BRIEF REPORT
Defining a Cognitive Function Decrement
in Schizophrenia
Richard S.E. Keefe, Charles E. Eesley, and Margaret P. Poe
Background: Although cognitive impairment is described as a core component of the characterization of schizophrenia, a sizable
percentage of patients are classified as unimpaired by traditional definitions of impairment. The purpose of this study was to determine
the percentage of patients with schizophrenia meeting criteria for a “cognitive function decrement” defined as a current level of
cognitive function that falls below the level predicted by premorbid estimates.
Methods: Linear regression analyses were performed on a healthy control population to determine a predicted composite cognitive
score based on maternal education, paternal education, and reading score as indicators of premorbid intellectual function. The
percentages of patients with current cognitive function above and below predicted values were calculated.
Results: When the Wide Range Achievement Test-3 (WRAT-3) score and maternal education are both used to predict current cognitive
performance, as expected, about half (42%) of control subjects fall below expectations. However, 98.1 % of patients fall below
expectations.
Conclusions: When cognitive function decrement is defined as a failure to reach the expected level of cognitive functioning, almost
all patients with schizophrenia meet this definition.
Key Words: Schizophrenia, cognition, premorbid, parental education
P
atients with schizophrenia are well known to have cognitive deficits (Saykin et al 1994; Heinrichs and Zakzanis
1998). However, which patients with schizophrenia are
the most appropriate candidates for cognitive enhancement is a
matter of controversy. The central issue of this controversy is
which patients can be identified as having a cognitive deficit.
Some conceptualizations of this issue have relied on approaches
from clinical neuropsychology that define cognitive impairment.
These definitions are based on a performance deficit compared
with a healthy control population, such as one standard deviation below the mean on one or more areas of cognitive function
(Bryson et al 1993; Heinrichs and Awad 1993; Palmer et al 1997).
The estimated percentage of schizophrenia patients who do not
have a cognitive deficit by these definitions has varied from 27%
(Palmer et al 1997) to 55% (Bryson et al 1993).
While these approaches are useful for clinical practice, they
may not be appropriate for the identification of which patients
may benefit from cognitive enhancement. The rationale that only
severely impaired patients should be included in a cognitive
enhancement trial is based on the idea that only those patients
could benefit from pharmacologic intervention. However, there
is little evidence to support this idea. Results from cognitive
enhancement trials with antipsychotic medications or adjunctive
agents are equivocal with regard to the relationship between
baseline severity of cognitive deficits and treatment-related cognitive improvement. In some studies, patients with the least
severe cognitive deficits have demonstrated the most benefit
from treatment (e.g., Friedman et al 2002; Harvey et al 2003).
There is limited evidence about the relationship between baseline cognitive impairment and response to cognitive training,
although one study suggested that cognitive training may im-
From Duke University, School of Medicine, Durham, North Carolina.
Address reprint requests to Dr. Richard Keefe, Box 3270, Duke University
Medical Center, Durham, NC 27710; E-mail: [email protected].
Received May 17, 2004; revised September 18, 2004; accepted January 5,
2005.
0006-3223/05/$30.00
doi:10.1016/j.biopsych.2005.01.003
prove working memory regardless of baseline cognitive impairment severity (Bell et al 2003).
It is possible that patients who do not meet clinical neuropsychology criteria for impairment may nonetheless have cognitive
deficits compared with what would be expected if they had
never developed schizophrenia. Patients identified as being
“within normal limits” by clinical neuropsychology criteria have
been reported to have parental education and reading scores that
are a full standard deviation above healthy control subjects with
similar cognitive performance (Kremen et al 2000). These data
suggest that patients with schizophrenia who perform at the
control mean on cognitive tests have actually suffered a significant decrement in their level of cognitive functioning. Samples of
monozygotic twins discordant for schizophrenia provide strong
support for this notion, as affected co-twins perform worse than
their unaffected co-twin in various cognitive domains (Goldberg
et al 1990). However, the accuracy of our expectations for
patients’ current level of cognitive functioning is less certain in
patients without a discordant monozygotic twin available for
comparison purposes.
While cognitive performance can be predicted statistically by
demographic factors such as age and personal education, these
variables either affect the course of a patient’s illness (age) or are
affected by the presence of schizophrenia (education). A preferred method for identifying which patients have suffered a
decrement in cognitive functioning compared with expectations
would be to determine the relationship between antecedent
factors such as parental education or premorbid intelligence
quotient (IQ) and current cognitive function in healthy control
subjects and then identify those patients whose current level of
cognitive functioning falls below these antecedent expectations.
