‘Chemobrain’ in breast carcinoma?

A prologue




Chemotherapy-induced cognitive dysfunction in patients with breast carcinoma has been described previously. However, those studies only assessed patients' postchemotherapy cognitive functioning and were not able to determine the relation between cognitive function and other treatments, such as surgery and radiotherapy, that often precede systemic chemotherapy.


Eighty-four women with breast carcinoma underwent a comprehensive neuropsychologic evaluation before receiving adjuvant therapy for nonmetastatic primary breast carcinoma.


Before the start of systemic therapy, 35% of women in the current cohort exhibited cognitive impairment. Verbal learning (18%) and memory function (25%) were impaired significantly more frequently relative to normative expectations. Although the impairments were not significant in the women who were examined, nonverbal memory (17%), psychomotor processing speed and attention (13%), confrontational naming (13%), visuoconstruction (13%), and upper-extremity fine motor dexterity (12%) were impaired more frequently than was expected. Affective distress was related significantly to cognitive impairment (Pearson chi-square = 9.90; P = 0.002). Given the conservative statistical approach employed, extent of surgery, hormone replacement therapy history, and current menopausal status failed to achieve statistical significance, but these variables did exhibit provocative trends with respect to cognitive impairment.


Cognitive impairment frequently is observed before the administration of systemic chemotherapy. Thus, investigations purporting to measure chemotherapy-induced cognitive dysfunction must employ study designs that incorporate prechemotherapy baseline assessments to accurately detect changes in cognitive function that are attributable to chemotherapy. Cancer 2004. © 2004 American Cancer Society.

Cancer and the therapeutic modalities used to treat cancer have been associated with alterations in patients' cognitive, emotional, and neurologic functioning. 1 It is widely accepted that disease involving the central nervous system (CNS) and treatments targeting the CNS (e.g., radiotherapy) may have adverse effects on cognitive functioning. There is growing awareness that malignant disease outside of the CNS and the treatment of such malignancies with biologic, immunologic, and/or hormonal techniques also may result in alterations in patients' mental status. 2

Breast carcinoma is the second most frequent cause of malignancy-related death among women in the United States. It is estimated that nearly one in nine women will develop some form of breast carcinoma in their lifetime. 3 Despite optimal surgical treatment for breast carcinoma, residual tumor cells may remain and either fail to be removed at the surgical site or disseminate to other areas of the body. For this reason, adjuvant therapy (e.g., radiotherapy, chemotherapy, hormonal therapies) has become a standard component of care for many women with breast carcinoma. The use of adjuvant chemotherapy is widely recognized as an important component of multimodal treatment. It was found that adjuvant chemotherapy increased disease-free and overall survival for many patients with breast carcinoma. 4, 5 However, chemotherapy has been associated with adverse effects, such as nausea, peripheral neuropathy, and encephalopathy. 6, 7 The relation between chemotherapy and cognitive functioning is an area of active interest in both clinical and research settings.

A clear understanding of the relation between chemotherapeutic treatments and cognitive functioning is critical, because there is a growing group of malignancy survivors who often want to return to their former occupational, scholastic, or familial activities. In many patients, the integrity of cognitive functioning is critical to the fulfillment of this goal. Recently, a number of studies of women with breast carcinoma have described changes in cognitive functioning that are purportedly associated with chemotherapy. Several retrospective studies have examined cognitive functioning months to years after chemotherapy and concluded that chemotherapy in patients with breast carcinoma is associated with persistent cognitive deficits after the completion of treatment. 8–13 The term chemobrain has been coined in the lay press to describe the cognitive decline associated with chemotherapy. However, to examine the cognitive effects of a therapeutic intervention rigorously, it is of paramount importance that investigators obtain a pretreatment measure of neuropsychologic functioning. Only in this manner can investigators determine whether the observed performance in subsequent evaluations (i.e., postchemotherapy) represents a change in functioning secondary to treatment or whether the observed deficits were present before the administration of the therapeutic agent(s). Remarkably, this critical assessment time point has been absent in all published neuropsychologic studies to date. Each of those studies concluded that there was clinically significant cognitive impairment associated with chemotherapy. However, conclusive support for this association requires longitudinal data that include objective measurement of patients' prechemotherapy cognitive functioning. Anecdotally, several patients at our institution (The University of Texas M. D. Anderson Cancer Center, Houston, TX) have expressed reservations regarding the receipt of agents with known beneficial effects secondary to concerns about acquiring ‘chemobrain’. This finding exemplifies emerging public awareness and concern surrounding the issue of chemobrain and underscores the importance of performing methodologically rigorous studies to help assess the frequency and nature, as well as the potential mechanisms, of chemotherapy-induced neurotoxicity.

