Prognostic Factors in Childhood T-cell Acute Lymphoblastic Leukemia:
A Pediatric Oncology Group Study
By Jonathan J. Shuster, John M. Falletta, D. Jeanette Pullen, William M. Crist, G. Bennett Humphrey, Barry L. Dowell,
Moody D. Wharam, and Michael Borowitz
Two hundred fifty-three children with newly diagnosed
T-cell acute lymphoblastic leukemia (ALL), who were
treated uniformly with modified LSA,L,
therapy, were
evaluated using univariate and recursive partition analyses
t o define clinical or biologic features associated with risk of
treatment failure. Overall event-free survival (EFS) a t 4
years was 43% (SE = 4%). Factors examined included
white blood cell (WBC) level, age, gender, race (black Y
other), presence of a mediastinal mass, hepatomegaly,
splenomegaly, marked lymphadenopathy, hemoglobin level,
platelet count, blast cell expression of antigens such as the
common acute lymphoblastic leukemia antigen (CALLA,
CD10). HLA-DR, and T-cell-associated antigens (CD3, CD4,
CD8, CD7, CD5, and THY). Univariate analysis showed that
age 5 5 or 5 7 years, WBC level 4 0 , 4 5 , 450 or 4 0 0 x
lo3/& and blast cell expression of CD4, CD8, or CALLA
were associated w i t h significantly better EFS, while
hepatomegaly and splenomegaly were associated with
worse EFS. Recursive partitioning analysis showed that the
most important single favorable prognostic factor was a
WBC level t 5 0 x 1 0 3 / ~ Land, for patients with WBC
counts below this level, the most important predictor of
EFS was blast cell expression of the pan-T antigen defined
by the monoclonal antibody (MoAb), L17F12 (CD5). For
patients with higher WBC levels, the most important
predictor of EFS was blast cell expression of THY antigen.
The recursive partitioning analysis defined three groups of
patients with widely varied prognoses identified as follows:
(1) those with a WBC count <50 x 1 0 3 / ~ Lwho lacked
massive splenomegaly and had blasts expressing CD5 had
the best prognosis (66%, SE = 7%. EFS 4 years, n = 84):
(2) those with ( b l ) WBC counts <50 x 103/pL with either
massive splenomegaly or who had blasts lacking CD5
expression, or (b2) WBC counts =.50 x 103/pL with
expression of the THY antigen had an intermediate prognosis (39%. SE = 7% EFS at 4 years, n = 94); (3)
those with
WBC counts >50 x 103/pL and whose blasts lacked
expression of THY antigen had the poorest outcome
(EFS = 19% at 4 years, SE = 8%. n = 63). A three-way
comparison of EFS according t o these groupings showed
significant differences among the three patient groups
(P< .001). The recursive partitioning was able t o classify
241 (95%) of the patients. Twelve patients could not be
grouped because of missing data. We conclude that this
prognostic factor analysis provides information of potential
importance for design of randomized clinical trials for
children with T-cell ALL (T-ALL), because appropriate
stratification depends on results of such analysis t o ensure
comparability of patient groups. In this study, recursive
partitioning analysis provides prognostic insight that was
not obtained by the stepwise Cox regression method.
However, these results should be considered tentative
until confirmed in a prospective clinical trial.
0 1990 by The American Society of Hematology.
T
children with T-ALL who have different clinical and biologic
features at diagnosis. Appropriate studies have been performed in children with B-cell progenitor malignancies
during the past 10 years, and have provided important
information regarding the prognostic strength of these features (white blood cell [WBC] level, age, gender, stage of cell
differentiation, genetic features, etc). 1-4,6-10 Information from
these studies has been valuable in improving the comparability of patient groups for clinical trials. Also, the identification
of groups of children with widely different prognoses has led
to tailored therapy designed to improve the therapeutic
index.
In part to address this need for information regarding
estimated prognosis in subgroups of children with T-ALL,
the Pediatric Oncology Group (POG) treated a relatively
large group (n = 253) of children registered between May
1981 and January 1986 with serologically well-defined
T-ALL uniformly on therapy modeled after the LSA,L,.".'*
Retrospective analysis of clinical and biologic features of
these children has provided important new information
regarding prognostic factors in T-ALL and forms the basis
for this report.
-CELL LEUKEMIA is relatively rare in children,
accounting for approximately 15% of newly diagnosed
cases of acute lymphoblastic leukemia (ALL)'.' Until recently, systematic immunologic typing of large cohorts of
children with ALL has not been undertaken; therefore,
children with T-cell ALL (T-ALL) have often not been
identified. Also, treatment of such children has varied widely
and generally has been given without regard to phenotype,
further hindering attempts to delineate prognostic factors in
From the University of Florida, POG Statistical Ofice, Gainesville; Duke University Medical Center, Durham, NC; the University
of Mississippi. Jackson; St Jude Children's Research Hospital and
the University of Tennessee, Memphis; University of Groningen,
The Netherlands; Abbott Laboratories, Abbott Park, IL; and Johns
Hopkins University School of Medicine, Baltimore, MD.
Submitted January 25, 1989; accepted August 25. 1989.
Supported in part by Grantsfrom the National Cancer Institute,
National Institutes of Health CA-29139. CA-15525. CA-15898,
CA-31566. CA-21765 (Core). CA-28476, CA-30969, and by the
American Lebanese Syrian Associated Charities (ALSAC).
Address reprint requests to Jonathan J. Shuster, PhD. (7837).
