1302

Clinical and Survival Covariates of Eight Classes of

Childhood Supratentorial Neuroglial Tumors
Floyd H. Gilles, M.D.1
Alan Leviton, M.D., S.M.2
C. Jane Tavare, M.S.1
Lester Adelman, M.D.3
Lucy B. Rorke, M.D.4
Eugene L. Sobel, Ph.D.5
E. Tessa Hedley-Whyte, M.D.6
Richard L. Davis, M.D.7
for The Childhood Brain Tumor
Consortium
1

Division of Neuropathology, Department of Pathology and Laboratory Medicine, Childrens Hospital Los Angeles and University of Southern California School of Medicine, Los Angeles, California.

Division of Neuroepidemiology, Childrens Hospital and Harvard Medical School, Boston, Massachusetts.

Neuropathology Laboratory, New England Medical Center Hospital and Tufts University, Boston,
Massachusetts.

BACKGROUND. In the current study, the authors investigated clinical, surgical, and
histologic characteristics (covariates) and their interactions in eight previously
identified classes of childhood supratentorial neuroglial tumors. The classes resulted from 5 factor score profiles on 703 supratentorial neuroglial tumors in the
Childhood Brain Tumor Consortium database.
METHODS. The Cox proportional models were used to identify class survival covariates.
RESULTS. Age was found to be a survival covariate only in Class 1, in which older
age increased the 5-year survival rate 73% from the first year (0.49) to the tenth year
(0.85). A greater amount of tumor removed improved survival in Classes 2 and 4
only. Rosenthal fibers improved survival in Class 2 and overrode the negative
effects of high Proliferative factor scores and pleomorphic nuclei. Survival for Class
3 children with high Proliferative factor scores improved from 0.60 to 0.95 as the
Spongy factor scores increased. Survival in Class 4 increased from 0.17 to 0.39 with
total tumor removal. Irregular nuclei and glomeruloid capillaries improved survival
in Class 5 patients. Class 6 survival improved with low cell density. Macrocysts in
tumors in Classes 1 and 5 were found to improve survival.
CONCLUSIONS. As a result of the current study, the authors conclude that survival
covariates differ with tumor class and may modify prognosis considerably. Cancer
2002;95:130210. 2002 American Cancer Society.
DOI 10.1002/cncr.10813

Division of Neuropathology, Childrens Hospital of

Philadelphia and University of Pennsylvania, Philadelphia, Pennsylvania.

Division of Biostatistics, Department of Preventive Medicine, University of Southern California

School of Medicine, Los Angeles, California.

Division of Neuropathology, Massachusetts General Hospital and Harvard Medical School, Boston,
Massachusetts.

Department of Pathology, University of California

San Francisco, School of Medicine, San Francisco,
California.
The Childhood Brain Tumor Consortium (Floyd H.
Gilles, principal investigator) is comprised of the
following institutions: Barnes and St. Louis Childrens Hospital, St. Louis, MO (William F. Blank
and Keith H. Fulling); Cardinal Glennon Memorial
Hospital, St. Louis, MO (John D. Blair [deceased]
and Daphne DeMello); Childrens Hospital, Boston,
MA (Floyd H. Gilles and Ana Sotrel); Childrens
Hospital of Denver, Denver, CO (Edmund N. Orsini,
Jr.); Childrens Hospital of Los Angeles, Los Angeles, CA (Hart Isaacs); Childrens Hospital of Philadelphia, Philadelphia, PA (Lucy B. Rorke); Childrens Hospital of Pittsburgh, Pittsburgh, PA
(Yoshie Hashida, Robert A. Price, Eduardo J. Yunis,
and Barbara W. Zaias); Hospital for Sick Children,

2002 American Cancer Society

KEYWORDS: brain tumor, child, supratentorial, survival.

t generally is accepted that survival is improved by certain covariates, such as greater amounts of tumor removed, older patient age,
and the abscence of pleomorphic nuclei, regardless of tumor type.
The objectives of the current study were to investigate this assump-

Toronto, Ontario, Canada (Larry E. Becker); Los

Angeles County-University of Southern California
Medical Center, Los Angeles, CA (Richard L.
Davis); and St. Christophers Hospital for Children,
Philadelphia, PA (Guillermo A. de Leon, Dale Huff,
Gladys Mestre, and Massoud Shamszadeh). The
following are slide readers for the Childhood Brain
Tumor Consortium: Lester S. Adelman (Tufts University Medical School, Boston, MA); R. Damon
Averill, Jr. (University of Michigan Medical School,
Ann Arbor, MI); Richard L. Davis (University of
California-San Francisco, San Francisco, CA); Umberto DeGirolami (University of Massachusetts
Medical School, Boston, MA); E. Tessa HedleyWhyte (Harvard Medical School, Boston, MA);
George M. Kleinman (Yale Medical School, University of Connecticut Medical School, New Haven,
CT); James H. Morris (Harvard Medical School,
Boston, MA); E. P. Richardson, Jr. (deceased) (Har-

