The n e w e ng l a n d j o u r na l of m e dic i n e

Review Article

Dan L. Longo, M.D., Editor

Brain Change in Addiction as Learning,

Not Disease
Marc Lewis, Ph.D.​​

D
uring the past 30 years, the assumption that addiction is a dis- From the University of Toronto, Toronto.
ease or pathology has crystallized into the “brain disease model of addic- Address reprint requests to Dr. Lewis at
Klingelbeekseweg 24, 6812DH Arnhem, the
tion.”1 This trend was driven by the convergence of 12-step thinking with Netherlands, or at ­m​.­lewis@​­psych​.­ru​.­nl.
residential treatment approaches in the latter half of the 20th century,2 the explo-
N Engl J Med 2018;379:1551-60.
sion of neuroimaging technologies that began in the 1990s, and promotion by pro- DOI: 10.1056/NEJMra1602872
fessional organizations3 and community groups.4 According to the brain disease Copyright © 2018 Massachusetts Medical Society.

model, addiction is a chronic disease brought about by changes in the brain sys-
tems that mediate the experience and anticipation of reward and in higher-order
systems that underlie judgment and cognitive control.1,5 The proponents of the
model propose that these changes are driven by exposure to drugs of abuse or
alcohol, though links with behavioral addictions have also been explored.6
The brain disease model is the most prevalent model of addiction in the western
world. Particularly in the United States, it dominates professional and public dis-
course on prevention, treatment, research agendas, and policy issues. Because the
disease model focuses on brain change, it has helped explain why persons with
addictions find it difficult to change their thoughts and behaviors quickly or easily.6
Because it focuses on biologic factors rather than moral arguments, it has helped
reduce the stigma faced by those with addictions and their families, at least in
some respects. (See Table 1 for a broader discussion of stigma.) The brain disease
model has also legitimized the role of doctors and other medical professionals in
addiction treatment and driven research on new drugs to combat addiction, and it
has been used to advocate for access to treatment and care rather than segregation
and punishment.
These aims and outcomes are well intended, and they have been beneficial in
some contexts, but the narrow focus of the disease model on the neurobiologic
substrates of addiction has diverted attention (and research funding) from other
models.10 Alternatives to the brain disease model often highlight the social and
environmental factors that contribute to addiction, as well as the learning pro-
cesses that translate these factors into negative outcomes.11-15 For example, it has
been shown repeatedly that adverse experiences in childhood and adolescence in-
crease the probability of later addiction.13,14 Also, exposure to physical, economic,
or psychological trauma greatly increases susceptibility to addiction.14-17 Learning
models propose that addiction, though obviously disadvantageous, is a natural,
context-sensitive response to challenging environmental contingencies, not a dis-
ease.18,19 Yet the brain disease model construes addictive learning in terms of patho-
logic brain changes triggered mainly by substance abuse. Learning models also
favor individual solutions for overcoming addiction, facilitated by cognitive modi-
fications and personal agency. (See Table 2 for a discussion of empowerment.)
Learning models can include multiple levels of analysis: societal, social, psycho-
logical, and biologic. According to experts both inside and outside the medical

n engl j med 379;16 nejm.org  October 18, 2018 1551

The New England Journal of Medicine
Downloaded from nejm.org at KAOHSIUNG MEDICAL UNIVERSITY on October 17, 2018. For personal use only. No other uses without permission.
Copyright © 2018 Massachusetts Medical Society. All rights reserved.
 The n e w e ng l a n d j o u r na l of m e dic i n e

Table 1. Brain Disease Model and Stigma.

ing so, I hope to connect neurobiologic and en-
vironmental accounts to make sense of addiction
Proponents of the brain disease model of addiction have consistently claimed with a degree of depth and precision that could
that the disease definition has major social benefits for people with addic-
tion. Before addiction was defined as a disease, it was mostly viewed as a not be achieved by either one alone. I also inter-
moral failure, and “addicts” were reviled as self-indulgent, weak, dirty, or pret key neurocognitive findings from both learn-
malicious. But if addiction is viewed as a disease (like any other disease), ing and disease perspectives to highlight their
then the behaviors of people with addiction should not be seen as their
fault. In this way, the disease model was proposed to reduce stigma, blame, parallels as well as their disparities (Table 3).
and the assumption that people with addiction should be punished or re-
moved from society. The disease model should be commended for even
partial success in achieving these humanitarian goals. A ddic t ion a s L e a r ning
Yet the disease definition can replace one kind of stigma with another. The
­notion of a mental illness or disease can hurt more than help those with
Psychologists have historically divided learning
behavioral problems such as addiction, because it fuels discrimination into operant conditioning, by which animals
and alienation of another sort. The disease designation can reinforce the work to receive rewards predicted by specific
belief that an inviolable or essentialist “badness” is built in and perma-
nent, resulting in a sense that one is fundamentally different from “normal”
cues, and Pavlovian conditioning, by which ani-
people, with concomitant feelings of inferiority and shame.7,8 The label can mals respond automatically to the stimulus prop-
also curtail attempts to improve one’s functioning without medical care. erties of cues themselves. Advances in cognitive
Biogenetic explanations carry the implication that people with addictions
are not really trustworthy, now or in the future, because of a biologic pro-
psychology reveal that learning also involves
clivity they cannot control.9 Not only does this fuel one kind of stigmatiza- planning, decision making, inhibitory control,
tion; it also helps rationalize a long-standing policy of withholding employ- and strings of cues that eventually lead to pre-
ment benefits and positions of authority from anyone who has ever been
labeled an addict.
dicted rewards. The contemporary view from
cognitive science has extended this understand-
It is true that some people with addiction feel consoled by the disease label.
In fact, psychiatric classifications have provided people who have diverse ing with models of “embodied cognition,” which
emotional and mental problems with a label and (sometimes) a hypotheti- propose that all cognitive activity (including
cal explanation for adversities that can otherwise seem indefinable, amor- learning) results from iterative, self-perpetuat-
phous, and yet blameworthy. Distinct categories with concrete labels can
help provide closure, context, and even a sense of belonging (to a particu- ing interactions (i.e., feedback) between the ani-
lar group). mal and the environment.29 From this perspec-
Yet many people with addiction recoil from the disease label. Especially when tive, learning occurs when the animal’s neural
they are successful in galvanizing their willpower and rejigging their habits capacities become entrained with an environ-
(i.e., recovering), they often find it confusing and debilitating to be told they
are chronically ill. People with previous addictions (“recovered addicts”)
mental context. Thus, learning is not just a re-
usually want to feel that they have developed beyond their addiction and sponse to stimuli but active engagement with
become better people as a result. Many would prefer respect for that meaningful aspects of the environment.30
achievement over the pity bequeathed by the disease definition.
The brain disease model does not dismiss the
importance of learning but views this learning
as pathologic. Addictive behaviors are proposed
field,27 these levels of analysis should ideally be to begin as impulsive bids for highly motivating
integrated for a comprehensive understanding of rewards, consolidated through operant condi-
addiction. Unfortunately, however, the neural level tioning, but to end up as automatic (Pavlovian)
of analysis is almost always ignored by nondis- responses that bypass intention, augmented by
ease models that emphasize learning. (Work by a loss of inhibitory control and a capacity for
Szalavitz is a notable exception.28) Rather than choice. This observation is consistent with mod-
ignore (or dispute) evidence of brain change in els of “delay discounting,” which propose that
addiction, the current learning model reinterprets immediate payoffs are inflated in their perceived
such evidence. Psychological change, develop- value, whereas longer-term rewards are “dis-
ment, and indeed all learning involve brain counted” (devalued).31,32 Psychologists view delay
change. It is therefore unnecessary and perhaps discounting as an intrinsic cognitive bias, not
unreasonable for a learning model of addiction only in humans but in other mammals as well.
to dismiss neural findings. Yet delay discounting seems to be augmented in
In this review, I examine addiction within a addiction, with long-term rewards falling off the
learning framework, informed by classic and radar almost entirely. “Dual process” models of
contemporary cognitive principles, which can addiction may help to explain this phenomenon,33
incorporate the brain changes seen in addiction in that a cognitive “overseer” loses the capacity
without reference to pathology or disease. In do- to override impulsive choices.34 Although none

