Improved Survival for Children and Adolescents With Acute Lymphoblastic Leukemia Between 1990 and 2005: A Report From the Children's Oncology Group

INTRODUCTION

Acute lymphoblastic leukemia (ALL) is the most common pediatric malignancy, comprising 25% of cancers occurring before age 15 years and 19% among those younger than age 20 years.¹ The 5-year survival rate increased from less than 10% in the 1960s to 77% in 1985 to 1994.¹ Survival rate has continued to increase over the past 10 to 15 years.^2–11 The National Cancer Institute (NCI) SEER Program reported that 5-year survival for US patients younger than age 15 years with ALL increased from 80.2% to 87.5% between 1990-1994 and 2000-2004.¹² Five-year survival rates for adolescents age 15 to 19 years increased from 41.0% in 1980-1984 to 61.1% in 2000-2004.¹³

The Children's Oncology Group (COG) includes more than 200 member institutions in the United States, Canada, Australia, and New Zealand. Unlike the SEER system, which tracks outcome in five representative states and four metropolitan areas that include approximately 10% of the US population, COG data include patients from all areas of the United States and Canada and provide an opportunity to assess outcome for children with ALL throughout these countries and to examine the prognostic impact of covariates not included in registry data. We report changes in survival among children enrolled onto COG ALL clinical trials between 1990 and 2005 and the extent to which different clinical and biologically defined patient subgroups benefited from treatment improvements.

PATIENTS AND METHODS

RESULTS

Overall 5- and 10-year survival rates increased significantly over time (Fig 1; P < .001). Five-year survival increased from 83.7% (SE, 0.4%) in 1990-1994 to 87.7% (SE, 0.4%) in 1995-1999 and to 90.4% (SE, 0.5%) in 2000-2005 (Table 1). Similar increases were seen in 10-year survival between 1990-1994 (80.1%; SE, 0.8%) and 1995-1999 (83.9%; SE, 1.3%; P < .001). For the eras from 1990 to 1999, approximately 84% of deaths occurred within 5 years of diagnosis and only 1% occurred more than 10 years following diagnosis (Appendix Tables A2 to A6, online only). Because of these factors and the more limited follow-up for the 2000-2005 era, subsequent analyses focused on 5-year survival.

Survival improved significantly in all subgroups examined except for infants ≤ 1 year old (Table 1 and Appendix Figs A1 to A6, online only): ages 1 to 9 years, 10+ years, and 15+ years; males and females; self-reported whites, blacks, and other races; self-reported Hispanics, non-Hispanics, and persons of unknown ethnicity; those with B-cell and T-cell ALL; and those with standard-risk (age 1 to 9.99 years; initial WBC < 50,000/μL) or high-risk (age ≥ 10 years and/or initial WBC ≥ 50,000/μL) features by using NCI/Rome²⁵ criteria. The relative reductions in 5-year risk of death between the 1990-1994 and 2000-2005 eras were similar in all non–infant subgroups examined, ranging from 30% to 50% (Table 1). Because the results for infants differed from those of older children, we also analyzed data separately for infants and non–infants (Appendix Tables A7 and A8, online only).

Deaths that occurred after induction failure or relapse were classified as leukemia related, and those that occurred without prior induction failure or relapse were deemed treatment related (Table 2 and Appendix Tables A9 and A10, online only). Among all patients, the 5-year cumulative incidence of death decreased from 16.35% in 1990-1994 to 9.6% in 2000-2005 (P < .001), and the 10-year cumulative incidence of death decreased from 19.86% in 1990-1994 to 16.07% in 1995-1999 (P < .001). This decrease was primarily due to reduction in the 5-year cumulative incidence of death following relapse/disease progression from 12.83% in 1990-1994 to 7.22% in 2000-2005 (P < .001), with similar reductions in 10-year rates between 1990-1994 and 1995-1999. Among infants (Appendix Table A9), the 5-year cumulative incidence of death changed little between 1990-1994 and 2000-2005 (52.1% v 50.3%; P = .45), but the causes of death changed considerably. The 5-year cumulative incidence of death following relapse/disease progression decreased from 43% in 1990-1994 to 27.2% in 2000-2005 (P < .001), and the cumulative incidence of treatment-related death increased from 3.9% in 1990-1994 to 13.9% in 2000-2005 (P < .001).