The potential neurotoxic effects of adjuvant chemotherapeutic treatments are a serious concern for many women with breast carcinoma, and well designed investigations to help elucidate the potential risks and benefits of such treatments are necessary. Characterizing the neuropsychologic profiles of women with breast carcinoma who have not previously received chemotherapy will enhance our understanding of the multiple factors that may contribute to cognitive dysfunction in this population. In addition, these data should reinforce the importance of using prospective trials to assess change in cognitive function secondary to chemotherapy, as the use of retrospective analytic techniques is a methodologic limitation of all published trials to date. To this end, we have examined data that were generated by three separate prospective clinical research trials conducted at the University of Texas M. D. Anderson Cancer Center examining the effects of hormonal treatments (2001–2003) and chemotherapeutic treatments (1992–1995 and 1994–998) on cognitive functioning in a population of patients with nonmetastatic breast carcinoma. In each investigation, patients received a comprehensive neuropsychologic evaluation before the start of systemic chemotherapy or hormonal therapy.


Patients were recruited as part of three institutional review board–approved research protocols that were initiated collaboratively by the Neuropsychology Section and the Department of Breast Medical Oncology at the University of Texas M. D. Anderson Cancer Center. However, neuropsychologic outcome was a nonfunded endpoint in two of those trials. Patient enrollment and collection of data on this endpoint were terminated early due to funding issues; thus, the overall patient pool for the analysis of this endpoint was smaller than the patient pool that was available in the primary medical trials. All patients who met initial screening criteria and consented to participate in the trials were registered consecutively into a protocol for evaluation. All patients had a diagnosis of primary breast carcinoma and no evidence of metastatic disease, were age ≥ 18 years, had completed ≥ 8 years of formal education, and spoke fluent English. No patient had a history of other primary malignancy or a previous or current neurologic or psychiatric disorder, and no patient used substances that were active in the CNS and were believed to affect cognition (e.g., narcotics, antiemetics, or steroids) ≤ 1 week before testing. Ninety-seven patients met the initial screening criteria for participation. Eighty-four patients with a diagnosis of Stage I breast carcinoma (T1N0M0; n = 32), Stage II breast carcinoma (T2N0M0; n = 41), or Stage IIIA breast carcinoma (T2N2M0; n = 11) qualified for study entry and underwent neuropsychologic evaluation before beginning chemotherapy. Thirteen patients did not meet the inclusion criteria and were not part of the subsequent analyses. Of these, seven patients stated that they did not have time to participate, two patients had educational limitations that precluded valid assessment, two patients had histories of major depression, one patient had a history of substance abuse, and one patient had a history of previous breast carcinoma.

Of the 84 patients who were included in the analyses, 50% (n = 42) had undergone surgical resection for primary breast carcinoma, and 50% (n = 42) had undergone core-needle biopsy before the neuropsychologic evaluation in association with their participation in a trial that involved neoadjuvant chemotherapy after biopsy for pathologic verification. Approximately 12% of all patients (n = 10) had received adjuvant radiotherapy to the chest wall with or without the inclusion of regional lymphatic chains before neuropsychologic evaluation. The mean (± standard deviation [SD]) patient age was 50.4 years (± 9.1 years). On average, patients had completed 14 years of education (± 2.6). Ethnically, 69 patients (82%) were Caucasian, 8 (10%) were Hispanic, 6 (7%) were African American, and 1 (1%) was Asian. Data on menopausal status (based on review of patients' medical records) were available for all patients: approximately 42% (n = 35) were naturally postmenopausal, 15% (n = 13) had a history of surgically induced menopause, 2% (n = 2) were perimenopausal, and 40% (n = 34) were premenopausal at the time of neuropsychologic evaluation. Data on prior use of hormone replacement therapy (HRT) were available for 67 of 84 patients: Thirty patients (45%) had previously used HRT, whereas 37 (55%) had never used HRT. The median time between neuropsychologic evaluation and last use of HRT was 52 days (range, 4 days to 23.9 years).

