Scand J Haematol 1984;33:295300

Key words: acute lymphocytic leukaemia - prognostic factors - re1 apse-f ree survival

Prognostic factors for relapse-free survival in childhood acute lymphoblastic leukaemia

Niels Clausen, Elma Scheibel & Jerrgen Nyboe

Clinics of Paediatrics and Department of Statistics, University Hospital of Copenhagen, Denmark

11 variables determined at the onset of acute lymphocytic leukaemia were tested to predict disease-free survival in 267 children. All children had received similar induction treatment and basic maintenance therapy, whereas 5 different reinforce- ment treatments were administered irrespective of the severity of the disease at onset. Three modes of prophylactic central nervous system treatment were employed. The risk of relapse was significantly increased for leucocyte count above 50 x 109/1, lymphoblasts above 80% in peripheral blood, age below 2 years or over 5 years, and for males. The risk of relapse was increased by estimated factors of 1.9, 2.3, 1.7, and 1.7, respectively. When the above 4 variables were included in the model, no further prognostic value was demonstrated for haemoglobin concentration, platelet count, signs of bleeding in the skin, mucous membranes or intracranially. presence of leukaemic infiltrations outside the bone marrow, hepatosplenomegaly, mediastinal mass, or T- or B-cell leukaemia. However, evaluation of T- or B-lymphoblast type impact was hampered by small numbers.

Accepted for publication April 2 , 1984

More than 50% of children with acute lympho- blastic leukaemia (ALL) will survive today with- out a relapse for more than 2 years after the termination of complete treatment. Without any treatment, practically all patients will die. Thus, the antileukaemic treatment is the most im- portant prognostic factor at the onset of ALL.

The therapeutic dilemma arises between an intensification of the standard treatment in order

to reduce the number of relapses and the risk of fatal complications from overtreatment of those children who could be cured on the standard treatment. In other words, it is necessary to give different treatments according to the prognosis a t the onset of ALL.

Various data available a t the time of onset of ALL have been suggested as indicators of the chance of obtaining a first complete remission or

On behalf of the Danish Paediatric Society Oncology Group. Other participants: Dr. Bente Hamborg-Petersen, Dr. Henrik Schroder (Aarhus), Dr. Birgitte Djernes (Odense), Dr. Poul Bro (Aalborg), Dr. Helle Rotne (Glostrup), Dr. Birgit Peitersen and Dr. Anne-Marie Plesner (Copenhagen).

296 CLAUSEN ET AL

a maintained complete remission. They fall into 4 groups: general characteristics of the patient, haematological parameters, the extent of ALL in the individual, and characteristics of the lympho- blasts.

Since treatment is the dominant factor a t onset of ALL the finding of further prognostic factors must be seen in the light of the treatment prac- tice. The present investigation was performed to define the most important prognostic factors for disease-free survival among children with ALL in Denmark during 1973-81. During this period the treatment was administered without regard to possible prognostic factors.

Material and methods All 267 children, 153 boys and 114 girls, less than 15 years old, who were diagnosed with ALL in Denmark from July 1973 through June 1981 were included in this study. The 30 children with acute myeloblastic leukaemia and the 32 children with non-Hodgkin lymphomas diagnosed in Den- mark during the same period were not included. T h e criteria for the initial diagnosis and for re- lapses, as well as the clinical data, surveillance examinations, and treatments have been presented in detail elsewhere (I). Remission was achieved in 258 children.

Treatment All children received vincristine 2 mg/m2/week i.v. and prednisone 4 0 4 l mg/m2/d orally during the induction phase and oral methotrexate 15-20 mg/m2/week and mercaptopurine 50-90 mg/m2/d for the maintenance treatment.

The induction treatment was reinforced with asparaginase lo00 IU/kg/d i.v. for 5-10 d in 86 children, with adriamycin 3 0 4 5 mg/m2/d i.v. for 3 d in 32 children, or with both in 57 children, whereas cyclophosphamide 6 1 0 0 mg/m2 and cytosine arabinoside 60-100 mg/m2 were given i.v. 6 times once a week to 33 children. The other children received cyclophosphamide 200 mg/m2/ wk orally during the maintenance treatment.

Children without a relapse were treated for 3 to 5 years.

All children received prophylactic central ner- vous system treatment with intrathecally ad- ministered methotrexate. In 135 this was supple- mented with cranial irradiation with 24 Gy, in 46 with 3 courses of methotrexate 500 mg/m2 i.v. followed by folinic acid rescue.

The treatment scheme depended on the treat- ment centre and not on the severity of the dis- ease. The 7 treatment centres had an even pro- portion of severely diseased children as judged by the initial leucocyte count (WBC) and extra- medullary leukaemia involvement. The disease- free survival as of May 31st 1982 was registered.