1552 n engl j med 379;16 nejm.org  October 18, 2018

The New England Journal of Medicine

Downloaded from nejm.org at KAOHSIUNG MEDICAL UNIVERSITY on October 17, 2018. For personal use only. No other uses without permission.
Copyright © 2018 Massachusetts Medical Society. All rights reserved.
 Br ain Change in Addiction as Learning, Not Disease

of these learning mechanisms are necessarily Table 2. Learning Models and Empowerment.
unique, the brain disease model of addiction
views the progression of decreasing control as a Viewing addiction in terms of learning rather than disease may have direct
­advantages for those who are struggling. If people think that their addic-
reflection of pathologic brain changes. tion results from an underlying pathology, as implied by the brain disease
Addiction neuroscience explores these brain model, and that the pathology is chronic, as highlighted both by profes-
changes. The shift from impulsive (operant, sional bodies and by the 12-step movement, then they are less likely to
­believe they will ever be free of it, especially as a result of their personal ef-
reward-driven) actions to compulsive (automatic, forts.20 This characterization of addiction flies in the face of research show-
Pavlovian) associations is a case in point. When ing that a majority of persons with addictions recover without professional
drug taking is found to be highly rewarding, the treatment.21,22 In fact, addiction workers generally agree that personal mo-
tivation, a sense of empowerment, and belief in one’s own agency are the
ventral striatum (including the nucleus accum- most important psychological resources for overcoming addiction. These
bens) focuses attention on the desired goal, acti- qualities would seem peripheral rather than mandatory if addiction were
vates a behavioral sequence to achieve that goal, indeed a disease.
and produces a motivational urge to energize In response to this argument, proponents of the brain disease model have
pointed out that defining something as a disease does not exempt patients
that behavior.35 Over time, however, as behavior from responsibility for self-care (e.g., making lifestyle choices that improve
becomes more compulsive and less impulsive their prognosis). There is some truth to this counterargument; a sense
(less reward-driven), activation increases in the of empowerment can bolster self-care for patients with various medical
problems.
dorsal striatum, the region most associated with
automatic responses.10,33,36,37 This progression is Yet viewing oneself as a patient implies that one’s primary duty is to follow
the instructions of knowledgeable professionals rather than examine one’s
thought to eradicate willpower,38 because con- own motivations, beliefs, and intuitions. Taking on the role of a patient
scious choice is no longer driving the behavior. may be especially counterproductive in institutional settings, where people
The neurotransmitter dopamine has often with addictions tend to offload responsibility to treatment staff.23 More­
over, biogenetic explanations for psychological problems induce “prog-
been the focus of neural models of addiction.36 nostic pessimism.”9 People dealing with addiction will try to change only
But dopamine has many functions, both in the that which they feel is within their power to change.24 Thus, their own faith
striatum and in the prefrontal cortex, depending in their recovery and the confidence of those around them are hampered
by the disease definition.
partly on the receptor type absorbing it. For the
The choice of terminology suggests specific guidelines for treatment. If replac-
purposes of this discussion, we can think of ing the disease nomenclature with an emphasis on motivation and self-­
dopamine as activating synaptic activity and, direction increases the probability of successful outcomes, then treatment
over time, synaptic change, both in the ventral professionals (including doctors) should advise those seeking help that
they do not have a chronic disease. They should encourage people with ad-
and dorsal striatum and in the prefrontal cortex diction not to strive for obedience to a set of rules or pharmaceutical sub-
(partly through its effect on glutamate transmis- stitutes (unless heroin use prioritizes the need for medication-assisted
sion). The release of dopamine to these and treatment) but instead to seek counseling or psychotherapy to help them
organize and modify their own attentional and motivational habits. For ex-
other systems is triggered by the perception of ample, a psychotherapeutic technique called motivational interviewing has
cues paired with anticipated rewards (in the case been developed in which nonconfrontational counseling by the clinician
of operant learning) or with automatic responses encourages increased awareness of one’s own motives, conscious choices
that are consistent with one’s long-term goals, and reduced ambivalence;
(in the case of Pavlovian conditioning). Yet dopa- this approach is best known for its success in reducing substance use.25
mine metabolism also responds to the experi- More conventional psychotherapies such as cognitive behavioral therapy
ence of rewards, increasing when rewards ex- also show efficacy in overcoming addiction,26 and cognitively oriented
group interventions such as Self-Management and Recovery Training
ceed expectations and decreasing when they fall (SMART Recovery) are quickly gaining recognition.
short. Addiction neuroscientists highlight the
long-lasting sensitization of the dopamine sys-
tem to addictive rewards or the cues that predict ing (i.e., the proportion of impulsive choices).39
them, resulting in craving and narrowed atten- A key finding in support of the brain disease
tion6,37 as well as the subsequent blunting of the model is that drug use reduces connectivity be-
dopamine system over time.1 tween the prefrontal cortex and striatum, and
Striatal systems engage in constant cross-talk long-term addiction corresponds with reduced
with regions of the prefrontal cortex. Prefrontal gray-matter density (synaptic loss) in several
activation (in the orbitofrontal cortex) determines prefrontal and related regions. Such changes are
the attractiveness of potential rewards and also hypothesized to underlie diminished capacities
(in the dorsolateral prefrontal cortex) the exer- for judgment and self-control, or “impaired re-
cise of judgment and perspective shifting. In fact, sponse inhibition,” in people with addictions.5,40
disrupted activation of the lateral prefrontal cor- According to the brain disease model, the
tex has been shown to increase delay discount- cognitive and neural changes characterizing ad-