Due to the nature of the research studies and changes in clinical practice, patients received a variety of cognitive tests. To facilitate interpretation, these tests were grouped into domains based on previous neuropsychologic research 14 (Table 1). All patients underwent an assessment that included 5–14 different measures, which required approximately 40–120 minutes to be completed.

Table 1. Neuropsychologic Tests Grouped by Principle Cognitive Domain
Domain/test nameaTest abbreviation
  • WAIS-R/III: Wechsler Adult Intelligence Scale—Revised or Third Edition; WMS-III: Wechsler Memory Scale—Third Edition; HVLT: Hopkins Verbal Learning Test; VSRT: Verbal Selective Reminding Test; NVSRT: Nonverbal Selective Reminding Test; ROCFT: Rey–Osterreith Complex Figure Test; MAE: Multilingual Aphasia Examination; MMPI: Minnesota Multiphasic Personality Inventory.

  • a

    Superscript numerals refer to the list of references.

 WAIS-R 15/WAIS-III 16 Digit SpanWAIS-R/III DSpan
 WAIS-R Digit SymbolWAIS-R DSymbol
 WAIS-R ArithmeticWAIS-R Arith
 WAIS-III Letter-Number SequencingWAIS-III LN
 WMS-III 17 Mental ControlWMS-III MC
 Trail Making Test 18 Part ATMTA
 HVLT 19 Trials 1–3HVLT T1–T3
 HVLT Recognition DiscriminabilityHVLT DISCRIM
 VSRT 20 Long-Term StorageVSRT LTS
 VSRT Delayed RecallVSRT DR
 NVSRT 21 Long-Term StorageNVSRT LTS
 NVSRT Delayed RecallNVSRT DR
 ROCFT 22 Delayed RecallROCFT DR
 MAE 23 Controlled Oral Word AssociationCOWA
 Boston Naming Test 24BNT
 MAE Sequential CommandsMAE SC
 Trail Making Test Part BTMTB
 Category Test 25CT
 WAIS-III SimilaritiesWAIS-III Sim
 WAIS-R Block DesignWAIS-R BD
 Judgment of Line Orientation 26JLO
 Grip Strength (dominant hand) 27GRIP (dominant)
 Grooved Pegboard (dominant hand) 28PEG (dominant)
 Beck Depression Inventory 29BDI
 Beck Depression Inventory, Second Edition 30BDI-2
 MMPI (Depression scale) 31MMPI (Scale 2)
 State-Trait Anxiety Inventory—State score 32STAIS
 Beck Anxiety Inventory 33BAI
 MMPI (Psychasthenia scale)MMPI (Scale 7)

Patients' raw cognitive test score performances were converted to standardized scores (z scores: mean ± SD, 0 ± 1) based on published normative data to facilitate comparisons between measures. An Overall Cognitive Function Index (OCFI) for each patient's test performance was calculated. Impaired OCFI (OCFI-I) was ascertained using a 2-step criterion: 1) if a patient had multiple test performance z scores ≤ −1.5, then she was considered to have impaired functioning (OCFI-I); 2) if a patient had only 1 test that met the aforementioned criteria (i.e., z score ≤ −1.5), then that single z score had to be ≤ −2.0 for the patient to be classified as having impaired functioning. This two-step approach was used to minimize the number of false-positive errors due to multiple tests and to determine the frequency of impairment rather than the frequency of low performance levels. In a normal, healthy population, approximately 7% and 2% of the population would have scores < −1.5 and < −2.0, respectively, by chance alone. If a patient did not exhibit impaired performance, then the OCFI classification was not impaired (OCFI-NI).

For each cognitive measure, a binomial test was conducted to determine whether the frequency of impairment observed in the study group differed from normative expectations. To correct for multiple comparisons, α, the cutoff value for determining statistical significance, was set at 0.01. The overall frequency of cognitive impairment based on the OCFI and the frequency of impairment within each domain were then calculated.