Pediatric Oncology Group Operations Ofice, 4949 W Pine Blvd.
Suite 2A. Si Louis, MO 63108.
The publication costs of this article were defrayed in part by page
charge payment. This article must therefore be hereby marked
"advertisement" in accordance with 18 U.S.C.section 1734 solely to
indicate this fact.
0 1990 by The American Society ojHematology.
0006-4971/90/7501-OOO2$3.00/0
166
MATERIALS AND METHODS
Patients and treatment. Bone marrow samples from all untreated patients with ALL (n = 2,022) entered on the POG classification protocol (POG 8035) were sent to reference laboratories of
the POG for immunologic marker studies to define the major
immunophenotypes of ALL (early pre-B, pre-B, T, and B cell) using
previously published
For 496 patients with ALL, cell
Blood, Vol 75, No 1 (January 1). 1990: pp 166-173
167
PROGNOSTIC FACTORS IN CHILDHOOD T-CELL ALL
marker assessment was not adequate to permit definitive assignment
to a pre-B, B, T or early pre-B group (see Definitions section); hence,
these subjects were excluded from this analysis, leaving 1,526 with
adequate marker studies for further consideration. With the exception of children enrolled in pilot studies, all patients with non-B,
non-T-ALL were entered on POG treatment protocols according to
immunophenotype ( B T; or non-B, non-T). Two hundred fifty-three
eligible children had T-ALL and were treated uniformly on POG
7837 (Fig 1). General details of this treatment regimen have, for the
most part, been previously published." However, further modifications were made in this study as detailed in Fig 1. Informed consent
was obtained from parents, patients, or both, as deemed appropriate
for both treatment and laboratory studies. In summary, 253 patients
were used in the analysis, with 192 having complete data on all
covariates.
Laboratory evaluation. Leukemic cells were examined morphologically to establish the diagnosis of acute leukemia. In addition,
cells were examined for cytochemical reactivity to Sudan black B
and nonspecific esterase to exclude myeloid leukemia. Sheep erythrocyte receptors were also assayed with use of established and
T.I.T.
T.I.T.
I 1111
VCR
INDUCTION
T.I.T.
T.I.T.
I
I
BCNU
CONSOLIDATION
MAINTENANCE
Fig 1. Schema for modified LSA,L, therapy (POG protocol
7837). Remission induction: Cyclophosphamide(CP) 1,200 mg/mz
intravenously (IV) on day 1; vincristine (VCR) 2 mg/m2 (maximum
2 mg) IV weekly for four doses beginning on day 3 or 4 prednisone
(PRED) 60 mg/m2 (maximum 60 mg) PO beginning with VCR and
continuing for 28 days with 7 days of decremental dosage:
daunorubicin (DNR) 60 mg/m2 IV on 2 successive days beginning 2
weeks after CP. In patients with high WBC, CP may be delayed for
1 to 4 days, and VCR end PRED may be started on day 1.
Consolidation: Cytosine arabinoside (ARA-C) 100 mg/m' IV or
intramuscularly (IM) for 5 consecutive days of each of 4 weeks;
thioguanine (TG) 50 mg/m2 PO 8 to 12 hours after each dose of
ARA-C: L-asparaginase (A-ASE) 6,000 U/m2 IV or IM daily for 14
days; bis-nitrosuree (BCNU) 60 mg/mz IV within 3 days after the
last dose of A-ASE. Treatment is delayed until ANC >l,OOo/pL.
Maintenance (bday cycles with 7- to 10-day intervals between
cycles): Cycle I. TG 300 mg/m2 PO daily for 4 days followed by CP
600mg/m2 IV on day 5: cycle II. hydroxyurea (HU) 2,400mg/m2 PO
daily for 4 days followed by DNR 45 mg/m2 IV on day 5 (when
maximum DNR dose of 480 mg/m2 is reached, substitute CP. 600
mg/m2 IV on day 5); cycle 111, methotrexate (MTX) 10 mg/mz PO
daily for 4 days followed by BCNU 60 mg/m2 IV on day 5; cycle IV,
ARA-C 150 mg/m' IV for 4 days followed by VCR 2 mg/m2 IV
(maximum 2 mg) on day 5. Cycles are repeated in this order
throughout maintenance therapy; cycles I through IV-one course.
All patients received cranial irradiation (2.400 rad) and triple
intrathecal therapy (hydrocortisone 16 mg/m2, methotrexate 15
mglm', and cytosine 30 mg/m2) at the times indicated during the 3
years of treatment.
quality-controlled techniques at member institution^?^'^*'^ Cells were
stained for surface immunoglobulins (SIg) and centrifuged onto
glass slides.I6After fixation in cold 95% ethanol and 5% acetic acid,
they were washed and mailed submerged in phosphate-buffered
saline to the University of Alabama at Birmingham's Immunology
Reference Laboratory for cytoplasmic p (Cp) and SIg testing.
