CRITICAL REVIEW
What is cognitive reserve? Theory and research
application of the reserve concept
YAAKOV STERN
Cognitive Neuroscience Division, G.H. Sergievsky Center, The Taub Institute, and Departments of Neurology, Psychiatry,
and Psychology, Columbia University College of Physicians and Surgeons
(Received August 22, 2000; Revised February 26, 2001; Accepted February 28, 2001)
Abstract
The idea of reserve against brain damage stems from the repeated observation that there does not appear to be a
direct relationship between the degree of brain pathology or brain damage and the clinical manifestation of that
damage. This paper attempts to develop a coherent theoretical account of reserve. One convenient subdivision of
reserve models revolves around whether they envision reserve as a passive process, such as in brain reserve or
threshold, or see the brain as actively attempting to cope with or compensate for pathology, as in cognitive reserve.
Cognitive reserve may be based on more efficient utilization of brain networks or of enhanced ability to recruit
alternate brain networks as needed. A distinction is suggested between reserve, the ability to optimize or maximize
normal performance, and compensation, an attempt to maximize performance in the face of brain damage by using
brain structures or networks not engaged when the brain is not damaged. Epidemiologic and imaging data that help
to develop and support the concept of reserve are presented. (JINS, 2002, 8, 448– 460.)
Keywords: Functional imaging, Alzheimer’s disease, Compensation, Brain damage, Epidemiology
INTRODUCTION
The idea of reserve against brain damage stems from the
repeated observation that there does not appear to be a di-
rect relationship between the degree of brain pathology or
brain damage and the clinical manifestation of that damage.
For example, Katzman et al. (1989) described 10 cases of
cognitively normal elders who were discovered to have ad-
vanced Alzheimer’s disease (AD) pathology in their brains
at death. They speculated these women did not express the
clinical features of AD because their brains were larger
than average. Similarly, most clinicians are aware of the
fact that a stroke of a given magnitude can produce pro-
found impairment in one patient and while having minimal
effect on another. Something must account for the disjunc-
tion between the degree of brain damage and its outcome,
and the concept of reserve has been proposed to serve this
purpose.
There have been many attempts to produce a coherent
theoretical account of reserve. This paper will attempt to
review and synthesize concepts that have been suggested,
such as threshold, compensation, neuronal or brain reserve,
and cognitive reserve. Many of these terms have been used
interchangeably by previous authors, and they have not had
well-accepted definitions. Specific definitions will be of-
fered for these concepts that attempt to capture potential
theoretical distinctions between them.
The concept of reserve should be relevant to any situa-
tion where the brain sustains injury. In addition, it will be
argued that the concept of reserve should be extended to
encompass variation in healthy individuals’ performance,
particularly when they must perform at their maximum ca-
pacity. Nevertheless, many of the concrete examples will
be framed around AD, with the implicit assumption that the
discussion has implications for brain damage in general.
AD has some unique advantages for examining disease-
induced changes in brain function. AD pathology affects
cortical circuitry that subserves a wide range of cognitive
functions, and its pathology is more likely than conditions
Reprint requests to: Yaakov Stern, Sergievsky Center, 630 W. 168th
Street, New York, NY 10032. E-mail: ys11@columbia.edu
Journal of the International Neuropsychological Society (2002), 8, 448–460.
Copyright © 2002 INS. Published by Cambridge University Press. Printed in the USA.
DOI: 10.1017.S1355617701020240
448
such as stroke to affect similar anatomic sites across sub-
jects, allowing better generalization. AD is also slowly but
inexorably progressive, providing a more sensitive indica-
tor of the severity of brain insult required before cognitive
networks change. On the other hand, the potential for ad-
aptation of recovery might vary between slowly progres-
sive and acute pathologies, so studies of AD may not always
have direct implications for studies of other conditions.
Finally, it should be noted that this paper is not intended
to be a comprehensive review of all of the literature rele-
vant to the concept of reserve. For example, there is a large
body of work investigating the concept of reserve in the
context of HIV-related cognitive functioning that will not
be addressed here (Basso & Bornstein, 2000; Pereda et al.,
2000; Satz et al., 1993; Starace et al., 1998; Stern et al.,
1996). Rather, work has been selected that helps to develop
and support the ideas that will be presented.
