening array of ther Most tests come back negative,
roots of this shift run far deeper. apies and diagnostics fueled by yet misdiagnosis remains com-
Medical thinking has become advances in immunology, genet- mon.1 Patients seeking emergency
vastly more complex, mirroring ics, and systems biology. Patients care are often admitted to the
changes in our patients, our are older, with more coexisting hospital unnecessarily, yet many
health care system, and medical illnesses and more medications. also die suddenly soon after be-
science. The complexity of medi- They see more specialists and ing sent home.2 Overall, we pro-
cine now exceeds the capacity of undergo more diagnostic testing, vide far less benefit to our patients
the human mind. which leads to exponential accu- than we hope. These failures con-
Computers, far from being the mulation of electronic health rec- tribute to deep dissatisfaction
problem, are the solution. But us- ord (EHR) data. Every patient is and burnout among doctors and
ing them to manage the complex- now a big data challenge, with threaten the health care systems
ity of 21st-century medicine will vast amounts of information on financial sustainability.
require fundamental changes in past trajectories and current states. If a root cause of our chal-
the way we think about thinking All this information strains lenges is complexity, the solutions
and in the structure of medical our collective ability to think. are unlikely to be simple. Asking
education and research. Medical decision making has be- doctors to work harder or get

n engl j med 377;13 nejm.org September 28, 2017 1209


The New England Journal of Medicine
Downloaded from nejm.org on September 30, 2017. For personal use only. No other uses without permission.
Copyright 2017 Massachusetts Medical Society. All rights reserved.
PERS PE C T IV E Lost in Thought

smarter wont help. Calls to re- But there is hope. The same dead unexpectedly in any given
duce unnecessary care fall flat: computers that today torment us year. And they could guide basic
we all know how difficult its with never-ending checkboxes and research on the mechanisms of
become to identify what care is forms will tomorrow be able to newly discovered predictors.
necessary. Changing incentives process and synthesize medical Algorithms have also been de-
is an appealing lever for policy- data in ways we could never do ployed for an analysis of massive
makers, but that alone will not ourselves. Already, there are indi- amounts of EHR data whose re-
make decisions any easier: we can cations that data science can help sults suggest that type 2 diabetes
reward physicians for delivering us with critical problems. has three subtypes, each with its
own biologic signature and dis-
ease trajectory.4 Knowing which
type of patients were dealing
with can help us deliver treat-
ments to those who benefit most
and may help us understand why
some patients have complications
and others dont.
There is little doubt that algo-
rithms will transform the think-
ing underlying medicine. The only
question is whether this transfor-
mation will be driven by forces
from within or outside the field.
If medicine wishes to stay in con-
trol of its own future, physicians
will not only have to embrace
algorithms, they will also have
to excel at developing and evalu-
ating them, bringing machine-
learning methods into the medi-
cal domain.
Machine learning has already
spurred innovation in fields rang-
ing from astrophysics to ecology.
In these disciplines, the expert
less care, but the end result may Consider the challenge of read- advice of computer scientists is
simply be less care, not better care. ing electrocardiograms. Doctors sought when cutting-edge algo-
The first step toward a solution look for a handful of features to rithms are needed for thorny
is acknowledging the profound diagnose ischemia or rhythm problems, but experts in the field
mismatch between the human disturbances but can we ever astrophysicists or ecologists
minds abilities and medicines truly read the waveforms in a set the research agenda and
complexity. Long ago, we real- 10-second tracing, let alone the lead the day-to-day business of
ized that our inborn sensorium multiple-day recording of a Holter applying machine learning to rel-
was inadequate for scrutinizing monitor? Algorithms, by contrast, evant data.
the bodys inner workings can systematically analyze every In medicine, by contrast, clin-
hence, we developed microscopes, heartbeat. There are early signs ical records are considered trea-
stethoscopes, electrocardiograms, that such analyses can identify sure troves of data for research-
and radiographs. Will our inborn subtle microscopic variations ers from nonclinical disciplines.
cognition alone solve the myster- linked to sudden cardiac death.3 Physicians are not needed to en-
ies of health and disease in a If validated, such algorithms roll patients so theyre con-
new century? The state of our could help us identify and treat sulted only occasionally, perhaps
health care system offers little the tens of thousands of Ameri- to suggest an interesting outcome
reason for optimism. cans who might otherwise drop to predict. They are far from the

