Neurocognitive Consequences of Risk-Adapted Therapy for Childhood Medulloblastoma
http://www.100md.com
▲還散笫雖悝◎
the Division of Behavioral Medicine, Division of Radiation Oncology, Department of Biostatistics, Division of Neuro-Oncology, Department of Hematology Oncology, St Jude Children's Research Hospital, Memphis TN
Texas Children's Cancer Center, Baylor College of Medicine, Houston, TX
Royal Children's Hospital, Melbourne, Australia
ABSTRACT
PURPOSE: This prospective, longitudinal study examined the effects of risk-adapted craniospinal irradiation (CSI) dose and the interactions of dose with age and time from diagnosis on intelligence quotient (IQ) and academic achievement (reading, spelling, and math) among patients treated for medulloblastoma (MB).
PATIENTS AND METHODS: Patients received serial neurocognitive testing spanning from 0 to 6.03 years after diagnosis (median, 3.14 years). The multi-institutional study included 111 patients, who were 3 to 20 years of age at diagnosis (median age, 7.4 years), treated for MB with risk-adapted CSI followed by four cycles of high-dose chemotherapy (cyclophosphamide, cisplatin, and vincristine) with stem-cell support. High-risk patients (HR; n = 37) received CSI to 36 to 39.6 Gy and conformal boost treatment of the primary site to 55.8 to 59.4 Gy. Average-risk patients (AR; n = 74) received CSI to 23.4 Gy and conformal boost treatment of the posterior fossa to 36.0 Gy and primary site to 55.8 Gy.
RESULTS: Multivariate modeling revealed statistically significant declines in mean IQ (每1.59 points/yr; P = .006), reading (每2.95 points/yr; P < .0001), spelling (每2.94 points/yr; P < .0001), and math (每1.87 points/yr; P = .003) scores for the entire group. The effects of risk-adapted radiation therapy on IQ, reading, and spelling were moderated by age, with the greatest rates of decline observed for the HR patients who were younger (< 7 years old) at diagnosis.
CONCLUSION: Young age at diagnosis was the most prominent risk factor for neurocognitive deficits among survivors of MB despite reductions in CSI dosing and efforts to limit the boost volume. Younger patients exhibited substantial problems with the development of reading skills.
INTRODUCTION
Children who survive treatment for medulloblastoma (MB), which includes postoperative craniospinal irradiation (CSI) with or without chemotherapy, may experience disturbances in essential neurocognitive functions that result in impaired overall intelligence and a high rate of academic failure. Deficits in processing speed, memory ability, and attention have been prominent among patients treated for MB, with both treatment and patient factors attributing to eventual outcomes.1-3 Patients who receive higher dose CSI and who are younger at diagnosis experience greater deficits in neurocognitive performance.4-7 To reduce neurocognitive morbidity, contemporary protocols have been designed to limit radiation dose using risk-adapted CSI dosing and limited posterior fossa or primary site irradiation using three-dimensional conformal radiation therapy. With this approach, early results have proven successful, with evidence of excellent event-free survival for standard-risk patients receiving reduced-dose CSI (23.4 Gy) and adjuvant chemotherapy.8
The impact of the age of the patient at treatment and the level of radiation dose exposure on cognitive performance has been the focus of examination in several studies. Four of these studies are especially relevant to increasing our understanding of the influence of these potential risk factors. A retrospective Pediatric Oncology Group study of average-risk (AR) patients found that patients who were randomly assigned to receive reduced-dose CSI (23.4 Gy) and patients who were older at time of treatment (> 8.8 years) showed higher cognitive functioning than patients who were randomly assigned to receive standard-dose CSI (36 Gy) and patients who were younger at time of treatment.9
However, in a subsequent prospective study, even the AR patients who had received reduced-dose CSI (23.4 Gy) and adjuvant chemotherapy were found to experience a decline of 4.3 intelligence quotient (IQ) points per year over a 3-year period from treatment.10 Also using a longitudinal design, Palmer et al11 found a mean decline of 2.6 IQ points/yr among 44 children who were treated with CSI (24.3 to 39.6 Gy). Older patients (> 8 years) and patients with lower dose CSI (< 35.2 Gy) showed higher cognitive functioning than patients who were younger or who received a higher CSI dose. The same group later reported in more detail the patterns of change in intellectual functioning over a period of 7 years from diagnosis for 50 patients who were treated with conventional-dose (35 to 40 Gy) CSI.12 The overall rate of decline in IQ was 2.2 points/yr. A delayed decline in IQ was observed for patients who were treated at an older age. In contrast, younger patients showed a more immediate decline in IQ. Female patients were also at greater risk for IQ declines.
In addition to confirming the prevailing beliefs about the increased risks associated with younger age and higher CSI dose, these longitudinal studies were able to generate predictive models quantifying expected IQ loss for particular subgroups of patients. However, the potential neurocognitive benefits of reduced-dose CSI in MB are ambiguous. No prospective study of MB patients based on a single protocol has compared risk-adapted radiation therapy dose outcomes. In addition, the limited number of observations and patients has never allowed an analysis of the potential interactions among two or more risk factors (eg, whether the rate of IQ loss among younger patients treated with reduced-dose CSI is different from the rate of IQ loss among younger patients treated with conventional-dose CSI).
The purpose of the present study was to test the hypothesis, as posed in the original protocol, that significant differences in changes in neurocognitive function would be observed between the AR patients treated with reduced-dose CSI (23.4 Gy) and high-risk (HR) patients treated with CSI of 36 to 39.6 Gy from shortly after diagnosis to 2 years after diagnosis. It was believed that the AR patients would demonstrate better performance than the HR patients. In the original design, age was not proposed as a covariate. To analyze the maximal number of observations on patients over more prolonged time intervals and to take advantage of more contemporary methods of longitudinal analysis using multiple covariates and their interactions (ie, age, risk group, and time), the analytic approach was expanded.
PATIENTS AND METHODS
Patients
October 1996 to August 2002, 137 patients, who were 3 to 21 years of age with histologically proven MB, were enrolled onto a collaborative treatment protocol (SJMB96) involving the following five separate institutions: St Jude Children's Research Hospital (Memphis, TN), Texas Children's Hospital (Houston, TX), Royal Children's Hospital (Melbourne, Australia), New Children's Hospital (Sydney, Australia), and Royal Children's Hospital (Brisbane, Australia); the latter two sites did not participate in testing. Twenty-six patients (19.2%) did not have neurocognitive testing; 12 were enrolled at two sites that did not participate in neurocognitive testing, four had severe neurologic symptoms precluding testing, four had parents who refused testing, three lacked fluency in English, and three were never referred for testing.
A total of 111 patients were included in the final analysis (Table 1). The median age for the entire group was 7.4 years at the time of diagnosis. Patients were classified at the time of enrollment as AR (n = 74) based on the absence of metastatic disease using magnetic resonance imaging of the brain and spine and analysis of CSF and measurable residual primary tumor of less than 1.5 cm3. Median age at diagnosis for AR patients was 8.83 years (range, 3.11 to 20.1 years). Patients were classified as HR (n = 37) based on the presence of metastatic disease or measurable residual primary tumor 1.5 cm3. Median age at diagnosis for HR patients was 6.4 years (range, 3.06 to 16.89 years). Patients with brainstem invasion with no other HR features were eligible for the AR group.
All treatment sites followed the same protocol-driven guidelines for medical treatment. Postoperative radiation therapy was initiated within 28 days of definitive surgery and consisted of 23.4 Gy of CSI, 36 Gy of conformal posterior fossa boost, and 55.8 Gy of conformal primary site boost for AR patients. HR patients received 36 to 39.6 Gy of CSI and 55.8 Gy of conformal primary site boost. The clinical target volume (CTV) for the posterior fossa component of AR treatment was the anatomic posterior fossa including the cerebellar hemispheres, brainstem to the level of the midbrain, and upper cervical spinal cord at the level of the foramen magnum. The planning target volume (PTV) for the posterior fossa component included a geometric margin of 0.3 to 0.5 cm surrounding the CTV. The gross tumor volume for the primary site component of AR and HR treatment included the postoperative tumor bed; the CTV for the primary site included the gross tumor volume with an anatomically confined margin of 2 cm in adjacent brain; and the primary site PTV included a geometric margin of 0.3 to 0.5 cm surrounding the CTV. Conformal treatment consisted mainly of forward-planned, noncoplanar, individually shaped beam arrangements incident on the PTV for most patients. The field edge (cerrobend or multileaf collimator) was 0.6 to 0.8 cm outside the PTV. The homogeneity of dose (95% to 110%) was optimized across the PTV with beam modifiers that included physical or virtual wedges. The dose to the upper cervical spinal cord was limited to 54 Gy. Fractionation was standardized at 1.8 Gy/d. A small number of patients were treated with intensity-modulated radiation therapy using axial coplanar beams and a binary collimator (n = 22).