The highest relative risks of death occurred in known high-risk subgroups (Table 3). The relative risk of death was 2.3-fold (1990–1994) to 3.1-fold (2000–2005) higher for patients age 10+ years versus those age 1 to 9.99 years. The differences were even larger when those age 1 to 9.99 years were compared with infants age ≤ 1 year and adolescents age ≥ 15 years. We observed modest but statistically significant sex-based differences in survival, with males having a relative risk of death 1.2- to 1.3-fold higher than that of females. Self-described blacks had an increased relative risk of death compared with whites. Although information about ethnicity was not available for approximately 30% (6,509 of 21,626) of patients, self-described Hispanics (n = 2,589) had a higher relative risk of death than non-Hispanics. Leukemia immunophenotype was prognostic, with patients who had T-cell ALL having a higher relative risk of death than those with B-cell ALL. Patients with NCI high-risk ALL had a 2.4- to 3.6-fold higher relative risk of death than those with standard-risk ALL. However, because most patients have NCI standard-risk clinical features, a significant proportion of deaths occurred among the favorable prognosis subgroups (Table 4). Five-year survival of patients with NCI standard-risk ALL was 90% to 95% between 1990 and 2005 (Table 1 and Appendix Fig A6A), but approximately 36% of total deaths occurred among this subset.

For patients older than age 1 year, era, sex, race, immunophenotype, and NCI risk group were all significant prognostic factors in the multivariate Cox regression model (Table 5). To better understand the most important factors predicting risk of death, we performed survival tree regression modeling (Appendix Fig A7A, online only). This analysis showed that NCI risk group was the most significant overall prognostic factor. Among NCI standard-risk patients, era (2000-2005 v 1990-1999) was the most significant prognostic factor. For NCI high-risk patients, age (1 to 14.99 v ≥ 15 years) was the most prognostic factor. Among NCI high-risk patients younger than age 15 years, race was the most significant prognostic factor, with survival for black/other being inferior to that of whites. For adolescents age ≥ 15 years, era was the most significant prognostic factor, with better survival rates in 1995-2005 compared with 1990 to 1994.

For infants, age considered as a continuous variable was the most important prognostic factor (Appendix Fig A7B). The best age cutoff among infants was 92 days, with 5-year survival rates of 57.8% and 25.6% for age ≥ 92 days and less than 92 days, respectively (P < .001). Among infants ≥ 92 days old, WBC was the most significant predictor of survival, with higher WBC resulting in poorer outcome.

DISCUSSION

This study, which includes the largest childhood ALL cohort ever reported, documents progressive improvements in survival for children with ALL enrolled onto COG clinical trials between 1990-1994 and 2000-2005. Five-year survival increased from 83.7% to 90.4% during this time. The improved survival was explained primarily by an approximately 44% decrease in the risk of death following relapse/disease progression. Although we examined overall and not event-free survival (EFS) and cannot comment directly on changes in the incidence of relapse, a study of almost 10,000 children treated on COG ALL trials between 1988 and 2002, including 1,961 who relapsed, showed no significant improvements in survival after relapse over time.²⁶ Taken together with results of COG ALL clinical trials showing significant improvements in EFS during this period, we believe that the major reason for improved survival was decreased risk of relapse.^14,15,17

Our cohort includes 18,501 (55.8%) of 33,139 US ALL cases in persons age 0 to 14.99 years predicted to occur between 1990 and 2005. Thus, our results are representative of survival following contemporary therapy in the United States and are consistent with previous reports of outcomes for children younger than age 15 years diagnosed with cancer between 1990 and 1994 enrolled onto COG trials.^27,28 In contrast, only approximately 25% of adolescents age 15 to 19 years diagnosed with cancer were enrolled onto COG trials between 1990 and 1994.²⁷ Our data are similar, with 33.5% (1,392 of 4,159) of US adolescents age 15 to 19.99 years predicted to develop ALL between 1990 and 2005 enrolling onto COG trials. There are a variety of reasons that children and adolescents with ALL might not be included in COG ALL trials, including participation in trials conducted by other centers,^6,8 lack of an open study at the time of diagnosis, having the patient/parent decline participation, or failure to meet eligibility criteria. Most US children with ALL still enroll onto COG trials. In 2009, 1,951 (68%) of 2,869 US children and adolescents age 0 to 19.99 years predicted to develop ALL enrolled onto a COG trial, including 1,758 (69%) of 2,540 of those age 0 to 14.99 years and 168 (51%) of 329 of those age 15 to 19.99 years. The number of older adolescents with ALL enrolling onto COG trials has increased over time, which is an important trend, given the higher survival rates obtained with pediatric versus adult ALL trials for this age group.^29–31

