Relapse Thresholds (12/24 Mo) Define Survival Disparity in Pediatric B‐ALL

ABSTRACT

This study systematically analyzed relapse patterns in pediatric B‐cell acute lymphoblastic leukemia (B‐ALL) across 2930 patients from the TARGET and MP2PRT cohorts, with an additional 2972 patients from four independent external validation cohorts, to define clinically relevant prognostic thresholds. Monthly landmark‐based Cox analyses identified 13 months as the time point with the peak hazard ratio (HR = 31.97) and 27 months as the time point with the maximum −log10P value (158.08); given the small differences from 12 and 24 months and their greater clinical practicality, POD12 (progression of disease within 12 months) and POD24 (progression of disease within 24 months) were selected as clinically practical landmarks for subsequent analyses. In the TARGET and MP2PRT cohorts, POD12 occurred in 2.56% of patients and POD24 in 8.67%. Patients with POD12 had a 5‐year overall survival (OS) of 11.13% versus 90.89% in non‐POD12 patients, whereas patients with POD24 had a 5‐year OS of 36.19% versus 93.62% in non‐POD24 patients. In all four external validation cohorts, POD12 and POD24 were likewise associated with significantly inferior OS compared with their respective non‐POD groups. In univariate analyses, E2A‐PBX1 and MLL rearrangements were associated with increased risk of POD12 and POD24. These findings support POD12 and POD24 as clinically practical, data‐driven landmarks for identifying patients with adverse survival outcomes. They may also inform the exploratory evaluation of 1‐year and 2‐year progression‐free survival as hypothesis‐generating candidate early trial endpoints, pending prospective validation.

1. Introduction

Acute lymphoblastic leukemia is the most common hematological malignancy in children, with approximately 80% [1, 2, 3] being B‐cell acute lymphoblastic leukemia (B‐ALL). With the continuous refinement of risk‐stratified chemotherapy, the 5‐year overall survival (OS) rate for patients with B‐ALL has now exceeded 90% [4, 5, 6]. However, relapse remains the leading cause of treatment failure and mortality in B‐ALL. Approximately 15%–20% of children relapse after receiving first‐line therapy. These relapsed patients have a poor prognosis, with a 5‐year survival rate as low as 57.5% [7]. Employing cellular immunotherapy and allogeneic hematopoietic stem cell transplantation (allo‐HSCT) may improve long‐term survival in this patient cohort [8, 9].

Currently, the prognosis, risk stratification, and therapeutic strategies for relapsed B‐ALL are closely linked to the timing of relapse after initial diagnosis. In clinical practice and contemporary trials, relapse occurring within 18 and 36 months post‐diagnosis are widely adopted as pivotal time points defining “very early relapse” and “early relapse”, respectively [10, 11, 12]. Patients experiencing such relapses are associated with poor prognosis. However, a core limitation persists: current relapse time thresholds remain empirically derived from clinical practice rather than objective cutoffs established through large‐scale clinical data and rigorous statistical methods. These unvalidated definitions may compromise risk stratification accuracy and obscure distinct biological mechanisms of relapse. Moreover, these unvalidated thresholds may restrict our ability to implement more precise interventions.

Notably, the concept of “Progression of Disease within 24 months” (POD24) has been extensively validated in follicular lymphoma (FL) as a potent predictor of adverse outcomes [13, 14], clearly segregating patients into two distinct subgroups with divergent survival trajectories: those with POD24 exhibit a 5‐year OS of approximately 50%, compared with approximately 90% in non‐POD24 patients [15]. Subsequently, this strategy has been extended to extranodal natural killer/T‐cell lymphoma (ENKL) [16] and peripheral T‐cell lymphoma (PTCL) [17], validating that POD24 identifies patients with significantly inferior survival. In the present study, relapse was used as the sole criterion for defining POD. This proven approach serves as an important model for developing similar relapse timing thresholds in B‐ALL. Herein, by integrating multi‐source clinical datasets and applying rigorous statistical methods, we aimed to identify a relapse time threshold specific to B‐ALL that demonstrates sufficient discriminatory power to stratify patients into subgroups with significantly different survival outcomes. Moreover, through systematic integration of fundamental clinical and biological characteristics documented during diagnosis and treatment, we seek to identify risk factors associated with early relapse in B‐ALL (defined as relapse prior to the statistical time point).