Chemotherapy was initiated 6 weeks after the completion of radiation therapy and included four cycles of dose-intensive cyclophosphamide, cisplatin, and vincristine. Patients were observed every 3 months for 2 years and every 6 months thereafter. Audiograms were routinely conducted at enrollment; after completion of radiation therapy; at 1, 6, and 12 months after completion of treatment; and every 12 months thereafter for 5 years. Using National Cancer Institute Common Toxicity Criteria, if grade 3 ototoxicity was detected, the next dose of cisplatin was reduced by 50%, and if grade 4 ototoxicity was detected, cisplatin therapy was discontinued. Our institutional standard of care recommends hearing aids with hearing loss in either ear of 30 dB at 2,000 Hz. Endocrine testing was routinely provided yearly from enrollment, with appropriate replacement therapy if necessary.
Neurocognitive Testing
Neurocognitive testing was scheduled after surgical resection (shortly after the time of enrollment) and at approximately 1, 2, and 5 years after diagnosis. At St Jude Children's Research Hospital, this schedule was considered minimum follow-up, and every attempt was made to evaluate patients annually. An additional evaluation before starting chemotherapy was also conducted for some patients. Baseline was considered the highest score obtained between testing at enrollment and before chemotherapy. Depending on the age of the patient, each evaluation consisted of tests of intelligence and academic achievement. Children who were initially too young for testing would receive testing when appropriate for age.
No patients were excluded based on progressive disease, with only two patients receiving neuropsychology testing after recurrence. The first of these patients received one assessment of intelligence after progression, whereas the second patient received four assessments of intelligence and three assessments of achievement after progression. Although these two patients received surgical resection and additional chemotherapy, no additional radiation therapy was administered. To investigate whether there was a significant change in trend, changes in intelligence and achievement were compared between before progression and after progression. No change in trend was evident; therefore, these scores remained in the analyses.
Intelligence. Patients who were 3 years of age or older at the time of scheduled testing were evaluated with the age-appropriate Wechsler Intelligence Scale, which was administered in the standard format or in an abbreviated format at selected intervals (n = 104 patients, 264 observations).13-15 The abbreviated form administers three subtests from the standard administration (Information, Similarities, and Block Design) and, using a formula derived from the standard administration with persons from the general population, results in an estimation of full-scale IQ (FSIQ; estimated IQ [EIQ]).16 The EIQ was also calculated whenever the standard administration was performed. All quotients were adjusted for age using general population norms, with a mean of 100 and a standard deviation (SD) of 15, and scores were expected to be maintained over time (slope = 0). Because of concerns that testing patients with more than one version of the Wechsler scales would introduce significant error variance,17 we removed observations that involved a change in versions, while preserving the most observations possible for each patient. This resulted in the removal of 20 observations and left 244 observations for analysis. Multivariate models for EIQ over time were produced with and without the 20 observations involving a change in test version. The same conclusion was reached for each model (observations removed: intercept = 95.62, slope = 每1.59; observations included: intercept = 95.46, slope = 每1.41).
Earlier studies with children surviving MB have shown EIQ and FSIQ to be highly correlated (r = 0.91) and, in longitudinal analysis, to have comparable slopes.11,12 To assess the strength of association between EIQ and FSIQ, we derived the EIQ and FSIQ from 138 pairs of scores, regardless of test version and the timing; the resulting correlation coefficient was highly significant (r = 0.91, P < .0001). We also tested whether the difference between the EIQ and FSIQ slopes was significant, and it was not (mean = 0.34 points/yr; P = .216).
Academic achievement. For patients who were 5 years of age or older at the scheduled time of testing, academic achievement testing was conducted with the Wide Range Achievement Test每Revised (WRAT-R) or the Wechsler Individual Achievement Test (WIAT; basic reading, math, and spelling subtests).18,19 Patients who were younger than 5 years of age did not receive tests of academic achievement. Both the WRAT-R and the WIAT generate scores based on reading (decoding), spelling, and math (reasoning or computation). At selected intervals, 93 patients were administered the math subtest (n = 210 observations), 90 were administered the spelling subtest (n = 207 observations), and 91 were administered the reading subtest (n = 208 observations). All derived scores were adjusted for age using general population norms, with a mean of 100 and an SD of 15. It was expected that scores would be maintained over time, resulting in a slope of 0. Some patients received academic achievement testing with the WIAT, whereas others received the WRAT-R. No patient received both. When the slopes with each were compared, no significant differences were obtained (all P > .05), and therefore, the tests were combined in subsequent analyses.
Adherence. Adherence to the testing guidelines was variable, resulting in 207 to 244 observations for each outcome measure and a mean overall adherence of 54% over the study period. Additional analyses were completed to examine rates of adherence for AR and HR groups at baseline and at each year during the study period. There were no significant differences or trends between risk groups at any time point. Analyses also demonstrated that keeping values derived from patients with baseline observations only did not significantly affect the slope estimates. Therefore, those patients remained in the analyses, improving estimates and decreasing variability.
Statistical Approach
The original protocol design was planned to test the hypothesis that AR patients would show less IQ loss than HR patients in the interval from diagnosis to 2 years after diagnosis. With 19 patients planned in each risk group, the design had 80% power to detect a difference of 10 points or greater (one-sided test, = .05). The design was not planned to make comparisons of age at diagnosis and risk subgroups, and the small number of patients contributing to the original planned analysis did not allow for these comparisons to be made. However, assessing differences in characteristics between patients included in the original planned analysis and patients not included is important. Therefore, differences in risk group, age at diagnosis, and sex for patients included versus not included were examined. No significant differences were detected.
Subsequent exploratory analyses provided the opportunity to study the larger cohort of 111 patients and to analyze the longitudinal data using multivariate modeling procedures, which were frequently referred to as growth curve analyses.3,10-12 This approach treats data from different patients as statistically independent, while treating data from the same patient as correlated. Growth models emphasize the explanation of within-person variation over time by a natural developmental process. The growth model allows an estimation of the average rate of change in function (or slope) over time for patients within the total group, as well as an estimation of the differences in the slopes between subgroups.
Before any exploratory analysis, extensive visual inspection and spline smoothing of the scatter plots was completed. In addition, quadratic models were tested. Neither produced any suggestion of nonlinear patterns. Therefore, linear random coefficient models were used to analyze the data. For each outcome, we first developed a model that included time from diagnosis, age at diagnosis (< 7 v 7 years), and risk classification group (AR v HR). The age distribution at the time of diagnosis was divided at the age of 7 years because this enabled our results to be compared with previous studies that have used this convention10 and allowed for exploration of the effect of age and risk group interaction on IQ and achievement performance. When multiple observations were available on any particular outcome within 6 months of diagnosis, the highest score was chosen as the baseline value.
When slopes were compared with 0 (the normal expectation) or with each other, tests of statistical significance were conducted with the hypotheses that patients who were younger at diagnosis and patients who were treated with HR therapy would demonstrate the greatest decline. Our multivariate models were generally sensitive enough to detect a difference of 3 points/yr as a significant difference either from 0 or from the slope of another group. Additional exploratory analyses were conducted on the impact of shunting and posterior fossa syndrome on IQ. Proportions were compared between groups using the usual 2 test.
Missing values in research examining clinical populations is a common methodologic problem. However, the multivariate modeling procedures that we used account for such occurrences, and the theory on which mixed models is based remains intact. Therefore, no imputation for missing values was required, and patients were not excluded on this basis.20 The original hypothesis was the only preplanned analysis. Because the additional models were exploratory, no multiplicity corrections to the significance levels were made, and results should be interpreted with caution. The analyses for this study were carried out using Proc Mixed of the statistical computer package SAS System for Windows, Version 8.02 (SAS Institute, Inc, Cary, NC).20,21
RESULTS
Planned IQ Analysis
We first conducted the analysis as originally planned in the protocol by comparing change in IQ scores from the time of enrollment to the targeted 24 months (range, 16 to 28 months) from diagnosis. At the time of this analysis, 28 AR patients and 10 HR patients had complete data. AR patients had a mean change of 每0.4 points/yr (SD = 7.9 points/yr), and HR patients had a mean change of 每8.2 points/yr (SD = 9.3 points/yr), with the difference reaching statistical significance (P = .014) in a two-sample t test (two-tailed). However, an examination of these data revealed that age at the time of diagnosis was not proportionally distributed between the two risk groups; four of 28 patients in the AR group compared with seven of 10 patients in the HR group were 7 years of age or younger at the time of diagnosis. To adjust for age at diagnosis, an analysis of covariance model was used. Age at diagnosis was not found to be significant (P = .75), whereas the difference in decline between the two risk groups was still significant in this analysis of covariance model (P = .0215).