vard Medical School, Boston, MA); Lucy B. Rorke

(University of Pennsylvania Medical School, Philadelphia, PA); William C. Schoene (Harvard Medical
School, Boston, MA); Thomas W. Smith (University
of Massachusetts Medical School, Boston, MA);
and Raymond A. Sobel (Harvard Medical School,
Boston, MA). Administrative/data collection staff:
Carol Kenney and Donna Morelli. Editorial/Data
Analysis Committee: Floyd H. Gilles, E. Tessa HedleyWhyte, Alan Leviton, Eugene Sobel, and C. Jane
Tavare.
Address for reprints: Floyd H. Gilles, M.D., Division of
Neuropathology, Childrens Hospital Los Angeles,
4650 Sunset Blvd., MS #43, Los Angeles, CA 90027;
Fax: (323) 669-4553; E-mail: FGILLES@hsc.usc.edu
Received January 16, 2002; accepted April 24,
2002.

Covariates of Survival/Gilles et al.

TABLE 1
Factors and Their Defining Histologic Features
Factor

Defining histologic features

Jumbo

Nuclei; large nucleoli; prominent cells; multinucleated; (giant)

nuclei; intermediate abnormal neurons
Astrocytes; stroma elongated nuclei; low-cell density
Mitosis; necrosis; very high-cell density
Vacuoles; microcysts; spongy and compact pattern
Calcification; parenchymal; oligodendroglial foci

1303

TABLE 2
Tumor Classes and Their Factor Score Profiles: Mean Factor Scores
Compared with High/Low Cutpoints
Class

Fibrillary
Proliferative
Spongy
Oligo

tion by identifying survival covariates within homogeneous tumor classes and to examine their interactions
on survival.
Childhood neuroglial tumors are comprised of a
large number of differing microscopic features. Various patterns of histologic features differentiate these
tumors one from another. Although each tumor type
(astrocytoma, anaplastic astrocytoma, and glioblastoma) has broad survival expectations, other characteristics also modify survival expectation such as tumor location, amount of tumor removed, age of the
patient, etc. Relatively histologically homogenous
groups of tumors are essential to identify and compare
survival covariates. However, childhood neuroglial tumors of the same World Health Organization (WHO)
diagnostic name or Daumas-Duport grade do not capture all the histologic information available in a tumor.
They contain histologic subsets that differ markedly in
survival expectation.1,2 These observations reflect the
histologic heterogeneity within each diagnosis or
grade. Previously, we have used methods for assessing
extended childhood brain tumor histology that statistically minimize histologic variance, (the variation in
histologic features from tumor to tumor) and provide
eight classes of supratentorial neuroglial tumors3 with
relative histologic homogeneity. These procedures can
quantitatively fine-tune the diagnostic groups provided by the WHO classificaiton, by identifying the
subsets within broad diagnostic categories with different survival expectations.
Our alternative strategy for grouping childrens
neuroglial tumors was based on 26 histologic features
identified as reliably ascertained in childhood supratentorial neuroglial tumors.4 We used factor analysis
to identify significant histologic feature associations
(factors)and to quantify their relationships. Five factors requiring only 17 of the 26 features were identified.5 The factor names of Jumbo, Fibrillary, Proliferative, Spongy, and Oligo reflected the histologic
features important to each factor (Table 1). The group
of histologic features important for one factor is not
correlated with the group of histologic features important for another factor. A tumors score on a factor is a

Factor

Jumbo
Fibrillary
Proliferative
Spongy
Oligo

H
H

Mean score () not significantly different from score cutpoint; H: mean score significantly (P 0.05)
higher than score cutpoint; L: mean score significantly (P 0.05) lower than score cutpoint.

weighted linear combination of the absence or presence of the 17 histologic features. We calculated a
score on each of the 5 factors for each of the 703
tumors. The five scores characterize and summarize
the histology of the tumor. Mixed tumors have high
scores on more than one factor. Approximately 67% of
childhood supratentorial neuroglial tumors are mixed
when evaluated in this fashion. We used these 703
tumor factor score profiles in cluster analyses to form
8 relatively histologically homogeneous classes of tumors.3 Each class had a distinct pattern of high or low
factor score profiles that reflect its unique histology
(Table 2). The factor score profile and class assignment calculations are available as worksheets that are
easily programmable. The calculations of the five factor scores and class allocation for each tumor are
rigorous and completely without bias or subjectivity,
once the absence or presence of the reliable histologic
features are known, following strict definitions.4 This
reproducibility contrasts with the poor intraobserver
reliability of the diagnoses in the WHO classification6
when used for childhood brain tumors.
The eight classes had significantly different survival distributions (Table 3) and many WHO diagnoses
contained subsets that fell into two or more classes
(Table 4), reflecting the different distributions of histologic features in the subsets. For example, Class 4
had high mean factor scores on the Jumbo and Proliferative factors (reflecting the presence in these tumors of large numbers of large cells and mitosis, necrosis, and very high cell density) whereas Class 2 did
not. Classes 2 and 4 contained the two subsets of
anaplastic astrocytomas and had significantly different survival distributions. The anaplastic astrocytomas in Class 2 had a much higher estimated 5-year
survival probability than those in Class 4 (0.70 vs.
0.17). Two of the 3 classes that contained subependymal giant cell astrocytomas had similar survival probabilities (Classes 1 and 6 with estimated 5-year sur-

