Overall survival rates for children with B-cell acute lymphoblastic leukemia (ALL) exceed 85 to 90%, due to optimized risk-adapted multi-agent chemotherapy treatments; refinement of risk assessment based on leukemia-associated genetics; end-of-induction minimal residual disease (MRD) response; incorporation of relevant targeted therapies; and advances in supportive care.1,2 Current risk stratification uses Children’s Oncology Group (COG) and National Cancer Institute (NCI) threshold-defined dichotomous categories of clinically relevant risk factors, including age (>1 year and ≤10 years vs. <1 or >10); presenting white blood cell count (WBC) (<50x109/L vs. ≥50x109/L); extramedullary disease status; steroid pretreatment; favorable or unfavorable leukemia genetics; flow cytometric MRD of peripheral blood (PB) on day 8; and end-of-induction marrow MRD on day 29 (<0.01% vs. ≥0.01%). Assigning differing weights to individual risk factors or using continuous numerical rather than categorical values may allow for improved relapse risk prediction and therefore further improve risk stratification. The U.K. ALL (UKALL) group recently investigated MRD as a continuous score (prognostic index [PI]) and also incorporated favorable and unfavorable genetics and presenting WBC as continuous variables.3 An increase in the UKALL PI score was strongly associated with relapse risk in a validation cohort of three European pediatric ALL trials (combined n=2,313).3
Natalie Delrocco, PhD, and colleagues developed and validated a multivariable Cox model for relapse-free survival (RFS) in pediatric and young adult B-cell ALL patients by determining a PI using COG ALL trials. Patient outcomes were then compared between current risk stratification strategies and COG PI-derived stratification. The COG cohort included a total of 21,299 non-infant B-cell ALL patients enrolled in clinical trials between 2004 and 2019, including 13,875 NCI standard-risk (SR) patients from COG studies AALL03314 (n=5,099) and AALL09325 (n=8,776), and 7,324 NCI high-risk (HR) patients from COG studies AALL02326 (n=2,900) and AALL11317,8 (n=4,424). Patients with Down syndrome and Philadelphia chromosome-positive (Ph+) B-cell ALL were excluded. The COG PI model included the variables used in the UKALL PI (day 29 MRD, WBC, and favorable and unfavorable genetics) and also investigated the contributions of age, central nervous system (CNS) status, and day 8 PB MRD in predictive modeling for relapse risk. Among these additional variables, day 8 PB MRD was found to contribute independently to the model, indicating excellent expected outcomes for those who are day 8 PB MRD-negative. CNS2 and CNS3 patients were found to have similar hazard ratios.
Relapse risk was classified into four distinct groups, with 38.6% of patients defined as low-risk (RFS 96.8%); 33.1% of patients as SR (92.6%); 16.8% as intermediate-risk (84.9%); and 11.5% as HR (66.9%). When comparing COG PI scores to current COG clinical risk stratification, the two risk classification strategies generally agree when the risk is very high or very low. However, for other current COG risk classifications, including 63% of patients who fall into the SR-average, SR-high, and HR groups, there is a larger distribution of COG PI risk-group assignment (Table). Within the COG SR-average group, COG PI identified a low-risk subgroup with outstanding RFS of 95.5%, similar to outcomes seen in patients currently classified as SR-favorable. In the COG HR group, there was a larger range of outcome estimates, from a subgroup with RFS 95.5% to a subgroup with RFS of 66.0%, consistent with that expected in the very high-risk (VHR) group. Similar trends are seen for disease-free survival and overall survival.
Five-year RFS probability estimates for subgroups by COG retrospective and COG PI risk classifications
| . | COG PI Classification . | |||
|---|---|---|---|---|
| COG Risk Classification . | Low . | Standard . | Intermediate . | High . |
| SR Fav | 0.970 (0.003) | 0.927 (0.015) | -- | -- |
| SR Avg | 0.960 (0.008) | 0.937 (0.005) | 0.904 (0.018) | -- |
| SR High | 0.924 (0.052) * | 0.899 (0.012) | 0.831 (0.013) | 0.732 (0.022) |
| HR Fav | 0.982 (0.009) | 0.964 (0.014) | -- | -- |
| HR | 0.955 (0.014) | 0.903 (0.010) | 0.850 (0.010) | 0.660 (0.015) |
| VHR | -- | -- | -- | 0.534 (0.044) |
| . | COG PI Classification . | |||
|---|---|---|---|---|
| COG Risk Classification . | Low . | Standard . | Intermediate . | High . |
| SR Fav | 0.970 (0.003) | 0.927 (0.015) | -- | -- |
| SR Avg | 0.960 (0.008) | 0.937 (0.005) | 0.904 (0.018) | -- |
| SR High | 0.924 (0.052) * | 0.899 (0.012) | 0.831 (0.013) | 0.732 (0.022) |
| HR Fav | 0.982 (0.009) | 0.964 (0.014) | -- | -- |
| HR | 0.955 (0.014) | 0.903 (0.010) | 0.850 (0.010) | 0.660 (0.015) |
| VHR | -- | -- | -- | 0.534 (0.044) |
Empty cells indicate insufficient sample size for reliable estimation (<25 patients).
Patients in SR-Fav/Avg are missing MRD8, as such they are not represented in this table.
Large standard error reflects small sample size (n = 28) and hence broader uncertainty about the RFS estimate.
Abbreviations: Avg, average; COG, Children’s Oncology Group; Fav, favorable; HR, high-risk; PI, prognostic index; RFS, relapse-free survival; SR, standard-risk; VHR, very high-risk. (Table 4 from the original publication)
In Brief
The use of continuous information enhances traditional pediatric and young adult B-cell ALL risk assessment and can identify subsets of patients who would benefit from intensification or de-intensification of therapy in future clinical trials. For example, SR-average individuals who are COG PI low-risk (RFS 96.0%) could be considered for therapy de-intensification, while the subset of COG HR patients with predicted outcomes similar to the VHR group (RFS 66.0%) could be considered for novel interventions aimed at improving RFS. Additionally, the COG PI model-derived risk score will allow the timely incorporation of new information into risk stratification, such as novel genetic subtypes and results anticipated from ongoing trials (i.e., high-throughput sequencing MRD). This tool can be accessed in a web application for use by patients and practitioners at natalie-delrocco.shinyapps.io/COG_PI_Calculator. Future expansion of the tool will hopefully include additional patient subgroups, including T-cell ALL patients, ALL patients with Down syndrome, and those with Ph+ B-cell ALL.
Competing Interests
Dr. Pommert indicated no relevant conflicts of interest.