Multivariate Modeling
Patients received serial neurocognitive testing spanning from 0 to 6.03 years after diagnosis (median, 3.14 years). The results from multivariate modeling of the complete data set revealed statistically significant overall declines in mean IQ (每1.59 points/yr; P = .006), reading (每2.95 points/yr; P < .0001), spelling (每2.94 points/yr; P < .0001), and math (每1.87 points/yr; P = .003) scores for the entire group. To provide comparability to previous studies, the outcomes were subsequently analyzed by risk group and by age group (discussed in the following two sections). Sex effects were also explored with no significant effect. Therefore, sex was removed from further modeling and not reported in the present study. However, the unique quality and quantity of the data allowed for exploration of the interaction between risk group and age at diagnosis within a longitudinal model (Table 2).
Neurocognitive analysis by risk group. The results from the multivariate modeling of the complete data set by risk group (AR and HR) revealed no statistically significant overall declines in mean IQ for the AR group (每0.99 points/yr; P = .13), but the decline in the HR group was significant (每3.00 points/yr; P = .004), supporting the results of the originally planned analysis presented earlier. There was a trend toward a significant difference between the two groups (P = .097). Declines in the AR and HR groups for reading (每2.90 points/yr, P = .0001 and 每3.08 points/yr, P = .0003, respectively), spelling (每2.71 points/yr, P = .0003 and 每3.47 points/yr, P = .002, respectively), and math (每1.57 points/yr, P = .036 and 每2.49 points/yr, P = .024, respectively) were significant but did not differ significantly from each other.
Neurocognitive analysis by age group. The results from the multivariate modeling of the complete data set by age at diagnosis (< 7 years v 7 years) revealed statistically nonsignificant overall declines in mean IQ for the older group (每0.61 points/yr; P = .37), but the decline in the younger group was significant (每3.05 points/yr; P = .0005), with the two groups statistically different from each other (P = .026). For reading, declines in the younger group (每4.30 points/yr; P < .0001) and the older group (每1.87 points/yr; P = .001) were significant, and the differences between the two groups were significant (P = .005). For spelling, declines in the younger group (每4.03 points/yr; P < .0001) and the older group (每1.99 points/yr; P = .011) were significant, and the differences between the two groups trended toward significance (P = .083). For math, declines in the younger group (每2.32 points/yr; P = .016) were significant, with the older group (每1.46 points/yr; P = .076) trending toward significance, but the two groups were not significantly different (P = .49).
Neurocognitive analysis by age and risk group. Because of the unique quantity and quality of the data, interaction models were able to be developed to test age at diagnosis by risk group. Although the overall interaction models for IQ (Fig 1) and math were not statistically significant (P = .124 and P = .27, respectively), the overall interaction models for spelling and reading (Fig 2) were statistically significant (P = .034 and P = .036, respectively; Table 2).
For the older AR group, the mean change in IQ (每0.42 points/yr) was not statistically significant (P = .58). The mean declines in reading (每2.05 points/yr; P = .002), spelling (每2.62 points/yr; P = .002), and math (每1.84 points/yr; P = .042) were statistically significant.
For the older HR group, the mean changes in IQ (每1.56 points/yr; P = .35), reading (每1.05 points/yr; P = .43), spelling (1.02 points/yr; P = .56), and math (0.37 points/yr; P = .85) were not statistically significant. However, the reduced number of patients in this group (n = 13) limits the reliability of these outcomes.
For the younger AR group, the mean declines in IQ (每2.41 points/yr; P = .049), reading (每4.81 points/yr; P < .0001), and spelling (每2.60 points/yr; P = .039) were statistically significant. The mean change in math (每0.77 points/yr; P = .57) was not statistically significant.
For the younger HR group, the mean declines in IQ (每3.71 points/yr; P = .003), reading (每3.90 points/yr; P < .0001), and spelling (每5.31 points/yr; P < .0001) were statistically significant. The mean decline in spelling was also significantly greater than the decline for the older HR group (P = .004). The mean decline in math (每3.73 points/yr) was also statistically significant (P = .006).
Pairwise comparisons were conducted between differences in the change scores for the younger AR group and the younger HR group. For IQ (mean difference, 1.30 points/yr), spelling (mean difference, 2.71 points/yr), and math (mean difference, 2.96 points/yr), the models suggested an advantage for children treated in the AR group. This was not the case for reading (mean difference, 每0.91 points/yr). However, none of these differences were large enough to reach statistical significance (two-tailed t tests, all P .12). However, the magnitude of these changes do hold clinical significance. For example, over a 5-year period, the difference in spelling scores between the younger AR group and younger HR group could reach almost 15 points.
Impact of Clinical Events
We also explored the impact of shunt and posterior fossa syndrome on IQ development. First, differences in the frequency of these events were determined for the four age and risk groups. Then, analyses were conducted on IQ changes. There was no significant difference in the proportion of patients in the four age and risk groups with regard to shunting (P = .382; younger HR, 33%; younger AR, 33%; older HR, 15%; and older AR, 20%). Subsequent modeling of presence or absence of shunting revealed no significant impact on IQ change (P = .13).
Posterior fossa syndrome was defined by the postoperative appearance of pseudobulbar symptoms and mutism. There was a trend toward significance in the proportion of patients in the four age and risk groups (P = .077; younger HR, 38%; younger AR, 13%; older HR, 8%; and older AR, 18%), with the younger HR group having a higher incidence than the other age and risk groups. Subsequent modeling of presence or absence of posterior fossa syndrome revealed no significant impact on IQ change (P = .68).
DISCUSSION
The present study provides original, longitudinal neurocognitive data for patients treated for MB with postoperative risk-adapted CSI (23.4 v 36 to 39.6 Gy) and 55.8 Gy of conformal primary site boost using a 2-cm CTV margin before four cycles of dose-intensive chemotherapy. Notwithstanding this novel approach, for the group as a whole, statistically significant declines were observed for mean IQ, reading, spelling, and math. The overall magnitude of the negative IQ slope is commensurate with or somewhat smaller than the slopes previously reported with similar samples of MB patients,4,10-12 but these changes must now be interpreted with respect to the moderating influences of age and risk group that the present study demonstrates.
In terms of the initial hypothesis that AR patients would show less neurocognitive decline than HR patients, significant differences in the predicted direction were found for IQ when changes from baseline to 2 years after diagnosis were analyzed. These differences remained significant after adjusting for differences in age at diagnosis within the two risk groups. Furthermore, multivariate modeling revealed that the decline in IQ was significant for the HR patients but not for the AR patients. These results would seem to support the contention that the CSI dose reduction in the AR patients preserved neurocognitive function.
Because sufficient observations were available, modeling of the data for complex interactions was possible and makes a unique contribution to the current literature. Even among previous longitudinal studies that have examined age and radiation dose effects using multivariate methods, it has not been possible to identify interactions between risk group, age, and time.10-12 For two of the four neurocognitive outcomes (reading and spelling), significant interactions were found between risk group, age, and time. The presence of such interactions limits the value of simple interpretations of neurocognitive changes based only on differences in CSI between risk groups and underscores the need for multiple observations on large patient samples in future MB studies.
Several previous studies of patients treated for MB have shown young age to be a risk factor3,7,10-12; however, relatively few direct comparisons among young children treated with conventional-dose versus reduced-dose irradiation have been published. In the retrospective Pediatric Oncology Group series, which used conventional irradiation without chemotherapy, six or fewer patients were available in these two groups for analysis, necessitating nonparametric methods.9 Although the mean scores suggested a 10每to 15每IQ point advantage in favor of the reduced-dose group, the comparisons failed to achieve statistical significance. Furthermore, the most recent survival analysis in this cohort of patients treated with radiation therapy alone revealed an increased risk of early relapse and lower 5-year event free survival in the reduced-dose group.22 Although it is important to examine statistical significance, when sample sizes are small, effect size can also be relevant. A major finding of the present analysis was the failure to show a consistent benefit of the AR over the HR radiation therapy regimen. On examination of the slope magnitude or effect size, younger AR children did not show as much of a decline as younger HR children on measures of IQ, spelling, or math. However, for reading, this relationship was reversed. When statistically compared, differences between the slopes for the younger AR and HR groups did not reach significance for IQ, reading, and spelling.
A factor that may partially explain the difficulty in ascertaining neurocognitive differences among younger and, therefore, more vulnerable patients relates to radiation dosimetry. Intuitively, one would think that the 35% dose reduction from 36 to 23.4 Gy for the CSI dose would result in a commensurate reduction in brain volume exposed to higher doses of irradiation. However, as shown in Figure 3, the lower dose group nevertheless had 40% of total brain volume exposed to 36 Gy, and both regimens resulted in 20% of total brain volume exposure to 50 Gy or greater, largely because of the targeting guidelines used for this study and the minimum posterior fossa dose of 36 Gy.