1304

CANCER September 15, 2002 / Volume 95 / Number 6

TABLE 3
Estimated Survival Probability
Class
Survival

1 year
5 years

0.94
0.81

0.93
0.79

0.88
0.80

0.88
0.70

0.89
0.62

0.78
0.55

0.57
0.32

0.46
0.17

Groups of underlined Classes were found to have nonsignificant differences in estimated survival
distributions. Classes 3, 1, 6, and 2 had significantly different survival distributions than Classes 7, 5 and
4, and Classes 7 and 5 had significantly different survival distributions than Class 4 (P 0.01). Class 8
had a survival distribution that was significantly different than Classes 5 and 4. Children who survived
1 month were deleted from the survival analyses.

vival probabilities of 0.80), whereas the third had a

significantly lower estimated 5-year survival probability of 0.17 (Class 4). There were no significant differences with regard to the survival distributions for
classes containing the majority of pilocytic astrocytomas (Classes 1 and 2 with estimated 5-year survival
probabilities of 0.79 and 0.70), the other astrocytomas
(also included in Classes 1 and 2), oligodendrogliomas
(Classes 3 and 8 with estimated 5-year survival probabilities of 0.81 and 0.62), or ependymomas (Classes
5 and 7 with estimated 5-year survival probabilities of
0.32 and 0.55). Of the unclassifiable tumors (cases in
which the team of 2 readers agreed they could not
make any WHO diagnosis, 17.9% fell into Class 2,
16.1% into Class 4, and 46.4% into Class 5). Of the
ambiguous tumors (cases in which the team of 2 readers disagreed on a WHO diagnosis), 14% fell into Class
1, 43% into Class 2, and 23.7% into Class 5. Classes 1
and 2 had similar survival distributions; Class 5 was
significantly worse, and Class 4 had the worst survival
distribution among all supratentorial neuroglial tumors.

MATERIALS AND METHODS

The Childhood Brain Tumor Consortium (CBTC) database, which was presented in detail elsewhere,7 contains the CBTC WHO diagnosis as well as clinical,
surgical, tumor histology, treatment, and survival information for 3291 unselected children. To be included in the database, we required enough surgical
material to make a diagnosis. The number of slides
reviewed per case ranged from 1 to 22 slides, with a
median of 2.2 slides. Four teams of two neuropathologists reached consensus regarding the presence or
absence of a large number of histologic features in the
microscopic slides of each case. Another team came to
consensus regarding the appropriate WHO diagnosis.
The teams reviewed cases without clinical informa-

tion, primary diagnosis, or knowledge of whether the

case had been reviewed by them previously. The reliability judgments regarding individual histologic features or tumor diagnosis were based on the 25% of
cases reread by the same team.4,6,7 With a brain tumor
incidence of approximately 3 per 100,000 children per
year, our sample represents approximately 100,000,000
children-years of observation.8,9 The sample for the
current study was comprised of 703 children with
tumors confined to the supratentorial compartment
with complete information concerning 26 reliably
identified histologic features. We defined the caudal
anatomic limit of the supratentorial compartment as
the junction between the third ventricle and the aqueduct. We excluded from this analysis tumors that
involved both the infratentorial and supratentorial
compartments (n 120 tumors) or all three compartments (n 14 tumors) because of differences in associated clinical and histologic characteristics.10 Because the reliable histologic features in the
infratentorial and supratentorial compartments differed, the tumor factors, classes, and survival covariates apply only to neuroglial tumors in the supratentorial compartment. The operational definitions of
each of the histologic features were presented in Appendix 1 in our previous publication.4 The majority of
the survival covariates are self-explanatory; the estimation of the amount of tumor removed was the
surgeons estimate at the time of surgery. Because
magnetic resonance imaging (MRI) shows the amount
of proton density (or contrast enhancement) and not
the full extent of these often infiltrating tumors, we
used the surgeons estimate.
In the current study, we used the chi-square and
analysis of variance tests to evaluate whether the tumor class differences with regard to clinical, surgical,
and histologic characteristics achieved nominal statistical significance. We used Cox proportional hazards
models to identify tumor class survival covariates. For
each class, we allowed the factor scores and the seven
reliable histologic features unnecessary for the factors
into the model as covariates after first accounting for
associations with survival of the clinical features of
radiotherapy or rudimentary chemotherapy, amount
of tumor removed, patient age, decade of surgery,
location, and gender. Apart from the amount of tumor
removed, we did not include surgical observations in
our final model of survival expectation. This was because we made the assumption during data analysis
that if a surgical observation was absent from the
medical record it was not present in the tumor, which
may be incorrect, and because some classes were
small relative to the number of possible covariates. In
all survival analyses we excluded children who died