Self-reports of anxious and depressive symptoms warranted a classification of clinically significant distress–impaired (CSD-I) if the reports exceeded the limits recommended by the measures' manuals. Specifically, patients with raw scores ≥ 18 on the Beck Depression Inventory (BDI), the BDI Second Edition (BDI-2), and the Beck Anxiety Inventory (BAI) received a classification of CSD-I. Similarly, patients who had scores ≥ the 95th percentile on the ‘State’ subscale of the State-Trait Anxiety Inventory (STAIS) received a classification of CSD-I. Patients with scores lower than these specified values received a classification of not clinically significantly distressed (CSD-NI). All patients' Minnesota Multiphasic Personality Inventory (MMPI) profiles yielded adequate validity scale indices to be considered for further clinical scale interpretation. 34 Scale 2 of the MMPI measures depression through true/false statements regarding affective symptoms (e.g., poor morale), cognitive symptoms (e.g., hopelessness), and physical symptoms (e.g., sleep disturbances). Scale 7 of the MMPI measures anxiety across the domains of neuroticism, anxiety, withdrawal, poor concentration, agitation, psychotic tendencies, and poor physical health. Patients with T scores ≥ 70 on Scale 2 or Scale 7 of the MMPI received a classification of CSD-I; patients with scores below this cutoff received a classification of CSD-NI.

Pearson correlations between mood measures (i.e., BDI, BDI-2, and BAI raw scores; STAIS percentiles; and MMPI Scale 2 and Scale 7 T scores) and standard scores for marker variables from each domain were analyzed to decrease the number of statistical tests performed. Cognitive tests that appeared to be most sensitive were selected as marker variables (i.e., the Trailmaking Test Part A [TMTA]; the Verbal Selective Reminding Test, Long-Term Storage [VSRT LTS]; the VSRT Delayed Recall [DR]; the Nonverbal Selective Reminding Test [NVSRT] DR; the Boston Naming Test [BNT]; the Trailmaking Test Part B [TMTB]; the Rey–Osterreith Complex Figure Test [ROCFT] Copy; and the Grooved Pegboard Test [dominant hand]). Given the large number of statistical tests performed, α, the cutoff value for determining statistical significance, was set at 0.01.


Using the classification criteria described above, approximately 35% (29 of 84 patients; binomial test: P < 0.001) received a classification of OCFI-I before the initiation of adjuvant chemotherapy. Of the patients who had impaired functioning, approximately 31% (9 of 29) exhibited impairment on 1 test, whereas approximately 69% (20 of 29) exhibited impairment on 2 or more tests. The frequency of cognitive impairment within each domain varied as a function of the test that was used to make the measurement (Table 2). Women with breast carcinoma exhibited impaired verbal learning (VSRT LTS; Binomial test: P < 0.01) and verbal memory (VSRT DR; Binomial test: P < 0.01) significantly more frequently relative to normative expectations. In addition, 1 measure of psychomotor processing speed and attention approached significance (TMTA; P = 0.017). Several other measures of cognitive functioning (NVSRT DR, BNT, ROCFT Copy, and PEG [dominant hand]) that revealed levels of impairment similar to those revealed by the TMTA failed to approach significance, probably due to small sample sizes.

Table 2. Mean Scores and Impairment Frequencies Before Chemotherapy on Tests Grouped by Principle Cognitive Domain
Cognitive domainNo. of patients testedMean (SD)Impaired (%)
  • SD: standard deviation; WAIS-R/III: Wechsler Adult Intelligence Scale—Revised or Third Edition; DSpan: Digit Span; DSymbol: Digit Symbol; Arith: Arithmetic; LN: Letter-Number Sequencing; WMS-III MC: Wechsler Memory Scale—Third Edition, Mental Control; TMTA: Trail Making Test, Part A; HVLT T1–T3: Hopkins Verbal Learning Test (LVLT) Trials 1–3; VSRT: Verbal Selective Reminding Test; LTS: Long-Term Storage; NVSRT: Nonverbal Selective Reminding Test; DISCRIM: Recognition Discriminability; DR: Delayed Recall; ROCFT: Rey–Osterreith Complex Figure Test; COWA: Multilingual Aphasia Examination (MAE) Controlled Oral Word Association; BNT: Boston Naming Test; SC Sequential Commands; TMTB: Trail Making Test, Part B; CT: Category Test; Sim: Similarities; JLO: Judgment of Lone Orientation; BD: block design; GRIP (dominant): Grip Strength (dominant hand); PEG (dominant): Grooved Pegboard (dominant hand).

  • a

    Scaled scores: mean, 10; standard deviation, 3.