The apparent increased vulnerability of reading and spelling skills among the younger children in both risk groups may be explained by impairments of fundamental cognitive processes early in language development (orthographic and phonologic analysis). Deficits in phonologic analysis among otherwise healthy children represent the most specific and reliable predictors of later dyslexia. And dyslexic children, compared with normal readers, display a disrupted pattern of activation on functional magnetic resonance imaging of posterior brain regions and parietotemporal and occipitotemporal areas.23 For MB patients in both risk groups in the present study, these brain regions received relatively large total doses of irradiation from the posterior fossa boost, which is known to affect white matter development and, more specifically, posterior corpus callosum morphology.24-27 Whether cognitive and behavioral or pharmacologic interventions can effectively treat these and other neurocognitive sequelae is not yet known, although data from the short-term use of methylphenidate seems promising.28,29
The present study has several methodologic limitations. For ethical reasons, patients were not randomly assigned to treatment arms because of the known need for more aggressive therapy among HR patients. Missing values is a common difficulty when examining clinical populations, and the present study is no exception, but the robustness of the mixed models to missing values eased this limitation. In addition, a potentially confounding variable in the analysis of neurocognitive changes is the incidence of other clinical events that can impact these outcomes in addition to age and risk group. Our analyses of the impact of shunting did not reveal either imbalances among age and risk groups or consistent associations with IQ loss. In contrast, the incidence of posterior fossa syndrome was disproportionately high among children in the younger HR group. Because of limited sample size, we could not formally define the impact of posterior fossa syndrome on neurocognitive development. One would expect the presence of posterior fossa syndrome to depress neurocognitive development, therefore enhancing differences in neurocognitive trajectories between the younger patients in the HR and AR groups. However, except for math achievement, significant differences were not observed.
The ability to decode the written word depends on many interdependent skills. Decoding begins with the capacity to notice and manipulate the smallest units of sound in spoken language that can make a difference in the meaning of a word (eg, pup or cup). These units of sounds are called phonemes. The ability to distinguish between each phoneme is considered the building block for all other language and reading-related skills. A critical task for the early reader is not only to properly distinguish various phonemes but also to recognize that these auditory segments are represented by the alphabet in written form. By understanding that each letter represents a sound and by then using the sounds to blend phonemes into a recognizable word, a person can successfully decode written language. With a disturbance in any one of the interdependent skills, children may experience a slowed rate of knowledge acquisition and greater difficulty with reading and/or spelling. The potential for treatment-induced hearing loss to impact these skills, especially as these skills are still developing in the younger child, should be considered. Although audiograms were regularly scheduled and appropriate adjustment to chemotherapy dose and/or hearing aid support was provided when defined loss of function was detected, hearing ability was not included in the current analyses and constitutes another potential limitation. It is believed that a more detailed examination of audiology data and cochlear radiation dose within the MB population is important to furthering our understanding of the development of phonologic processing and reading decoding ability after treatment.
Despite limiting the psychometric evaluations to three experienced institutions with a strong commitment to the treatment protocol, adherence to the testing schedule was problematic and presents another potential limitation. Even among children who maintained their response to treatment for 2 years or more, adherence generally varied between 50% and 60% at each interval. The problem with adherence, which is long recognized in pediatric cooperative trials, limits both the sensitivity of statistical analyses and the ability to generalize findings. In our design, we generally had the power to detect slopes differing from 0 or from each other by 3 points/yr or more. Future studies may want to focus on even smaller differences between critical groups of patients, such as patients who are younger at diagnosis. Such studies will need to carefully consider the number of patients and observations needed to reliably test hypotheses of interest.
Authors' Disclosures of Potential Conflicts of Interest
The authors indicated no potential conflicts of interest.
NOTES
Deceased.
This research is supported by Musicians Against Childhood Cancer, the Noyes Foundation, the American Lebanese Syrian Associated Charities, National Cancer Institute grant Nos. R01CA78957 and U01CA81445, and Cancer Center Support grant No. P30CA21765.
Authors' disclosures of potential conflicts of interest are found at the end of this article.
REFERENCES
Mostow EN, Byrne J, Connelly RR, et al: Quality of life in long-term survivors of CNS tumors of childhood and adolescence. J Clin Oncol 9:592-599, 1991
Ris MD, Noll RB: Long-term neurobehavioral outcome in pediatric brain-tumor patients: Review and methodological critique. J Clin Exp Neuropsychol 16:21-42, 1994
Dennis M, Spiegler BJ, Hetherington CR, et al: Neuropsychological sequelae of the treatment of children with medulloblastoma. J Neurooncol 29:91-101, 1996
Spiegler BJ, Bouffet E, Greenberg ML, et al: Change in neurocognitive functioning after treatment with cranial radiation in childhood. J Clin Oncol 22:706-713, 2004
Grill J, Renaux VK, Bulteau C, et al: Long-term intellectual outcome in children with posterior fossa tumors according to radiation doses and volumes. Int J Radiat Oncol Biol Phys 45:137-145, 1999
Copeland DR, DeMoor C, Moore BD, et al: Neurocognitive development of children after a cerebellar tumor in infancy: A longitudinal study. J Clin Oncol 17:3476-3486, 1999
Walter AW, Mulhern RK, Gajjar A, et al: Survival and neurodevelopmental outcome of young children with medulloblastoma at St. Jude Children's Research Hospital. J Clin Oncol 17:3720-3728, 1999
Strother D, Ashley D, Kellie SJ, et al: Feasibility of four consecutive high-dose chemotherapy cycles with stem-cell rescue for patients with newly diagnosed medulloblastoma or supratentorial PNET after craniospinal radiotherapy: Results of a collaborative study. J Clin Oncol 19:2696-2704, 2001
Mulhern RK, Kepner JL, Thomas PR, et al: Neuropsychologic functioning of survivors of childhood medulloblastoma randomized to receive conventional or reduced-dose craniospinal irradiation: A Pediatric Oncology Group study. J Clin Oncol 16:1723-1728, 1998
Ris MD, Packer R, Goldwein J, et al: Intellectual outcome after reduced-dose radiation therapy plus adjuvant chemotherapy for medulloblastoma: A Children's Cancer Group study. J Clin Oncol 19:3470-3476, 2001
Palmer SL, Goloubeva O, Reddick WE, et al: Patterns of intellectual development among survivors of pediatric medulloblastoma: A longitudinal analysis. J Clin Oncol 19:2302-2308, 2001
Palmer SL, Gajjar A, Reddick WE, et al: Predicting intellectual outcome among children treated with 35-40 Gy craniospinal irradiation for medulloblastoma. Neuropsychology 17:548-555, 2003
Wechsler D: Wechsler Preschool and Primary Scale of Intelligence每Revised. New York, NY, The Psychological Corporation, 1989
Wechsler D: Wechsler Intelligence Scale for Children (ed 3). New York, NY, The Psychological Corporation, 1991
Wechsler D: Wechsler Adult Intelligence Scale每Revised. New York, NY, The Psychological Corporation, 1989
Sattler JM: Assessment of Children. San Diego, CA, Jerome Sattler, 1992, p 1170
Mulhern RK, Ochs J, Fairclough D: Deterioration of intellect among children surviving leukemia: IQ test changes modify estimates of treatment toxicity. J Consult Clin Psychol 60:477-480, 1992
Jastak S, Wilkenson GS: Wide Range Achievement Test每Revised. Wilmington, DE, Jastak Associates, Inc, 1984
Psychological Corporation: Wechsler Individual Achievement Test. New York, NY, Harcourt, Brace, Jovanovich, 1992
Littell RC, Milliken GA, Stroup W, et al: SAS System for Mixed Models. Cary, NC, SAS Institute, Inc, 1996
Laird NM, Ware JH: Random-effects models for longitudinal data. Biometrics 38:963-974, 1982
Thomas PRM, Deutsch M, Kepner J, et al: Low-stage medulloblastoma: Final analysis of trial comparing standard-dose with reduced-dose neuraxis irradiation. J Clin Oncol 18:3004-3011, 2000
Shaywitz BA, Shaywitz SE, Pugh KR, et al: Disruption of posterior brain systems for reading in children with developmental dyslexia. Biol Psychiatry 52:101-110, 2002
Mulhern RK, Palmer SL, Reddick WE, et al: Risks of young age for selected neurocognitive deficits in medulloblastoma are associated with white matter loss. J Clin Oncol 19:472-479, 2001
Reddick WE, White H, Glass JO, et al: Developmental model relating white matter volume with neurocognitive deficits in pediatric brain tumor survivors. Cancer 97:2512-2519, 2003
Palmer SL, Reddick WE, Glass JO, et al: Decline in corpus callosum volume among pediatric patients with medulloblastoma: Longitudinal MR imaging study. Am J Neuroradiol 23:1088-1094, 2002
Mulhern RK, Reddick WE, Palmer SL, et al: Neurocognitive deficits in medulloblastoma survivors and white matter loss. Ann Neurol 46:834-841, 1999
Butler RW, Copeland DR: Attentional processes and their remediation in children treated for cancer: A literature review and the development of a therapeutic approach. J Int Neuropsychol Soc 8:115-124, 2002
Mulhern RK, Khan RB, Kaplan S, et al: Short-term efficacy of methylphenidate: A randomized, double-blind, placebo-controlled trial among survivors of childhood cancer. J Clin Oncol 22:4795-4803, 2004(Raymond K. Mulhern, Shawn)
Texas Children's Cancer Center, Baylor College of Medicine, Houston, TX
Royal Children's Hospital, Melbourne, Australia
ABSTRACT
PURPOSE: This prospective, longitudinal study examined the effects of risk-adapted craniospinal irradiation (CSI) dose and the interactions of dose with age and time from diagnosis on intelligence quotient (IQ) and academic achievement (reading, spelling, and math) among patients treated for medulloblastoma (MB).