Covariates of Survival/Gilles et al.

1305

TABLE 4
Distribution of WHO Diagnoses by Class
Class
Row-percent
Diagnosis

Pilocytic astrocytoma
Subependymal giant cell astrocytoma
Anaplastic astrocytoma
Other astrocytoma
Oligodendroglioma
Ependymoma
Ganglioglioma
Glioblastoma
Giant cell glioblastoma
Medulloblastoma (PNET)
Unclassifiable
Ambiguous
Total no. of cases

44.4
13.3

46.0

24.5
53.7

29.6

17.9
49.0
1.5

6.6
35.6

60.7
1.5

3.7

2.2

45.2
71.4
78.6
17.9
43.0
218

14.0
128

16.1
45

93

1.5
28.9

25.8

14.3

76.2
14.3

73.3
46.4
23.7
99

58

40

22

Total
124
28
102
136
45
62
21
7
14
15
56
93
703

WHO: World Health Organization; PNET: primitive neuroectodermal tumor. The cutoff for a diagnosis to be included with a Class percent was a frequency of at least 10.

TABLE 5
Prevalence of Clinical and Surgeon Assessment Variables
Class
Variable
Location
Superficial
Radiation and/or chemotherapy
Gender:
Male
Age (yrs)
Mean
(SD)
Median
Surgeons assessment
Amount removed
Biopsy
Partial
Total

P value

84
46

891
49

692
332

76
63

77
721

83
292

80
43

682
55

0.003
0.001

48

52

36

55

53

53

60

55

NS

7.7
(4.5)
7.6

7.0
(4.8)
5.7

10.61
(4.9)
10.3

8.3
(4.9)
8.4

6.4
(4.2)
5.5

9.9
(4.9)
10.6

6.12
(4.2)
5.8

8.3
(5.1)
7.9

0.0001

20
63
17

351
51
142

122
59
29

18
69
132

20
56
24

152
53
331

162
53
32

52
58
371

0.001

NS: not significant; P-value: significance of overall difference; SD: standard deviation.
The arrows indicate a significantly (P 0.05) increased (1) or decreased (2) finding for the class.

within 1 month of surgery to avoid including deaths

from surgical complications. All reported significant
differences between clinical and histologic variables
and survival distributions were based on a P value of
0.01. A P value of 0.05 was used to identify
covariates with a significant association with survival
estimation.

RESULTS
There were no significant differences with regard to
male to female gender ratios among the classes (Table

5). Patients in two classes (Classes 3 and 6) were less

likely to have received auxiliary radiation therapy or
chemotherapy (33% and 29%, respectively), and those
in Class 5 were more likely to have received auxiliary
radiation therapy (72%). Approximately 40 55% of the
patients in the remaining classes received auxiliary
radiation therapy. Children in Class 3 were significantly older and children in Class 6 were significantly
younger. The surgeons estimate of total removal of
tumor was more likely in Classes 6 and 8, but nevertheless, it was reported to occur in only 30 40% of the

1306

CANCER September 15, 2002 / Volume 95 / Number 6

TABLE 6
Influence of Selected Covariate Values on Estimated Survival
Covariates

Estimated survival

Class

Age (yrs)

1 yr

5 yrs

1
5
10

0.81
0.91
0.95

0.49
0.71
0.85

0.83
0.92
0.90
0.87
0.94
0.77
0.94

0.62
0.81
0.76
0.70
0.86
0.53
0.85

0.84
0.98

0.60
0.95

0.18
0.74
0.85

0.01
0.39
0.61

0.75
0.89
0.49
0.75

0.48
0.75
0.17
0.49

0.75
0.97

0.60
0.94

Amount removed

Proliferative score

Rosenthal fibers

Pleomorphic nuclei

Biopsy
Partial
Partial
Partial
Total
Biopsy
Partial

Mean
Mean SD
Mean
Mean SD
Mean
Mean
Mean SD

Proliferative score

Spongy score

Mean SD
Mean SD

Mean SD
Mean SD

Amount removed

Proliferative score

Biopsy
Total
Total

Mean
Mean
Mean SD

Irregular nuclei

Glomeruloid capillaries

Very low cell density

SD: standard deviation: , negative; , positive.

