Cognitive functions in survivors of primary central nervous system lymphoma

Subjects.

Sixty-two patients who completed treatment for PCNSL were identified from a cohort of long-term survivors treated in the Department of Neurology at Memorial Sloan-Kettering Cancer Center between 1985 and 2002. Patients were contacted either by letter or by their treating physician. Among these patients, 10 were deceased, 15 were lost to follow-up, 4 were not eligible due to disease progression, and 5 refused to participate in the study. Twenty-eight patients diagnosed with PCNSL and treated with WBRT or chemotherapy or both who were in remission of their disease as defined by negative MRI or CSF cytology or both, had no psychiatric or other neurologic disorders, and were fluent in English participated in the study. The research protocol was approved by the Institutional Review Board of Memorial Sloan-Kettering Cancer Center, and informed consent was obtained from all participants.

All patients received a post-treatment baseline neuropsychological evaluation, and 14 patients were seen for an 8-month follow-up cognitive assessment to determine if there was a decline in their neuropsychological functioning. Fourteen patients did not return to complete the follow-up cognitive evaluation: 6 patients relapsed after the initial evaluation, 3 patients deteriorated and required nursing home placement, 3 patients were lost to follow-up due to relocation, and 2 patients refused to return for another testing session. Among the three patients who deteriorated, two had received combined modality therapy, one had chemotherapy only, and none showed evidence of tumor progression; ages at the time of the baseline evaluation were 53, 65, and 73 years.

Measures.

A battery of standardized neuropsychological tests was used to evaluate attention (Digit Span of the Wechsler Memory Scale-III [DF/DB],22 Brief Test of Attention [BTA]23), executive (Trail Making Test Parts A and B [TMTA, TMTB],24 Stroop Color-Word Test-Interference [SCWTI],25 Phonemic Verbal Fluency Test [VF]26), psychomotor (Grooved Pegboard Test, Dominant and Nondominant [GPTD, ND]27), and learning and memory abilities (Hopkins Verbal Learning Test-Revised, Learning [HVLTL], Delayed Recall [HVLTD], and Discrimination Index [HVLTDI]28). Tests of language (Boston Naming Test [BNT],29 Category Fluency Test [AF]26) and visual-construction (Clock Drawing Test, CDTO-Command and CDTY-Copy30) skills were administered. Screening cognitive and performance tests such as the Mini-Mental State Examination (MMSE)31 and the Karnofsky Performance Scale (KPS) were obtained for all patients. Mood and quality of life were assessed using the Beck Depression Inventory (BDI)32 and the Functional Assessment of Cancer Therapy-Brain (FACT-BR Version 4).33 The full battery was administered in a single 90-minute session either by a neuropsychologist (D.D.C.) or by a trained psychometrist. The same test battery was used at follow-up, and alternate test forms were used when available.

Raw cognitive test scores were compared with published normative values according to age, and when available, to age and education, and subsequently converted into z-scores. Moderate impairment was defined as a z-score equal to or less than 2 standard deviations below the mean of the normative sample; mild impairment was defined as a z-score between 1.4 and 1.9 standard deviations below the normative sample. Given the large number of cognitive tests and the small number of patients, composite scores were calculated by adding the z-scores for each test within a standard cognitive domain and dividing the sum by the number of tests. Five cognitive domains were calculated: attention/executive (ZDF, ZDB, ZBTA, ZTMTA, ZTMTB, ZSCWTI, ZFAS), memory (ZHVLTL, ZHVLTD, ZHVLTDI), psychomotor (ZGPTD, ND), language (ZBNT, ZAF), and visual-construction (ZCDTY, ZCDTO).

Neuroimaging.

The extent of white matter disease on MRI scans of the brain performed within a maximum of 3 months of each cognitive evaluation was rated by the same neurologist (L.E.A.) who was blind to the cognitive test results. Radiographic endpoints were measured according to the classification of Fazekas et al.34 and included no white matter change (grade 0), minimal patchy white matter foci (grade 1), start of confluence of white matter disease (grade 2), large confluent areas (grade 3), confluence with cortical and subcortical involvement (grade 4), leukoencephalopathy (grade 5), and possible radiation necrosis (grade 6).

Statistical analyses.

Internal consistency reliability among subtests that comprised each cognitive domain was assessed using Cronbach’s alpha and Spearman correlation. Imputation was used for a small number of missing values on cognitive tests, and was based on linear regression. Within-subjects multivariate analysis of variance (ANOVA) followed by post hoc comparisons were performed to assess whether the composite mean z-scores and other measures differed among patients at post-treatment baseline, and whether subtests within each cognitive domain differed from each other. The relation between specific demographic and clinical variables and neuropsychological composite test scores at post-treatment baseline was assessed using Spearman correlation, t-tests, ANOVA followed by post hoc comparisons when appropriate, and nonparametric tests. Paired t-tests were performed to compare cognitive test performance between post-treatment baseline (time 1) and follow-up (time 2) evaluations. Kruskal-Wallis test with Monte Carlo estimates for the exact test was used for group comparisons that involved very small sample sizes and skewed distributions.