Lymphocyte-associated antigens were identified at Duke University's Immunology Reference Laboratory (Durham, NC) by microcytotoxicity assays using heteroantisera (pT, THY, DR, and CALLA)
and monoclonal antibodies (MoAbs) (3A1 [CD7], 17F12 [CD5],
OKT3 [CD3], OKT4 [CD4], OKT8 [CDS], CDlO [CALLA, J51,
and HLA-DR) as previously described.''.'*
Definitions. All cases were classified as pre-B, B, T, or early
pre-B ALL according to the following criteria: pre-B if more than
10% of marrow lymphoblasts contained Cw; B if more than 10% of
marrow lymphoblasts had SIg without Cp; and T if more than 40%
of marrow lymphoblasts were lysed by pan-T (pT) heteroantisera
(40% above control lysis by cytotoxicity testing). The early pre-B
group included all patients with ALL who had complete immunophenotyping and were not classified by the above criteria, irrespective of
the expression of CALLA and HLA-DR antigens. Significant expression of THY, CD3, CD4, CD8, and pan-T cell antigens (pT,
CD5, CD7) was defined by cytotoxicity testing more than 40% above
control. Cases were defined as CALLA-positive by J5 if J5 was
greater than 40% cell lysis as compared with control, or by CALLA
heteroantiserum if CALLA heteroantiserum was greater than 40%.
CALLA definitions were used separately, but had strong correlation.
Nearly all cases (more than 95%) were studied using both heteroantiserum and monoclonal antibodies. Event-free survival (EFS), the
dependent variable of this study, is the time from achievement of a
complete remission (CR) to relapse, death, or last contact, whichever
occurs first. Induction failures are scored as failures at time zero.
Study design and statistical analysis. The POG 7837 study was
originally randomized, but the experimental comparison arm was
closed in March 1981 after accrual objectives were achieved.
Thereafter, all patients were treated identically on the LSA,L,-like
regimen (Fig 1) to answer ancillary questions. Of 253 children who
were confirmed as T-ALL by reference laboratory results after
March 1981, 192 had complete clinical and laboratory data,
permitting adequate evaluation of the impact of these parameters on
outcome. The following factors were analyzed univariately for their
effect on EFS: age at entry (25.0, 27.0, 29.0 and 215.0 years);
WBC (210, 25, 50, and 100 x 103/pL); blast cell E-rosette
formation (greater than 10,20,30, and 40%); presence or absence of
a mediastinal mass, marked lymphadenopathy; French-AmericanBritish (FAB) (L1 v L2); gender; race (black v other); massive
splenomegaly (below umbilicus); massive hepatomegaly (below
umbilicus); hemoglobin (28 or 210 g/pL); platelet count ( r 2 5 x
103/wL); significant expression of acid phosphatase; and significant
expression of antigens recognized by monoclonal or heterologous
antibodies with specificity for HLA-DR, CD5, CD3, CD4, CD8,
CALLA (by J5 and ANTI-CA), and THY antigens.
Recursive partitioning" is used to construct a treelike diagram
that attempts to place patients into subsets of comparable risks.
Using binary (yes/no) independent variables, partitions (branches)
are made according to the variable that has the smallest P value
(which must also be less than .Ol) for all patients as the "node" (the
point in the tree where a partition is to be made). Recursive partition
methods also allow amalgamation (or combination) of adjacent risk
groups. After each partition, groups are ranked starting from the
lowest, and combined if P > .01. This procedure continues until no
further partition can be achieved. The entire analysis is done in a
stepwise manner, with each step attempting to partition each defined
subset and amalgamate the new subsets created by the partition.
This procedure differs slightly from that reported in Ciampi et all9in
168
SHUSTER ET AL
that amalgamation is part of our recursive process, while Ciampi et
a1 only amalgamate at the end. Ciampi et a1 recommend recursive
partitioning over stepwise Cox regression” because of its ability to
detect interactions (eg, subset-specific factors) and because of its
final result that produces clearer descriptionsin direct clinical terms.
A further advantage of recursive partitioning using the logrank tests
is that it provides correct P values whether or not proportional
hazards assumptions, required by Cox regression, hold. This is
discussed in Statistical Note 5 of Pet0 et a1.*’Both methods are most
sensitive to detecting true differences under proportional hazards,
and therefore may enter less prognostically significant factors,
satisfying the assumption ahead of factors that do not. In the absence
of interactions, however, the stepwise Cox procedure would be
superior because it looks for prognostic factors across the entire data
set rather than within subsets. Similar results were obtained from
this analysis, whether the 192 patients with all variables measured
were used or the entire group of 253 patients who were eligible for
analysis of outcome (EFS) was examined, eliminatingonly those for
whom a specific variable is not available. In order to classify the
maximum number of patients, the latter was used in the analysis
(Table 1 and Fig 2). Kaplan-Meier curves,zzwith standard errors of
Pet0 et al,’-’ are presented for the final risk groupings (good,
intermediate, and poor) in Fig 3. The reader is cautioned that
prognostic factor analysis, whether done by Cox regression or
recursive partitioning, is an exploratory analysisinvolving numerous
significance tests. Spurious association is possible. In order to reduce
this problem somewhat, a P value of .01 (rather than the more
conventional .05) was used for entry into the model.
RESULTS
Table 1 provides the results of univariate analysis of each
univariately significant (P< .05) prognostic indicator.
Younger age ( 5 7 years) and lower WBC level ( t 1 0 , 2 5 , 50,
and 100 x 103/pL) are seen to be significantly associated
with better EFS. Lower hemoglobin level ( 5 8 g/pL) and
leukemic cell expression of CD4 or CALLA are also associated with a favorable outcome. Conversely, hepatomegaly
and splenomegaly are associated with inferior prognosis.