PATIENTS AND METHODS: Patients received serial neurocognitive testing spanning from 0 to 6.03 years after diagnosis (median, 3.14 years). The multi-institutional study included 111 patients, who were 3 to 20 years of age at diagnosis (median age, 7.4 years), treated for MB with risk-adapted CSI followed by four cycles of high-dose chemotherapy (cyclophosphamide, cisplatin, and vincristine) with stem-cell support. High-risk patients (HR; n = 37) received CSI to 36 to 39.6 Gy and conformal boost treatment of the primary site to 55.8 to 59.4 Gy. Average-risk patients (AR; n = 74) received CSI to 23.4 Gy and conformal boost treatment of the posterior fossa to 36.0 Gy and primary site to 55.8 Gy.
RESULTS: Multivariate modeling revealed statistically significant declines in mean IQ (每1.59 points/yr; P = .006), reading (每2.95 points/yr; P < .0001), spelling (每2.94 points/yr; P < .0001), and math (每1.87 points/yr; P = .003) scores for the entire group. The effects of risk-adapted radiation therapy on IQ, reading, and spelling were moderated by age, with the greatest rates of decline observed for the HR patients who were younger (< 7 years old) at diagnosis.
CONCLUSION: Young age at diagnosis was the most prominent risk factor for neurocognitive deficits among survivors of MB despite reductions in CSI dosing and efforts to limit the boost volume. Younger patients exhibited substantial problems with the development of reading skills.
INTRODUCTION
Children who survive treatment for medulloblastoma (MB), which includes postoperative craniospinal irradiation (CSI) with or without chemotherapy, may experience disturbances in essential neurocognitive functions that result in impaired overall intelligence and a high rate of academic failure. Deficits in processing speed, memory ability, and attention have been prominent among patients treated for MB, with both treatment and patient factors attributing to eventual outcomes.1-3 Patients who receive higher dose CSI and who are younger at diagnosis experience greater deficits in neurocognitive performance.4-7 To reduce neurocognitive morbidity, contemporary protocols have been designed to limit radiation dose using risk-adapted CSI dosing and limited posterior fossa or primary site irradiation using three-dimensional conformal radiation therapy. With this approach, early results have proven successful, with evidence of excellent event-free survival for standard-risk patients receiving reduced-dose CSI (23.4 Gy) and adjuvant chemotherapy.8
The impact of the age of the patient at treatment and the level of radiation dose exposure on cognitive performance has been the focus of examination in several studies. Four of these studies are especially relevant to increasing our understanding of the influence of these potential risk factors. A retrospective Pediatric Oncology Group study of average-risk (AR) patients found that patients who were randomly assigned to receive reduced-dose CSI (23.4 Gy) and patients who were older at time of treatment (> 8.8 years) showed higher cognitive functioning than patients who were randomly assigned to receive standard-dose CSI (36 Gy) and patients who were younger at time of treatment.9
However, in a subsequent prospective study, even the AR patients who had received reduced-dose CSI (23.4 Gy) and adjuvant chemotherapy were found to experience a decline of 4.3 intelligence quotient (IQ) points per year over a 3-year period from treatment.10 Also using a longitudinal design, Palmer et al11 found a mean decline of 2.6 IQ points/yr among 44 children who were treated with CSI (24.3 to 39.6 Gy). Older patients (> 8 years) and patients with lower dose CSI (< 35.2 Gy) showed higher cognitive functioning than patients who were younger or who received a higher CSI dose. The same group later reported in more detail the patterns of change in intellectual functioning over a period of 7 years from diagnosis for 50 patients who were treated with conventional-dose (35 to 40 Gy) CSI.12 The overall rate of decline in IQ was 2.2 points/yr. A delayed decline in IQ was observed for patients who were treated at an older age. In contrast, younger patients showed a more immediate decline in IQ. Female patients were also at greater risk for IQ declines.
In addition to confirming the prevailing beliefs about the increased risks associated with younger age and higher CSI dose, these longitudinal studies were able to generate predictive models quantifying expected IQ loss for particular subgroups of patients. However, the potential neurocognitive benefits of reduced-dose CSI in MB are ambiguous. No prospective study of MB patients based on a single protocol has compared risk-adapted radiation therapy dose outcomes. In addition, the limited number of observations and patients has never allowed an analysis of the potential interactions among two or more risk factors (eg, whether the rate of IQ loss among younger patients treated with reduced-dose CSI is different from the rate of IQ loss among younger patients treated with conventional-dose CSI).
The purpose of the present study was to test the hypothesis, as posed in the original protocol, that significant differences in changes in neurocognitive function would be observed between the AR patients treated with reduced-dose CSI (23.4 Gy) and high-risk (HR) patients treated with CSI of 36 to 39.6 Gy from shortly after diagnosis to 2 years after diagnosis. It was believed that the AR patients would demonstrate better performance than the HR patients. In the original design, age was not proposed as a covariate. To analyze the maximal number of observations on patients over more prolonged time intervals and to take advantage of more contemporary methods of longitudinal analysis using multiple covariates and their interactions (ie, age, risk group, and time), the analytic approach was expanded.
PATIENTS AND METHODS
Patients
October 1996 to August 2002, 137 patients, who were 3 to 21 years of age with histologically proven MB, were enrolled onto a collaborative treatment protocol (SJMB96) involving the following five separate institutions: St Jude Children's Research Hospital (Memphis, TN), Texas Children's Hospital (Houston, TX), Royal Children's Hospital (Melbourne, Australia), New Children's Hospital (Sydney, Australia), and Royal Children's Hospital (Brisbane, Australia); the latter two sites did not participate in testing. Twenty-six patients (19.2%) did not have neurocognitive testing; 12 were enrolled at two sites that did not participate in neurocognitive testing, four had severe neurologic symptoms precluding testing, four had parents who refused testing, three lacked fluency in English, and three were never referred for testing.
A total of 111 patients were included in the final analysis (Table 1). The median age for the entire group was 7.4 years at the time of diagnosis. Patients were classified at the time of enrollment as AR (n = 74) based on the absence of metastatic disease using magnetic resonance imaging of the brain and spine and analysis of CSF and measurable residual primary tumor of less than 1.5 cm3. Median age at diagnosis for AR patients was 8.83 years (range, 3.11 to 20.1 years). Patients were classified as HR (n = 37) based on the presence of metastatic disease or measurable residual primary tumor 1.5 cm3. Median age at diagnosis for HR patients was 6.4 years (range, 3.06 to 16.89 years). Patients with brainstem invasion with no other HR features were eligible for the AR group.
All treatment sites followed the same protocol-driven guidelines for medical treatment. Postoperative radiation therapy was initiated within 28 days of definitive surgery and consisted of 23.4 Gy of CSI, 36 Gy of conformal posterior fossa boost, and 55.8 Gy of conformal primary site boost for AR patients. HR patients received 36 to 39.6 Gy of CSI and 55.8 Gy of conformal primary site boost. The clinical target volume (CTV) for the posterior fossa component of AR treatment was the anatomic posterior fossa including the cerebellar hemispheres, brainstem to the level of the midbrain, and upper cervical spinal cord at the level of the foramen magnum. The planning target volume (PTV) for the posterior fossa component included a geometric margin of 0.3 to 0.5 cm surrounding the CTV. The gross tumor volume for the primary site component of AR and HR treatment included the postoperative tumor bed; the CTV for the primary site included the gross tumor volume with an anatomically confined margin of 2 cm in adjacent brain; and the primary site PTV included a geometric margin of 0.3 to 0.5 cm surrounding the CTV. Conformal treatment consisted mainly of forward-planned, noncoplanar, individually shaped beam arrangements incident on the PTV for most patients. The field edge (cerrobend or multileaf collimator) was 0.6 to 0.8 cm outside the PTV. The homogeneity of dose (95% to 110%) was optimized across the PTV with beam modifiers that included physical or virtual wedges. The dose to the upper cervical spinal cord was limited to 54 Gy. Fractionation was standardized at 1.8 Gy/d. A small number of patients were treated with intensity-modulated radiation therapy using axial coplanar beams and a binary collimator (n = 22).