2. Patients and Methods

2.1. Patients

This study established an integrated cohort using data from two public datasets and four Chinese hospital‐based cohorts. The Therapeutically Applicable Research to Generate Effective Treatments (TARGET) dataset, acquired from the Genomic Data Commons portal (https://portal.gdc.cancer.gov/projects/TARGET‐ALL), initially included 1700 patients. The Molecular Profiling to Predict Response to Treatment (MP2PRT) dataset, obtained from the same Genomic Data Commons repository (https://portal.gdc.cancer.gov/projects/MP2PRT‐ALL) and supplemented with data from Chang et al. [18], initially comprised 1510 patients. In addition, a multicenter Chinese pediatric B‐ALL cohort was constructed by collecting patient data from four hospitals from 2015 to 2025: Children's Hospital of Soochow University (CHSOU, n = 1306), Children's Hospital Affiliated to Zhengzhou University (CHAZU, n = 737), Union Hospital, Tongji Medical College, Huazhong University of Science and Technology (UTH, n = 533), and The First Affiliated Hospital of University of Science and Technology of China (FAHU, n = 396). To assemble the final analytical cohort, stringent eligibility criteria were uniformly applied: (1) confirmed diagnosis of B‐ALL, (2) age ≤ 18 years at diagnosis, (3) availability of definitive survival status, and (4) documented relapse status. After applying these filters, the eligible cohorts comprised 1437 patients from TARGET, 1493 from the MP2PRT dataset, and a total of 2972 patients from the four Chinese centers (CHSOU, CHAZU, UTH, and FAHU). After integration, the overall study population comprised 5902 pediatric patients with B‐ALL. For analyses of relapse time point identification, the TARGET and MP2PRT cohorts were used as the discovery cohort, whereas the four Chinese cohorts served as the external validation cohort (Figure 1).

FIGURE 1.

Cohort assembly and selection flow diagram. Flow diagram illustrating cohort construction from the TARGET and MP2PRT datasets and four independent Chinese cohorts (CHSOU, CHAZU, UTH, and FAHU), including inclusion and exclusion criteria and the final analytical populations for discovery and validation analyses. [Color figure can be viewed at wileyonlinelibrary.com]

2.2. Identification of Key Time Points

To determine the optimal critical threshold, we employed a sliding‐window landmark‐based framework to dynamically profile relapse‐associated mortality risk over monthly intervals from 6 to 48 months after diagnosis (43 time points in total). The initial discovery cohort comprised 2930 pediatric B‐ALL patients (combining the TARGET and MP2PRT datasets), with exclusions applied at each candidate time point: (1) deaths from non‐relapse causes prior to the candidate time point, and (2) patients without documented relapse whose follow‐up duration was shorter than the candidate time point. For each candidate time point, Cox proportional hazards models were established to compare mortality outcomes between patients who had relapsed by the candidate time point (POD group) and those who remained relapse‐free at the candidate time point, including patients who relapsed thereafter. At each candidate time point, patients were classified according to whether relapse had occurred by that time. Survival was evaluated from the time at which group membership was defined, with the date of relapse used for patients in the POD group and the corresponding candidate time point used for patients in the non‐POD group. Wald tests were used to assess the significance of hazard ratios. To account for multiple testing across the 43 candidate time points, Bonferroni correction was applied to the corresponding p values. The trajectories of hazard ratios (HRs) and transformed significance values (−log₁₀P) were visualized using dual‐axis plots. Candidate time points were identified based on the joint evaluation of HR magnitude and statistical significance across serial landmark‐based evaluations.