DEFINING RESERVE
One convenient, although not entirely accurate, subdivision
of reserve models revolves around whether they envision
reserve as a passive process, or see the brain as actively
attempting to cope with or compensate for pathology. In
passive models, reserve is defined in terms of the amount of
damage that can be sustained before reaching a threshold
for clinical expression. In the active models, reserve re-
volves around differences in how the task is processed. These
two approaches are not mutually exclusive. Ultimately, some
combination of these two approaches might best describe
the empirical observations that have prompted us to de-
velop the concept of reserve.
Passive Models: Brain Reserve or Threshold
Many investigators have proposed passive models includ-
ing Katzman (1993; brain reserve) and Mortimer et al. (1981;
neuronal reserve). This type of model has also long been
implicitly adopted by most clinicians. The threshold model,
critically reviewed by Satz (1993), is one of the best artic-
ulated passive models. The threshold model revolves around
the construct of brain reserve capacity (BRC). While BRC
is a hypothetical construct, concrete examples of BRC might
include brain size or synapse count. The model recognizes
that there are individual differences in BRC. It also presup-
poses that there is a critical threshold of BRC. Once BRC is
depleted past this threshold, specific clinical or functional
deficits emerge.
This formulation, illustrated in Figure 1 (derived from
Satz, 1993b), is sufficient to account for many clinical ob-
servations. Assume that two patients have two different
amounts of BRC. A lesion of a particular size might result
in a clinical deficit in a person with less BRC (Patient 2),
because it exceeds the threshold of brain damage sufficient
to produce that deficit. However, an individual with greater
BRC (Patient 1) could remain unaffected, because this
threshold is not exceeded. Thus, more BRC can be consid-
ered protective factor, while less BRC would impart vul-
nerability. An apparently intact individual with pre-existing
brain damage can tolerate less new brain damage than an-
other individual without this underlying pathology: the pre-
existing damage reduces the amount of remaining BRC, so
the new lesion is sufficient to exceed the functional impair-
ment cutoff.
Many observations about the prevalence and incidence
of AD are consistent with the threshold model. Figure 2
(based on Katzman, 1993) illustrates that the progression of
AD pathology and the clinical expression of AD can be
discontinuous. AD pathology probably begins to develop
many years before the disease is expressed clinically and
Fig. 1. The threshold or brain reserve model. In 2 patients with
different amounts of brain reserve capacity (BRC), a lesion of a
particular size results in a clinical deficit in a person with less
BRC (Patient 2), because it exceeds the threshold of brain damage
sufficient to produce that deficit. However, an individual with
greater BRC could remain unaffected.
Fig. 2. AD pathology probably begins to develop many years be-
fore the disease is expressed clinically and slowly becomes more
severe. At some point symptoms of sufficient severity allow the
diagnosis of AD. The arrows surrounding the point in the figure
where clinical symptoms appear denote the fact that there are
individual differences in reserve capacity, and these differences
result in later or earlier expression of clinical symptoms.
Cognitive reserve 449
slowly becomes more severe. The threshold model would
assume that when synapses are depleted beyond some crit-
ical point
1
the initial symptoms of dementia will appear. At
some point after this, depletion will result in symptoms of
sufficient severity to allow the diagnosis of AD. The arrows
surrounding the point in the figure where clinical symp-
toms appear denote the fact that there are individual differ-
ences in reserve capacity, and these differences result in
later or earlier expression of clinical symptoms. In patients
with more reserve, synapse loss must be more severe before
clinical symptoms appear and the symptoms appear later.
Conversely, symptoms would appear earlier in a patient
with less reserve.
The threshold approach can be extended to account for
more than just differences in the onset of a clinical out-
come. Because reserve mediates between the pathology and
its clinical outcome, the level of reserve should also influ-
ence the severity of clinical symptoms after the threshold
for their appearance has occurred. This is demonstrated sche-
matically in Figure 3 with regard to AD. Almost all patients
in this scheme are demented, except those with mild pathol-
ogy and high levels of reserve. Two levels of pathologic
severity are illustrated. Within any level, patients with more
reserve show less severe clinical signs of AD as assessed by
global measures such as mental status tests or by more fo-
cused measures such as memory tests. Still, at any level of
reserve, more severe pathology results in more severe clin-
ical deficits.