In a sample of 141 neurologically healthy subjects, premorbid IQ,
as measured by a reading score combined with the Wechsler
Adult Intelligence Scale-Revised (WAIS-R) score, predicted cognitive test performance beyond the accuracy achieved by demographics alone (Gladsjo et al 1999). Thus, reading score is a
viable predictor of whether an individual’s current cognitive
function meets earlier expectations.
The current study investigates the percentage of patients
whose current level of cognitive function fails to match what is
expected of them based on the antecedent factors of parental
education and estimates of IQ score based on reading tests
BIOL PSYCHIATRY 2005;57:688 – 691
© 2005 Society of Biological Psychiatry
BIOL PSYCHIATRY 2005;57:688 – 691 689
R.S.E. Keefe et al
(Kremen et al 1996, 2000). We hypothesize that when a cognitive
function decrement is defined as worse cognitive function than
predicted by parental education and/or premorbid IQ estimates,
nearly all patients with schizophrenia will meet this definition.
Methods and Materials
Data collection procedures, which were approved by the
Duke University Medical Center Institutional Review Board, have
been described in detail previously (Keefe et al 2004) and will be
summarized briefly here. Fifty healthy control subjects and 150
patients were recruited from the inpatient facilities at Duke
University, John Umstead Hospital, the University of North
Carolina Neurosciences Hospital, and Dorothea Dix Hospital. All
participants provided written informed consent prior to study
entry. Patients were required to meet DSM-IV criteria for schizophrenia, schizoaffective illness, or schizophreniform disorder.
Premorbid intelligence was estimated using the Wide Range
Achievement Test, third edition (WRAT-3), Reading Subtest,
which measures recognition and pronunciation of printed words
considered to be an estimate of premorbid IQ (Gladsjo et al
1999). Maternal education data were available from 107 patients
and paternal education data were available for 105 patients.
Healthy control subjects and patients did not differ significantly
on measures of parental education, age, and ethnic background
as described previously (Keefe et al 2004, Table 1, p. 286).
Subjects received a 2.5-hour cognitive battery, as described in
Keefe et al (2004), assessing the cognitive constructs of verbal
memory, working memory, motor function, processing speed,
attention, and reasoning and problem solving. Good performance was associated with positive scores. A composite score
was calculated as follows: A z-score was calculated for the
primary measure for each test based on the healthy control
standard deviation and mean. To compute a standardized score
for each construct, the average of the z-scores for each of the
measures that comprised each construct was calculated. The
composite score was a z score of the mean of the construct
scores.
Multiple linear regression analyses were performed on the
control population to determine the predicted cognitive composite score based on maternal education, paternal education,
and/or WRAT-3 Reading score. Cognitive function decrement
was defined as a composite score falling below the regression
line. The percentage of patients meeting this definition was
calculated. To determine the bounds within which we could
assert expected performance, a one-tailed 95% confidence interval was calculated for the regression line. This analysis calculated
the range for which the regression line was accurate, with 95%
probability. We used a one-tailed confidence interval because
our hypothesis regarded only the accuracy of our estimates for
one side of the regression line—those patients who performed
worse than expectations. The purpose of this study was not to
determine whether patients performing above the regression line
could be within a 95% interval around the regression line.
of the cognitive composite score with paternal education are as
follows: [␤] ⫽ .699, SE ⫽ .028; t ⫽ 6.482, df ⫽ 45, p ⬍ .001. The
statistics for the prediction of cognitive composite score with
maternal education are as follows: [␤] ⫽ .712, SE ⫽ .026; t ⫽
7.030, df ⫽ 49, p ⬍ .001 (see Figure 1).
One hundred three (96.2%) of 107 patients and 24 (48%) of 50
control subjects met the definition of cognitive function decrement in that they did not reach the cognitive score predicted by
the level of maternal education (88.8% of patients and 34% of
control subjects with a one-tailed 95% confidence interval).
Ninety-eight (93.3%) of 105 patients and 20 (43.5%) of 46 control
subjects did not reach the cognitive score predicted by paternal
education (88.6% of patients and 34.8% of control subjects with
a one-tailed 95% confidence interval).
When WRAT-3 Reading score was used in the regression
equation to predict expected ability ([␤] ⫽ .777, SE ⫽ .010; t ⫽
8.538, df ⫽ 49, p ⬍ .001), 141 (95.3%) of 148 patients and 20
(40%) of 50 control subjects had cognitive composite scores
falling below the regression line (91.2% of patients and 30% of
control subjects with a one-tailed 95% confidence interval).