1210 n engl j med 377;13 nejm.org September 28, 2017

The New England Journal of Medicine


Downloaded from nejm.org on September 30, 2017. For personal use only. No other uses without permission.
Copyright 2017 Massachusetts Medical Society. All rights reserved.
PE R S PE C T IV E Lost in Thought

intellectual center of the work or gender biases. Ignoring these icine, are now at the core of
and rarely engage meaningfully facts will result in automating medical research, and medical
in thinking about how algo- and even magnifying problems education has made all doctors
rithms are developed or what in our current health system.5 into informed consumers of these
would happen if they were ap- Noticing and undoing these prob- fields. Similar efforts in data sci-
plied clinically. lems requires a deep familiarity ence are urgently needed. If we
But ignoring clinical thinking with clinical decisions and the lay the groundwork today, 21st-
is dangerous. Imagine a highly data they produce a reality century clinicians can have the
accurate algorithm that uses EHR that highlights the importance tools they need to process data,
data to predict which emergency of viewing algorithms as thinking make decisions, and master the
department patients are at high partners, rather than replace- complexity of 21st-century patients.
risk for stroke. It would learn ments, for doctors. Disclosure forms provided by the authors
todiagnose stroke by churning Ultimately, machine learning are available at NEJM.org.
through large sets of routinely in medicine will be a team sport,
collected data. Critically, all these like medicine itself. But the team From Brigham and Womens Hospital and
Harvard Medical School, Boston (Z.O.,
data are the product of human will need some new players: clini- T.H.L.), and Press Ganey, Wakefield (T.H.L.)
decisions: a patients decision cians trained in statistics and com- both in Massachusetts.
toseek care, a doctors decision to puter science, who can contribute
order a test, a diagnosticians meaningfully to algorithm devel- 1. Institute of Medicine. Improving diag-
nosis in health care. Washington, DC:Na-
decision to call the condition a opment and evaluation. Todays tional Academies Press, 2015.
stroke. Thus, rather than predict- medical education system is ill 2. Obermeyer Z, Cohn B, Wilson M, Jena AB,
ing the biologic phe- prepared to meet these needs. Cutler DM. Early death after discharge from
An audio interview emergency departments: analysis of national
with Dr. Obermeyer
nomenon of cerebral Undergraduate premedical re- US insurance claims data. BMJ 2017;356:j239.
is available at NEJM.org ischemia, the algo- quirements are absurdly outdat- 3. Syed Z, Stultz CM, Scirica BM, Guttag JV.
rithm would predict ed. Medical education does little Computationally generated cardiac biomark-
ers for risk stratification after acute coronary
the chain of human decisions to train doctors in the data sci- syndrome. Sci Transl Med 2011;3:102ra95.
leading to the coding of stroke. ence, statistics, or behavioral sci- 4. Li L, Cheng W-Y, Glicksberg BS, et al.
Algorithms that learn from ence required to develop, evalu- Identification of type 2 diabetes subgroups
through topological analysis of patient sim-
human decisions will also learn ate, and apply algorithms in ilarity. Sci Transl Med 2015;7:311ra174.
human mistakes, such as over clinical practice. 5. Mullainathan S, Obermeyer Z. Does ma-
testing and overdiagnosis, failing The integration of data science chine learning automate moral hazard and
error? Am Econ Rev 2017;107:476-80.
to notice people who lack access and medicine is not as far away
to care, undertesting those who as it may seem: cell biology and DOI: 10.1056/NEJMp1705348
cannot pay, and mirroring race genetics, once also foreign to med- Copyright 2017 Massachusetts Medical Society.
Lost in Thought

The Paradox of Coding

The Paradox of Coding Policy Concerns Raised


by Risk-Based Provider Contracts
BruceE. Landon, M.D., M.B.A., and RobertE. Mechanic, M.B.A.

T he medical claims that health


care providers submit to in-
surers generally include a Current
surers have traditionally focused
on accurate CPT coding to ensure
that reimbursement matches the
The purpose of risk adjustment
is to ensure that health plans that
enroll sicker patients and provid-
Procedural Terminology (CPT) services provided. In recent years, ers who care for such patients are
code, which describes the medi- however, the medical diagnoses compensated fairly. Health plans
cal, surgical, or diagnostic service listed in claims have taken on in- participating in Medicare Advan-
provided to the patient, as well as creasing importance as capitated tage, for example, are reimbursed
a series of International Classification and risk-based payment systems at higher rates for enrolling peo-
of Diseases diagnostic codes. Un- have begun to use these codes ple with multiple conditions and
der fee-for-service reimbursement, to adjust the payments made to more complex diagnoses. Without
health care organizations and in- health plans and providers. risk adjustment, plans would have

n engl j med 377;13 nejm.org September 28, 2017 1211


The New England Journal of Medicine
Downloaded from nejm.org on September 30, 2017. For personal use only. No other uses without permission.
Copyright 2017 Massachusetts Medical Society. All rights reserved.

You might also like