Figures 2 and 3 show results of recursive partitioning
Table 1. Relationship of Pretreatment Characteristics to EFS in
253 Children W i t h T-ALL (Univariate Logrank Analysis)
Better
Factor
Age 1 5
Age 5 7
WBC <10K
WBC t 2 5 K
WBC t 5 0 K
WBC < 1OOK
Hemoglobin ~ 8 . g/dL
0
Hepatomegalyt
Splenomegalyt
CD4
CALLA (Anti-CA)
CALLA (J5)
Category*
P Value
Sample
Size
55
.035
.042
.014
.oo 1
<.001
.003
.046
.040
,043
.004
.002
,049
253
253
253
253
253
253
244
253
252
234
240
239
17
<10K
<25K
<50K
< 1OOK
t8.0
-
-
+
+
+
Abbreviation: K, x 103/pL.
*If a factor did not attain a significance level of P t .05,a “better” or
”worse” risk category was not indicated and the variable was not
included in this table. Other factors analyzed are listed in the Materials
and Methods section.
tlnvolved below umbilicus (by palpation).
analysis. Figure 2 tracks the recursive partitioning analysis
through its tree diagram. The most significant prognostic
variable was WBC count (t50v >50 x 103/pL). This result
partitioned the patients into two groups defined by WBC
count. The most significant factor in the higher WBC count
group was THY (positive being favorable), while within the
lower WBC count group, the most significant factor was
CD5 (positive being favorable). This step resulted in a total
of four potential prognostic groups, two for each WBC count
defined subgroup. However, two groups could be amalgamated, since there was no significant difference in outcome
between the groups (WBC t 5 0 x 103/pL and CD5-)
versus (WBC >50 x 103/pL and T H Y + ) . The three resultant groups were each studied for significant factors. The only
subgroup revealing a significant factor (P t .01) was in the
(WBC t 5 0 x 103/pL and CD5+) group (spleen). Finally,
the (WBC <50% x 103/pL, CD5+, spleen+) group was
amalgamated with the group (WBC 4 0 x 103/pL and
CD5-) plus (WBC >50 x 103/pL and T H Y + ) , on the
basis of no significant difference in outcome.
No further significant prognostic factors were uncovered
within any of the three resulting groups. The outcome is
defined in Figs 2 and 3. The relatively good prognosis group
includes patients who (1) have a WBC count t 5 0 x 103/pL;
(2) are CD5 positive; and (3) have no splenomegaly. This
group has 84 patients with 4-year EFS of 66% (SE = 7%).
The poor prognosis group includes patients who (1) have a
WBC count >50 x 103/pL and (2) are T H Y negative. The
poor prognosis group has 63 patients with 4-year EFS of 19%
(SE = 8%). All other patients (provided that they can be
classified) are considered to fall in the intermediate prognosis
group, which consists of 94 patients, who have 4-year EFS of
39% (SE = 7%). The three-way logrank analysis of EFS for
these three patient groups shows a significant difference
(Pc .001). Also, two-way analyses between the good- and
intermediate-risk patients or the intermediate- versus poorrisk patients also reveal significant differences in EFS
(Pc .001 and P = .007, respectively). The overall EFS
within the 253 eligible patients is 43% (SE = 4%) at 4 years.
For interested readers, a forward stepwise Cox model
entered only WBC, subdivided at 50 x 103/pL at P t .01.
DISCUSSION
Our results show that children with T-ALL treated with a
single aggressive multimodal therapy are at widely varied
risk of treatment failure, depending on clinical and laboratory parameters that are readily measured a t the time of
diagnosis. This is the first large study of prognostic factors in
children with T-ALL and its interpretation is facilitated by
use of uniform treatment. Very little information is available
regarding prognostic factors in T-ALL because it is relatively rare, often not well-identified, and usually not treated
uniformly.
Similar analysis of children with non-T, non-B-cell ALL
have demonstrated a wide range of clinical and biologic
features that are associated with treatment outcome.’-’0The
strength of each factor in predicting outcome varies; the age
of the child and extent of tumor bulk at diagnosis have
consistently been among the strongest of prognostic indica-
169
PROGNOSTIC FACTORS IN CHILDHOOD T-CELL ALL
1411253
(WBC: P < ,-O
}O
.l
WBC < 50 x 103/pL
WBC > 5 0 x 103/pL
I
80/117 [55.3]
61/136 [85.7]
kD5: P
CD5?
I
=
THY
CD5-
,0051
+
I
{THY P = ,011
THY?
THY 619
2 1/34 [ 12.21
34/93[42.8]
30/5 1.[4 1.31
I
(Spleen P
=
48/63 [36.7]
(Poor)
213
.002)
28/84 [32.1]
(Good)
6/9[1.9]
(Intermediate)
Fig 2. Recursive partitioning analysis of event-free survival in 253 children with T-ALL. Entries are failed/N [expected t o fail]. This tree
diagram tracks the 253 patients (141 failures) through the recursive partitioning sequence. A t the top of each branching point, the
branching variable is givdn along with the Pvalue. Within the branches. details of the logrank test are provided. For example, 136 patients
had WBC t50 x 103/pL, 61 failed versus 85.7 expected by the logrank test. By contrast, 117 had W8C >50 x 103/pL, 80 failed versus 55.3
expected by the logrank test.
tors in these children. Therefore, it is not surprising that our
univariate analysis revealed that higher WBC levels and
splenomegaly, as well as older age, are associated with worse
EFS in children with T-ALL. A lower hemoglobin level was
associated with better treatment outcome in this study when
analyzed univariately, but not in the recursive partition
analysis, indicating that its independent prognostic importance is not strong. It is known that adolescents have higher
hemoglobin levels compared with prepubescent children
100 r
41
an
(n 84)
(n = 94)
(n = 63)
I
----
peo.001
Poor
01
I
I
I
I
0
1
2
3
4
1
5
Years
Fig 3. EFS of children with T-ALL in subgroups designated as
prognostic groups. Good-risk patients have a WBC at diagnosis
450 x 103/pL, lack of splenomegaly, and have blasts that express
CD5. Patients with WBC levels >50 x 10’lpL and whose blasts
lacked expression of THY antigen were designated poor-risk.