cases in each class. Total tumor removal was less likely

in Classes 2 and 4. There were no significant differences with regard to prevalence among classes in tumor colors gray/white (range, 19 41%) or purple/red
(range, 1738%) or the surgeons ability to distinguish
margins (range, 1227%); whether the tumors were
firm (range, 18 33%), vascular (range, 229%), friable
(range, 310%), or contained trabeculae (range,
3 6%); or whether there was a ventricular attachment
(range, 1129%). Class 2 tumors tended to be located
superficially and the tumors in Classes 3 and 8 tended
to be located deep in the brain. Leptomeningeal deposits were found at the time of surgery in 10% of
cases in all classes.

In six of the eight classes, additional clinical, surgical, or histologic characteristics changed survival expectancy from the average for the class (Table 6). The
survival covariates varied from class to class, had opposite effects in some classes, and interacted in occasionally surprising ways.
Class 1 was the only class in which the age of the
child made a significant difference (P 0.0006) to the
survival of the child. The overall 5-year survival expectation for children with a Class 1 tumor was 0.81. If the
child was in the first year of life (0 years), the 5-year
expected survival was 0.49 compared with a child age
10 years with the same tumor, in whom the survival
expectancy was 0.85 (Table 6) (Fig. 1). These tumors

Covariates of Survival/Gilles et al.

1307

FIGURE 1. Class 1 survival by patient age at the time of first surgery. This

FIGURE 2. Class 3 survival by Proliferative and Spongy factor scores. The

is the only class in which the age of the child was a covariate. If the child was
in the first year of life, the 5-year survival expectancy was 0.49 (lowest curve);
for the same tumor type, the 5-year survival expectancy for a child age 10
years was 0.85 (upper curve).

overall class 5-year survival estimate was 0.81. In this class, higher Spongy
factor scores improved survival expectancy whereas higher Proliferative factor
scores worsened survival expectancy. However, if the Proliferative factor
scores were held at the mean for the class plus 1 standard deviation (SD), the
5-year survival expectancy increased significantly as the Spongy factor scores
increased. Interactions among covariates, such as in this example, are important in estimating survival expectancy using quantitative histologic methods.

had significantly higher percentages of macrocysts

(53%) and were macroscopically yellow or tan in color
(39%). Classes 1 and 2 contained the majority of pilocytic astrocytomas (approximately 45% in each class)
(those in Class 1 containing regions of spongy and
compact pattern; those in Class 2 without this histologic feature). Class 1 also included approximately
15% of the subependymal giant cell astrocytomas and
30% of the grouped other astrocytomas.
Class 2 tumors were astroglial tumors without significantly high or low factor scores for any factor. In
addition to the pilocytic astrocytomas, Class 2 also
contained 25% of the anaplastic astrocytomas, 50%
of the other astrocytomas, and nearly 20% of the
unclassifiable tumors. For children with Class 2 tumors, there were four survival covariates. Increased
amounts of tumor removed and Rosenthal fibers were
reported to improve survival. The overall 5-year survival expectancy for this class was 0.70. If these tumors
had been biopsied only and had the class mean Proliferative score with neither Rosenthal fibers nor pleomorphic nuclei present, the 5-year survival expectation was 0.62 (Table 6). If the surgeon believed the
same tumor had been partially removed, the 5-year
survival improved to 0.76, and if the surgeon believed
the tumor had been removed totally, the 5-year survival was 0.86. Higher scores on the Proliferative factor
(P 0.0163) and the presence of pleomorphic nuclei
(P 0.0467) worsened survival expectancy. Adding
pleomorphic nuclei to the original tumor reduced the
5-year survival expectancy to 0.53. Holding the
amount of tumor removed at partial removal and both
histologic features absent, increasing scores on the
Proliferative factor from the class mean Proliferative
score less 1 standard deviation to the class mean score

plus 1 standard deviation reduced the 5-year survival

expectancy from 0.81 to 0.70. Surprisingly, the presence of Rosenthal fibers, even with high Proliferative
scores and pleomorphic nuclei, overrode the negative
effects of the latter on survival expectancy and elevated survival expectation to 0.85. This phenomenon demonstrates the interaction of survival covariates and the importance of recognizing the presence
of individual covariates (in this case, histologic features) within a tumor class or group. Only approximately 25% of Class 2 tumors were found to contain
macrocysts at the time of surgery.
Many of the tumors in Class 3 were oligodendrogliomas, and high scores on the Oligo and Proliferative
factors characterized these tumors. The class 5-year
survival estimate was 0.81. In this class, higher scores
on the Spongy factor (P 0.0133) improved survival
expectancy whereas high Proliferative factor scores (P
0.0495) worsened survival. For instance, the 5-year
survival expectancy for Class 3 children with high
scores on the Proliferative factor improved from 0.60
to 0.95 as the score on the Spongy factor increased
(Fig. 2). They also were more frequently yellow or tan
in color (42%) and gritty or calcified (20%).
The overall 5-year survival expectancy for children
with Class 4 tumors was 0.17. Class 4 tumors, characterized by high scores for the Jumbo and Proliferative
factors, included approximately 50% of the anaplastic
astrocytomas and all the glioblastomas. More tumor
removed, low Proliferative factor scores, and the absence of pleomorphic nuclei increased survival (P values of 0.0037, 0.0001, and 0.0009, respectively). For