n engl j med 379;16 nejm.org  October 18, 2018 1553

The New England Journal of Medicine
Downloaded from nejm.org at KAOHSIUNG MEDICAL UNIVERSITY on October 17, 2018. For personal use only. No other uses without permission.
Copyright © 2018 Massachusetts Medical Society. All rights reserved.
 The n e w e ng l a n d j o u r na l of m e dic i n e

Table 3. Comparison of Claims Made by Disease and Learning Models of Addiction and Sample Evidence for Learning.*

Disease Model Learning Model Evidence for Learning

Addiction is characterized by a shift from All behavioral habits devolve to stimulus– Dorsal striatal activation or behavioral automati­
impulsive to compulsive processing, response mechanisms; automatization zation is seen with practice of even simple
loss of free will, and a shift of activation is a normal outcome of learning. (e.g., motor) tasks; for people with addiction,
to dorsal striatum. operant contingencies facilitate the choice to
abstain from using drugs.
Functional connectivity between striatum When planning and decision making are Immediate or valued rewards lead to increased
and PFC is lost, with reduced synaptic bypassed, PFC demand is reduced; ex- striatal activation and decreased dorsolateral
density in specific PFC regions. tended plasticity is normal; underused PFC activation and cognitive control; synaptic
synapses may be pruned. density in the PFC has been shown to rebound
with recovery.
Sensitization to drug cues is increased Sensitization to valued rewards is normal; Motivated goal pursuit leads to increased dopa-
(and enduring), mediated by increased an ongoing need or desire leads to on­ mine, cue sensitization, and learning; high
mesolimbic dopamine uptake. going sensitization (e.g., love, attachment, emotional salience facilitates lasting synaptic
wealth acquisition, religious practice). alterations (e.g., after trauma).
Ongoing drug use leads to loss of receptor Adversity, trauma (with or without drug Loss of social status or trauma leads to reduced
availability or sensitivity and reduced use), isolation, and overstimulation D2 or D3 receptor availability; high levels of
pleasure (dopaminergic blunting). lead to reduced dopamine-receptor mating behavior, eating, engagement with
­response or pleasure. ­pornography, and Internet use lead to a hypo­
dopaminergic system.

* PFC denotes prefrontal cortex.

diction are unique and pathologic. Some theories creased and enduring sensitivity (i.e., sensitiza-
highlight distinct phases or stages: drug taking tion) to cues predicting addictive rewards, under-
is driven by positive reinforcement at first, then pinned by mesolimbic dopamine.37 The fourth
by negative reinforcement (underpinned by re- change is a decrease in sensitivity, not only to
duced dopamine signaling and blunted receptor alternative rewards but even to addictive rewards
responses), and finally by the loss of prefrontal themselves.1 I argue that these four neurocogni-
control.1,41 A closely related theory suggests that tive changes are not specific to addiction and do
addictive urges are increasingly driven by the not indicate a disease process.
brain’s rebound from drug stimulation — an
“antireward” effect resulting from an overactive R ein ter pr e t ing
stress-response system, dopamine blunting, and the Neuro c o gni t i v e Data
physical withdrawal symptoms.42 These theories
emphasize repeated episodes of negative reinforce- Role of Compulsive or Automatic Responses
ment (learning to avoid an aversive outcome) and According to the brain disease model, impulsive
positive reinforcement, plus changes in neuro- drug seeking and use are linked with activation
chemistry and circuitry. of the ventral striatum or nucleus accumbens at
But are the neurocognitive processes that give first, but these behaviors become compulsive and
rise to addiction actually pathologic, or are they automatic with activation of the dorsal striatum
constituents of normal learning with detrimen- over time.35,43 Yet behavior generally becomes
tal consequences? To help resolve this question, more automatic with practice, as novelty is re-
I examine four neurocognitive changes central to placed by familiarity, and dorsal striatal (includ-
brain disease models. The first is the hypothe- ing globus pallidus) involvement underlies this
sized shift from impulsive behavior mediated by automatization even in a simple finger-tapping
the ventral striatum to compulsive responses task.44 As Everitt and Robbins, acknowledged
mediated by the dorsal striatum.35 The second experts on the ventral-to-dorsal shift, state, “There
change, which also supports the presumption of is nothing aberrant or unusual about devolving
involuntary behavior, is a reduction in functional behavioural control to a dorsal striatal S-R [stimu-
and structural connectivity between the striatum lus–response or Pavlovian] ‘habit’ mechanism.”35
and prefrontal cortex.1,5 The third change is in- They assert that this shift is to be expected in

1554 n engl j med 379;16 nejm.org  October 18, 2018

The New England Journal of Medicine

Downloaded from nejm.org at KAOHSIUNG MEDICAL UNIVERSITY on October 17, 2018. For personal use only. No other uses without permission.
Copyright © 2018 Massachusetts Medical Society. All rights reserved.
 Br ain Change in Addiction as Learning, Not Disease