Chemotherapy was initiated 6 weeks after the completion of radiation therapy and included four cycles of dose-intensive cyclophosphamide, cisplatin, and vincristine. Patients were observed every 3 months for 2 years and every 6 months thereafter. Audiograms were routinely conducted at enrollment; after completion of radiation therapy; at 1, 6, and 12 months after completion of treatment; and every 12 months thereafter for 5 years. Using National Cancer Institute Common Toxicity Criteria, if grade 3 ototoxicity was detected, the next dose of cisplatin was reduced by 50%, and if grade 4 ototoxicity was detected, cisplatin therapy was discontinued. Our institutional standard of care recommends hearing aids with hearing loss in either ear of 30 dB at 2,000 Hz. Endocrine testing was routinely provided yearly from enrollment, with appropriate replacement therapy if necessary.
Neurocognitive Testing
Neurocognitive testing was scheduled after surgical resection (shortly after the time of enrollment) and at approximately 1, 2, and 5 years after diagnosis. At St Jude Children's Research Hospital, this schedule was considered minimum follow-up, and every attempt was made to evaluate patients annually. An additional evaluation before starting chemotherapy was also conducted for some patients. Baseline was considered the highest score obtained between testing at enrollment and before chemotherapy. Depending on the age of the patient, each evaluation consisted of tests of intelligence and academic achievement. Children who were initially too young for testing would receive testing when appropriate for age.
No patients were excluded based on progressive disease, with only two patients receiving neuropsychology testing after recurrence. The first of these patients received one assessment of intelligence after progression, whereas the second patient received four assessments of intelligence and three assessments of achievement after progression. Although these two patients received surgical resection and additional chemotherapy, no additional radiation therapy was administered. To investigate whether there was a significant change in trend, changes in intelligence and achievement were compared between before progression and after progression. No change in trend was evident; therefore, these scores remained in the analyses.
Intelligence. Patients who were 3 years of age or older at the time of scheduled testing were evaluated with the age-appropriate Wechsler Intelligence Scale, which was administered in the standard format or in an abbreviated format at selected intervals (n = 104 patients, 264 observations).13-15 The abbreviated form administers three subtests from the standard administration (Information, Similarities, and Block Design) and, using a formula derived from the standard administration with persons from the general population, results in an estimation of full-scale IQ (FSIQ; estimated IQ [EIQ]).16 The EIQ was also calculated whenever the standard administration was performed. All quotients were adjusted for age using general population norms, with a mean of 100 and a standard deviation (SD) of 15, and scores were expected to be maintained over time (slope = 0). Because of concerns that testing patients with more than one version of the Wechsler scales would introduce significant error variance,17 we removed observations that involved a change in versions, while preserving the most observations possible for each patient. This resulted in the removal of 20 observations and left 244 observations for analysis. Multivariate models for EIQ over time were produced with and without the 20 observations involving a change in test version. The same conclusion was reached for each model (observations removed: intercept = 95.62, slope = 每1.59; observations included: intercept = 95.46, slope = 每1.41).
Earlier studies with children surviving MB have shown EIQ and FSIQ to be highly correlated (r = 0.91) and, in longitudinal analysis, to have comparable slopes.11,12 To assess the strength of association between EIQ and FSIQ, we derived the EIQ and FSIQ from 138 pairs of scores, regardless of test version and the timing; the resulting correlation coefficient was highly significant (r = 0.91, P < .0001). We also tested whether the difference between the EIQ and FSIQ slopes was significant, and it was not (mean = 0.34 points/yr; P = .216).
Academic achievement. For patients who were 5 years of age or older at the scheduled time of testing, academic achievement testing was conducted with the Wide Range Achievement Test每Revised (WRAT-R) or the Wechsler Individual Achievement Test (WIAT; basic reading, math, and spelling subtests).18,19 Patients who were younger than 5 years of age did not receive tests of academic achievement. Both the WRAT-R and the WIAT generate scores based on reading (decoding), spelling, and math (reasoning or computation). At selected intervals, 93 patients were administered the math subtest (n = 210 observations), 90 were administered the spelling subtest (n = 207 observations), and 91 were administered the reading subtest (n = 208 observations). All derived scores were adjusted for age using general population norms, with a mean of 100 and an SD of 15. It was expected that scores would be maintained over time, resulting in a slope of 0. Some patients received academic achievement testing with the WIAT, whereas others received the WRAT-R. No patient received both. When the slopes with each were compared, no significant differences were obtained (all P > .05), and therefore, the tests were combined in subsequent analyses.
Adherence. Adherence to the testing guidelines was variable, resulting in 207 to 244 observations for each outcome measure and a mean overall adherence of 54% over the study period. Additional analyses were completed to examine rates of adherence for AR and HR groups at baseline and at each year during the study period. There were no significant differences or trends between risk groups at any time point. Analyses also demonstrated that keeping values derived from patients with baseline observations only did not significantly affect the slope estimates. Therefore, those patients remained in the analyses, improving estimates and decreasing variability.
Statistical Approach
The original protocol design was planned to test the hypothesis that AR patients would show less IQ loss than HR patients in the interval from diagnosis to 2 years after diagnosis. With 19 patients planned in each risk group, the design had 80% power to detect a difference of 10 points or greater (one-sided test, = .05). The design was not planned to make comparisons of age at diagnosis and risk subgroups, and the small number of patients contributing to the original planned analysis did not allow for these comparisons to be made. However, assessing differences in characteristics between patients included in the original planned analysis and patients not included is important. Therefore, differences in risk group, age at diagnosis, and sex for patients included versus not included were examined. No significant differences were detected.
Subsequent exploratory analyses provided the opportunity to study the larger cohort of 111 patients and to analyze the longitudinal data using multivariate modeling procedures, which were frequently referred to as growth curve analyses.3,10-12 This approach treats data from different patients as statistically independent, while treating data from the same patient as correlated. Growth models emphasize the explanation of within-person variation over time by a natural developmental process. The growth model allows an estimation of the average rate of change in function (or slope) over time for patients within the total group, as well as an estimation of the differences in the slopes between subgroups.
Before any exploratory analysis, extensive visual inspection and spline smoothing of the scatter plots was completed. In addition, quadratic models were tested. Neither produced any suggestion of nonlinear patterns. Therefore, linear random coefficient models were used to analyze the data. For each outcome, we first developed a model that included time from diagnosis, age at diagnosis (< 7 v 7 years), and risk classification group (AR v HR). The age distribution at the time of diagnosis was divided at the age of 7 years because this enabled our results to be compared with previous studies that have used this convention10 and allowed for exploration of the effect of age and risk group interaction on IQ and achievement performance. When multiple observations were available on any particular outcome within 6 months of diagnosis, the highest score was chosen as the baseline value.
When slopes were compared with 0 (the normal expectation) or with each other, tests of statistical significance were conducted with the hypotheses that patients who were younger at diagnosis and patients who were treated with HR therapy would demonstrate the greatest decline. Our multivariate models were generally sensitive enough to detect a difference of 3 points/yr as a significant difference either from 0 or from the slope of another group. Additional exploratory analyses were conducted on the impact of shunting and posterior fossa syndrome on IQ. Proportions were compared between groups using the usual 2 test.
Missing values in research examining clinical populations is a common methodologic problem. However, the multivariate modeling procedures that we used account for such occurrences, and the theory on which mixed models is based remains intact. Therefore, no imputation for missing values was required, and patients were not excluded on this basis.20 The original hypothesis was the only preplanned analysis. Because the additional models were exploratory, no multiplicity corrections to the significance levels were made, and results should be interpreted with caution. The analyses for this study were carried out using Proc Mixed of the statistical computer package SAS System for Windows, Version 8.02 (SAS Institute, Inc, Cary, NC).20,21
RESULTS
Planned IQ Analysis
We first conducted the analysis as originally planned in the protocol by comparing change in IQ scores from the time of enrollment to the targeted 24 months (range, 16 to 28 months) from diagnosis. At the time of this analysis, 28 AR patients and 10 HR patients had complete data. AR patients had a mean change of 每0.4 points/yr (SD = 7.9 points/yr), and HR patients had a mean change of 每8.2 points/yr (SD = 9.3 points/yr), with the difference reaching statistical significance (P = .014) in a two-sample t test (two-tailed). However, an examination of these data revealed that age at the time of diagnosis was not proportionally distributed between the two risk groups; four of 28 patients in the AR group compared with seven of 10 patients in the HR group were 7 years of age or younger at the time of diagnosis. To adjust for age at diagnosis, an analysis of covariance model was used. Age at diagnosis was not found to be significant (P = .75), whereas the difference in decline between the two risk groups was still significant in this analysis of covariance model (P = .0215).
Multivariate Modeling
Patients received serial neurocognitive testing spanning from 0 to 6.03 years after diagnosis (median, 3.14 years). The results from multivariate modeling of the complete data set revealed statistically significant overall declines in mean IQ (每1.59 points/yr; P = .006), reading (每2.95 points/yr; P < .0001), spelling (每2.94 points/yr; P < .0001), and math (每1.87 points/yr; P = .003) scores for the entire group. To provide comparability to previous studies, the outcomes were subsequently analyzed by risk group and by age group (discussed in the following two sections). Sex effects were also explored with no significant effect. Therefore, sex was removed from further modeling and not reported in the present study. However, the unique quality and quantity of the data allowed for exploration of the interaction between risk group and age at diagnosis within a longitudinal model (Table 2).