Disease and treatment history.

Table 2 presents disease and treatment history of patients who completed baseline (time 1) and follow-up (time 2) evaluations; treatment history reflects all therapy received including treatment at relapse, if applicable.

Sixteen patients received combined modality treatment (WBRT + chemotherapy) at initial diagnosis. Among these, 14 patients received both pre- and post-WBRT chemotherapy. Pre-WBRT chemotherapy was comprised of a high-dose MTX regimen in all cases; 10 patients received methotrexate, procarbazine, and vincristine, whereas 4 were treated with methotrexate alone or in combination with other agents. Post-WBRT chemotherapy was comprised of high-dose cytarabine (ARA-C), either alone (n = 12) or in combination with other agents (n = 2). Among the 16 patients treated with combined modality therapy at initial diagnosis, 2 patients received only post-WBRT ARA-C chemotherapy.

Two patients were treated with WBRT at initial diagnosis, and one of these patients received chemotherapy at relapse. Among the 10 patients treated with chemotherapy alone, all received a high-dose MTX based regimen and 8 also received ARA-C. Four patients had a history of disease relapse prior to study enrollment, and were treated with either methotrexate-based chemotherapy (n = 3) or WBRT and nitrosourea (CCNU).

Baseline analyses.

Baseline neuropsychological test performance revealed composite z-scores in the moderately impaired range on cognitive domains involving memory and psychomotor speed, and in the mildly impaired range in the language domain (figure). Performance on the attention/executive domain was in the low average range, and visual-construction skills were within the average range. Table 3 shows mean z-scores for the five cognitive domains and individual tests for all subjects. Internal consistency reliability estimates (Cronbach’s alpha) were obtained for the z-scores within cognitive domains that included more than two tests, and Spearman correlation was performed for domains including two tests. Cronbach’s alpha values were as follows: attention/executive = 0.89; memory = 0.89. Correlations between the two subtests that comprised each of the other domains were as follows: language = 0.29; psychomotor = −0.63; visual-construction = 0.83.

• Download figure
• Open in new tab
• Download powerpoint

Figure. Mean domain z-scores at baseline by treatment groups. WBRT = whole brain radiotherapy. *p < 0.004. Error bars = 1 SE.

The results of repeated measures ANOVA showed that the mean z-scores of the seven subtests in the attention/executive domain were significantly different [Wilk’s lambda = 0.291, F(6,16) = 6.5, p < 0.0013]. Post hoc pairwise comparisons with Bonferroni adjustment (paired t-test) showed that z-scores on the TMTB and BTA were significantly (p < 0.05) lower than the z-scores on DF and DB. The mean z-scores of the three subtests that comprised the memory domain were not significantly different [Wilk’s lambda = 0.828, F(2,25) = 2.58, p = 0.096]; patients performed in the mild to moderately impaired range on learning and recall of the word list, delayed recall, and discrimination on recognition testing. Patients performed in the moderately impaired range on the two subtests that comprised the psychomotor and language domains, and there were no significant differences between the two subtests on each domain. Performance on the two subtests of the visual-construction domain was in the average range; patients obtained significantly higher scores on the copy trial than they did on the command trial (t[25] = 5.18, p < 0.0001). Tests’ mean z-scores are presented in table 3.

Quality of life and occupational status at baseline.

Mean scores on the FACT-BR were comparable to those of a large group of primary brain tumor patients33; mean z-scores were as follows: total = −0.06, social well-being subscale = 0.06, functional well-being subscale = −0.32, physical well-being subscale = 0.19, emotional well-being subscale = 0.80, additional concerns subscale = 0.27. BDI mean score revealed that patients were not experiencing depressed mood (table 4). Comparisons between test performance of patients who were working (n = 9) and patients who were not working due to their illness (n = 9) at the time of the baseline evaluation revealed that patients who were employed had significantly higher scores on the attention/executive domain (t[16] = 2.32, p < 0.04) and FACT-BR (t[16] = 2.28, p < 0.05) and lower scores on the BDI (t[16] = −2.25, p < 0.04) than did patients who were not working. There were no significant correlations between the FACT-BR and clinical variables.

Lesion number/white matter disease.

Cognitive test performance differed significantly on the following domains according to number of lesions: attention/executive [F(2,25) = 4.77, p < 0.018], language [F(2,23) = 23.02, p < 0.0001], and visual-construction [F(2,23) = 8.83, p < 0.002]. Post hoc comparisons with corrections for multiple comparisons (Student Newman-Keuls test) showed that the performance of patients with three or more lesions was significantly (p < 0.05) more impaired on these three domains than that of patients with one or two lesions.