many aspects of our lives, including eating and counting39 and more generally in effortful reward
other habitual activities. “Automatisation of be- seeking.56 But would this loss of functional con-
haviour frees up cognitive processes,” Everitt and nectivity normally lead to structural changes?
Robbins continue, which explains why we can Indeed, the elimination, or “pruning,” of under-
talk, eat, and drive at the same time. used synapses is considered a key mechanism of
Not only is normal behavior partly automatic, learning.57,58 Massive cortical pruning has tradi-
but also addictive behavior, even in its later stages, tionally been associated with adolescence,59 when
remains partly operant (reward-driven).45 Support- most addictions develop. However, since pruning
ing evidence comes from numerous studies in makes the brain more efficient when new skills
which the reward value of the addictive goal are practiced and consolidated, it is now thought
(e.g., the amount of drug offered) shifts in rela- to underpin learning over the lifespan.57,60
tion to the reward value of an alternative goal Synaptic density in certain prefrontal regions
(e.g., money).45-49 In fact, these studies show that decreases with the duration of drug use, but a
the probability of abstaining is proportional to contrasting increase in synaptic density (in simi-
the relative reward value of the two choices; this lar but not identical regions) correlates with the
sensitivity to environmental contingencies is the number of weeks of abstinence.61 In studies using
hallmark of operant learning. Contingency man- functional magnetic resonance imaging (fMRI),
agement programs, based on these principles, “cocaine-dependent” participants who became
have shown a consistent effect in the reduction abstinent no longer differed from controls with
of drug use.26,49 The ventral striatum continues respect to the activation of inhibitory control
to be involved in reward seeking in later-stage networks in the prefrontal cortex or the perfor-
addiction, even when the dorsal striatum domi- mance of motor-inhibition tasks.62 Thus, reduc-
nates behavior control.43 In sum, a combination tions in prefrontal involvement and synaptic
of deliberate and automatic neurobehavioral density appear to be restricted to the period of
mechanisms characterizes both addiction and habitual drug use, which may be followed by a
“normal” habitual behavior. period of synaptic growth when a new skill —
abstinence — is learned. This two-way street in
Loss of Prefrontal Connectivity frontal neuroplasticity is consistent with evidence
and Synaptic Pruning that most people with addiction recover,21,22 and
Evidence of a functional and (in some studies) most of those who recover do so without treat-
structural disconnection between the prefrontal ment.21,63 This finding would seem to be impos-
cortex and striatum has been pivotal for defin- sible if prefrontal changes were permanent and
ing addiction as a brain disease.40 Unfortunately, therefore pathologic.
these findings come from cross-sectional, not
longitudinal, research, so some cortical differ- Sensitization to Cues
ences must precede rather than follow addiction, People with drug addiction are highly sensitive
as acknowledged by the researchers. Yet even to drug-related cues, even after they quit using
cortical changes that arise from (or with) addic- drugs. To account for this sensitization, the brain
tive drug use need not be considered pathologic. disease model points to a sharp rise in mesolim-
When skills become streamlined with prac- bic (reward-related) dopamine uptake.37 The moti-
tice, they no longer engage conscious, reflective, vational drive provided by mesolimbic dopamine
or effortful control. In fact, higher-order cogni- is essential for survival, because it ensures that
tion is unnecessary once behavior becomes habit- we prioritize eating, social relationships, and pro-
ual, as any professional musician or athlete can creation. Addiction neuroscientists acknowledge
demonstrate. Also, rewards perceived as both that the levels of cue-triggered dopamine seen in
immediate and valuable often bypass cognitive addiction can parallel those related to “natural”
control, as seen in the reduction of planning, rewards.64 Indeed, romantic relationships de-
decision making, and concomitant prefrontal in- pend on motivational dopamine uptake,50,65 and
volvement when it comes to sex, gambling, and desire after romantic rejection matches the crav-
eating fast food.50-55 Research points to an inverse ing for cocaine.50 Motivated pursuits (natural or
correlation between striatal activation and dorso- otherwise), including shopping, sports, religious
lateral prefrontal engagement, both in delay dis- practice, wealth acquisition, gambling, binge eat-

n engl j med 379;16 nejm.org  October 18, 2018 1555

The New England Journal of Medicine
Downloaded from nejm.org at KAOHSIUNG MEDICAL UNIVERSITY on October 17, 2018. For personal use only. No other uses without permission.
Copyright © 2018 Massachusetts Medical Society. All rights reserved.
 The n e w e ng l a n d j o u r na l of m e dic i n e

ing, romantic love, and pornography, correspond chemical rebound effect, independent of addictive
with cue sensitization and increased activation learning.
of striatal dopamine.35,50-55,64-66 Even a simple in- But even if addictive learning results in dopa-
crease in reward availability on a computer screen minergic blunting, it need not denote pathologic
is sufficient to increase mesolimbic dopamine, brain change. Poverty, trauma, and diminished
with a concomitant increase in effort.67 social status reduce the availability of the D2 and
Proponents of the brain disease model empha- D3 dopamine receptors in humans and nonhu-
size that cue sensitivity in addiction is not only man primates.72 In fact, a reduction in D2 or D3
extreme but also prolonged, whereas cue sensitiv- receptor availability has been shown to corre-
ity returns to normal levels in relation to natural spond with reduced social dominance or isola-
reinforcers, once the need has been met.1 This tion, driving drug or alcohol use as a means of
prolonged sensitization is seen as the cause of countering anxiety or distress.73-76 As noted above,
relapse.37,68 Yet prolonged sensitization also results early adversity and trauma are reliable predictors
from normal learning of emotionally salient of subsequent drug use.13,16,17,77 However, social
associations, through synaptic alterations in re- adversity may also result from drug use itself.
gions that process emotion, such as the amyg- Society responds to illicit drug use by excluding
dala.69,70 Stimuli associated with past triumphs or or punishing users, which in turn leads to bro-
traumas or even a once-loved song will reliably ken relationships and erosion of self-esteem.
trigger strong feelings. Because these cues refer Thus, social and psychological hardships may
to still-meaningful experiences, dopamine uptake result in dopaminergic blunting, which then en-
remains adaptive (rather than pathologic) for courages addictive activities, amplifying these
ongoing behavioral adaptations. hardships.
Perhaps the most parsimonious explanation Dopaminergic blunting can also result from
for enduring cue sensitivity is that, in addiction, nondrug rewards. Mating behavior in rats reduces
goal seeking remains unfulfilled. The drug or dopamine output in mesolimbic dopamine cir-
activity that was pursued to satisfy emotional cuitry, leading to “a hypodopaminergic system,”
needs may have lost its effect because of a short and identical changes result from prolonged ex-
duration of action, chemical tolerance, or habitu- posure to opiates.78 In addition, obesity has been
ation. The value of addictive rewards is always linked to reduced dopamine receptivity, with the
determined by context, including both the strength hypothetical explanation that dopaminergic blunt-
of aversive feelings and the effectiveness of drugs, ing leads to increased food consumption.79,80 Ex-
for example, in quelling them. Unresolved needs posure to other potentially habit-forming plea-
can make drug taking relevant indefinitely. surable activities also leads to dopaminergic
blunting, as shown with pornography use81 and
Desensitization to Drug-Related extensive Internet use.82 Thus, it seems that dopa-
and Natural Rewards minergic blunting can result from frequent acti-
In parallel with cue sensitization and increased vation of the mesolimbic dopamine system by
levels of dopamine release, there is an appar- any repetitive reward-seeking behavior rather than
ently paradoxical decrease in sensitivity to alter- by drug exposure itself. Kent Berridge, a renowned
native rewards and even to drugs themselves.1,68 addiction neuroscientist, views dopaminergic
This reward desensitization is thought to con- suppression as a temporary effect of overstimu-
tribute to increasing drug consumption. Brain lation, which may result from drug addiction but
disease models ascribe this blunting to the down- does not cause it.68
regulation (reduced availability or responsiveness)
of dopamine receptors (e.g., D2 and D3 recep- A ddic t ion a s Org a nism –
tors), a pathologic process that may be mani- En v ironmen t En t r a inmen t
fested as tolerance or withdrawal effects.37,42 Yet
many studies of addiction use psychostimulants Most alternatives to the brain disease model
(e.g., cocaine and methamphetamine), serious- of addiction share the view that explanations of
ly confounding this observation.71 The buildup addiction should include societal, social, and fa-
(e.g., delayed reuptake) of dopamine resulting milial factors that predict drug misuse. The brain
from psychostimulants may directly trigger a disease model has acknowledged these factors,