2.3. Statistical Analyses

All statistical analyses were performed using R version 4.4.2. To identify factors associated with POD status, three univariable analytic approaches were performed, with odds ratios (ORs) and corresponding 95% confidence intervals (CIs) calculated for each potential predictor across all three methods. The three univariable analytic approaches included standard univariable logistic regression, univariable logistic regression after random hot‐deck imputation for missing data, and univariable analysis using a multinomial regression framework to account for competing outcomes, in which non‐relapse death was treated as a separate competing event category. For hot‐deck imputation, missing values in each variable were imputed by random sampling with replacement from the observed non‐missing values of that variable using the sample function, with the random seed set to 123. In addition, the cumulative incidence of relapse was estimated using competing‐risks methods. Survival probabilities were generated via the Kaplan–Meier estimator and compared using the log‐rank test (significance threshold, p < 0.05). For analyses related to POD‐based time point identification and validation, the same landmark‐based classification framework and time origin definition were applied. This framework was used for monthly candidate time points from 6 to 48 months after diagnosis, the two critical POD thresholds identified in this study, and additional comparator thresholds. For analyses comparing survival by relapse timing, the same time origin definition was used, with 5 years after diagnosis serving as the time origin for patients without relapse within 5 years. The timing of survival tracking initiation and patient exclusion criteria were consistently applied as specified in the Identification of Key Time Points section. To assess the robustness of the identified time points, we additionally performed a sensitivity analysis using an alternative survival time definition, in which survival time for POD patients was calculated as OS minus relapse time and survival time for non‐POD patients was calculated as OS minus the median relapse time of patients who relapsed by each time point.

3. Results

3.1. Demographic and Clinical Characteristics

The combined cohort of 2930 pediatric B‐ALL patients demonstrated heterogeneity in baseline clinical and demographic features (Table S1). Patients aged 1–10 years predominated in MP2PRT (96.72%), while infants (< 1 year) and adolescents (> 10 to ≤ 18) accounted for 42.52% of the TARGET cohort. Both cohorts were predominantly White (TARGET: 75.09%, MP2PRT: 77.43%). Non‐CNS3 status was prevalent in TARGET (96.24%), with no CNS involvement data available for MP2PRT. MP2PRT utilized a two‐tiered risk stratification (standard‐risk, SR; high‐risk, HR), with 92.50% classified as SR and 7.50% as HR. Hyperleukocytosis (white blood cell, WBC ≥ 100 × 10⁹/L at diagnosis) was observed in 21.16% of TARGET patients. Consistent with these baseline differences, cohort‐specific univariate Cox regression analyses for overall survival showed that prognostic associations differed between TARGET and MP2PRT, as reflected by differences in HRs and 95% CIs (Figure S1).

3.2. Early Relapse Is Strongly Associated With Inferior Survival

The cumulative relapse curve over time (Figure 2A) revealed a 5‐year cumulative incidence of relapse of 24.52% (95% CI: 22.95%–26.10%) in the entire cohort. Notably, the slope of relapse incidence markedly flattened beyond 5 years (average annual increase < 2.00%), suggesting significantly reduced long‐term relapse risk. The instantaneous relapse hazard function (Figure 2B) showed a sharp increase in the first 3 years after diagnosis, reaching a peak hazard rate of 0.077 at 3.3 years. Following this peak, the hazard rate declined rapidly, dropping below 0.02 by 5.8 years post‐diagnosis.

FIGURE 2.

Relapse dynamics and survival outcomes in B‐ALL patients from the TARGET and MP2PRT cohorts. (A) Cumulative relapse incidence. (B) Hazard rate of relapse over time in B‐ALL patients. (C) Overall survival by relapse timing; survival time was measured from relapse for relapsed patients and from 5 years after diagnosis for patients without relapse within 5 years. [Color figure can be viewed at wileyonlinelibrary.com]

Survival analysis based on time to relapse (Figure 2C), with survival time measured from relapse for patients who relapsed within 5 years and from 5 years after diagnosis for those without relapse during this period, revealed a significant gradient in OS rates across relapse‐interval groups. Patients relapsing within 1 year had the poorest OS (5‐year OS: 11.13%; 95% CI, 5.64%–21.96%). Later relapse timing was associated with progressively higher 5‐year OS: 46.55% (95% CI, 39.70%–54.59%) for 1–2 years, 53.33% (95% CI, 46.26%–61.48%) for 2–3 years, 61.80% (95% CI, 54.81%–69.69%) for 3–4 years, 75.91% (95% CI, 67.11%–85.87%) for 4–5 years. Patients without relapse within 5 years exhibited the highest 5‐year OS rate (98.87%; 95% CI, 98.30%–99.44%). These findings suggest that earlier relapse timing is strongly associated with reduced OS.