Statistical methods The Cox regression model (2) for life table data was used for multivariate analysis of the risk of relapse in the 258 children, who were brought into the first remission. In the present context, the period of observation started on the date of remission, and terminated in the event of relapse or in case of either death without relapse (1 child) or at the conclusion of the study period on May 31st 1982.

The regression model was employed to inves- tigate which of the available variables gave the best prediction of risk of relapse. The variables investigated, which were determined at diagnosis, included age, sex, haemoglobin concentration, platelet count, WBC, percentage of lymphoblasts in venous blood, signs of bleeding in the skin or mucous membranes, leukaemic infiltrations out- side .the bone marrow, hepatomegaly or spleno- megaly, mediastinal mass, and immunological lymphoblast type. Variables expressed in con- tinuous or in interval scales were analyzed to determine the cut-off point which best separated groups with good or poor prognoses. Data were complete in respect to all variables except mediastinal mass in 10 cases, lymphoblast per- centage in 12, and lymphoblast type in 160. The clinical characteristics of the patients with in- complete data were similar to those with com- plete data.

The Cox regression model implicitly assumes

PROGNOSTIC FACTORS IN CHILDHOOD ALL 297

that the variation of the risk in respect to time is identical for all subgroups, which may be formed from the descriptive variables. It is further as- sumed that differences in the risk with respect to a given variable can be described by a constant ratio regardless of the values which the other descriptive variables attain. The first of these assumptions was examined by means of graphical methods. The second assumption was tested by addition of first order interaction terms to the model to see whether this gave a better fit to the data.

Results Among the 11 variables examined, 4 were found to be of significant value for predicting the risk of relapse. These included the WBC, the lym- phoblast percentage, age and sex. When these 4 variables were included in the model, none of the other 7 variables was of significant im- portance. Details of the results of the analysis are shown in Table 1. The risk of relapse for children with WBC over 50 x 10y/l at diagnosis may be estimated to be 1.9 (95% confidence limits 1.3-2.9) times higher than that of children with lower counts. An analysis of the risk of relapse with a more detailed grouping of WBC for 0-10, 11-20, 21-50, 51-100, and over 100 x 10y/l showed estimated relative risks of 1.0, 0.7, 0.9, 1.6 and 2.0, respectively. There was no

TABLE 1 Significant predictive variables

apparent trend for the 3 first groups, but the group with more than 100 x 10y/l showed a somewhat, but not significantly higher risk than the group with 51-100 x 10y/l.

The percentage of lymphoblasts in the periph- eral blood was a variable of independent sig- nificant importance for values of 80 % and more, raising the risk of relapse by an estimated factor of 2.3 in comparison with the risk in children with lower lymphoblast percentages. Even in children with values below 80%, the risk of relapse was slightly correlated with high lymphoblast percent- ages, but in attempt to use the percentage itself as a variable in the model was not successful.

Age groups 0-1, 2-3, 4-5, 6-9, and from 10 years and older demonstrated relative risks of relapse of 1 .O, 0.6, 0.7, 1 .O, and 1.3, respectively. The most favourable prognosis was seen among patients aged 2-3 and 4-5 years. When these age-groups were combined, a significantly higher risk of relapse was found for children outside this age range, estimated to ,be 1.7 times greater than that of the 2-5-year-olds.

Boys had 1.7 times (95% confidence limits 1.2-2.4) higher rate of relapse compared with that of girls.

A more detailed analysis has shown minor de- viations from the model presented above. Thus, a graphical representation of the changes in relapse risk with observation time indicated that the risk for boys and girls was nearly the same during the first year of observation and only after 2 years of

Variable (number + vs number -)

risK ratio regression Standard coefficient error

,1111115 b eb

Estimated 95 % Estimated

confidence . . ..

,:-:.-

Leucocyte count above vs. below 50 x lO9/l (60: 198) 0.654 0.213 I .9 1.3-2.9

Lymphoblast % above vs. below 80 % (52:194) 0.817 0.221 2.3 1.5-3.5

Age less than 2 ys or 6 ys or more vs. 2-5 ys (110:148) 0.549 0.174 1.7 1.2-2.5

Male vs. female (147:lll) 0.532 0.180 1.7 1.2-2.4

298 CLAUSEN ET AL

3 - > tY 3 u)

w w Df LL w u)

I

a a w tY I- Z w 0 tY W a

X X

YEARS AFTER REMISSION Figure 1. Estimated relapse-free survival ( x ) and 95 % confidence limits (.) among the patient groups with the best prognosis (upper curve) and the worst prognosis (lower curve). The best prognostic grpup comprised girls of 2-5 years who had WBC of 50 x 10911 or less, and less than 80 % lymphoblasts in the peripheral blood. The worst prognosis occurred in boys less than 2 or more than 5 years of age with WBC above 50 x 10y/l and a peripheral blast count of 80 % or more.

observation was the risk for boys on a constantly higher level than the risk for girls.