1308

CANCER September 15, 2002 / Volume 95 / Number 6

FIGURE 3. Class 4 survival by amount of tumor removed and Proliferative

score. The overall class 5-year survival estimate was 0.17. In the lower two
curves, the score for the Proliferative factor was held at the mean for the class
and illustrates the influence of biopsy (lower curve) and total tumor removal
(middle curve) on survival expectancy (5-year survival estimates of 0.01 and
0.39, respectively). The upper curve demonstrates the effect on survival of the
combination of an estimated total tumor removal and a low Proliferative factor
score (5-year survival expectancy of 0.61). SD: standard deviation.

example, holding the Proliferative factor score at the

Class 4 mean, the 5-year survival probability increased
from 0.01 for biopsy only to 0.39 for children in whom
the surgeon believed that the tumor was removed
totally. Conversely, for any amount of tumor removed,
as the score for the Proliferative factor increased, survival probability decreased. In fact, for tumors believed to be removed totally, if the score for Proliferative factor was low, the 5-year survival probability
was 0.61 (Fig. 3). Class 4 tumors had a higher percentage of soft tumors (39%).
The overall 5-year survival expectancy for children
with Class 5 tumors was 0.32. Low scores on the Fibrillary factor characterized this tumor class. Large proportions of the ependymomas and unclassifiable tumors as well as the majority of the primitive
neuroectodermal tumors (medulloblastomas) fell into
this class. Irregular nuclei (P 0.00029) and glomeruloid capillaries (P 0.0234) were found to be associated significantly with survival. If neither irregular nuclei nor glomeruloid capillaries were present in the
tumor, the 5-year survival expectancy improved to
0.48. If just irregular nuclei were present without glomeruloid capillaries, then survival expectancy fell to
0.17. If just glomeruloid capillaries were present, then
survival expectancy improved to 0.75. If both histologic features were present, the glomeruloid capillaries balanced the survival effect of irregular nuclei and
the 5-year survival expectancy returned to 0.49. Class
5 tumors had a significantly lower percent of gritty or
calcified tumors (2%).
The overall 5-year survival expectancy of children
with Class 6 tumors was 0.80. Class 6 tumors included

FIGURE 4.

Class 6 survival by low cell density. The overall class 5-year

survival estimate was 0.80. If the histologic feature of low cell density was
absent in a tumor in this class, the survival expectancy decreased significantly
(lower curve) to 0.60. If the same histologic feature was present, the survival
expectancy increased to 0.94 (upper curve).

the majority of the subependymal giant cell astrocytomas and gangliogliomas. High scores on the Jumbo
factor characterized this class. Very low cell density
was identified as a covariate (P 0.0083). The 5-year
survival expectancy improved to 0.94 if very low cell
density was present (Fig. 4).
Classes 7 and 8 had no survival covariates in the
final model. Class 7 tumors had high mean scores on
Proliferative and Oligo factors, occurred in younger
children, and were less likely to have been biopsied
only. The overall 5-year survival expectancy for Class 7
tumors was 0.55. Class 8 tumors differed from Class 3
tumors by having a lower mean score on the Fibrillary
factor and an average score of zero on the Spongy
factor. Similar to Class 3 tumors, tumors in Class 8
were more likely to be gritty or calcified (18%). A
tumor with oligodendroglial foci, parenchymal calcification, astrocytes, and stroma would be assigned to
Class 8. If this tumor had the additional features of
necrosis, intermediate-size nuclei, and mitosis, it
would fall into Class 7. Class 8 tumors were less likely
than other tumors to have been located superficially,
and were more likely to have been totally removed.
Approximately 29% of oligodendrogliomas fell into
Class 8.