Some research that supports the threshold or brain re-
serve model in AD will be reviewed below. To give one
concrete example, several studies have found that individ-
uals with larger brain size or head circumference have less
severe AD or are less likely to develop AD (Graves et al.,
1996; Schofield et al., 1997), or are more likely to have less
severe AD. Ostensibly, individuals with larger brain size
would have more synapses to lose before the critical thresh-
old for AD is reached.
There are several reasons why threshold or brain reserve
models can be termed passive models of reserve. First, this
type of model assumes that there is some fixed cut-off or
threshold at which functional impairment will occur for
everyone. In the case of AD, this threshold might be deple-
tion of synapses to the point where only a specific number
remain. Second, threshold models are essentially quantita-
tive models. They assume that a specific type of brain dam-
age will have the same effect in each person, and that
repeated instances of brain damage sum together. Individ-
uals differ only in their overall brain capacity, and brain
damage is either sufficient or insufficient to deplete BRC to
some critical level. The model does not account for individ-
ual differences in how the brain processes cognitive or func-
tional tasks in the face of the disruption caused by brain
damage. It also does not address potential qualitative dif-
ferences between different types of brain damage.
These observations do not negate the importance of the
threshold model. They just suggest that this model alone is
probably not sufficient to explain all features of reserve and
that extensions of the threshold model need to be considered.
Active Models
The active models of reserve suggest that the brain actively
attempts to compensate for brain damage. I suggest that in
its active form, there can be at least two types of reserve.
The first is cognitive reserve. This could take the form of
using brain networks or cognitive paradigms that are less
susceptible to disruption. I propose that this type of reserve
is a normal process used by healthy individuals when cop-
ing with task demands. The second is compensation: using
brain structures or networks not normally used by individ-
uals with intact brains in order to compensate for brain
damage.
Cognitive reserve
Cognitive reserve parallels the concept of brain reserve in
that it is a potential mechanism for coping with brain dam-
age. In the threshold model, the reserve capacity typically
consists of additional synapses or an increased number of
redundant neuronal networks. Cognitive reserve focuses
more on the “software.” This could consist of the ability of
the cognitive paradigm underlying a task to sustain disrup-
tion and still operate effectively. Alternately, this could con-
sist of the ability to use alternate paradigms to approach a
problem when the more standard approach is no longer
operational.
The concept of cognitive reserve provides a ready expla-
nation for why many studies have demonstrated that higher
1
For the purposes of discussion, we can treat the advancing AD
pathology as loss of synapses. Loss of synapses is the facet of AD pathol-
ogy that has been most reliably linked to cognitive change and disease
severity (DeKosky & Scheff, 1990; Terry et al., 1991).
Fig. 3. Because reserve mediates between the pathology and its
clinical outcome, the level of reserve should also influence the
severity of clinical symptoms after the threshold for their appear-
ance has occurred. Here, at any level of disease pathology, pa-
tients with more reserve evidence more mild levels of clinical
severity
450 Y. Stern
levels of intelligence, and of educational and occupational
attainment are good predictors of which individuals can
sustain greater brain damage before demonstrating func-
tional deficit. Rather than positing that these individuals’
brains are grossly anatomically different than those with
less reserve (e.g., they have more synapses), the cognitive
reserve hypothesis posits that they process tasks in a more
efficient manner.
The concept of cognitive reserve also differs from the
passive threshold approach in other important ways. Recall
that in the passive model, individuals may have different
levels of BRC and a lesion of the same size is sufficient to
deplete BRC below the critical threshold in some individ-
uals but not others (see Figure 1). The cognitive reserve
model is illustrated in Figure 4. Here the 2 patients have the
same amount of BRC (again, let’s say, the same number of
synapses). However, Patient 1 has more cognitive reserve
than Patient 2, in that Patient 1 uses more efficient process-
ing mechanisms. As a result, Patient 1 can tolerate a larger
lesion than Patient 2 before functional impairment is appar-
ent. Thus, the cognitive reserve model does not assume that
there is some fixed cut-off or threshold at which functional
impairment will occur. The critical threshold differs from
one person to the next, depending on how efficient or resil-
ient the “software” is in using the remaining neural sub-
strate. Putting it another way, the threshold approach
supposes that the person with more BRC has more to lose
before they reach some clinical cut-point. The cognitive
reserve hypothesis focuses less on what is lost and more on
what is left. In the case of AD, one individual may begin to
express clinical features when synapses are depleted to a
particular number, while an individual with more cognitive
reserve may be able to operate effectively with the same
number of synapses.