3.3. POD12 And POD24 Identify Patients With Inferior Survival

Comprehensive analysis of relapse timing across monthly intervals from 6 to 48 months post‐diagnosis (Figure 3A), using overall survival measured from relapse for POD patients and from the corresponding cutoff time after diagnosis for non‐POD patients, revealed that relapse occurring within 13 months post‐diagnosis corresponded to the peak observed hazard ratio (HR = 31.97, 95% CI, 20.49–49.89, for relapse within 13 months vs. no relapse within 13 months), while relapse within 27 months demonstrated the most statistically significant survival discrimination (−log10P = 158.08; relapse within 27 months vs. no relapse within 27 months). Given the small differences between adjacent time points and clinical practicality, we evaluated the use of 12‐month and 24‐month thresholds for defining early disease progression. Specifically, the HR for relapse at 12 months was 31.20 (95% CI, 19.71–49.40), compared with 31.97 at 13 months, representing a numerical difference of 0.77. Similarly, the −log10P value for relapse at 24 months was 147.61, versus 158.08 at 27 months, with a difference of 10.47. Both time points remained statistically significant after Bonferroni correction (adjusted p < 0.05). A sensitivity analysis using an alternative survival time origin, with non‐POD patients anchored to the median relapse time among patients who relapsed by each time point, yielded results similar to those observed in the primary analysis (Figure S4). Quantitative analysis of restricted mean survival time at 6 years (RMST₆) further supported the similarity of these time points. The difference in RMST₆ (ΔRMST₆) between non‐relapse and relapse patients was 4.29 years at 12 months (Figure 3B) and 4.28 years at 13 months (Figure S5A), showing a small numerical difference. Likewise, the ΔRMST₆ between non‐relapse and relapse patients was 2.98 years at both 24 and 27 months (Figures 3C and S5B). In addition, post‐relapse survival analyses using the adjacent 13‐ and 27‐month thresholds showed similar patterns to those observed with the 12‐ and 24‐month cutoffs (Figures 3D,E and S5C,D). Clinically, the transition from 13 to 12 months affected only 8 of 2930 patients (0.27%), and the adjustment from 27 to 24 months involved 43 patients (1.47%). Furthermore, the use of 13 or 27 months as thresholds would likely necessitate more frequent patient follow‐up. Accordingly, considering both the prognostic performance and clinical practicality of these time points, 12 months (POD12) and 24 months (POD24) were adopted as data‐driven, clinically practical landmarks for characterizing early disease progression in B‐ALL.

FIGURE 3.

Defining POD12 and POD24 as critical early progression thresholds and their association with survival outcomes in the TARGET and MP2PRT B‐ALL cohorts. (A) Optimal time point identification for early relapse based on overall survival analyses using each candidate cutoff (Bonferroni‐adjusted p‐values and hazard ratio; dashed line: Bonferroni‐adjusted p = 0.05). (B) Overall survival: POD12 (red) vs. non‐POD12 (blue). RMST₆, restricted mean survival time at 6 years; HR, hazard ratio. (C) Overall survival: POD24 (red) vs. non‐POD24 (blue). (D) Post‐relapse survival: POD12 (red) vs. relapse after 12 months (black). (E) Post‐relapse survival: POD24 (red) vs. relapse after 24 months (black). For panels A–C, survival time was measured from relapse for POD patients and from the corresponding cutoff time after diagnosis for non‐POD patients; for panels D and E, post‐relapse survival was measured from relapse. [Color figure can be viewed at wileyonlinelibrary.com]

To ensure comparability of outcome assessment, we excluded patients with non‐relapse deaths within 12 months (n = 26) or insufficient follow‐up without relapse (n = 26) for POD12 analysis. Similarly, POD24 analysis further excluded those with non‐relapse deaths within 24 months (n = 40) or inadequate relapse‐free follow‐up (n = 39) (Figure 1). Kaplan–Meier analyses using the same survival time definition corroborated the prognostic power of these cutoffs: POD12 patients exhibited drastically inferior 5‐year OS versus non‐POD12 counterparts (11.13% vs. 90.89%; Figure 3B), with consistent findings in the POD24 cohort (36.19% vs. 93.62%; Figure 3C). Crucially, early progression within 12 months portended worse survival than later progression after 12 months (5‐year OS: 11.13% vs. 60.00%; Figure 3D), and patients with POD24 fared significantly worse than those with progression beyond 24 months (5‐year OS: 36.19% vs. 64.48%; Figure 3E). Collectively, these data indicate that POD12 and POD24 are robust indicators of inferior survival. Consistent with these overall results, separate analyses in the TARGET and MP2PRT cohorts identified similar candidate cutoff time points for early relapse (Figures S2A and S3A), and further validated POD12 and POD24 as robust prognostic factors associated with inferior survival (Figures S2B–E and S3B–E).