Furthermore, the interaction analysis has in- dicated an interaction, just significant at the 5 % level, between lymphoblast percentage and age: the individual effect of these variables is probably

T A B L E 2 Estimated risk ratios for the combined effect og age and lymphoblast percentage obtained by inclusion of interaction term (italic) and without such a term (in parenthesis)

Lymphoblast % Lymphoblast % less than 80 80 or more

2-5 ys 1.0 (1.0) 1.4 (1.7) Less than 2 or over 5 ys 4.6 (3.9) 1.4 (2.3)

somewhat less than indicated by the risk ratios given in Table 1 and the combined effect on risk of relapse of lymphoblasts over 80% and age outside the 2-5-year range is somewhat greater than the product of the two risk ratios (Table 2).

Table 3 gives the estimated regression coeffi- cients for the 7 variables which did not demon- strate significant predictive values.

Although not statistically significant, the data show a tendency towards an increased risk of relapse for presence of mediastinal mass, haemo- globin concentration less than 4 mmol/l, and bleeding in skin, mucous membranes or intra- cranially. Evaluation of the impact of T or B lymphoblast type was difficult because only 97 analyses were available.

PROGNOSTIC FACTORS IN CHILDHOOD ALL 299

To visualize the importance of the prognostic factors, the estimated relapse-free survival for the patient groups with the best and the worst combinations of the 4 significant variables are presented in Figure 1. The difference corre- sponds to a 13-fold increment in the risk of relapse.

Discussion The present study involved a complete popula-

tion of children with ALL from a geographically- defined area. In future it will become increasing- ly difficult to find patient groups which have not been selected for different treatment schedules by virtue of their risk factors at diagnosis. The interaction between the treatment and the prog- nostic factors will seriously compromise the anal- ysis of prognostic factors.

Induction of first complete remission was achieved in 96% of the children in the present study. Induction failure was associated with the presence of congenital leukaemia, and haemor- rhagic tendency.

TABLE 3 Non-significanr predictive variables The results refer to a model comprising the 4 significant variables plus the variable concerned

Estimated Standard Variable regression

coefficient (number +:number - )

~~

Hb concentration less than 4 mmol/l vs. 4 or more (112: 146) 0.254 0.179

Platelet count 10 x iO9/i or more vs. less than 10 x lO9/l (30:228) 0.296 0.252

(126: 132) 0.230 0.179

mia + vs. - (23:235) 0.194 0.305

megaly + vs. - (160:98) 0.073 0.194

(29 : 219) 0.419 0.277

+ VS. - (15:83) 0.446 0.377

Bleeding tendency + vs. -

Extramedullary leukae-

Hepato- and/or spleno-

Mediastinal mass + vs. -

Lymphoblast type T o r B

The outcome during and after treatment for A L L is consistently found to be poor in groups of children with a high initial WBC ( S 5 ) . Vari- ous investigators have operated with different classes of WBC (47) . In accordance with our results, the Memphis study demonstrated no dif- ference in prognosis between groups with WBC less than 50 x 109/l. However the groups with WBC from 50 to 100 x lo9/] and WBC greater than 100 x 109/1 had significantly increasing pro- portions of patients with a poor prognosis (8). The present study demonstrated the same ten- dency without attaining statistical significance.

The prognostic significance of the peripheral lymphoblast percentage has not previously been observed. However, a total peripheral blast cell count under 10 x 109/1 gave a better prognosis than higher total blast cell counts, and peripheral WBC and blast cell count were closely correlated in one study (5 ) . In the present study there was also a correlation but no interaction was found between WBC and lymphoblast percentage. They were each of prognostic significance. It may be speculated whether a high lymphoblast percent- age indicates that a large proportion of the lym- phocytes belong to an emigrating clone, which is of high proliferative activity and relatively re- sistant to treatment. This hypothesis found sup- port in a study which showed a significant corre- lation between a poor prognosis and a high pre- treatment proliferative activity of blood and mar- row blasts determined by autoradiography (9).

Age at onset of ALL has also been found to be of prognostic significance in all previous studies (3, 5 , 8-10). Even when the congenital cases in the present study were excluded, the youngest children had a poorer prognosis than did the intermediate age group from 2-5 years.

It is reasonable to assume that WBC, lympho- blast percentage, and age at onset influence the prognosis in a continuous manner, and therefore the limits used for definition of prognostic groups are necessarily arbitrary and express a condi- tioned truth.