Other patients. definitively not good-risk or poor-risk. were
designated as intermediate risk. Numbers in parentheses on the
curves reflect number at risk at the time shown.
because of expected developmental changes, indicating that
hemoglobin and age are interrelated factors and providing an
explanation for the favorable effect of lower hemoglobin on
EFS. Splenomegaly was noted to be associated with a worse
prognosis in this analysis. Since it reflects increased tumor
burden, its negative predictive effect on prognosis is again
expected. Splenomegaly has been reported to have a similar
effect on prognosis in children with non-B, ~ o ~ - T - A L L . ~ . ’ ~
The expression of CALLA on leukemic blast cells was noted
to be associated with a better treatment response, as has been
previously reported by us for children with T-ALL” and by a
number of investigators for children with non-T, non-B(early pre-B, and pre-B cell) ALL.25*26
The biologic explanation for this association with favorable outcome is unknown,
although recent work defining the enzymatic function of
CALLA27.28may help to define its role in leukemia.
By recursive partitioning analysis, the WBC count is
shown to be the strongest predictor of EFS in children with
T-ALL in this study. We obtained comparable results in a
much smaller study that used similar treatment.I2 However,
our finding of prognostic importance of higher WBC levels (a
continuous variable) extends our previous findings by demonstrating the independent and strong influence of this factor as
compared with all other clinical and biologic variables
measured. In this regard, the finding is consistent to that
observed for children with non-B, non-T-ALL, a disease in
which numerous studies have documented its strong independent predictive prognostic p ~ w e r . ~ * However,
’ ~ * ’ ~ our findings
regarding the importance of WBC count in predicting
prognosis differ from those reported in this issue by Pui et
ai.” An age 215 years, L2 blast morphology, an abnormal
karyotype and cell-surface expression of CD3 are found to
infer an increased risk of treatment failure in that study. W e
170
do not have sufficient information regarding the karyotype in
our study and therefore cannot evaluate this feature. Data
from Pui et a1 are in general agreement with ours in regard to
the negative effect of increasing age as a continuous variable.
However, they report an even worse prognosis for adolescents
between 15 and 21 years. Also, L2 morphology and CD3
expression on blast cells are found to be important in their
study, but not in ours. Conversely, we find that blast cell
expression of either L17F12 or T H Y antigen is associated
with a better treatment outcome. Neither of these antibodies
are directly examined in the study by Pui et al. However, we
have shown that the leukemias identified by these heteroantisera are nearly the same as those recognized by CD5 or
CD7, respectively; both were used in the study by Pui et al.
Reasons for clear differences in our findings are speculative.
The study by Pui et a1 had a higher proportion of subjects
with blasts expressing CD5, age over 15 years, WBC counts
over 100 x 103/pL, and treatment varied considerably,
although overall treatment outcome was similar. These
differences suggest important differences in the patient
populations studied and may explain in part some of the
differences observed. Our patient group was much larger and
uniformly treated.
Our findings regarding the independent prognostic importance of expression of certain T-cell associated surface
antigens and of splenomegaly have not been reported for
children with T-ALL and therefore will require prospective
confirmation. The association of leukemic cell expression of
certain T-cell antigens recognized by the monoclonal or
heteroclonal antibodies (L17F12 or anti-THY, respectively)
with better EFS in subgroups of children with T-ALL is of
interest, but an explanation for this association is not
apparent. Both antigens are expressed relatively early during
T-cell ontogeny, although they are absent a t the earliest
stages of thymocyte differentiation. The T H Y antisera reacts
with thymocytes but not peripheral T cells, and precipitates a
45,000 to 48,000 dalton common thymocyte antigen that is
similar to that detected by OKT6 (CDl).’* L17F12 is an
MoAb that recognizes an antigen in the CD5 cluster group
which has pan-T reactivity, and used in conjunction with
3A1 (CD7), identifies virtually all pediatric cases of TThe function of neither antigen is presently known.
Discovery of the function of these molecules on normal
thymocytes m a y provide insight into the mechanism by
which these antigens confer a more favorable outlook on
certain subgroups of children with T-ALL.
The relatively poor prognosis conferred by lack of CD5
expression (in the low WBC count patients), and the lack of
THY expression (in the high WBC patients) would be
consistent with the general notion that tumors of an immature thymic phenotype have a poor prognosis. In our own
previous experience, an early stage of thymocyte differentiation had a negative impact on remission induction” with
“intermediate”-stage patients (analogous to THY + patients
here) and tended to have the most favorable prognosis. In
other studies of non-T-ALL, antigen phenotype including
expression of CALLA and myeloid antigens has also been
linked to prognosis.3.25~26*29.34~35
The prognosis for children with T-ALL has improved
dramatically over the past 15 years, and 40% to 70% of
SHUSTER ET AL
children are expected to survive after treatment with modern
therapy.’2*35-37
This dramatic improvement in outlook for
children with T-ALL has led investigators to ask why some
patients with T-ALL fail treatment while others are cured.