When WRAT-3 Reading score and maternal education were both
used in the regression equation to predict expected ability ([␤] ⫽
⫺3.858, SE ⫽ .431; t ⫽ ⫺8.943, df ⫽ 49, p ⬍ .001), 105 (98.1%)
of 107 patients and 21 (42%) of 50 control subjects failed to score
at the expected level on the cognitive composite score (92.5% of
patients and 32% of control subjects with a one-tailed 95%
confidence interval). The combination of WRAT-3 reading score
and maternal education was chosen post hoc based on the
superior classification rates of maternal education as compared
with paternal education and because more patients (107 compared with 105) and control subjects (50 compared with 46)
knew their mother’s education compared with their father’s
education.
Discussion
The purpose of the present study was to determine the
frequency of a cognitive function decrement in schizophrenia
patients as defined by a failure to meet the expected level of
Results
All regression equations predicting current cognitive functioning based on parental education and estimates of premorbid
intellectual functioning (WRAT-3 Reading) were calculated using
control subjects and were statistically significant. These regression models were then used for the prediction of cognitive
composite scores using the patients’ parental education and
reading scores. The statistics from the equation for the prediction
Figure 1. Cognitive score predicted by maternal education. The regression
line describes the relationship between maternal education (in years) and
current cognitive composite score. The curved line represents the 95%
confidence interval of the regression line. Those individuals falling above
the regression line perform above the expectations generated by maternal
education. Those individuals falling below the regression line perform below expectations. *Maternal education values jittered for clarity.
www.elsevier.com/locate/biopsych
690 BIOL PSYCHIATRY 2005;57:688 – 691
cognitive functioning predicted by parental education and/or
WRAT-3 Reading score. Almost all patients (up to 98%) met this
definition, compared with 40% of healthy control subjects. The
percentage of patients is far higher than has previously been
reported when cognitive impairment has been characterized as a
cognitive performance that is not within the normal range
(Bryson et al 1993; Palmer et al 1997). The present study suggests
that while schizophrenia patients may be found who perform
within the normal range on cognitive tests, it is likely that these
patients have suffered a decrement in their cognitive functions.
For any regression strategy, about half of the sample used to
define the regression line will fall above and below the line.
Thus, unsurprisingly, about 40% of healthy control subjects in
this study performed below expectations. Our strategy suggests
that these individuals, like those with schizophrenia, did not
function on cognitive assessments at the level predicted by their
WRAT-3 Reading scores or parental education. One limitation of
using this approach to defining a cognitive function decrement is
that failing to meet the expectations forwarded by our predictor
variables does not in itself suggest that healthy control subjects
are cognitively unhealthy. Typical clinical estimates of impairment based on a definition of one standard deviation below the
normal mean usually include a lower number—about 15%— of
the general population as impaired. Thus, our approach identifies a larger proportion of the healthy control population as not
meeting average expectations, which may be seen as a disadvantage of this scheme. However, the advantage of our approach
over traditional approaches is that it allows the identification of
anyone who has a cognitive function decrement relative to an
estimate of their inherited or premorbid capability. Therefore, it
is not based on current level of functioning itself but rather the
difference between current functioning and expectations based
on antecedent factors. This difference may be the most relevant
measure of which patients with schizophrenia have suffered
cognitively as a result of their illness.
Cognitive impairment has increasingly been described as a
core feature of schizophrenia that should be a target for treatment (Hyman and Fenton 2003). If cognitive impairment is
indeed considered as a core feature, it implies that very few
patients are able to maintain the cognitive functions that they
would have had if they had not developed schizophrenia and
that cognitive impairment is not simply the result of symptoms or
treatment (Elvevag and Goldberg 2000). The results of the
current study point to a way of resolving the apparent contradiction between the idea that cognitive impairment is a core
feature of schizophrenia and the characterization of sizable
subgroups of schizophrenia patients with normal cognitive abilities. Since the current cognitive functions of almost all patients
suggest a failure to meet premorbid expectations, almost all
patients would be considered to have a cognitive performance
decrement.
Previous definitions of schizophrenia patients who perform
within the normal range as neuropsychologically normal may
possibly bias treatment strategies against improving cognition in
high-functioning schizophrenia patients who have nonetheless
suffered a decline in cognition and possibly in quality of life.
