A robust and validated integrated prognostic index for defining risk groups in adult acute lymphoblastic leukemia: an EWALL collaborative study

Abstract

Risk stratification is crucial to the successful treatment of acute lymphoblastic leukemia (ALL). Although numerous risk factors have been identified, an optimal prognostic model for integrating variables has not been developed. We used individual patient data from 4 contemporary academic national clinical trials, UKALL14, NILG-ALL10/07, GIMEMA-LAL1913, and PETHEMA-ALL-HR2011, to generate and validate the European Working Group for Adult ALL prognostic index (EWALL-PI), which is based on white blood cell count, genetics, and end of induction minimal residual disease (MRD). Individual patient risk scores were calculated for 778 patients aged 15 to 67 years in complete remission using the validated UKALL-PI formula, applying minor modifications to reflect differences between pediatric and adult ALL. Per-trial analysis revealed that EWALL-PI correlated with relapse and death. Regression analysis revealed that each unit increase in EWALL-PI increased the risk of relapse or death by ∼30% with no evidence of heterogeneity across trials or patient subgroups. EWALL-PI-defined risk models outperformed the stratification algorithms used by each trial. Threshold analysis revealed an EWALL-PI threshold that divided patients with B cell and T cell into standard (EWALL-PI <2.50) and high (EWALL-PI ≥2.50) risk groups, respectively. Per-trial analysis showed that patients at high risk had a significantly increased relapse rate and inferior survival compared with patients with standard risk (subdistribution hazard ratio for relapse, ranged from 1.85 to 3.28; hazard ratio for death, 1.73 to 3.03). Subgroup analysis confirmed the robustness of these risk groups by sex, age, white blood cell count, and lineage. In conclusion, we validated an integrated risk model across 4 independent adult ALL clinical trials, demonstrating its utility defining clinically relevant risk groups.

Graphical abstract

Figure 1.

Flow diagram describing the method for calculating the EWALL-PI for a patient.^§τ(MRD) was defined as the negative natural log of the absolute end of induction MRD level with undetectable MRD and MRD outside the quantitative range being assigned values of 1 × 10^–6 and 1 × 10^–5 respectively, whereas values ≥1 were rounded down to 0.99999. ^†For this study, τ(WCC) was defined as log(WCC + 1) rather than log(WCC) + 1 to avoid negative numbers resulting from WCC values below 0.4 × 10⁹/L. ^ŦGR-GEN and HR-GEN is coded as “1” if any of the listed genetic abnormalities is present, else “0.” ^∗The EWALL-PI is standardized to make it range from 0 to 10, as follows: EWALL-PI = ([actual EWALL-PI − EWALL-PI minimum]/[EWALL-PI maximum − EWALL-PI minimum]) × 10; where EWALL-PI minimum = −3.6381537 and EWALL-PI maximum = 2.0882617. The minimum and maximum value of EWALL-PI was derived based on existing data sets. High hyperdiploidy, 51 to 67 chromosomes; low hypodiploidy, 30 to 39 chromosomes, including masked low hypodiploidy with 60 to 78 chromosomes; near-haploidy, <30 chromosomes; complex karyotype, ≥5 chromosomal abnormalities detected by karyotype analysis excluding established ploidy subgroups; JAK-STAT, IGH::CRLF2, P2RY8::CRLF2, JAK2 fusions.

Figure 2.

Distribution of EWALL-PI by trial and within each trial immunophenotype, relapse, and death. (A) The dotted vertical line shows the median EWALL-PI for each trial. The UKALL14 distribution was significantly different to GIMEMA-LAL1913 (P < .001) and PETHEMA-ALL-HR2011 (P < .001) but not NILG-ALL10/07 (P = .73) (P values from a Kolmogorov–Smirnov test for equality of distribution). (B) There was no difference between the EWALL-PI distributions for patients with B-ALL and T-ALL within each trial. (C) The median EWALL-PI values for patients treated on UKALL14, GIMEMA-LAL1913, and PETHEMA-ALL-HR2011 who relapse was significantly greater than those patients who did not relapse. (D) Similarly, the median EWALL-PI values for patients treated on UKALL14, NILG-ALL10/07, and PETHEMA-ALL-HR2011 who died was significantly greater than those patients who did not die. The P values reported in boxes B, C and D are from a Mann-Whitney U test because the EWALL-PI was not normally distributed using a Shapiro-Wilk Test for normality.

Figure 3.

Forest plots and tests of heterogeneity comparing risk of relapse and death across the 4 different trials. (A) Subdistribution hazard ratios from Fine-Gray competing risk models for risk of relapse. (B) Hazard ratios from Cox regression analysis for risk of death.

Figure 4.

Optimal threshold selection in the UKALL14 cohort. To determine the optimal EWALL-PI threshold for defining risk groups, we assessed 14 selected thresholds using a cohort of 79 patients treated on UKALL14 with chemotherapy only. The cumulative incidence of relapse (CIR [A]) and OS rates (B) were calculated for patients above (red) and below (blue) each threshold. (C-D) show the proportion of patients with EWALL-PI scores above (red) and below (blue) each of the selected EWALL-PI thresholds. Threshold analysis was performed by sorting the PI, dividing the cohort into bins comprising ∼5 cases (∼6% cohort) and sequentially testing each threshold until the exemplar clinical criteria were met. The most discriminatory threshold (ie, the one with the maximum difference between CIR and OS rates) was 2.50, which divided the chemotherapy cohort 75%/25% but the whole UKALL14 cohort 52%/48%. Patients with a EWALL-PI score <2.50 had a CIR at 3 years of 15% vs 35% for patients with a score ≥2.50 (P < .001). The equivalent OS rates at 3 years were 90% vs 59% (P < .001).

Figure 5.

Validation of EWALL-PI–defined risk groups across the 4 trials. UKALL14 (A-B), NILG-ALL10/07 (C-D), GIMEMA-LAL1913 (E-F) and PETHEMA-ALL-HR2011 (G-H). Panels A,C,E,G show Kaplan-Meier plots for CIR and the subdistribution hazard ratio for increased risk of relapse for patients at HR vs patients at SR. Panels B,D,F,H show the hazard ratio for OS. Patients with a EWALL-PI <2.50 were assigned to the SR group (blue lines) whereas patients at HR had EWALL-PI ≥2.50 (red lines).

Figure 6.

Forest plot showing the subdistribution hazard ratios for risk of relapse derived from Fine-Gray competing risk model and hazard ratios for death derived from Cox regression analysis, comparing patients in the EWALL-PI–defined HR and SR groups stratified by key patient subgroups within each trial and using a combined data set.