Prognostic factor analyses are useful in providing a t least
partial answers to these questions. Our findings regarding the
prognostic importance of both clinical and laboratory features of children with T-ALL are not surprising, since similar
findings have been reported for virtually all childhood and
adult neoplasms. Since only approximately 15% to 20% of
children with ALL have T-cell leukemia, and approximately
2,000 cases of pediatric ALL occur annually in the US, it is
apparent that accrual of sufficient subjects for studies of
prognostic factors is diffi~ult.~‘
Very few single institutions
can accrue sufficient numbers of patients to fully evaluate the
independent importance of prognostic variables, and use of
potentially confounding varied treatments for these patients
further hinders efforts to identify prognostic factors. Also,
when new factors are identified, prospective verification is
recommended.
The findings from logrank and recursive partitioning
analyses in this study provide a mechanism for dividing the
patient population with T-ALL into groups a t strikingly
varied risk of relapse in the context of therapy. Although not
fully reported here, it is of interest to note that a Cox forward
regression (using P -= .01 for entry) does not find any
significant factor, after adjusting for WBC count > or <50 x
103/pL. Although T H Y and CD5 were not univariately
significant, even at P = .lo, each was a powerful prognostic
indicator within a WBC count defined subset. Use of easily
measured clinical features, including WBC level and spleen
size in conjunction with biologic features of leukemic cells, is
relatively easy and has been exploited extensively for identical purposes in clinical trials for children with B-cell progenitor ALL? The information from this study, if confirmed, has
potential importance in the stratification of future randomized clinical trials, since comparability of patient groups
enhances the power of such trials.
Multivariate analyses require large numbers of patients to
have adequate power, but are essential in establishing the
independent importance of clinical and biologic variables on
prognosis. Because patients with T-cell leukemia tend to
have clinical features of high-risk ALL, most investigators
have included such patients in therapy protocols based on
clinical risk factors without regard to phenotype and, in fact,
many have not even systematically identified such children.
However, in current leukemia trials worldwide, most children with T-ALL are being identified and it is expected that
more information regarding this phenotype of ALL will be
available soon. However, stratification and treatment assignment are presently done generally without regard to immunophenotype, making evaluation of the prognostic importance of clinical or laboratory variables difficult or impossible.
To address this problem, the POG has arbitrarily taken a
different approach to evaluation and treatment of these
children, using a single aggressive multiagent, multimodal
approach. The aim of this strategy was to assess prognostic
factors in T-ALL while testing proven effective therapy,
modified primarily to strengthen its central nervous system
prophylaxis component. This approach has permitted a
171
PROGNOSTIC FACTORS IN CHILDHOOD T-CELL ALL
detailed assessment of the relative importance of prognostic
factors in T-ALL, resulting in new information which, if
confirmed, could prove useful to leukemia therapists as they
develop randomized trials using risk-appropriate therapy.
We are currently prospectively evaluating the importance of
these prognostic factors in our front-line study for children
with T-ALL. Stratification for WBC count (at 50 x 103/L)
is used and all other factors noted in this analysis are being
further studied. The MoAbs, CD5 and CD7, have been
substituted for the heteroantisera, L17F12 and THY, because they are abundant and recognize approximately the
same cases.
Appendix. Principal Investigators of PO0 Participating in This Study
Institution
Alberta Pediatric Oncology Consortium
Edmonton, Alberta, Canada
Baylor College of Medicine
Houston, TX
Bowman Gray School of Medicine
Winston-Salem, NC
Children's Hospital of Michigan
Detroit, MI
City of Hope
Duarte, CA
Cleveland Clinic Foundation
Cleveland, OH
Duke University Medical Center
Durham, NC
Emory University School of Medicine
Atlanta, GA
Johns Hopkins University
Baltimore, MD
McGill University
Montreal, Quebec, Canada
Medical College of Virginia
Richmond, VA
Medical University of South Carolina
Charleston, SC
Midwest Children's Cancer Center
Milwaukee, WI
Mt Sinai School of Medicine
New York, NY
New England Pediatric Oncology Consortium
Providence, RI
Oklahoma University Health Science Center
Oklahoma City, OK
Roswen Park Memorial Institute
Buffalo, NY
Southwestern Medical School
Dallas, TX
Stanford University Medical Center
Palo Alto, CA
State University of New York
Syracuse, NY
Swiss Pediatric Oncology Group
Bern, Switzerland
UniformedServices Oncology Consortium
Washington, DC
University of Alabama
Birmingham, AL
University of Arkansas
Little Rock, AR
University of California
San Diego. CA
University of Florida
Gainesville, FL
University of Kansas Medical Center
Kansas City, KS
Investigator
John Akabutu
Grant
-
Donald Fernbach
CA-03 16 1
Richard Patterson
-
Y. Ravindranath
CA-29691
Patricia Konrad
-
Donald Norris
-
John Falletta
CA-15525
Abdel Ragab
CA-20549
Brigid Leventhal
CA-28476
V. Michael Whitehead
CA-33587
Harold Maurer
CA-28 53 0
H. Biemann Othersen
-
Bruce Camitta
CA-32053
Jeffery Lipton
CA-38859
Edwin Forman
CA-29293
Rupecht Nitschke
CA- I 1233
Martin Brecher
CA-28383
George Buchanan
CA-33625
Michael Link
CA-33603
Ronald Dubowy
CA-41721
Hans Wagner
-
David Maybee
CA-28572
Robert Castleberry
CA-25408
D.H. Berry
CA-4 1 188
Faith Kung
CA-28439
Samuel Gross
CA-29281
Tribhawan Vats
CA-2884 1
(Continuedon following page)
172
SHUSTER ET AL
Appendix. Principal Investigators of POG Participating in This Study (Cont'd)
Institution
Investigator
Grant
University of Miami School of Medicine
Miami, FL
University of Mississippi
Jackson, M S
Stuart Toledano
CA-4 1082
Jeanette Pullen
CA-15989
University of South Florida Medical Center
Tampa, FL
University of Texas Cancer Center
Houston, TX
University of Virginia School of Medicine
Charlottesville, VA
Washington University Medical Center
St Louis, MO
Eva Hvizdala
-
Donald Pinkel
CA-037 13
R. Beverly
Vita Land
Raney
CA-05587
REFERENCES
1. Pullen DJ (for the POG): A Pediatric Oncology Group
classification protocol for acute lymphocytic leukemia (AlinC 13).