3.4. 12/24‐Month Relapse Thresholds Show Greater Survival Separation Than 18/36‐Month Cutoffs

Relapse in pediatric B‐ALL is associated with poor survival outcomes, particularly for patients experiencing early disease progression. Given the dismal outcomes associated with relapse, we sought to identify clinically practical, data‐informed relapse landmarks for stratifying patient subgroups with different survival outcomes. We compared the ability of 12/24‐month and 18/36‐month relapse thresholds to stratify patient outcomes. Hypothetical curves (Figure 4A,B) illustrate idealized scenarios of maximal POD/non‐POD separation versus no meaningful distinction. In the combined TARGET and MP2PRT B‐ALL cohorts, the 12‐month threshold was associated with greater survival separation than the 18‐month cutoff, with a larger ΔRMST₆ (4.29 vs. 3.35 years; Figure 4C). Similarly, the 24‐month threshold was associated with greater survival separation than the 36‐month cutoff, with a larger ΔRMST₆ (2.98 vs. 2.66 years; Figure 4D). Notably, the 12‐month and 24‐month thresholds permit earlier classification of relapse timing compared with the 18‐month and 36‐month cutoffs. A Sankey diagram (Figure 4E) illustrates the overlap between conventional relapse timing categories and POD12/POD24‐based categories. Among patients classified as very early relapse by conventional thresholds, 82 corresponded to the early POD category, whereas 172 patients classified as early relapse corresponded to the late POD category. Landmark analysis (Figure 4F) showed the cumulative incidence of relapse and non‐relapse mortality across landmark time points, with cumulative relapse rates of 2.58% and 6.40% at 1 and 2 years, respectively.

FIGURE 4.

Comparison of 12/24‐ and 18/36‐month relapse thresholds in TARGET and MP2PRT B‐ALL cohorts. (A) Hypothetical survival curve with maximal POD/non‐POD separation. (B) Hypothetical survival curve with no POD/non‐POD separation. (C) Overall survival by 12/18‐month relapse status: No relapse by 12 months (light orange), relapse by 12 months (dark orange), no relapse by 18 months (light blue), relapse by 18 months (dark blue). (D) Overall survival by 24/36‐month relapse status: No relapse by 24 months (light orange), relapse by 24 months (dark orange), no relapse by 36 months (light blue), relapse by 36 months (dark blue). For panels C and D, survival time was measured from relapse for patients who relapsed within the indicated time window and from the corresponding cutoff time after diagnosis for patients without relapse within that window. ΔRMST₆, difference in restricted mean survival time at 6 years. (E) Overlap between conventional relapse timing categories and POD12/POD24‐based categories, shown using a Sankey diagram. Values represent the number of patients in each category overlap. (F) Landmark analysis of cumulative incidence: Relapse (blue), non‐relapse mortality (red). [Color figure can be viewed at wileyonlinelibrary.com]

3.5. 12/24‐Month Relapse Thresholds Were Validated Across Four Independent Cohorts

To validate the prognostic stratification ability of the 12/24‐month relapse thresholds, we performed external validation in four independent pediatric B‐ALL cohorts (CHSOU, CHZAU, UTH, FAHU). Consistent with the primary cohort findings, POD12 and POD24 status were associated with significantly inferior overall survival across all four cohorts (Figure 5). At 4 years, OS rates for POD12 versus non‐POD12 patients were 50.87% vs. 96.70% in the CHSOU cohort, 50.63% vs. 97.74% in the CHZAU cohort, 60.00% vs. 98.05% in the UTH cohort, and 66.67% vs. 98.36% in the FAHU cohort. Similarly, 4‐year OS rates for POD24 versus non‐POD24 patients were 63.28% vs. 97.53% in the CHSOU cohort, 31.77% vs. 98.24% in the CHZAU cohort, 68.69% vs. 98.50% in the UTH cohort, and 60.00% vs. 98.92% in the FAHU cohort. Across all four cohorts, POD12 and POD24 were each associated with a more than 20‐fold higher hazard of death than the corresponding non‐POD groups (all p < 0.0001). These results consistently demonstrated robust survival separation between POD and non‐POD patients, confirming the generalizability of the 12/24‐month thresholds across distinct patient populations.