DISCUSSION
The results of the current study demonstrate that generalizations regarding survival covariates of childhood
neuroglial tumors (e.g., a greater amount of tumor
removed improves survival, older patient age improves survival, or the presence of pleomorphic nuclei
in a tumor worsens survival) do not apply to all statistically identified classes of supratentorial neuroglial

Covariates of Survival/Gilles et al.

tumors. The classes resulted from clustering each tumors extended histologic profile quantitatively represented in five uncorrelated factor scores. The calculation of the factor scores and class allocation
incorporate the information concerning all reliable
features in a reproducible and unbiased manner using
easily programmable worksheets3 or an automated
spreadsheet.
Age, a frequently cited survival covariate for adult
neuroglial tumors (e.g., in studies by Kowalczuk et al.
and Levin et al.11,12) and for children,13 was found to
apply only to Class 1 tumors, approximately 50% of
the pilocytic astrocytomas (those with spongy and
compact regions), and approximately 25% of the anaplastic astrocytomas. Older age at surgery in children
with Class 1 neuroglial tumors resulted in a 73% increase in survival expectancy at 5 years after first surgery from the first to the tenth year. Conversely, older
age at surgery was not found to be a survival covariate
for the remainder of the pilocytic astrocytomas (those
without spongy and compact regions), remaining anaplastic astrocytomas, or other supratentorial tumors
in the remaining seven classes (specifically, the glioblastomas or the primitive neuroectodermal tumors
[medulloblastomas]).
Another survival covariate that is cited frequently
is the amount of tumor removed (e.g., in studies by
Levin et al., Hess, and Groves et al.14 16). In children
with supratentorial neuroglial tumors, the amount of
tumor removed was found to be an important covariate in only two classes, Classes 2 and 4 (but not in
Class 1) (Table 6). This observation demonstrates that
the probability of survival for the two varieties of pilocytic astrocytomas found in Classes 1 and 2 is improved by greater tumor removal for only the second
variety. Greater removal of supratentorial juvenile
pilocytic astrocytomas (those with alternating spongy
and compact regions found in Class 1) did not appear
to improve survival probability but greater removal of
the other variety of pilocytic astrocytomas (those without alternating spongy and compact regions in Class
2) did appear to improve survival probability. Thus,
only for the class containing the more malignant tumors (Class 4) and another class containing some of
the more benign tumors (Class 2) was additional tumor removal found to be a significant factor. The
amount of tumor removed is, of course, a surgical and
clinical decision depending on many considerations,
including tumor location and surgical findings. These
findings do not, of course, assist the surgeon in any
way in deciding at the time of surgery whether to
attempt to remove additional tumor. That decision is
dictated by the biology of the tumor as revealed at the
time of surgery. Furthermore, the histologic factoring

1309

and grouping of the tumors into classes requires permanent sections, not the intraoperative frozen sections.
Individual factors and individual histologic features also had an effect on survival. Higher scores on
Proliferative factor were associated with decreased
survival probability in Classes 2, 3, and 4, whereas
higher scores on Spongy factor increased the survival
probability in Class 3. Individual reliable histologic
features, not represented in the factors, were important in survival probability models of several classes;
the presence of Rosenthal fibers in Class 2, pleomorphic nuclei in Class 4, glomeruloid capillaries in Class
5, and very low cell density in Class 6 were associated
with improved survival probability. Pleomorphic nuclei in Class 2 and irregular nuclei in Class 5 were
associated with worsened survival. The presence of
pleomorphic nuclei had opposite effects in two classes
(improved survival probability in Class 4 and decreased survival probability in Class 2). Deep or superficial location within the hemisphere, treatment
with radiation therapy or rudimentary chemotherapy,
decade of surgery, and gender were not found to be
associated with changes in survival probability.
These observations appear to extend the usefulness of the WHO definitions. They emphasize the importance of regarding neuroglial neoplastic processes
as containing many different histologic patterns, some
of which occur commonly in different kinds of childhood brain tumors. Only by quantifying the full histologic content of brain tumors and using the measured
relations among their histologic features can we recognize groupings of histologic features that are important for prognosis and identify associated survival covariates. We have shown that there are two or more
varieties of many of the supratentorial neuroglial tumors, and that for these varieties of tumor, different
covariates are associated with survival, (e.g., that maximal tumor removal of some pilocytic astrocytomas
improves survival expectation, but not all). We also
have shown that the same histologic features may
have opposite effects on survival expectation in the
varieties of anaplastic astrocytoma that fall into different classes.
The reasons why tumors of the same name, but
different classes, often differ with regard to survival
covariates are complex. They are in different classes
because they differ with regard to their extended histologic content (used in the current series of studies)
that is not reflected in the conventional diagnosis.2
Differing extended histologic content among tumors
of the same name is important to measure because it
can result in the assignment to classes that may differ
with regard to estimated survival distributions and

1310

CANCER September 15, 2002 / Volume 95 / Number 6

survival covariates. For instance, the anaplastic astrocytomas fell into two different classes with markedly different survival expectations (0.17 in Class 4 and
0.70 in Class 2 at 5 years after first surgery). The results
of the current study demonstrate that neuroglial tumors of the same name in the supratentorial compartment in children are not histologically or biologically
homogeneous, but often are divisible into subclasses.
Conversely, tumors with different WHO diagnoses
may fall into the same class because they contain
common patterns of histologic features other than the
few features used for naming the tumor type.
Histology and tumor locations remain the dominant variables influencing survival.17,18 After the supratentorial neuroglial tumors in the current study
were histologically grouped statistically based on a
relatively large number of reliable features, only a few
covariates remained significant. The most important
observations are that the covariates were important
for only some of the classes, had opposite effects in
some of the classes, and interacted in surprising ways
in some of the classes. We suggest that these tumor
classes, identified by factor and cluster analyses on a
relatively large number of reliable features, provide
the researcher with an improved method for comparing therapies in clinical trials of childhood tumors in
the supratentorial compartment.