  • b

    z scores: mean, 0; standard deviation, 1.

  • c

    Binomial test (P < 0.01).

  • d

    Binomial test (P < 0.001).

 WAIS-R/III DSpana8410.0 (3.0)5
 WAIS-R DSymbola6011.7 (2.5)2
 WAIS-R Aritha1810.2 (2.6)6
 WAIS-III LNa2411.7 (3.3)4
 WAIS-III MCa2411.2 (2.6)0
 TMTAb84−0.11 (1.37)13
 HVLT T1–T3b42−0.09 (0.90)10
 VSRT LTSb40−0.27 (1.20)18c
 NVSRT LTSb180.11 (0.83)6
 HVLT DISCRIMb420.29 (0.71)2
 VSRT DRb40−0.54 (1.80)25d
 NVSRT DRb180.00 (1.04)17
 ROCFT DRb240.13 (0.82)4
  COWAb84 6
  BNTb24 13
  MAE SCb24 0
Executive function   
 TMTBb840.13 (1.90)6
 CTb410.22 (0.91)0
 WAIS-III Sima4212.1 (3.4)5
  JLOb240.91 (0.38)0
  WAIS-R BDa1811.4 (2.9)0
  ROCFT Copyb240.31 (1.34)13
 GRIP (dominant)b180.11 (0.98)0
 PEG (dominant)b420.02 (1.14)12

In the current cohort, 26% of patients (20 of 78) reported symptoms of anxiety and/or depression that met the criteria for a classification of CSD-I. Chi-square analysis demonstrated that patients who were classified as having reported significant affective distress (i.e., CSD-I) were significantly more likely to be cognitively impaired (i.e., OCFI-I; chi-square = 9.90; P ≤ 0.005). Among patients with a classification of OCFI-I who completed mood assessment (n = 28), < 50% (13 of 28) received a classification of CSD-I. Scores on the self-report measures of affective distress for patients with a classification of CSD-I were as follows: BDI raw score range, 24–28; BDI-II raw score, 20; BAI raw score range, 19–29; STAIS percentile range, < 1 to 6; MMPI Scale 2 T score range, 71–80; and MMPI Scale 7 T score, 74. Overall, these scores are consistent with mild-to-moderate ranges of affective distress, and no patient received a psychiatric diagnosis of major depression or an anxiety disorder. Analysis of Pearson correlations between domain-specific cognitive marker variables and mood measures revealed that STAIS scores were significantly correlated with TMTB scores (Table 3). No other significant associations between mood and cognitive function were detected. Differences in the frequency of distress within the domains of mood function also varied, depending on the measure used (Table 4), with the MMPI appearing most sensitive to depression and roughly equal to the STAIS in terms of sensitivity to symptoms of anxiety.

Table 3. Coefficients for Correlations between Mood Measures and Cognitive Marker Variables
Cognitive testMood measurea
  • BDI: Beck Depression Inventory; BDI-2: Beck Depression Inventory—Second Edition; BAI: Beck Anxiety Inventory; STAIS: State-Trait Anxiety Inventory—State scale; MMPI: Minnesota Multiphasic Personality Inventory; TMTA: Trail Making Test, Part A; VSRT: Verbal Selective Reminding Test; LTS: Long-Term Storage; DR: Delayed Recall; NVSRT: Nonverbal Selective Reminding Test; BNT: Boston Naming Test; TMTB: Trail Making Test, Part B; ROCFT: Rey–Osterreith Complex Figure Test; PEG: Grooved Pegboard Test; NA: not administered (due to variation between protocols, neither measure was administered).

  • a

    Metric: raw score for BDI, BDI-2, and BAI; percentile for STAIS; T score for MMPI Scale 2 and Scale 7; and z score for cognitive tests.

  • b

    Pearson correlation: P < 0.01.

 ROCFT CopyNA−0.32−0.30NANANA
Table 4. Frequency of Clinically Significant Before Chemotherapy Distress on Self-Report Measures by Domain
Psychologic domainNo. of patients assessedCSD-I (%)
  1. CSD-I: clinically significant distress; MMPI: Minnesota Multiphasic Personality Inventory; BDI-2: Beck Depression Inventory—Second Edition; BDI: Beck Depression Inventory; STAIS: State-Trait Anxiety Inventory—State scale; BAI: Beck Anxiety Inventory.