FIGURE 5.

Overall survival according to POD12 and POD24 status across four independent pediatric B‐ALL cohorts. Kaplan–Meier curves comparing overall survival between POD12 and non‐POD12 groups (left panels) and between POD24 and non‐POD24 groups (right panels) in each cohort. (A, B) CHSOU cohort; (C, D) CHAZU cohort; (E, F) UTH cohort; (G, H) FAHU cohort. Survival time was measured from relapse for patients who relapsed within 12 or 24 months and from the corresponding cutoff time after diagnosis for patients without relapse within those windows. POD12/POD24 groups are shown in red, and non‐POD12/non‐POD24 groups in blue. RMST₄, restricted mean survival time at 4 years. [Color figure can be viewed at wileyonlinelibrary.com]

We further evaluated post‐relapse survival outcomes in a combined cohort of patients from the four independent B‐ALL cohorts to assess the prognostic impact of different relapse timing thresholds. The overlap between conventional relapse timing categories and POD12/POD24‐based categories in the combined cohort is summarized in Figure S6. In the combined cohort, patients with POD12 exhibited significantly inferior 4‐year post‐relapse survival compared with those who relapsed after 12 months (52.53% vs. 68.19%, p = 0.0027; Figure 6A). A similar, statistically significant difference was observed between patients with POD24 and those who relapsed after 24 months (60.18% vs. 69.40%, p = 0.022; Figure 6B). The 18‐month cutoff also identified a significant difference in post‐relapse survival (59.43% vs. 67.74%, p = 0.021; Figure 6C). However, the prognostic effect decreased, with a lower HR of 1.68 (95% CI, 1.08–2.64) and a smaller ΔRMST₄ (the difference in restricted mean survival time at 4 years) of 0.44 compared with the 12‐month threshold (HR = 2.06, 95% CI, 1.27–3.34; ΔRMST₄ = 0.68). In contrast, the 36‐month cutoff failed to detect a statistically meaningful difference in post‐relapse survival (63.96% vs. 67.76%, p = 0.27; Figure 6D), with an HR of 1.32 (95% CI, 0.80–2.16) and a lower ΔRMST₄ of 0.18 compared with the 24‐month threshold (HR = 1.68, 95% CI, 1.07–2.62; ΔRMST₄ = 0.39). Collectively, these findings suggest that the 12/24‐month POD landmarks provide additional clinically relevant post‐relapse risk stratification within the conventional 18/36‐month relapse timing framework.

FIGURE 6.

Post‐relapse survival by different relapse timing thresholds in combined CHSOU, CHAZU, UTH, and FAHU B‐ALL cohorts. (A) Post‐relapse survival: Relapse within 12 months (red) vs. relapse after 12 months (black). (B) Post‐relapse survival: Relapse within 24 months (red) vs. relapse after 24 months (black). (C) Post‐relapse survival: Relapse within 18 months (red) vs. relapse after 18 months (black). (D) Post‐relapse survival: Relapse within 36 months (red) vs. relapse after 36 months (black). Post‐relapse survival was calculated from the date of relapse for all patients. [Color figure can be viewed at wileyonlinelibrary.com]