Pulte et al^12,13 reported survival for US children and adults diagnosed with ALL between 1990 and 2004 by using the SEER 9 Registries database, which includes about 30 million people. Our results show higher survival rates than the SEER-estimated 5-year survival of 87.5% from 2000-2004 for children younger than age 15 years and 61.1% for those age 15 to 19 years. We found 5-year survival rates of 91.4% for US children younger than age 15 years and 74.5% for those age 15 to 19 years in 2000-2005. It is unlikely that the slight difference in time period analyzed (SEER: 2000-2004 v COG: 2000-2005) accounts for these differences. There may be differences based on clinical trial enrollment because the SEER data include all patients reported to tumor registries in the specified areas, although the COG data are based on patients who met eligibility criteria and were offered and accepted clinical trial enrollment. The SEER 9 population is more urban and includes a higher percentage of foreign-born persons than the overall US population; these differences might contribute to observed differences in survival. There was a 13% absolute survival advantage for older adolescents in this COG cohort compared with the SEER estimates, consistent with the significant survival advantages for older adolescents with ALL treated on COG versus adult cooperative group trials.²⁹ These data emphasize that optimal treatment for an older adolescent with ALL is referral to a pediatric center and enrollment onto a pediatric cooperative group trial.

In our analyses, survival improvements occurred in every subgroup analyzed with the exception of infants age ≤ 1 year (Table 1 and Appendix Figs A1 to A6). The magnitude of the decrease in risk of death between 1990-1994 and 2000-2005 was generally similar among non–infant subgroups and ranged from approximately 30% to 50%.

There were no survival improvements for infants enrolled onto COG ALL trials between 1990 and 2005 (Appendix Fig A1A). Infants contributed disproportionately to deaths because they accounted for only 2.1% (461 of 21,626) of patients but 8.0% (231 of 2,878) of deaths (P < .001). During this period, the COG pursued several strategies to attempt to increase survival for infants with ALL. Chemotherapy treatment was intensified significantly in the Children's Cancer Group (CCG) 1953 and COG P9407 trials, and the use of stem-cell transplantation in first remission was explored for those with MLL gene rearrangements. Stem-cell transplantation was not beneficial for infants in these COG ALL trials,¹⁶ and treatment intensification shifted the causes of death, with a significant decrease in death following relapse/disease progression but a parallel increase in death related to toxicity, with no net improvement in survival (Appendix Table A9). The Interfant99 infant ALL trial (1999–2005) obtained results similar to those of the COG trials, with 4-year EFS of only 48%.³² Although Interfant99 showed a benefit for stem-cell transplantation among a select high-risk subgroup of infants, survival was still poor for these patients.³³ Infant ALL is a unique high-risk subset that requires new therapeutic strategies.

Survival for self-reported blacks with ALL improved significantly during the eras examined. The absolute difference in 5-year survival between blacks and whites decreased from 11.0% in 1990-1994 to 3.3% in 2000-2005. There are race-based differences in ALL biology, with blacks having a higher incidence of T-cell ALL and other high-risk features.³⁴ Among the COG patients, 17.7% of blacks versus 9.5% of whites had T-cell ALL (P < .001) and 44.5% of blacks had NCI high-risk features versus 32.9% of whites (P < .001). Thus, black children and adolescents with ALL are predicted to have an inferior survival compared with whites because of the different percentages of ALL subtypes in the two racial groups. However, within these ALL subtypes, although patient numbers are small, our results show decreases in the racial outcome gap between 1990-1994 and 2000-2005. The absolute difference in 5-year survival for blacks versus whites with T-cell ALL decreased from 5.0% in 1990-1994 (94 blacks) to 0.02% in 2000-2005 (57 blacks). Similarly, for children with NCI high-risk B-cell ALL, the gap decreased from 11.1% in 1990-1994 (129 blacks) to 6.6% in 2000-2005 (53 blacks).