 MMPI (Scale 2)1735
 MMPI (Scale 7)1735

Exploratory analyses were conducted to identify demographic, disease-related, and treatment-related factors that may be associated with the presence of cognitive impairment (i.e., OCFI-I) before chemotherapy. An adjusted α level of 0.01 was chosen for the assessment of statistical significance, given our desire to correct for multiple comparisons and minimize the possibility of Type I error. Association between OCFI and age and education were explored using independent-sample t tests. Association of disease-related and treatment-related variables with OCFI were examined using chi-square analyses. All disease- and treatment-related variables were dichotomized. For disease stage, women were divided into 2 groups: women with Stage I or II disease (n = 74) and women with Stage IIIA disease (n = 10). Women also were divided into 2 groups on the basis of extent of surgery: women who underwent surgical resection (n = 41) and women who underwent biopsy only (n = 43). Women who previously had received chest wall irradiation (n = 10) were distinguished from women who had never received irradiation (n = 74). Women who were postmenopausal either naturally or due to surgery were grouped together (n = 48) and were compared with women who were either premenopausal or perimenopausal (n = 36). Postmenopausal women with available data on HRT use were classified as HRT users (n = 30) if they had any history of HRT use or as HRT nonusers (n = 18) if they had never used HRT.

Patients in the OCFI-I and OCFI-NI groups did not differ significantly with regard to age (t[82] = 2.02; P = 0.05; mean ± SD: OCFI-I, 53.1 ± 9.8 years; OCFI-NI, 48.9 ± 8.4 years) or education (t[82] = 1.92; P = 0.06; mean ± SD: OCFI-I, 13.2 ± 2.7 years; OCFI-NI, 14.4 ± 2.4 years). Women who were postmenopausal (Fig. 1), had no history of HRT use (Fig. 2), or had undergone lumpectomy/mastectomy (Fig. 3) were nearly twice as likely to be cognitively impaired compared with women who did not have such characteristics; however, for none of these differences was P ≤ 0.01. Women with a classification of OCFI-I did not differ significantly from women with a classification of OCFI-NI in terms of any other demographic, disease-related, or treatment-related variable (4).

Figure 1.

Overall Cognitive Function Index (OCFI) in relation to menopausal status. Postmeno: postmenopausal; Premeno: premenopausal; OCFI-I: OCFI-impaired; OCFI-NI: OCFI–not impaired. Pearson χ2(1, N = 84) = 4.22; P = 0.04.

Figure 2.

Overall Cognitive Function Index (OCFI) in relation to hormone replacement therapy (HRT) status. HRT nonusers had no history of HRT use, whereas HRT users did have a history of HRT use. OCFI-I: OCFI-impaired; OCFI-NI: OCFI–not impaired. Pearson χ2(1, N = 48) = 2.29; P = 0.13.

Figure 3.

Overall Cognitive Function Index (OCFI) in relation to extent of surgery. Lump/mast: lumpectomy or mastectomy; OCFI-I: OCFI-impaired; OCFI-NI: OCFI–not impaired. Pearson χ2(1, N = 84) = 4.95; P = 0.03.

Figure 4.

Overall Cognitive Function Index (OCFI) in relation to affective status. CSD-I: clinically significant distress–impaired; CSD-NI: clinically significant distress–not impaired; OCFI-I: OCFI-impaired; OCFI-NI: OCFI–not impaired. Pearson χ2(1, N = 78) = 9.90; P = 0.002.


Adjuvant chemotherapy has been reported to be associated with the development of both early and delayed cognitive dysfunction in a population of women with breast carcinoma. However, methodologic limitations in all studies on this topic that have been published to date make these conclusions controversial. 35, 36 Perhaps most significantly, previous studies failed to include a pretreatment neuropsychologic evaluation of cognitive functioning against which postchemotherapy changes in performance could be compared. A longitudinal design of this nature is necessary to clearly demonstrate the relation between adjuvant chemotherapy and cognitive functioning.