1556 n engl j med 379;16 nejm.org  October 18, 2018

The New England Journal of Medicine

Downloaded from nejm.org at KAOHSIUNG MEDICAL UNIVERSITY on October 17, 2018. For personal use only. No other uses without permission.
Copyright © 2018 Massachusetts Medical Society. All rights reserved.
 Br ain Change in Addiction as Learning, Not Disease

but its emphasis on brain pathology sidelines synchronized through mutual adjustments. Be-
their causal status and their relevance to preven- havioral outcomes continue to shape a social en-
tion and treatment efforts. Yet viewing addiction vironment that progressively narrows behavioral
solely as the product of environmental forces options. For example, the social environment may
tends to ignore the properties of the organism, become increasingly limited to people who can
its nervous system, and its response proclivities. supply drugs (dealers or doctors), people with
A comprehensive, balanced model of addiction whom to take drugs, and “friends” who remain
needs to recognize that the organism and its apathetic and disengaged. Behavioral proclivities
environment are connected at every level, from will change accordingly. Besides the increasing
perception to cognition to behavior, and interact habit strength of drug pursuit itself, there is
continuously as an open system. likely to be increased lying to avoid rejection or
I have presented arguments and evidence that punishment, as well as disengagement from
automatization, reduced neural flexibility, endur- romantic partners and family members, further
ing cue sensitization, and reward desensitization limiting the chance to feel connected and pro-
are normal features of learning highly motivat- tected. These changes would be mediated by cog-
ing, repetitive, and habitual behavioral patterns. nitive modifications — changes in attentional
Thus, I dispute the idea that addiction is patho- foci, belief systems, identity, and self-esteem — as
logic. Nevertheless, there is considerable poten- well as by immature habits of emotion regula-
tial for reconciliation between aspects of the brain tion (e.g., suppression or denial) more generally.
disease model and an environmental model of But how might this addiction spiral get
addiction, given that both view a rigidified be- started? The embodied-cognition view encour-
havioral pattern as learned, and learned deeply. ages us to look for biologic and environmental
Classic learning models have limited value for vulnerabilities that amplify and reinforce each
this synthesis, since they view the learner as an other. The goal here is not to list organismic
independent agent responding to a static environ- (e.g., genetic) and environmental risk factors and
ment. In contrast, principles of embodied cogni- add them together, but instead to track the inter-
tion construe learning as a process of reciprocal action of factors that reciprocally influence each
adjustments between the activities of the organ- other. I suggest that the addiction spiral gets
ism and meaningful features of the environment. started with early psychosocial adversity. First,
What is meaningful is assumed to be con- we already know that early adversity and trauma
strained by biologic antecedents and emerging are strong predictors of later addiction.13,16,17,77
biologic sensitivities, as well as the stimulus Second, developmental psychologists have shown
properties of the animal’s environment (i.e., that early trauma (physical, emotional, or sexual)
features of the environment that afford or invite leaves enduring effects on nervous system func-
specific actions, known as affordances). tion, such as sympathetic or parasympathetic
For a young human, the range of potentially overattunement (causing hyperreactivity or hypo-
meaningful environmental features can be vast, reactivity), oversensitivity to threat based on ac-
at least until social, familial, and psychological celerated amygdala development, and hippocam-
setbacks narrow it down to a small subset of pal damage resulting from excessive cortisol
suboptimal rewards. For example, many children levels. Third, in animal models, researchers have
grow up with an unpredictable, disengaged, or pinpointed epigenetic changes (e.g., methylation
violent parent. As adolescents, they may face of a gene that tunes the glucocorticoid feedback
disruptions in education, employment, or rela- loop) that take place in utero or the first year of
tionships as a result of financial or other disad- life in response to inadequate nurturing. But
vantages. These persons tend to find increased these neuropsychological insults do not emerge
meaning in drugs that reduce stress or promote in a vacuum. Both trauma and “stress methyla-
feelings of security and well-being, especially tion” can begin with overstressed parents and
because these effects can be attained without even grandparents83,84 in families challenged by
mediation by other people. As drug use pro- unemployment, marital discord, histories of abuse,
gresses and becomes a more consistent focus of or alienation from the community, affecting the
attention and behavior, the properties of the in- stress response in childhood and throughout life.84
dividual and of the environment tend to become From these beginnings, a narrowing spiral of

n engl j med 379;16 nejm.org  October 18, 2018 1557

The New England Journal of Medicine
Downloaded from nejm.org at KAOHSIUNG MEDICAL UNIVERSITY on October 17, 2018. For personal use only. No other uses without permission.
Copyright © 2018 Massachusetts Medical Society. All rights reserved.
 The n e w e ng l a n d j o u r na l of m e dic i n e