Self-described ethnicity is another important risk factor in childhood ALL. Hispanics have inferior outcomes to non-Hispanics.³⁵ Our study confirms this observation, with some differences in the magnitude of differences in the three eras (Tables 1 and 3). Importantly, there was much greater capture of information regarding self-described ethnicity in 2000-2005, with only 3.6% (261 of 7,153) of unknown ethnic group compared with 43.2% (6,284 of 14,473) in 1990-1999. We observed a 1.5-fold higher risk of death for Hispanics versus non-Hispanics in 2000-2005. A variety of reasons might account for the inferior outcome of Hispanics. For example, recent investigations have shown a much higher incidence of certain high-risk leukemia cell genomic alterations in Hispanics enrolled onto COG ALL trials.³⁶

Simple demographic and clinical prognostic factors can identify patient subsets with significantly increased risk of death: infants, adolescents age ≥ 10 or ≥ 15 years, T-cell ALL, and NCI high-risk ALL. The relative risks of death for these subsets ranged from about two- to eight-fold higher than for lower-risk subsets (Table 3). However, 36% of total deaths occurred among patients with NCI standard-risk ALL. Thus, efforts to decrease ALL deaths must focus both on high-risk patient subsets and on the large subset of patients with favorable clinical characteristics. Detailed biologic characterization of lymphoblasts and host germline variability and sophisticated measurements of early treatment response can improve identification of ultra-low-risk patient subsets and identify patients at high risk of treatment failure.^18,36–47

We analyzed whether death occurred as a first event or after relapse/induction failure to investigate whether observed survival improvements were due to better front-line antileukemia therapy, better supportive care leading to a decrease in non–leukemia-related death, or both. Five to six times as many deaths occurred following relapse/disease progression compared with toxicity (Table 2). Although the cumulative incidence of deaths related to toxicity is relatively low at approximately 2%, it accounted for a higher percentage of overall deaths as the rate of death from leukemia decreases and was a particular problem among infants. Prevention of treatment-related deaths must be a critical component of efforts to improve childhood ALL survival.

Ten-year survival was 3% to 4% lower than 5-year survival in 1990-1999, when 84% of deaths occurred within 5 years, and only 1% occurred more than 10 years following diagnosis (Appendix Tables A2 to A4). Given this lower 10-year survival rate, the death rates observed in the 2000-2005 era (Appendix Tables A5 and A6), and the shape of the survival curves (Fig 1), we believe that it is extremely unlikely that there will be a significant increase in deaths beyond 5 years for patients diagnosed in 2000-2005, and we anticipate significant improvements in 10-year survival. We anticipate that the 10-year survival rate for children treated on COG ALL trials in 2006-2010 will approach or exceed 90%. The trend from 1990 to 2005 predicts an absolute 2% to 3% increment in survival during each 5-year era. More importantly, randomized COG ALL clinical trials conducted between 1995 and 2005 established superior treatment regimens that then became the baseline therapy for COG trials conducted in 2006 to 2010.^4,5,14,15

This report underscores the remarkable improvements in the outcomes for childhood ALL since Farber et al⁴⁸ first described temporary remissions in 1948. The ongoing discovery of important biologic subsets of ALL^{36,38–40,46} will further refine risk stratification and facilitate the combination of molecularly targeted therapies with chemotherapy. As proof of principle, addition of imatinib to chemotherapy resulted in a dramatic increase in survival for pediatric Philadelphia chromosome–positive ALL.⁴⁹ As these changes occur, it will remain essential to closely assess the survival for the majority of US children, adolescents, and young adults with ALL who are treated on COG clinical trials.

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

Although all authors completed the disclosure declaration, the following author(s) indicated a financial or other interest that is relevant to the subject matter under consideration in this article. Certain relationships marked with a “U” are those for which no compensation was received; those relationships marked with a “C” were compensated. For a detailed description of the disclosure categories, or for more information about ASCO's conflict of interest policy, please refer to the Author Disclosure Declaration and the Disclosures of Potential Conflicts of Interest section in Information for Contributors.

Employment or Leadership Position: None Consultant or Advisory Role: Paul S. Gaynon, EUSA Pharma, sanofi-aventis, Teva Pharmaceutical Industries, Bristol-Myers Squibb (C) Stock Ownership: None Honoraria: Paul S. Gaynon, Enzon Pharmaceuticals, Sigma Tau Pharmaceuticals, Genzyme Research Funding: Paul S. Gaynon, Genzyme Expert Testimony: None Other Remuneration: None

AUTHOR CONTRIBUTIONS

Conception and design: Stephen P. Hunger, Meenakshi Devidas, Bruce M. Camitta, Naomi J. Winick, Gregory H. Reaman, William L. Carroll

Provision of study materials or patients: Naomi J. Winick

Collection and assembly of data: Meenakshi Devidas, Gregory H. Reaman, William L. Carroll

Data analysis and interpretation: All authors

Manuscript writing: All authors

Final approval of manuscript: All authors