Immunologic phenotypes and correlation with treatment results, in
Murphy SB, Gilbert J R (eds): Leukemia Research: Advances in
Cell Biology & Treatment. New York, NY, Elsevier, 1983, p 221
2. Pullen DJ, Falletta JM, Crist WM, Vogler LB, Dowell B,
Humphrey GB, Blackstock R, van Eys J, Cooper MD, Metzgar RS,
Meydrech EF: Southwest Oncology Group experience with immunologic phenotyping in acute lymphocytic leukemia of childhood.
Cancer Res41:4801, 1981
3. Pullen DJ, Boyett JM, Crist WM, Falletta JM, Roper M,
Dowel1 B, van Eys J, Jackson JF, Humphrey GB, Metzgar RS,
Cooper MD: Pediatric Oncology Group utilization of immunologic
markers in the designation of acute lymphocytic leukemia subgroups. Influence on treatment response. Ann NY Acad Sci 428:26,
1984
4. Crist W, Boyett J, Pullen J, van Eys J, Vietti T Clinical and
biologic features predict poor prognosis in acute lymphoid leukemias
in children and adolescents: A Pediatric Oncology Group Review.
Med Pediatr Oncol 14:135, 1986
5. Crist WM, Grossi CE, Pullen DJ, Cooper MD: Immunologic
markers in childhood acute lymphocytic leukemia. Semin Oncol
12:105,1985
6. Miller DR, Leikin S, Albo V: Intensive therapy and prognostic
factors in acute lymphoblastic leukemia of childhood: CCG 141, in
Neth R, Gallo RC, Graf H (eds): Haematology and Blood Transfusion: Modern Trends in Human Leukemia IV. Berlin, SpringerVerlag, 1981
7. Crist W, Boyett J, Roper M, Pullen J, Metzgar R, van Eys J,
Rogab A, Starling K, Vietti T, Cooper M: Pre-B cell leukemia
responds poorly to treatment. Blood 63:407,1984
8. Williams DL, Harber J, Murphy SB, Look AT, Kalwinsky
DK, Rivera G, Melvin SL, Stass S, Dahl GV: Chromosomal
translocations play a unique role in influencing prognosis in childhood acute lymphoblastic leukemia. Blood 68:205, 1986
9. Williams DL, Tsiatis A, Brodeur GM, Look AT, Melvin SL,
Bowman WP, Kalwinsky DK, Rivera G, Dahl GV: Prognostic
importance of chromosome number of 136 untreated children with
acute lymphoblastic leukemia. Blood 602364, 1982
10. Crist W, Boyett J, Jackson J, Vietti T: Results of a prospective randomized clinical trial demonstrates the independent prognostic importance of the pre-B cell immunophenotype in childhood
lymphocytic leukemia. A Pediatric Oncology Group Study. Blood
(submitted)
1 1. Haghbin M, Tan CC, Clarkson BD, Mikie V, Burchenal JH,
Murphy ML: Intensive chemotherapy in children with acute lymphoblastic leukemia (L-2 Protocol). Cancer 33:1491, 1974
12. Pullen DJ, Sullivan MP, Falletta JM, Boyett, JM, Humphrey
GB, Starling KA, Land VJ, Dyment PG, Vats T, Duncan MH:
Modified LSA,L, treatment in 5 3 children with E-rosette positive
T-cell leukemia: Results and prognostic factors. Blood 60:1159,
1982
13. Sullivan MP, Pullen J, Crist W, Head D, Brecher M,
Ramirez I, Sabio H, Borowitz M, Shuster J, Murphy S: Clinical and
biological heterogeneity of childhood B-cell leukemia: Implications
for clinical trials. Leukemia (in press)
14. Kersey J, Nesbit M, Hallgren H, Sabad A, Yunis E, GajlPeczalska K Evidence for origin of certain childhood acute lymphoblastic leukemias and lymphomas in thymus-derived lymphocytes.
Cancer 36:1348,1975
15. Kersey JH, Nesbit ME, Luckasen JR: Acute lymphoblastic
leukemia and lymphoma cells with thymus-derived T-markers.