ineffective coregulation emerges between devel- utors to addiction. Addictive activities are deter-
oping children and their caregivers, leading even- mined neither solely by brain changes nor solely
tually to entrainment between drug seeking and by social conditions. Although they indeed result
its environmental concomitants. From the Rat from and contribute to brain changes, addictive
Park studies of the 1970s and 1980s, in which activities also feed back to the social environ-
even addicted rats avoided ingesting morphine ment, further narrowing what are often already
when allowed to socialize and play,85 to contem- limited opportunities for well-being, which in
porary evidence of the adverse consequences of turn further narrows cognitive and neural flex-
socioeconomic fragmentation,11 Bruce Alexander ibility. It follows that the narrowing seen in ad-
has shown that addiction emerges universally diction takes place within the behavioral reper-
as a response to the disruption of normal social toire, the social surround, and the brain — all
interactions. Therefore, models of addiction pred- at the same time. It also follows that growth
icated on embodied cognition should focus on beyond addiction can be facilitated by improved
environments in which social stressors affect social support, extended behavioral opportuni-
early neuropsychological development, as a gate- ties, targeted pharmacologic interventions, or
way to ongoing reciprocal adjustments between some combination of these strategies.
disadvantageous organismic adaptations and nar- Disclosure forms provided by the author are available with the
rowing environmental opportunities. full text of this article at NEJM.org.
In summary, the embodied-cognition frame- I thank Shaun Shelly (University of Pretoria, Department of
Family Medicine) for providing information and references sup-
work can help model the interaction between porting the arguments reviewed here, as well as feedback on
neurobiologic and social-environmental contrib- previous versions of the manuscript.

References
1. Volkow ND, Koob GF, McLellan AT. disease model of addiction: is it support- 19. Heather N, Best D, Kawalek A, et al.
Neurobiologic advances from the brain ed by the evidence and has it delivered Challenging the brain disease model of
disease model of addiction. N Engl J Med on its promises? Lancet Psychiatry 2015;​2:​ addiction: European launch of the Addic-
2016;​374:​363-71. 105-10. tion Theory Network. Addict Res Theory
2. Fletcher AM. Inside rehab:​the sur- 11. Alexander BK. The globalization of 2018;​26:​249-55.
prising truth about addiction treatment addiction:​a study in the poverty of spirit. 20. Wiens TK, Walker LJ. The chronic dis-
— and how to get help that works. New Oxford, England:​Oxford University Press, ease concept of addiction: helpful or
York:​Penguin Books, 2013. 2008. harmful? Addict Res Theory 2015;​23:​309-
3. American Society of Addiction Medi- 12. Heilig M, Epstein DH, Nader MA, 21.
cine. Definition of addiction. April 12, Shaham Y. Time to connect: bringing so- 21. Heyman GM. Quitting drugs: quanti-
2011 (http://www​.asam​.org/​quality​-­practice/​ cial context into addiction neuroscience. tative and qualitative features. Annu Rev
definition​-­of​-­addiction). Nat Rev Neurosci 2016;​17:​592-9. Clin Psychol 2013;​9:​29-59.
4. Narcotics Anonymous World Servic- 13. Dube SR, Felitti VJ, Dong M, Chap- 22. Lopez-Quintero C, Hasin DS, de Los
es. What is addiction? Narcotics Anony- man DP, Giles WH, Anda RF. Childhood Cobos JP, et al. Probability and predictors
mous Bulletin no. 16. 2017 (https://na​.org/​ abuse, neglect, and household dysfunction of remission from life-time nicotine, alco-
?ID=bulletins​-­bull17​-­r). and the risk of illicit drug use: the adverse hol, cannabis or cocaine dependence: re-
5. Goldstein RZ, Volkow ND. Drug ad- childhood experiences study. Pediatrics sults from the National Epidemiologic
diction and its underlying neurobiologi- 2003;​111:​564-72. Survey on Alcohol and Related Conditions.
cal basis: neuroimaging evidence for the 14. Enoch MA. The role of early life stress Addiction 2011;​106:​657-69.
involvement of the frontal cortex. Am J as a predictor for alcohol and drug depen- 23. Bonnie RJ. Responsibility for addic-
Psychiatry 2002;​159:​1642-52. dence. Psychopharmacology (Berl) 2011;​ tion. J Am Acad Psychiatry Law 2002;​30:​
6. Volkow ND, Baler RD, Goldstein RZ. 214:​17-31. 405-13.
Addiction: pulling at the neural threads of 15. Buchanan J. Understanding problem- 24. Pickard H, Ward L. Responsibility with-
social behaviors. Neuron 2011;​69:​599-602. atic drug use:​a medical matter or a social out blame:​philosophical reflections on
7. Buchman D, Reiner PB. Stigma and issue? Br J Community Justice 2006;​4:​47-61 clinical practice. In:​Fulford KWM, Davies
addiction: being and becoming. Am J Bio- (http://epubs​.glyndwr​.ac​.uk/​siru/​18). M, Gipps RGT, et al., eds. The Oxford
eth 2009;​9:​18-9. 16. Valentine K, Fraser S. Trauma, dam- handbook of philosophy and psychiatry.
8. Haslam N. Genetic essentialism, neu- age and pleasure: rethinking problematic Oxford, England:​Oxford University Press,
roessentialism, and stigma: commentary drug use. Int J Drug Policy 2008;​19:​410-6. 2013 (http://oxfordhandbooks​.com/​v iew/​
on Dar-Nimrod and Heine (2011). Psychol 17. McDevitt-Murphy ME, Murphy JG, 10​.1093/​oxfordhb/​9780199579563​.001​
Bull 2011;​137:​819-24. Monahan CM, Flood AM, Weathers FW. .0001/​oxfordhb​-­9780199579563​-­e​-­066).
9. Kvaale EP, Haslam N, Gottdiener WH. Unique patterns of substance misuse asso- 25. Dunn C, Deroo L, Rivara FP. The use
The ‘side effects’ of medicalization: a meta- ciated with PTSD, depression, and social of brief interventions adapted from moti-
analytic review of how biogenetic expla- phobia. J Dual Diagn 2010;​6:​94-110. vational interviewing across behavioral
nations affect stigma. Clin Psychol Rev 18. Lewis M. Addiction and the brain: de- domains: a systematic review. Addiction
2013;​33:​782-94. velopment, not disease. Neuroethics 2017;​ 2001;​96:​1725-42.
10. Hall W, Carter A, Forlini C. The brain 10:​7-18. 26. Dutra L, Stathopoulou G, Basden SL,

1558 n engl j med 379;16 nejm.org  October 18, 2018

The New England Journal of Medicine

Downloaded from nejm.org at KAOHSIUNG MEDICAL UNIVERSITY on October 17, 2018. For personal use only. No other uses without permission.
Copyright © 2018 Massachusetts Medical Society. All rights reserved.
 Br ain Change in Addiction as Learning, Not Disease