3.6. Key Clinical Variables Are Associated With POD12 and POD24

Using two independent datasets (TARGET and MP2PRT), we evaluated baseline predictors of POD12 and POD24 by univariate logistic regression, including genetic markers and clinical response indicators. Results from the TARGET cohort (Table 1) confirmed that both E2A‐PBX1 (OR = 4.38, 95% CI, 2.20–8.34, p < 0.001) and MLL rearrangements (OR = 3.75, 95% CI: 1.47–8.38, p = 0.008) were significant risk factors for POD12. Additionally, hyperleukocytosis (white blood cell, WBC ≥ 100 × 10⁹/L) at diagnosis and age group (< 1 or > 10 years) showed significant associations exclusively with POD24 (p < 0.001 and p = 0.007, respectively), with no significant links to POD12 (all p > 0.05). ETV6‐RUNX1 functioned as a protective factor against both POD12 (OR = 0.08, 95% CI, 0.004–0.35, p < 0.001) and POD24 (OR = 0.13, 95% CI, 0.05–0.29, p < 0.001). Hyperdiploidy (51–65 chromosomes) significantly reduced risks for POD12 (OR = 0.07, 95% CI, 0.004–0.33, p < 0.001) and POD24 (OR = 0.32, 95% CI, 0.16–0.60, p < 0.001). Trisomies 4 and 10 conferred consistent protection at POD12 (OR = 0.11, 95% CI, 0.006–0.49, p = 0.001) and POD24 (OR = 0.24, 95% CI, 0.09–0.51, p < 0.001); similarly, trisomy 21 reduced risks for POD12 (OR = 0.24, 95% CI, 0.08–0.56, p < 0.001) and POD24 (OR = 0.39, 95% CI, 0.22–0.64, p < 0.001). D29 minimal residual disease (MRD) positivity increased risks for both POD12 (OR = 2.07, 95% CI, 1.21–3.54, p = 0.009) and POD24 (OR = 2.56, 95% CI, 1.80–3.62, p < 0.001), whereas CNS involvement remained non‐significant (p > 0.05).

TABLE 1.

Univariate logistic regression analysis for POD12 and POD24 relapse in TARGET patients with B‐ALL.

Variables	POD12 (N = 1385)				POD24 (N = 1358)
	N	Missing (%)	OR (95% CI)	p	N	Missing (%)	OR (95% CI)	p
WBC, ≥ 100 × 109/L (vs. < 100 × 109/L)	291	0	1.52 (0.83–2.65)	0.17	284	0	2.26 (1.56–3.24)	< 0.001
Age, < 1 or > 10 years (vs.1–10 years)	585	0	1.60 (0.95–2.69)	0.08	566	0	1.60 (1.14–2.24)	0.007
Gender, male (vs. female)	787	0	0.99 (0.59–1.69)	0.98	773	0	1.26 (0.89–1.79)	0.19
MLLr (yes vs. no)	44	23.18	3.75 (1.47–8.38)	0.008	43	23.05	3.72 (1.89–7.07)	< 0.001
E2A‐PBX1 (yes vs. no)	93	28.23	4.38 (2.20–8.34)	< 0.001	92	28.13	2.64 (1.55–4.37)	< 0.001
BCR‐ABL1 (yes vs. no)	37	0.29	1.29 (0.21–4.38)	0.74	31	0.29	1.55 (0.52–3.78)	0.40
ETV6‐RUNX1 (yes vs. no)	217	18.19	0.08 (0.004–0.35)	< 0.001	217	17.97	0.13 (0.05–0.29)	< 0.001
CNS3 (vs. Non‐CNS3)	53	0	1.34 (0.32–3.80)	0.64	52	0	1.25 (0.51–2.64)	0.61
D29‐MRD (≥ 0.01% vs. < 0.01%)	405	0.87	2.07 (1.21–3.54)	0.009	392	0.81	2.56 (1.80–3.62)	< 0.001
High hyperdiploidy (51–65 chromosomes, yes vs. no)	206	29.39	0.07 (0.004–0.33)	< 0.001	205	29.46	0.32 (0.16–0.60)	< 0.001
Trisomies 4 and 10 (yes vs. no)	153	29.39	0.11 (0.006–0.49)	0.001	152	29.46	0.24 (0.09–0.51)	< 0.001
Trisomy 21 (yes vs. no)	299	29.39	0.24 (0.08–0.56)	< 0.001	296	29.46	0.39 (0.22–0.64)	< 0.001

Abbreviations: CNS3, central nervous system leukemia status 3; MLLr, MLL rearrangement; MRD, minimal residual disease; WBC, white blood cell.

Univariate analysis of the MP2PRT cohort (Table S2) demonstrated that E2A‐PBX1 significantly increased risks of both POD12 (OR = 11.66, 95% CI, 1.60–57.86, p = 0.02) and POD24 (OR = 3.82, 95% CI, 1.21–10.22, p = 0.02), whereas ETV6‐RUNX1 conferred protection against POD12 (OR = 0.25, 95% CI, 0.04–0.92, p = 0.04) and POD24 (OR = 0.49, 95% CI, 0.30–0.77, p = 0.002), aligning with findings from the TARGET cohort. Trisomy 21 reduced risks for POD12 (OR = 0.08, 95% CI, 0.005–0.44, p = 0.001) and POD24 (OR = 0.62, 95% CI, 0.39–0.97, p = 0.04). Hyperdiploidy and trisomies 4 and 10 significantly reduced POD24 risk (hyperdiploidy: OR = 0.49, 95% CI, 0.28–0.81, p = 0.005; trisomies 4 and 10: OR = 0.24, 95% CI, 0.10–0.49, p < 0.001). In contrast, impacts of age, risk stratification, hyperdiploidy, and trisomies 4 and 10 on POD12 could not be reliably assessed due to inadequate sample size in the POD12 subgroup.