From a strict point of view, the differences in cognitive
processing envisioned by the cognitive reserve model must
also have a physiologic basis, in that the brain must ulti-
mately mediate all cognitive function. The difference is in
terms of the level of analysis. Presumably, the physiologic
variability subsumed by cognitive reserve is at the level of
variability in synaptic organization, or in relative utiliza-
tion of specific brain regions. Thus cognitive reserve im-
plies anatomic variability at the level of brain networks,
while brain reserve implies differences in the quantity of
available neural substrate.
The cognitive reserve model also does not assume that a
specific type of brain damage will have the same effect in
each person. Because of individual variability in how they
cope with brain damage, the same amount of damage will
have different effects on different people, even if BRC is
held constant.
A proposed definition of cognitive reserve is: the ability
to optimize or maximize performance through differential
recruitment of brain networks, which perhaps reflect the
use of alternate cognitive strategies. Since the changes in
brain recruitment associated with reserve are a normal re-
sponse to increased task demands, this definition suggests
that cognitive reserve is present in both healthy individuals
and those with brain damage, and is reflected in the modu-
lation of the same brain networks. In essence, an individual
who uses a brain network more efficiently, or is more ca-
pable of calling up alternate brain networks or cognitive
strategies in response to increased demand may have more
cognitive reserve. The definition encompasses two possi-
bilities, differences in recruitment of the same network, and
differential ability to recruit alternate networks. These pos-
sibilities will be discussed in turn. This discussion is ex-
tremely speculative, although some evidence to support these
speculative lines will be reviewed.
More efficient use of brain networks
This idea is based on studies of how normal individuals
respond as tasks are made increasingly difficult and of in-
dividual differences in task performance. The rough paral-
lel here is that, in effect, brain damage acts to increase task
difficulty. Several functional imaging studies suggest that a
common response to increasing task difficulty in normal
individuals is increased activation of areas involved in an
easier version of the task and0or the recruitment of addi-
tional brain areas (Grady et al., 1996; Grasby et al., 1994;
Gur et al., 1988; Rypma et al., 1999). There are also indi-
vidual differences in how this additional recruitment oc-
curs. For any level of task difficulty, more skilled individuals
typically show less task-related recruitment than less skilled
individuals. When the more skilled individual exerts her-
self maximally, she can perform better than the less skilled
one. This increased efficiency and larger dynamic range
can be considered reserve. If brain damage is considered a
form of demand (similar to increased task difficulty), then a
person with more cognitive reserve would be able to cope
with more brain damage and still maintain effective func-
tioning. Often, increased task difficulty results in the re-
cruitment of additional brain areas or networks. We might
Fig. 4. The cognitive reserve model. Two patients have the same
amount of brain reserve. However, Patient 1 has more cognitive
reserve than Patient 2, in that Patient 1 uses more efficient pro-
cessing mechanisms. As a result, Patient 1 can tolerate a larger
lesion than Patient 2 before functional impairment is apparent.
Cognitive reserve 451
speculate that, in a person with more reserve, this addi-
tional recruitment would occur at a higher difficulty level.
Differential ability to recruit
alternate networks
This possibility is more speculative, but is consistent with
the concept of cognitive reserve. The point is simply that a
person with more reserve might be able to call on a larger
array of alternate networks for solving the problem at hand.
As a concrete example, a trained mathematician might be
able to solve a mathematics problem many different ways,
while a less experienced individual might have only one
possible solution strategy available. The mathematician
would have more flexibility in solving the problem if any
particular solution strategy was precluded. This built in re-
dundancy would permit greater resilience in the face of
brain damage.