When a definition of cognitive impairment is based on expected
level of cognitive functioning, all patients suffering from a deficit
or decline in cognitive function are identified for potential
cognition-improving treatments. This strategy would target almost all patients with schizophrenia for cognitive enhancement,
consistent with the philosophy that patients who do not meet the
expected level of cognitive function may be able to benefit from
www.elsevier.com/locate/biopsych
R.S.E. Keefe et al
cognitive intervention, even if their cognitive functions are in the
normal range.
Theoretically, this strategy would permit the inclusion in
treatment studies of subjects who might perform at ceiling for the
cognitive measures used as primary outcomes. However, welldesigned batteries of tests will not include measures that are
susceptible to ceiling effects in patients with schizophrenia. Less
than 1% of 1493 patients in the schizophrenia trial of the Clinical
Antipsychotic Trials of Intervention Effectiveness (CATIE) Project
(Keefe et al 2003), performed within one standard deviation of
the ceiling on trial one of the Hopkins Verbal Learning Test. A
recent Food and Drug Administration-National Institute of Mental
Health (FDA-NIMH) panel on designs for cognitive enhancement
trials in schizophrenia recommended that the only patients who
should be excluded from these trials are those few who perform
at ceiling at baseline (Buchanan et al, in preparation) but
estimated that this percentage would be very small when carefully designed batteries are used.
This approach to defining a cognitive decrement raises ethical
issues regarding the risk/benefit ratio of pharmacotherapy. Patients who fall below predicted cognitive performance but who
are otherwise classified as unimpaired may be viewed as not able
to benefit enough from additional pharmacotherapy to warrant
the potential associated risks. However, since as few as 10% of
patients with schizophrenia are employed (Marwaha and Johnson 2004) and less than 5% have a college degree (Keith et al
1991), it is clear that being neuropsychologically normal (which
includes at least 27% of patients) is not adequate for achieving
healthy functional outcomes in schizophrenia patients. Cognitive
enhancement for these neuropsychologically normal patients,
who would have been at the higher end of the cognitive and
functional distributions in the normal population if they had not
developed schizophrenia, may potentially make a substantial
contribution to their quality of life.
This study is limited by the small number of control subjects
that were used to establish the relationships between predictive
factors and current cognitive functioning. A study with a larger
number of control subjects will provide more definitive data
about the relationship between current cognitive functioning and
antecedent factors in the healthy population. However, the 98%
classification of patients with a performance decrement will be
difficult to improve on. Further, the amount of decrement from
premorbid measures such as reading score and parental education is probably underestimated in the schizophrenia sample.
First, since young people destined to develop schizophrenia
already have experienced mild cognitive impairment at the early
stages of life when they are learning to read words (Davidson
et al 1999; Fuller et al 2002) and since parents of patients with
schizophrenia may also have reduced education due to the
presence of schizophrenia or related phenotypes in the family
(Keefe et al 1994), the antecedent estimates used in this study are
already affected by mild cognitive impairments associated with
the prodrome of schizophrenia or its heritability. Second, for the
63% of patients who were receiving atypical antipsychotic treatment, which may have had a mild beneficial effect on their
cognitive deficits (Harvey and Keefe 2001), the cognitive performance was less severe than if untreated or treated with typical
antipsychotics, and the amount of cognitive decrement may have
been further underestimated.
This study was presented in part at the 59th meeting of the
Society of Biological Psychiatry in New York City, May, 2004.
Neurocognitive data were collected by Leigh Coughenour, Adam
R.S.E. Keefe et al
Vaughn, Susan Shortel, Matthew Dukes, Trina Walker, and
Joseph Kang.
Bell M, Bryson G, Wexler BE (2003): Cognitive remediation of working memory deficits: Durability of training effects in severely impaired and less
severely impaired schizophrenia. Acta Psychiatr Scand 108(2):101–109.
Bryson GJ, Silverstein ML, Nathan A, Stephen L (1993): Differential rate of
neuropsychological dysfunction in psychiatric disorders: Comparison
between the Halstead-Reitan and Luria-Nebraska batteries. Percept Mot
Skills 76(1):305–306.
Buchanan RW, Davis M, Goff D, Green MF, Keefe RSE, Leon AC, et al (in press):
Report from the FDA-NIMH-MATRICS workshop on clinical trial designs
for neurocognitive drugs for schizophrenia. Schizophr Bull.