Leyro TM, Powers MB, Otto MW. A meta- seeking habits and drug memories — in- through the lifespan: mechanisms of
analytic review of psychosocial interven- dications for novel treatments of addic- change. Trends Cogn Sci 2006;​10:​131-8.
tions for substance use disorders. Am J tion. Eur J Neurosci 2014;​40:​2163-82. 58. Petanjek Z, Judaš M, Šimic G, et al.
Psychiatry 2008;​165:​179-87. 44. Lehéricy S, Benali H, Van de Moortele Extraordinary neoteny of synaptic spines
27. Satel S, Lilienfeld SO. Addiction and P-F, et al. Distinct basal ganglia territories in the human prefrontal cortex. Proc Natl
the brain-disease fallacy. Front Psychiatry are engaged in early and advanced motor Acad Sci U S A 2011;​108:​13281-6.
2014;​4:​141. sequence learning. Proc Natl Acad Sci U S A 59. Selemon LD. A role for synaptic plas-
28. Szalavitz M. Squaring the circle: addic- 2005;​102:​12566-71. ticity in the adolescent development of
tion, disease and learning. Neuroethics 45. Heather N. Is the concept of compul- executive function. Transl Psychiatry 2013;​
2017;​10:​83-6. sion useful in the explanation or descrip- 3:​e238.
29. Cowart M. Embodied cognition. In- tion of addictive behaviour and experi- 60. Segal I, Korkotian I, Murphy DD. Den-
ternet Encyclopedia of Philosophy, 2018 ence? Addict Behav Rep 2017;​6:​15-38. dritic spine formation and pruning: com-
(https://www​.iep​.utm​.edu/​embodcog/​). 46. Cohen M, Liebson IA, Faillace LA. The mon cellular mechanisms? Trends Neuro-
30. Wilson AD, Golonka S. Embodied cog- role of reinforcement contingencies in sci 2000;​23:​53-7.
nition is not what you think it is. Front chronic alcoholism: an experimental analy- 61. Connolly CG, Bell RP, Foxe JJ, Gara-
Psychol 2013;​4:​58. sis of one case. Behav Res Ther 1971;​9:​ van H. Dissociated grey matter changes
31. Ainslie G. Précis of Breakdown of Will. 375-9. with prolonged addiction and extended
Behav Brain Sci 2005;​28:​635-50. 47. Hart C. High price:​a neuroscientist’s abstinence in cocaine users. PLoS One
32. Bickel WK, Marsch LA. Toward a be- journey of self-discovery that challenges 2013;​8(3):​e59645.
havioral economic understanding of drug everything you know about drugs and so- 62. Bell RP, Garavan H, Foxe JJ. Neural
dependence: delay discounting processes. ciety. New York:​Harper, 2013. correlates of craving and impulsivity in
Addiction 2001;​96:​73-86. 48. Hart CL, Haney M, Foltin RW, Fisch­ abstinent former cocaine users:​towards
33. Wiers RW, Bartholow BD, van den man MW. Alternative reinforcers differ- biomarkers of relapse risk. Neuropharma-
Wildenberg E, et al. Automatic and con- entially modify cocaine self-administra- cology 2014;​85:​461-70.
trolled processes and the development of tion by humans. Behav Pharmacol 2000;​ 63. Dawson DA, Grant BF, Stinson FS,
addictive behaviors in adolescents: a review 11:​87-91. Chou PS. Maturing out of alcohol depen-
and a model. Pharmacol Biochem Behav 49. Menza TW, Jameson DR, Hughes JP, dence: the impact of transitional life
2007;​86:​263-83. Colfax GN, Shoptaw S, Golden MR. Con- events. J Stud Alcohol 2006;​67:​195-203.
34. Stacy AW, Wiers RW. Implicit cogni- tingency management to reduce metham- 64. Olsen CM. Natural rewards, neuroplas-
tion and addiction: a tool for explaining phetamine use and sexual risk among ticity, and non-drug addictions. Neuro-
paradoxical behavior. Annu Rev Clin Psy- men who have sex with men: a random- pharmacology 2011;​61:​1109-22.
chol 2010;​6:​551-75. ized controlled trial. BMC Public Health 65. Fisher HE, Xu X, Aron A, Brown LL.
35. Everitt BJ, Robbins TW. From the ven- 2010;​10:​774. Intense, passionate, romantic love: a nat-
tral to the dorsal striatum: devolving 50. Fisher HE, Brown LL, Aron A, Strong ural addiction? How the fields that inves-
views of their roles in drug addiction. G, Mashek D. Reward, addiction, and emo- tigate romance and substance abuse can
Neurosci Biobehav Rev 2013;​37:​9 Pt A:​ tion regulation systems associated with inform each other. Front Psychol 2016;​7:​
1946-54. rejection in love. J Neurophysiol 2010;​104:​ 687.
36. Volkow ND, Fowler JS, Wang G-J, 51-60. 66. Sussman S, Lisha N, Griffiths M.
Swanson JM, Telang F. Dopamine in drug 51. Noori HR, Cosa Linan A, Spanagel R. Prevalence of the addictions: a problem of
abuse and addiction: results of imaging Largely overlapping neuronal substrates the majority or the minority? Eval Health
studies and treatment implications. Arch of reactivity to drug, gambling, food and Prof 2011;​34:​3-56.
Neurol 2007;​64:​1575-9. sexual cues: a comprehensive meta-analy- 67. Hamid AA, Pettibone JR, Mabrouk
37. Robinson TE, Berridge KC. The neural sis. Eur Neuropsychopharmacol 2016;​26:​ OS, et al. Mesolimbic dopamine signals
basis of drug craving: an incentive-sensi- 1419-30. the value of work. Nat Neurosci 2016;​19:​
tization theory of addiction. Brain Res 52. Rømer Thomsen K, Fjorback LO, 117-26.
Brain Res Rev 1993;​18:​247-91. Møller A, Lou HC. Applying incentive sen- 68. Berridge KC. Is addiction a brain dis-
38. Volkow ND. Addiction is a disease of sitization models to behavioral addiction. ease? Neuroethics 2017;​10:​29-33.
free will. Bethesda, MD:​National Insti- Neurosci Biobehav Rev 2014;​45:​343-9. 69. Phelps EA, Lempert KM, Sokol-Hess-
tute on Drug Abuse, 2015 (https://www​ 53. Joyner MA, Kim S, Gearhardt AN. ner P. Emotion and decision making: mul-
.drugabuse​.gov/​about​-­nida/​noras​-­blog/​2015/​ Investigating an incentive-sensitization tiple modulatory neural circuits. Annu
06/​addiction​-­disease​-­free​-­will). model of eating behavior: impact of a Rev Neurosci 2014;​37:​263-87.
39. Figner B, Knoch D, Johnson EJ, et al. simulated fast-food laboratory. Clin Psy- 70. Panno A, Lauriola M, Figner B. Emo-
Lateral prefrontal cortex and self-control chol Sci 2017;​5:​1014-26. tion regulation and risk taking:​predict-
in intertemporal choice. Nat Neurosci 54. Balodis IM, Kober H, Worhunsky PD, ing risky choice in deliberative decision
2010;​13:​538-9. Stevens MC, Pearlson GD, Potenza MN. making. Cogn Emot 2013;​27:​326-34.
40. Volkow ND, Fowler JS. Addiction, a dis- Diminished frontostriatal activity during 71. Nutt DJ, Lingford-Hughes A, Erritzoe
ease of compulsion and drive: involve- processing of monetary rewards and loss- D, Stokes PRA. The dopamine theory of
ment of the orbitofrontal cortex. Cereb es in pathological gambling. Biol Psychia- addiction: 40 years of highs and lows. Nat
Cortex 2000;​10:​318-25. try 2012;​71:​749-57. Rev Neurosci 2015;​16:​305-12.
41. Volkow ND, Morales M. The brain on 55. Robinson MJF, Fischer AM, Ahuja A, 72. Masten CL, Eisenberger NI, Borofsky
drugs: from reward to addiction. Cell 2015;​ Lesser EN, Maniates H. Roles of “want- LA, et al. Neural correlates of social exclu-
162:​712-25. ing” and “liking” in motivating behavior:​ sion during adolescence: understanding
42. Koob GF. Neurobiological substrates gambling, food, and drug addictions. Curr the distress of peer rejection. Soc Cogn
for the dark side of compulsivity in addic- Top Behav Neurosci 2016;​27:​105-36. Affect Neurosci 2009;​4:​143-57.
tion. Neuropharmacology 2009;​56:​Suppl 1:​ 56. Westbrook A, Braver TS. Dopamine 73. Czoty PW, Gage HD, Nader MA. Dif-
18-31 does double duty in motivating cognitive ferences in D2 dopamine receptor avail-
43. Everitt BJ. Neural and psychological effort. Neuron 2016;​89:​695-710. ability and reaction to novelty in socially
mechanisms underlying compulsive drug 57. Craik FIM, Bialystok E. Cognition housed male monkeys during abstinence