In the current study, we found that 36% of women with breast carcinoma demonstrated impaired overall cognitive function before the initiation of adjuvant chemotherapy. Cognitive impairment in the domains of verbal learning and memory was observed significantly more frequently in the study cohort than in the comparison group of normal control patients. In addition, differences in psychomotor processing speed and attention, nonverbal memory, confrontational naming, complex visuoconstruction, and fine motor dexterity approached significance. In five previously published cross-sectional studies that assessed patients' cognitive status after chemotherapy, cognitive dysfunction was defined inconsistently and often far less rigorously than in the current investigation, with these definitions sometimes including performance scores that were within the ranges for normal control patients. The finding that many patients who were treated with chemotherapy had performance scores that fell within normal ranges makes attributions of neurotoxicity in these patients particularly dubious. Across these studies, the frequency of cognitive dysfunction ranged from 17% to 75%. Given the current documentation of objective cognitive impairment before adjuvant systemic therapy, a large proportion of patients in previously published reports who performed at levels below what was expected when they were assessed after chemotherapy may well have performed at that same level before chemotherapy. Those previous studies, therefore, may have overestimated the true incidence of decreases in cognitive functioning secondary to chemotherapy.

The assessment of cognitive impairment in the current study involved comparison of patients' scores on measures of cognitive skills with published normative data. We propose that test performance scores for a group of individuals without breast carcinoma constituted a valid control for the purposes of the study. Women diagnosed with nonmetastatic breast carcinoma are not commonly believed to experience neurologic dysfunction. Thus, their performance is expected to be similar to that of individuals with similar demographic backgrounds (e.g., in terms of age and education). The use of normative data means that study patients are being compared with just such a control group. Normative data often have the advantage of representing much larger populations than the control populations recruited by individual investigators on a study-by-study basis. This leads to two important design features: 1) stratification according to relevant demographic variables (e.g., age and education) can be performed, so that the study group is compared with appropriately matched control patients; and 2) the larger cohort associated with the normative data provides a more stable estimate of the mean and SD in the population from which the representative sample was drawn, and thus a more accurate and valid estimate of the true score for the control group with which the study cohort is compared.

This approach has been criticized, because it fails to control for potential differences in emotional distress that may accompany being diagnosed with a medical illness. Thus, a similarly affected control group often is desirable to control for this potentially confounding influence on cognitive function. However, because the current cohort was untreated, it was not possible to recruit another breast carcinoma cohort that would control for issues of emotional distress while not also controlling for our variable of interest, cognitive function in women with breast carcinoma before treatment. The use of another medically affected population also is problematic, because there is evidence that the presence of disease without direct CNS involvement may nonetheless have effects on the CNS in some instances. For example, newly diagnosed (i.e., untreated) patients with small cell lung carcinoma who had no evidence of CNS involvement exhibited cognitive dysfunction before the initiation of any therapy, 37 further underscoring the importance of obtaining pretreatment measures of cognitive function. Thus, the use of another population that is affected by disease may also obscure the very phenomenon that is of interest to the investigator. Due to these considerations, we chose to use relevant normative data as our control and supplemented our assessment with measures of emotional well-being to investigate the correlation between emotional distress and cognitive function.

Methodologic limitations notwithstanding, a consistent pattern of cognitive dysfunction attributable to chemotherapy has not emerged from previously published studies. The involved domains of cognitive function have included learning and memory, processing speed, attention, and visuoperception. The variability in findings of cognitive dysfunction may be attributable in part to the choice of measures used in each study and the differential sensitivity of each measure, as suggested in the current investigation. In general, researchers interested in studying perturbations of the underlying neural networks that are responsible for various cognitive functions may wish to employ the most sensitive measure for each specific cognitive domain. The effects of a treatment (or any neurologic condition), as demonstrated previously, 38 may be observable only when a cognitive system is stressed sufficiently so that it unmasks the dysfunction and overwhelms potential compensatory mechanisms. Relatively insensitive tests, such as cognitive screening tools (e.g., the Mini-Mental State Examination 39), tend to underestimate the nature and extent of cognitive dysfunction, often lack specificity for any particular cognitive domain, and may fail to capture subtle but meaningful changes in a patient's cognitive status. 40

We also attempted to discern factors related to the occurrence of cognitive impairment in patients with breast carcinoma. Although correlations between cognitive function before chemotherapy and demographic, disease-related, and clinical characteristics did not exceed our adjusted α value for determining statistical significance, several provocative trends emerged. Patients who underwent more invasive surgery (lumpectomy or mastectomy), patients who were postmenopausal, and patients who had not previously used any HRT appeared to have a greater risk of presenting with cognitive impairment.