Mayo Clin Proc 49:584,1974
16. Vogler LB, Crist WM, Bockman DE, Pearl ER, Lawton AR,
Cooper MD: Pre-B cell leukemia: A new phenotype of childhood
lymphoblastic leukemia. N Engl J Med 298:872, 1978
17. Metzgar RS, Dowell BL, Lachman LB, Jones NH, George
FW IV: Classification of human leukemia by membrane antigen
analysis with xenoantisera. Cancer Res 198 1:4785
18. Dowell BL, Falletta JM, Moore JO, Metzgar RS: Detection
and partial characterization of human thymus-leukemia antigens. J
Natl Cancer Inst 65:691, 1980
19. Ciampi A, Hogg SA, Kates L Stratification by stepwise
regression, correspondence analysis, and recursive partitioning: A
comparison of three methods of survival analysis with covariates.
Comput Stat Data Anal 4:185, 1986
20. Bartolucci AA, Fraser MD: Comparativestep-up and composite tests for selecting prognostic indicators associated with survival.
Biometrical J 19:437, 1977
21. Pet0 R, Pike MC, Armitage P, Brewlow NE, Cox DR,
Howard SV, Mantel N, McPherson K, Pet0 J, Smith PG: Design
and analysis of randomized clinical trials requiring prolonged
observation of each patient. I. Introduction and design. Br J Cancer
34:585,1976
22. Kaplan EL, Meier P: Nonparametric estimation from incomplete observations. J Am Stat Assoc 53:457, 1958
23. Pet0 R, Pike MC, Armitage P, Brewlow NE, Cox DR,
Howard SV, Mantel N, McPherson K, Pet0 J, Smith PG: Design
and analysis of randomized clinical trials requiring prolonged
observation of each patient. 11. Analysis and examples. Br J Cancer
35:1, 1977
24. Miller D R Acute lymphoblastic leukemia. Pediatr Clin N
AM 27:269,1980
25. Pullen J, Boyett J, Borowitz M, Dowell B, Falletta J, Krejmas
PROGNOSTIC FACTORS IN CHILDHOOD T-CELL ALL
N, Sexauer C, Crist W: How important is common acute lymphocytic leukemia antigen (CALLA) negativity as a prognostic factor in
B-precursor acute lymphocytic leukemia (ALL)? A Pediatric Oncotogy Group study. ASCO6:151, 1987
26. Greaves M, Paxton A, Janossy G, Pain C, Johnson S, Lister
TA: Acute lymphoblastic leukemia associated antigen. 111. Alterations in expression during treatment and in relapse. Leuk Res 4:1,
1979
27. Shipp MA, Richardson NE, Sayre PH, Brown NR, Masteller
EL: Molecular cloning of the common acute lymphoblastic leukemia
antigen (CALLA) identifies a type I1 integral membrane protein.
Proc Natl Acad Sci USA 854819,1988
28. Letarte M, Vera S, Tran R, Addis JBL, Onizuka RJ,
Quackenbush EJ, Jongeneel CV, McInnes RR: Common acute
lymphocytic leukemia antigen is identical to neutral endopeptidase.
J Exp Med 168:1247,1988
29. Dow LW, Borella L, Sen L, Aur RJ, George SL, Mauer AM,
Simone JV: Initial prognostic factors and lymphoblasts-erythrocyte
rosette formation in 109 children with acute lymphoblastic leukemia. Blood 50671, 1977
30. Pui CH, Behm FG, Singh B, Schell MJ, Williams DL, Rivera
GK, Kalwinsky DK, Sandlund JT, Crist WM, Raimondi SC:
Heterogeneity and prognostic features and their relation to treatment outcome in 120 children with T-cell acute lymphoblastic
leukemia. Blood 75:174, 1990
3 1. Borowitz MJ, Dowell BL, Boyett JM, Falletta JM, Pullen DJ,
Crist WM, Humphrey GB, Metzger RS: Monoclonal antibody
definition of T cell leukemia. Blood 65:785, 1985
173
32. Link M, Warnke R, Finlay J, Amylon M, Miller R, Dilley J,
Levy R A single monoclonal antibody identifies T-cell lineage of
childhood lymphoid malignancies. Blood 62:722, 1983
33. Crist WM, Shuster JJ, Falletta J, Pullen DJ, Berard C, Vietti
TJ, Alvarado C, Roper MA, Prasthofer E, Grossi CE: Clinical
features and outcome in childhood T cell leukemia-lymphoma
according to stage of thymocyte differentiation. A Pediatric Oncology Groupstudy. Blood 72:1891,1988
34. Mirro J, Zipf TF, Pui CH, Kitchingman G, Williams D,
Melvin S, Murphy SB, Stass S: Acute mixed lineage leukemia:
Clinicopathologic correlations and prognostic significance. Blood
66:1115, 1985
35. Borella L, Sen L: T cell surface markers on lymphoblasts
from acute lymphocytic leukemia. J Immunol 11:1257,1973
36. Riehm H, Feickert H-J, Lampert F: Acute lymphoblastic
leukemia, in Voute PA, Barrett A, Bloom J, et a1 (eds): UICC
Cancer in Children: Clinical Management, ed 2. Berlin, SpringerVerlag, 1986, p 101
37. Clavell LA, Gelber RD, Cohen HJ, Hitchcock-Bryan S,
Cassady JR, Tarbell NJ, Blattner SR, Tantravahi R, Leavitt P,
Sallan S E Four-agent induction and intensive asparaginase therapy
for treatment of childhood acute lymphoblastic leukemia. N Engl J
Med 315:657, 1986
38. Young J, Ries LG, Silverberg E, Horn J, Miller R: Cancer
incidence, survival and mortality in children younger than 15 years.
J Clinicians 58598, 1986