Davidson M, Reichenberg A, Rabinowitz J, Weiser M, Kaplan Z, Mark M
(1999): Behavioral and intellectual markers for schizophrenia in apparently healthy male adolescents. Am J Psychiatry 156(9):1328 –1335.
Elvevag B, Goldberg TE (2000): Cognitive impairment in schizophrenia is the
core of the disorder. Crit Rev Neurobiol 14:1–21.
Friedman JI, Adler DN, Howanitz E, Harvey PD, Brenner G, Temporini H, et al
(2002): A double blind placebo controlled trial of donepezil adjunctive
treatment to risperidone for the cognitive impairment of schizophrenia.
Biol Psychiatry 51:349 –357.
Fuller R, Nopoulos P, Arndt S, O’Leary D, Ho B, Andreasen NC (2002): Longitudinal assessment of premorbid cognitive functioning in patients with
schizophrenia through examination of standardized scholastic test performance. Am J Psychiatry 159(7):1183–1189.
Gladsjo JA, Heaton RK, Palmer BW, Taylor MJ, Jeste DV (1999): Use of oral
reading to estimate premorbid intellectual and neuropsychological
functioning. J Int Neuropsychol Soc 5:247–254.
Goldberg TE, Ragland JD, Torrey EF, Gold JM, Bigelow LB, Weinberger DR
(1990): Neuropsychological assessment of monozygotic twins discordant for schizophrenia. Arch Gen Psychiatry 47:1066 –1072.
Harvey PD, Green MF, McGurk SR, Meltzer HY (2003): Changes in cognitive
functioning with risperidone and olanzapine treatment: a large-scale,
double-blind, randomized study. Psychopharmacology 169:404 – 411.
BIOL PSYCHIATRY 2005;57:688 – 691 691
Harvey PD, Keefe RSE (2001): Studies of cognitive change in patients with
schizophrenia following novel antipsychotic treatment. Am J Psychiatry
158(2):176 –184.
Heinrichs RW, Awad AG (1993): Neurocognitive subtypes of chronic schizophrenia. Schizophr Res 9(1):49 –58.
Heinrichs RW, Zakzanis KK (1998): Neurocognitive deficit in schizophrenia: A
quantitative review of the evidence. Neuropsychology 12:426 – 445.
Hyman SE, Fenton WS (2003): What are the right targets for psychopharmacology? Science 299(5605):350 –351.
Keefe RSE, Goldberg TE, Harvey PD, Gold JM, Poe M, Coughenour L (2004):
The Brief Assessment of Cognition in Schizophrenia: Reliability, sensitivity, and comparison with a standard neurocognitive battery. Schizophr
Res 68(2–3):283–297.
Keefe RSE, Mohs RC, Bilder RM, Harvey PD, Green MF, Meltzer HY, et al (2003):
Neurocognitive assessment in the CATIE Project schizophrenia trial: Development, methodology and rationale. Schizophr Bull 29:45–55.
Keefe RSE, Silverman JM, Lees Roitman SE, Harvey PD, Duncan A, Alroy
D, et al (1994): Performance of nonpsychotic relatives of schizophrenic
patients on cognitive tests. Psychiatry Res 53:1–12.
Keith SJ, Regier DA, Rae DS (1991): Schizophrenic disorders. In: Robins LN,
Regier DA, editors. Psychiatric Disorders in America: The Epidimiologic
Catchment Area Study. New York: The Free Press.
Kremen WS, Seidman LJ, Faraone SV, Pepple JR, Lyons MJ, Tsuang MT (1996):
The “3 Rs” and neuropsychological function in schizophrenia: An empirical test of the matching fallacy. Neuropsychology 10:22–31.
Kremen WS, Seidman LJ, Faraone SV, Toomey R, Tsuang MT (2000): The
paradox of normal neuropsychological function in schizophrenia. J Abnorm Psychol 109:743–752.
Marwaha S, Johnson S (2004): Schizophrenia and employment: A review. Soc
Psychiatry Psychiatr Epidemiol 39(5):337–349.
Palmer BW, Heaton RK, Paulsen JS, Kuck J, Braff D, Harris MJ, et al (1997): Is it
possible to be schizophrenic yet neuropsychologically normal? Neuropsychology 11:437– 446.
Saykin AJ, Shtasel DL, Gur RE, Kester DB, Mozley LH, Stafiniak P, et al (1994):
Neuropsychological deficits in neuroleptic naive patients with first episode schizophrenia. Arch Gen Psychiatry 51:124 –131.
www.elsevier.com/locate/biopsych