Neurocognitive analysis by risk group. The results from the multivariate modeling of the complete data set by risk group (AR and HR) revealed no statistically significant overall declines in mean IQ for the AR group (每0.99 points/yr; P = .13), but the decline in the HR group was significant (每3.00 points/yr; P = .004), supporting the results of the originally planned analysis presented earlier. There was a trend toward a significant difference between the two groups (P = .097). Declines in the AR and HR groups for reading (每2.90 points/yr, P = .0001 and 每3.08 points/yr, P = .0003, respectively), spelling (每2.71 points/yr, P = .0003 and 每3.47 points/yr, P = .002, respectively), and math (每1.57 points/yr, P = .036 and 每2.49 points/yr, P = .024, respectively) were significant but did not differ significantly from each other.
Neurocognitive analysis by age group. The results from the multivariate modeling of the complete data set by age at diagnosis (< 7 years v 7 years) revealed statistically nonsignificant overall declines in mean IQ for the older group (每0.61 points/yr; P = .37), but the decline in the younger group was significant (每3.05 points/yr; P = .0005), with the two groups statistically different from each other (P = .026). For reading, declines in the younger group (每4.30 points/yr; P < .0001) and the older group (每1.87 points/yr; P = .001) were significant, and the differences between the two groups were significant (P = .005). For spelling, declines in the younger group (每4.03 points/yr; P < .0001) and the older group (每1.99 points/yr; P = .011) were significant, and the differences between the two groups trended toward significance (P = .083). For math, declines in the younger group (每2.32 points/yr; P = .016) were significant, with the older group (每1.46 points/yr; P = .076) trending toward significance, but the two groups were not significantly different (P = .49).
Neurocognitive analysis by age and risk group. Because of the unique quantity and quality of the data, interaction models were able to be developed to test age at diagnosis by risk group. Although the overall interaction models for IQ (Fig 1) and math were not statistically significant (P = .124 and P = .27, respectively), the overall interaction models for spelling and reading (Fig 2) were statistically significant (P = .034 and P = .036, respectively; Table 2).
For the older AR group, the mean change in IQ (每0.42 points/yr) was not statistically significant (P = .58). The mean declines in reading (每2.05 points/yr; P = .002), spelling (每2.62 points/yr; P = .002), and math (每1.84 points/yr; P = .042) were statistically significant.
For the older HR group, the mean changes in IQ (每1.56 points/yr; P = .35), reading (每1.05 points/yr; P = .43), spelling (1.02 points/yr; P = .56), and math (0.37 points/yr; P = .85) were not statistically significant. However, the reduced number of patients in this group (n = 13) limits the reliability of these outcomes.
For the younger AR group, the mean declines in IQ (每2.41 points/yr; P = .049), reading (每4.81 points/yr; P < .0001), and spelling (每2.60 points/yr; P = .039) were statistically significant. The mean change in math (每0.77 points/yr; P = .57) was not statistically significant.
For the younger HR group, the mean declines in IQ (每3.71 points/yr; P = .003), reading (每3.90 points/yr; P < .0001), and spelling (每5.31 points/yr; P < .0001) were statistically significant. The mean decline in spelling was also significantly greater than the decline for the older HR group (P = .004). The mean decline in math (每3.73 points/yr) was also statistically significant (P = .006).
Pairwise comparisons were conducted between differences in the change scores for the younger AR group and the younger HR group. For IQ (mean difference, 1.30 points/yr), spelling (mean difference, 2.71 points/yr), and math (mean difference, 2.96 points/yr), the models suggested an advantage for children treated in the AR group. This was not the case for reading (mean difference, 每0.91 points/yr). However, none of these differences were large enough to reach statistical significance (two-tailed t tests, all P .12). However, the magnitude of these changes do hold clinical significance. For example, over a 5-year period, the difference in spelling scores between the younger AR group and younger HR group could reach almost 15 points.
Impact of Clinical Events
We also explored the impact of shunt and posterior fossa syndrome on IQ development. First, differences in the frequency of these events were determined for the four age and risk groups. Then, analyses were conducted on IQ changes. There was no significant difference in the proportion of patients in the four age and risk groups with regard to shunting (P = .382; younger HR, 33%; younger AR, 33%; older HR, 15%; and older AR, 20%). Subsequent modeling of presence or absence of shunting revealed no significant impact on IQ change (P = .13).
Posterior fossa syndrome was defined by the postoperative appearance of pseudobulbar symptoms and mutism. There was a trend toward significance in the proportion of patients in the four age and risk groups (P = .077; younger HR, 38%; younger AR, 13%; older HR, 8%; and older AR, 18%), with the younger HR group having a higher incidence than the other age and risk groups. Subsequent modeling of presence or absence of posterior fossa syndrome revealed no significant impact on IQ change (P = .68).
DISCUSSION
The present study provides original, longitudinal neurocognitive data for patients treated for MB with postoperative risk-adapted CSI (23.4 v 36 to 39.6 Gy) and 55.8 Gy of conformal primary site boost using a 2-cm CTV margin before four cycles of dose-intensive chemotherapy. Notwithstanding this novel approach, for the group as a whole, statistically significant declines were observed for mean IQ, reading, spelling, and math. The overall magnitude of the negative IQ slope is commensurate with or somewhat smaller than the slopes previously reported with similar samples of MB patients,4,10-12 but these changes must now be interpreted with respect to the moderating influences of age and risk group that the present study demonstrates.
In terms of the initial hypothesis that AR patients would show less neurocognitive decline than HR patients, significant differences in the predicted direction were found for IQ when changes from baseline to 2 years after diagnosis were analyzed. These differences remained significant after adjusting for differences in age at diagnosis within the two risk groups. Furthermore, multivariate modeling revealed that the decline in IQ was significant for the HR patients but not for the AR patients. These results would seem to support the contention that the CSI dose reduction in the AR patients preserved neurocognitive function.
Because sufficient observations were available, modeling of the data for complex interactions was possible and makes a unique contribution to the current literature. Even among previous longitudinal studies that have examined age and radiation dose effects using multivariate methods, it has not been possible to identify interactions between risk group, age, and time.10-12 For two of the four neurocognitive outcomes (reading and spelling), significant interactions were found between risk group, age, and time. The presence of such interactions limits the value of simple interpretations of neurocognitive changes based only on differences in CSI between risk groups and underscores the need for multiple observations on large patient samples in future MB studies.
Several previous studies of patients treated for MB have shown young age to be a risk factor3,7,10-12; however, relatively few direct comparisons among young children treated with conventional-dose versus reduced-dose irradiation have been published. In the retrospective Pediatric Oncology Group series, which used conventional irradiation without chemotherapy, six or fewer patients were available in these two groups for analysis, necessitating nonparametric methods.9 Although the mean scores suggested a 10每to 15每IQ point advantage in favor of the reduced-dose group, the comparisons failed to achieve statistical significance. Furthermore, the most recent survival analysis in this cohort of patients treated with radiation therapy alone revealed an increased risk of early relapse and lower 5-year event free survival in the reduced-dose group.22 Although it is important to examine statistical significance, when sample sizes are small, effect size can also be relevant. A major finding of the present analysis was the failure to show a consistent benefit of the AR over the HR radiation therapy regimen. On examination of the slope magnitude or effect size, younger AR children did not show as much of a decline as younger HR children on measures of IQ, spelling, or math. However, for reading, this relationship was reversed. When statistically compared, differences between the slopes for the younger AR and HR groups did not reach significance for IQ, reading, and spelling.
A factor that may partially explain the difficulty in ascertaining neurocognitive differences among younger and, therefore, more vulnerable patients relates to radiation dosimetry. Intuitively, one would think that the 35% dose reduction from 36 to 23.4 Gy for the CSI dose would result in a commensurate reduction in brain volume exposed to higher doses of irradiation. However, as shown in Figure 3, the lower dose group nevertheless had 40% of total brain volume exposed to 36 Gy, and both regimens resulted in 20% of total brain volume exposure to 50 Gy or greater, largely because of the targeting guidelines used for this study and the minimum posterior fossa dose of 36 Gy.