Debate continues about the adverse effects of cardiac and noncardiac surgery on patients' postacute cognitive function, with some investigations demonstrating declines in postoperative cognitive function 41, 42 and others reporting no significant decline approximately 1 month after surgery. 43 All women in the current study who underwent lumpectomy or mastectomy subsequently underwent neuropsychologic evaluation an average of 53 days (SD, 39.8 days) after surgery. Follow-up analyses in the current sample did not detect significant differences between patients who exhibited cognitive impairment (OCFI-I; mean ± SD, 63.8 ± 50.3 days) and patients with a classification of cognitively not impaired (OCFI-NI; mean ± SD, 44.4 ± 25.8 days) in terms of the length of time between surgery and neuropsychologic evaluation. Thus, there is no evidence that women who were assessed more acutely postoperatively were more likely to exhibit cognitive impairment.

Alterations in women's hormonal milieu reportedly have been associated with changes in cognitive function. Although controversy remains regarding the cognitive effects of HRT, 44 there is evidence that surgical or chemical induction of menopause is associated with impairments in cognitive function and mood. 45, 46 The presence of affective distress was associated with cognitive impairment in the current investigation; however, it was not causally related to cognitive impairment. Thus, depression, anxiety, and cognitive dysfunction appear to be separable, albeit comorbid, conditions in this population.

We recognize that a short screening measure that can be administered and interpreted by a variety of health care professionals with minimal training in neuropsychology would be highly desirable. Unfortunately, the sensitivity of these measures (e.g., the Mini-Mental State Examination) in detecting cognitive disturbances that are not severe enough to be considered dementia is poor. 40 Currently, clinicians should continue discussing cognitive functioning with their patients and should refer patients for neuropsychologic evaluation if they or their family members report changes in their ability to think. In this way, a comprehensive evaluation may be undertaken that can yield differential diagnostic information and provide treatment recommendations.

To our knowledge, the current report is the first to document the presence of cognitive impairment in a cohort of patients with breast carcinoma without CNS involvement before the initiation of systemic adjuvant therapy. Although the current study does not afford a comprehensive examination of the many potential etiologic agents responsible for this dysfunction, it is speculated that both host-related factors and disease-related factors may be involved. Research is actively pursuing host-related, or soil, characteristics, such as genetic polymorphisms, immune reactivity, nutritional factors, hormonal histories, or lack of cognitive reserve, that are associated with an increased likelihood of developing cognitive dysfunction either before or in concurrence with the administration of adjuvant therapy. In addition, disease-related, or seed, factors, including tumor gene mutations, induction of proinflammatory cytokines, and paraneoplastic disorders, as well as interactions between disease-related factors and host-related factors, may contribute to the development of cognitive dysfunction. Interdisciplinary investigations of both the behavioral expression of perturbations within the neural networks responsible for cognition and the potential mechanistic contributors to the observed neurotoxicity will assist in the development of rational, targeted therapeutic options (e.g., cytokine antagonists or neuroprotectants) in the future.

There appears to be justification for concern regarding the potential neurotoxicities associated with systemic antitumor agents used therapeutically to treat and cure breast carcinoma. However, our current understanding of the cognitive and neurobehavioral effects of these treatments is extremely limited. This issue is of significant clinical importance given the prevalence of breast carcinoma; the increased use of chemotherapy as adjuvant therapy; the increasing use of more aggressive dosing schedules; the increasing survival rates being achieved; and patients' natural desire to return to their normal occupational, academic, and social pursuits. Thus, it is imperative that future investigations use well designed longitudinal methodologies that will assist in defining the relative risks and benefits inherent in the available treatments so that truly informed treatment decisions can be made. The results of our own prospective longitudinal trials will be available soon.


The authors thank Gabriel N. Hortobagyi, M.D. (Nellie B. Connally Chair in Breast Cancer, Department of Breast Medical Oncology, The University of Texas M. D. Anderson Cancer Center), for the participation of his department in ongoing collaborations; and Lori Smythe (Research Coordinator, Department of Neuro-Oncology, The University of Texas M. D. Anderson Cancer Center) and Carol Ann Long, Ph.D. (Director, Medical Affairs Oncology, Amgen, Inc.), for their technical assistance.