These two ideas about reserve might form a heuristic
framework for designing studies about cognitive reserve.
Studies can be aimed at behavior in unimpaired individu-
als, taking advantage of inherent inter-individual differ-
ences in skills or intelligence. Predictions can then be made
about how these individual differences might affect re-
sponse to brain damage.
Compensation
The term cognitive reserve can be limited to the variability
seen in non-brain damaged individuals, which distin-
guishes it from compensation, which might be reserved
for a specific response to brain damage. This distinction
emerges from the consideration of findings in functional
imaging studies that compared task-related activation in
impaired and unimpaired groups with a more critical eye.
For example, several functional neuroimaging studies com-
paring task-related activation in AD patients and controls
found more marked and extensive activation in the pa-
tients (Becker et al., 1996; Deutsch et al., 1993; Grady
et al., 1993). These findings have been interpreted as evi-
dence that the patients compensated for AD pathology.
That is, since pathology impaired the patients’ ability to
mediate the task through the same brain network used by
the controls, the patients compensated by engaging alter-
nate brain areas during task performance. One may ask
whether this observed “compensation,” as the investiga-
tors called it, is the same thing as cognitive reserve. If
compensation truly represents a change that is induced by
brain damage, then it might be important to distinguish
between compensation and cognitive reserve. This distinc-
tion has not been commonly used in the reserve literature.
However, recent functional imaging studies are often care-
fully designed to ensure that observed group differences
are not simply a function of task difficulty.
Compensation is thus first defined in the negative, in that
it cannot be simply a normal response to difficulty. In ad-
dition, the term compensation implies an attempt to maxi-
mize performance in the face of brain damage by using
brain structures or networks not engaged when the brain is
not damaged.
One of the theoretical reasons why discriminating be-
tween reserve and compensation might be important is that
it helps critically evaluate the results of studies comparing
impaired and unimpaired populations. For example, one
study (Becker et al., 1996) reported a comparison of PET
rCBF in AD patients and elderly controls performing a ver-
bal list-learning task. Three list-lengths, of one, three, and
eight were used. In the eight-word task (compared to the
three-word task), patients showed decreased activation of
the lateral frontal cortex relative to controls. However, dor-
solateral prefrontal cortex and areas surrounding the angu-
lar gyrus were more active than in the controls. The authors
suggested that this may represent a response to neuropath-
ologic changes that is specific to AD patients. In the pro-
posed classification scheme, this would be considered
compensation. Later, using another analytic technique, the
same authors concluded that both patients and controls used
the same underlying network to mediate the memory task
(Herbster et al., 1996). They reported that observed group
differences they originally reported were a result of differ-
ential activation of this network, probably because the task
was more difficult for the AD patients. In the proposed
classification scheme, this would be considered cognitive
reserve. In assessing the brain’s response to damage, it clearly
will be important to know which responses are within the
range of normal behavior and which only occur in the pres-
ence of pathology.
Disentangling compensation and reserve presents a spe-
cific experimental design problem. Studies must use tasks
that allow for systematic manipulation of task difficulty.
Ideally, task difficulty should be equated across individu-
als, not just subject groups. Once task difficulty is equated,
group differences in patterns and levels of functional acti-
vation are more likely to represent compensation, and not
reserve.
Stern et al. (2000) tried to determine whether the pathol-
ogy of Alzheimer’s disease (AD) alters the brain networks
subserving performance on a memory task, while carefully
controlling for task difficulty. H
2
[
15
O] PET was used to
measure regional cerebral blood flow in patients and healthy
elders during the performance of a verbal recognition task.
Task difficulty was matched across participants by adjust-
ing the size of the list that each subject had to remember
such that each subject’s recognition accuracy was 75%. In
the healthy elders, a network of brain areas involving left
anterior cingulate, anterior insula, and left basal ganglia
was activated during task performance. Higher study list
size (SLS) was associated with increased recruitment of
this network, indicating that this network was associated
with task performance and that subjects who could recruit
the network to a greater degree could perform the task bet-
ter. Only 3 AD patients also expressed this network in a
similar manner. This network used by the controls and a
minority of AD patients may mediate reserve, in that it
452 Y. Stern