The apparent increased vulnerability of reading and spelling skills among the younger children in both risk groups may be explained by impairments of fundamental cognitive processes early in language development (orthographic and phonologic analysis). Deficits in phonologic analysis among otherwise healthy children represent the most specific and reliable predictors of later dyslexia. And dyslexic children, compared with normal readers, display a disrupted pattern of activation on functional magnetic resonance imaging of posterior brain regions and parietotemporal and occipitotemporal areas.23 For MB patients in both risk groups in the present study, these brain regions received relatively large total doses of irradiation from the posterior fossa boost, which is known to affect white matter development and, more specifically, posterior corpus callosum morphology.24-27 Whether cognitive and behavioral or pharmacologic interventions can effectively treat these and other neurocognitive sequelae is not yet known, although data from the short-term use of methylphenidate seems promising.28,29
The present study has several methodologic limitations. For ethical reasons, patients were not randomly assigned to treatment arms because of the known need for more aggressive therapy among HR patients. Missing values is a common difficulty when examining clinical populations, and the present study is no exception, but the robustness of the mixed models to missing values eased this limitation. In addition, a potentially confounding variable in the analysis of neurocognitive changes is the incidence of other clinical events that can impact these outcomes in addition to age and risk group. Our analyses of the impact of shunting did not reveal either imbalances among age and risk groups or consistent associations with IQ loss. In contrast, the incidence of posterior fossa syndrome was disproportionately high among children in the younger HR group. Because of limited sample size, we could not formally define the impact of posterior fossa syndrome on neurocognitive development. One would expect the presence of posterior fossa syndrome to depress neurocognitive development, therefore enhancing differences in neurocognitive trajectories between the younger patients in the HR and AR groups. However, except for math achievement, significant differences were not observed.
The ability to decode the written word depends on many interdependent skills. Decoding begins with the capacity to notice and manipulate the smallest units of sound in spoken language that can make a difference in the meaning of a word (eg, pup or cup). These units of sounds are called phonemes. The ability to distinguish between each phoneme is considered the building block for all other language and reading-related skills. A critical task for the early reader is not only to properly distinguish various phonemes but also to recognize that these auditory segments are represented by the alphabet in written form. By understanding that each letter represents a sound and by then using the sounds to blend phonemes into a recognizable word, a person can successfully decode written language. With a disturbance in any one of the interdependent skills, children may experience a slowed rate of knowledge acquisition and greater difficulty with reading and/or spelling. The potential for treatment-induced hearing loss to impact these skills, especially as these skills are still developing in the younger child, should be considered. Although audiograms were regularly scheduled and appropriate adjustment to chemotherapy dose and/or hearing aid support was provided when defined loss of function was detected, hearing ability was not included in the current analyses and constitutes another potential limitation. It is believed that a more detailed examination of audiology data and cochlear radiation dose within the MB population is important to furthering our understanding of the development of phonologic processing and reading decoding ability after treatment.
Despite limiting the psychometric evaluations to three experienced institutions with a strong commitment to the treatment protocol, adherence to the testing schedule was problematic and presents another potential limitation. Even among children who maintained their response to treatment for 2 years or more, adherence generally varied between 50% and 60% at each interval. The problem with adherence, which is long recognized in pediatric cooperative trials, limits both the sensitivity of statistical analyses and the ability to generalize findings. In our design, we generally had the power to detect slopes differing from 0 or from each other by 3 points/yr or more. Future studies may want to focus on even smaller differences between critical groups of patients, such as patients who are younger at diagnosis. Such studies will need to carefully consider the number of patients and observations needed to reliably test hypotheses of interest.
Authors' Disclosures of Potential Conflicts of Interest
The authors indicated no potential conflicts of interest.
NOTES
Deceased.
This research is supported by Musicians Against Childhood Cancer, the Noyes Foundation, the American Lebanese Syrian Associated Charities, National Cancer Institute grant Nos. R01CA78957 and U01CA81445, and Cancer Center Support grant No. P30CA21765.
Authors' disclosures of potential conflicts of interest are found at the end of this article.
REFERENCES
Mostow EN, Byrne J, Connelly RR, et al: Quality of life in long-term survivors of CNS tumors of childhood and adolescence. J Clin Oncol 9:592-599, 1991
Ris MD, Noll RB: Long-term neurobehavioral outcome in pediatric brain-tumor patients: Review and methodological critique. J Clin Exp Neuropsychol 16:21-42, 1994
Dennis M, Spiegler BJ, Hetherington CR, et al: Neuropsychological sequelae of the treatment of children with medulloblastoma. J Neurooncol 29:91-101, 1996
Spiegler BJ, Bouffet E, Greenberg ML, et al: Change in neurocognitive functioning after treatment with cranial radiation in childhood. J Clin Oncol 22:706-713, 2004
Grill J, Renaux VK, Bulteau C, et al: Long-term intellectual outcome in children with posterior fossa tumors according to radiation doses and volumes. Int J Radiat Oncol Biol Phys 45:137-145, 1999
Copeland DR, DeMoor C, Moore BD, et al: Neurocognitive development of children after a cerebellar tumor in infancy: A longitudinal study. J Clin Oncol 17:3476-3486, 1999
Walter AW, Mulhern RK, Gajjar A, et al: Survival and neurodevelopmental outcome of young children with medulloblastoma at St. Jude Children's Research Hospital. J Clin Oncol 17:3720-3728, 1999
Strother D, Ashley D, Kellie SJ, et al: Feasibility of four consecutive high-dose chemotherapy cycles with stem-cell rescue for patients with newly diagnosed medulloblastoma or supratentorial PNET after craniospinal radiotherapy: Results of a collaborative study. J Clin Oncol 19:2696-2704, 2001
Mulhern RK, Kepner JL, Thomas PR, et al: Neuropsychologic functioning of survivors of childhood medulloblastoma randomized to receive conventional or reduced-dose craniospinal irradiation: A Pediatric Oncology Group study. J Clin Oncol 16:1723-1728, 1998
Ris MD, Packer R, Goldwein J, et al: Intellectual outcome after reduced-dose radiation therapy plus adjuvant chemotherapy for medulloblastoma: A Children's Cancer Group study. J Clin Oncol 19:3470-3476, 2001
Palmer SL, Goloubeva O, Reddick WE, et al: Patterns of intellectual development among survivors of pediatric medulloblastoma: A longitudinal analysis. J Clin Oncol 19:2302-2308, 2001
Palmer SL, Gajjar A, Reddick WE, et al: Predicting intellectual outcome among children treated with 35-40 Gy craniospinal irradiation for medulloblastoma. Neuropsychology 17:548-555, 2003
Wechsler D: Wechsler Preschool and Primary Scale of Intelligence每Revised. New York, NY, The Psychological Corporation, 1989
Wechsler D: Wechsler Intelligence Scale for Children (ed 3). New York, NY, The Psychological Corporation, 1991
Wechsler D: Wechsler Adult Intelligence Scale每Revised. New York, NY, The Psychological Corporation, 1989
Sattler JM: Assessment of Children. San Diego, CA, Jerome Sattler, 1992, p 1170
Mulhern RK, Ochs J, Fairclough D: Deterioration of intellect among children surviving leukemia: IQ test changes modify estimates of treatment toxicity. J Consult Clin Psychol 60:477-480, 1992
Jastak S, Wilkenson GS: Wide Range Achievement Test每Revised. Wilmington, DE, Jastak Associates, Inc, 1984
Psychological Corporation: Wechsler Individual Achievement Test. New York, NY, Harcourt, Brace, Jovanovich, 1992
Littell RC, Milliken GA, Stroup W, et al: SAS System for Mixed Models. Cary, NC, SAS Institute, Inc, 1996
Laird NM, Ware JH: Random-effects models for longitudinal data. Biometrics 38:963-974, 1982
Thomas PRM, Deutsch M, Kepner J, et al: Low-stage medulloblastoma: Final analysis of trial comparing standard-dose with reduced-dose neuraxis irradiation. J Clin Oncol 18:3004-3011, 2000
Shaywitz BA, Shaywitz SE, Pugh KR, et al: Disruption of posterior brain systems for reading in children with developmental dyslexia. Biol Psychiatry 52:101-110, 2002
Mulhern RK, Palmer SL, Reddick WE, et al: Risks of young age for selected neurocognitive deficits in medulloblastoma are associated with white matter loss. J Clin Oncol 19:472-479, 2001
Reddick WE, White H, Glass JO, et al: Developmental model relating white matter volume with neurocognitive deficits in pediatric brain tumor survivors. Cancer 97:2512-2519, 2003
Palmer SL, Reddick WE, Glass JO, et al: Decline in corpus callosum volume among pediatric patients with medulloblastoma: Longitudinal MR imaging study. Am J Neuroradiol 23:1088-1094, 2002
Mulhern RK, Reddick WE, Palmer SL, et al: Neurocognitive deficits in medulloblastoma survivors and white matter loss. Ann Neurol 46:834-841, 1999
Butler RW, Copeland DR: Attentional processes and their remediation in children treated for cancer: A literature review and the development of a therapeutic approach. J Int Neuropsychol Soc 8:115-124, 2002
Mulhern RK, Khan RB, Kaplan S, et al: Short-term efficacy of methylphenidate: A randomized, double-blind, placebo-controlled trial among survivors of childhood cancer. J Clin Oncol 22:4795-4803, 2004(Raymond K. Mulhern, Shawn)