Patient sex has been demonstrated to have little or no relation to a 3-year continuous remis- sion or to survival (8, 10, 11). A difference

300 CLAUSEN ET AL

in relapse rates for boys and girls appeared in groups of patients surviving longer than previous groups due to the introduction of prophylactic central nervous system treatment. The difference was apparent after 12 months and was further accentuated after discontinuation of therapy (12, 13). In agreement with the present data, a recent study demonstrated that sex was an independent predictor of disease-free survival of marginal sig- nificance, but a sex-related difference for survival was not found (3).

When multivariate analysis for significant in- dependent variables to predict disease-free sur- vival was employed, mediastinal mass and central nervous system disease at diagnosis were not sig- nificant indicators in the present or in other studies (3, 14).

No uniform agreement exists as to the prog- nostic significance of an enlarged liver or spleen. One study concluded that modern treatment has almost obliterated the significance of clinical or- gan enlargement (12) in accordance with the present results. A recent study, however, found massive splenomegaly of importance but hepato- megaly without significance (3). Yet another study demonstrated prognostic significance of both liver and spleen enlargement (15). A low platelet count was associated with a

poor prognosis in previous studies using uni- variate analysis (5, 7, 8), but not when analyzed together with other variables (3).

For all investigations, it may be concluded that the important prognostic factors for disease-free survival after the first remission are the initial WBC, the age, and possibly the sex. The demon- stration of a significant prognostic importance of the initial lymphoblast percentage is new. Other factors may contribute to refine the prediction of an increased relapse rate, but such additional information is marginal.

The concept that risk factors may not be merely multiplicative but synergistic when they occur together has not previously been demon- strated.

References I . Clausen N, Scheibel E, Bro P et al. Acute lymphoblastic

leukaemia in children in Denmark in the period 1973-81. (English abstract). Ugeskr L z g 1984;146: 1028-32.

2. Cox DR. Regression models and life tables. J R Stat SOC (B ) 1972 ;34: 187-220.

3. Miller D R , Leikin S , Albo V, Sather H , Karon M, Ham- mond D. Prognostic factors and therapy in acute lympho- blastic leukemia of childhood: CCG-141. Cancer 1983;51: 1041-9.

4. Zuelzer WW. Implications of long-term survival in acute stem cell leukemia of childhood treated with composite cyclic therapy. Blood 1964;24:477-94.

5 . Hardisty RM, Till MM. Acute leukaemia 195944: Factors affecting prognosis. Arch Dis Child 1968;43: 107-15.

6. Miller DR, Leikin S , Albo V et al. Use of prognostic factors in improving the design and efficiency of clinical trials in childhood leukemia: Children’s cancer study group report. Cancer Treat Rep 1980;64:381-92.

7. George SL, Fernbach DJ, Vietti TJ et al. Factors influ- encing survival in pediatric acute leukemia: The SWCCSG experience 195g1970. Cancer 1973;32: 1542-53.

8. Sirnone JV, Verzosa MS, Rudy JA. Initial features and prognosis in 363 children with acute lymphocytic leukemia. Cancer 1975;36:2099-108.

9. Dow LW, Chang LJA, Tsiatis A A , Melvin SL, Bowman WP. Relationship of pretreatment lymphoblast proliferative activity and prognosis in 97 children with acute lympho- blastic leukemia. Blood 1982;59: 1197-202.

10. George SL, Fernbach DJ, Lee ET. Early deaths in newly diagnosed cases of pediatric acute leukemia. Cancer 1978; 42:7814.

11. Steinhorn SC, Myers MH. Progress in the treatment of childhood acute leukemia: A review. Med Pediatr Oncol 1981 ;9:33346.

12. Hardisty RM, Till MM, Peto J. Acute lymphoblastic leukaemia: four-year survivals old and new. J Clin Pathol 1981 ;34:249-53.

13. Sather H, Miller D, Nesbit M, Heyn R , Hammond D . Differences in prognosis for boys and girls with acute lym- phoblastic leukaemia. Lancet 1981;1:73943.

14. Chilcote RR, Coccia P, Sather H, Baehner R , Nesbit M, Hammond D. Mediastinal mass and prognosis in acute lymphocytic leukemia (Abstr ‘2-224). Proc Am SOC Clin Oncol 1976;17:292.

15. Heme G, Langermann HJ, Rhiem H. Acute lymphoblastic leukemia therapy study BFh4 81. In: Mandelli F et al, eds. 3rd international symposium on therapy of acute leukemias. Pisa: Officine Grafiche d. Pacini Editore 1983.

Correspondence to: Niels Clausen, M.D. Department of Paediatrics G 4064 Rigshospitalet DK-2100 Copenhagen, Denmark