The results of Spearman’s correlations with Bonferroni adjustment for multiple comparisons showed that more extensive white matter changes on MRI at baseline were significantly associated with greater impairment on the attention/executive (r = −0.42, p < 0.03), memory (r = −0.52, p < 0.004), and language (r = −0.42, p < 0.03) domains, and on the MMSE (r = −0.61, p < 0.0007). MRI white matter ratings at baseline and follow-up are reported in table 5; there was no significant increase in white matter abnormalities within an 8-month period. Baseline MMSE and KPS scores were only mildly reduced, in comparison to the degree of impairment shown on some of the cognitive domains (see table 4).

There were no significant differences on any of the baseline composite z-scores between the patients who returned for follow-up assessment (n = 14) and those who either relapsed (n = 6) or deteriorated (n = 3) after the baseline (i.e., within an 8-month period).

Comparisons by treatment modality.

Table 6 presents demographic and clinical characteristics of patients at baseline according to treatment modality received. Patients treated with WBRT ± chemotherapy were significantly younger than patients who received chemotherapy only both at diagnosis [t(26) = −5.68, p < 0.0001] and baseline evaluation [t(26) = −4.73, p < 0.0001]; this is consistent with current treatment practice in which most patients who are 60 years of age or older are treated with chemotherapy only. Patients treated with WBRT ± chemotherapy also had significantly longer time intervals between diagnosis and baseline evaluation [t(26) = 5.39, p < 0.0001] and between treatment completion and baseline evaluation [t(26) = 4.00, p < 0.0005] than did patients treated with chemotherapy alone; this is consistent with the relatively recent trend to eliminate WBRT in older patients. The calculation of time since completion of treatment was based on all therapy received, including treatment at relapse if applicable. There were no significant differences in education or estimated premorbid verbal IQ.

Patients who received chemotherapy alone had significantly higher scores than did patients treated with WBRT ± chemotherapy on the memory domain [t(26) = −3.21, p < 0.004]. A similar trend was detected on the attention/executive domain but the comparison did not reach significance (p = 0.06) (see table 3). In order to assess whether time since completion of treatment rather than type of treatment could account for the findings, patients who received WBRT ± chemotherapy were divided into two groups according to time since treatment completion (≤60 months, n = 9; >60 months, n = 9) and compared to patients treated with chemotherapy only (≤60 months). ANOVA results showed that performance on the memory domain differed by treatment type/interval (i.e., chemotherapy only: ≤60 months; combined: ≤60 months; combined: >60 months) [F(2,25) = 5.27, p < 0.013]. Post hoc contrasts with corrections for multiple comparisons (Student Newman-Keuls test) showed that the performance of patients who received WBRT ± chemotherapy did not differ on the memory domain according to time since completion of treatment (mean z-score >60 months = −2.59; ≤60 months = −3.08), but was significantly (p < 0.05) more impaired than the performance of patients treated with chemotherapy alone (mean z-score ≤60 months = −0.84).

In order to assess whether age differences between the two groups rather than type of treatment could account for the findings, further analyses were performed. Among the 15 patients who were 60 years of age or older, those treated with chemotherapy only (n = 10) displayed significantly less impairment in the memory domain (χ² = 6.0, df = 1, p < 0.015) than did older patients treated with WBRT ± chemotherapy (n = 5). Patients treated with chemotherapy only (all 60 years and older) were also significantly less impaired in the memory domain (χ² = 4.19, df = 1, p < 0.041) than patients younger than 60 years of age (n = 13) who received WBRT ± chemotherapy. In patients treated with WBRT ± chemotherapy, there were no significant differences in memory performance between younger patients (<60 years, n = 13) and older patients (≥60 years, n = 5). In a multivariate regression model, patients treated with chemotherapy only displayed significantly less impairment on the memory domain than did patients treated WBRT ± chemotherapy after adjusting for age, KPS, and time since completion of treatment (p < 0.003). No other comparisons (e.g., surgery, intrathecal chemotherapy, history of relapse, medications) reached statistical significance.

MRI ratings for extent of white matter abnormalities according to treatment modality are presented in table 4. Approximately 50% of patients who received WBRT ± chemotherapy showed extensive abnormalities, whereas 70% of patients treated with chemotherapy only had mild white matter changes on MRI. The two treatment groups differed significantly on the KPS [t(26) = −2.06, p < 0.05] and MMSE [t(26) = −2.37, p < 0.03], but not on the BDI or FACT-BR (see table 4).

Baseline and follow-up analyses.

Results of paired t-tests comparing composite z-scores at baseline (time 1) and follow-up (time 2) for those subjects who completed both visits (n = 14) showed no significant differences on any of the cognitive domains (see table 3). Scores on the KPS, MMSE, BDI, and FACT-BR total and subscale scores remained stable at follow-up with no significant differences between the two assessments (see table 4). Considering the small number of subjects, no statistical analysis was performed to assess possible differences between treatment groups on the cognitive domains at follow-up.