n engl j med 379;16 nejm.org  October 18, 2018 1559

The New England Journal of Medicine
Downloaded from nejm.org at KAOHSIUNG MEDICAL UNIVERSITY on October 17, 2018. For personal use only. No other uses without permission.
Copyright © 2018 Massachusetts Medical Society. All rights reserved.
 Br ain Change in Addiction as Learning, Not Disease

from cocaine. Psychopharmacology (Berl) and overdose among persons with opioid pornography consumption: the brain on
2010;​208:​585-92. use disorder. Drug Alcohol Depend 2017;​ porn. JAMA Psychiatry 2014;​71:​827-34.
74. Czoty PW, Nader MA. Effects of dopa- 179:​325-9. 82. Kim SH, Baik SH, Park CS, Kim SJ,
mine D2/D3 receptor ligands on food- 78. Pitchers KK, Coppens CM, Beloate LN, Choi SW, Kim SE. Reduced striatal dopa-
cocaine choice in socially housed male et al. Endogenous opioid-induced neuro- mine D2 receptors in people with Internet
cynomolgus monkeys. J Pharmacol Exp plasticity of dopaminergic neurons in the addiction. Neuroreport 2011;​22:​407-11.
Ther 2013;​344:​329-38. ventral tegmental area influences natural 83. Meaney MJ. Maternal care, gene ex-
75. Morgan D, Grant KA, Gage HD, et al. and opiate reward. J Neurosci 2014;​34:​ pression, and the transmission of individ-
Social dominance in monkeys: dopamine 8825-36. ual differences in stress reactivity across
D2 receptors and cocaine self-administra- 79. Volkow ND, Wang GJ, Telang F, et al. generations. Annu Rev Neurosci 2001;​24:​
tion. Nat Neurosci 2002;​5:​169-74. Low dopamine striatal D2 receptors are 1161-92.
76. Blum K, Chen ALC, Chen TJH, et al. associated with prefrontal metabolism in 84. Keating DP. Born anxious:​the life-
Putative targeting of dopamine D2 recep- obese subjects: possible contributing fac- long impact of early life adversity — and
tor function in reward deficiency syn- tors. Neuroimage 2008;​42:​1537-43. how to break the cycle. New York:​St.
drome (RDS) by synaptamine complexTM 80. Stice E, Spoor S, Bohon C, Small DM. Martin’s Press, 2017.
variant (KB220): clinical trial showing Relation between obesity and blunted 85. Alexander BK, Beyerstein BL, Hada­
anti-anxiety effects. Gene Ther Mol Biol striatal response to food is moderated by way PF, Coambs RB. Effect of early and
2009;​13:​214-30. TaqIA A1 allele. Science 2008;​322:​449- later colony housing on oral ingestion of
77. Stein MD, Conti MT, Kenney S, et al. 52. morphine in rats. Pharmacol Biochem
Adverse childhood experience effects on 81. Kühn S, Gallinat J. Brain structure and Behav 1981;​15:​571-6.
opioid use initiation, injection drug use, functional connectivity associated with Copyright © 2018 Massachusetts Medical Society.

images in clinical medicine

The Journal welcomes consideration of new submissions for Images in Clinical
Medicine. Instructions for authors and procedures for submissions can be found
on the Journal’s website at NEJM.org. At the discretion of the editor, images that
are accepted for publication may appear in the print version of the Journal,
the electronic version, or both.

1560 n engl j med 379;16 nejm.org  October 18, 2018

The New England Journal of Medicine

Downloaded from nejm.org at KAOHSIUNG MEDICAL UNIVERSITY on October 17, 2018. For personal use only. No other uses without permission.
Copyright © 2018 Massachusetts Medical Society. All rights reserved.