6.

7.

8.
9.

10.

11.

12.

13.

14.

REFERENCES
1.

2.

3.

4.

5.

Brown WD, Gilles FH, Tavare CJ, et al. Prognostic limitations

of the Daumas-Duport grading scheme in childhood supratentorial astroglial tumors. J Neuropathol Exp Neurol. 1998;
57:10351040.
Gilles F, Brown W, Leviton A, et al. Limitations of the World
Health Organization classification of childhood supratentorial astrocytic tumors. Cancer. 2000;88:14771483.
Gilles F, Leviton A, Tavare C, et al. Definitive classes of
childhood supratentorial neuroglial tumors. Pediatr Dev
Pathol. 2000;3:126 139.
Gilles FH, Sobel EL, Leviton A, Tavare CJ, Hedley-Whyte ET.
Histologic feature reliability in childhood neural tumors.
J Neuropathol Exp Neurol. 1994;53:559 571.
Gilles FH, Sobel EL, Leviton A, Tavare CJ, Hedley-Whyte ET.
Quantitative histologic factors for grouping childhood supratentorial neuroglial tumors. Pediatr Pathol Lab Med.
1997;17:729 754.

15.
16.

17.

18.

Childhood Brain Tumor Consortium. Intraobserver reproducibility in assigning brain tumors to classes in the World
Health Organization diagnostic scheme. J Neurooncol. 1989;
7:211224.
Childhood Brain Tumor Consortium. A study of childhood
brain tumors based on surgical biopsies from ten North
American institutions: sample description. J Neurooncol.
1988;6:9 23.
Young JL Jr., Miller RW. Incidence of malignant tumors in
US children. J Pediatr. 1975;86:254 258.
Leviton A. Principles of epidemiology. In: Cohen ME, Duffner PK, editors. Brain tumors in children: principles of
diagnosis and treatment. New York: Raven Press, Ltd., 1994:
27 49.
Gilles FH, Leviton A, Hedley-Whyte ET, et al. Childhood
brain tumors that occupy more than one compartment at
presentation. Multiple compartment tumors. J Neurooncol.
1992;14:4556.
Kowalczuk A, Macdonald RL, Amidei C, et al. Quantitative
imaging study of extent of surgical resection and prognosis of
malignant astrocytomas. Neurosurgery. 1997;41:10281036;
discussion, 3638.
Levin VA, Prados MR, Wara WM, et al. Radiation therapy
and bromodeoxyuridine chemotherapy followed by procarbazine, lomustine, and vincristine for the treatment of anaplastic gliomas. Int J Radiat Oncol Biol Phys. 1995;32:75 83.
Levin VA, Edwards MS, Wara WM, Allen J, Ortega J, Vestnys
P. 5-fluorouracil and 1-(2-chloroethyl)-3-cyclohexyl-1-nitrosourea (CCNU) followed by hydroxyurea, misonidazole, and
irradiation for brain stem gliomas: a pilot study of the Brain
Tumor Research Center and the Childrens Cancer Group.
Neurosurgery. 1984;14:679 681.
Levin VA, Maor MH, Thall PF, et al. Phase II study of accelerated fractionation radiation therapy with carboplatin followed by vincristine chemotherapy for the treatment of
glioblastoma multiforme. Int J Radiat Oncol Biol Phys. 1995;
33:357364.
Hess KR. Extent of resection as a prognostic variable in the
treatment of gliomas. J Neurooncol. 1999;42:227231.
Groves MD, Maor MH, Meyers C, et al. A phase II trial of
high-dose bromodeoxyuridine with accelerated fractionation radiotherapy followed by procarbazine, lomustine,
and vincristine for glioblastoma multiforme. Int J Radiat
Oncol Biol Phys. 1999;45:127135.
Tooth HH. Some observations on the growth and survival
period of intracranial tumours based on the review of 500
cases, with special reference to the pathology of gliomata.
Brain. 1912;35:61108.
Wong ET, Hess KR, Gleason MJ, et al. Outcomes and prognostic factors in recurrent glioma patients enrolled onto
phase II clinical trials. J Clin Oncol. 1999;17:2572.