Sensitivity analyses, including random imputation for missing data and treating non‐relapse death as a competing event, yielded results consistent with the aforementioned univariate analysis (TARGET cohort: Tables S3 and S4; MP2PRT cohort: Tables S5 and S6).

4. Discussion

Through a multi‐cohort analysis of large‐scale datasets, this study demonstrates the prognostic significance of POD12 and POD24 in pediatric B‐ALL. Comprehensive statistical evaluation identified these time points as clinically informative, data‐driven thresholds with distinct prognostic value and potential clinical relevance. Most importantly, in the combined analysis of the TARGET and MP2PRT cohorts, patients experiencing POD12 or POD24 had markedly inferior long‐term outcomes, with 5‐year OS of 11.13% and 36.19%, respectively, compared with > 90% survival in non‐POD12/POD24 patients. These findings were further validated in four independent external cohorts, which consistently confirmed worse survival among patients with POD12 or POD24 compared with non‐POD12/POD24 patients. While POD12 occurred in 2.56% and POD24 in 8.67% of TARGET and MP2PRT patients, these thresholds precisely delineated a subgroup with exceptionally poor prognosis. Taken together, our findings provide robust evidence that POD12 and POD24 are highly informative prognostic thresholds for identifying a small subset of pediatric B‐ALL patients at ultra‐high risk. POD12 and POD24 represent data‐driven, clinically practical landmarks associated with marked survival differences between POD and non‐POD patient subgroups and complement established relapse‐timing definitions.

The adverse outcomes associated with POD12 and POD24 are biologically and clinically plausible, given their close association with established high‐risk disease features. In the TARGET cohort, the POD12/24 groups were enriched for established high‐risk factors associated with relapse, including MLL rearrangements [19, 20], which are notably associated with poor prognosis in infants [21, 22, 23, 24], elevated WBC count [25, 26], extreme age groups (infants and adolescents) [25, 26], and positive MRD status [27, 28]. Across the TARGET and MP2PRT cohorts, E2A‐PBX1 was consistently associated with increased risk, whereas ETV6‐RUNX1 and favorable cytogenetic features, including high hyperdiploidy, trisomies 4 and 10, and trisomy 21, were associated with reduced risk of POD12 or POD24. These findings are consistent with the report by Reismüller et al. [29], which showed that patients with MLL rearrangements or E2A‐PBX1 tend to relapse earlier, and with prior studies linking ETV6‐RUNX1 to more favorable outcomes and later relapse [19, 30, 31, 32]. Collectively, these results indicate that POD12 and POD24 are not merely temporal cutoffs but clinically meaningful markers that reflect the interplay between relapse timing and underlying disease biology.

Within the 2026 NCCN framework for pediatric ALL, post‐relapse management is not determined by relapse timing alone, but rather by an integrated assessment of relapse setting, response to reinduction, MRD status, disease biology, and transplant eligibility [33]. In this context, incorporating 12‐month and 24‐month landmarks into the conventional 18/36‐month relapse timing framework may help refine post‐relapse risk assessment and clinical management. Specifically, the narrower thresholds may more precisely identify a particularly high‐risk subgroup that should be considered for earlier clinical trial referral, earlier transplant evaluation, and earlier use of immunotherapy, targeted agents, or other intensified salvage approaches, rather than reliance on conventional salvage chemotherapy alone. In selected relapse settings, patients experiencing POD12 or POD24 may be considered for salvage strategies that include blinatumomab [34, 35, 36]. Conversely, patients relapsing between 12 and 18 months or between 24 and 36 months may no longer be uniformly grouped within the highest‐risk category defined by relapse timing, allowing post‐relapse treatment intensity to be better individualized according to MRD response, reinduction sensitivity, and underlying disease biology.

Several limitations warrant consideration. First, the retrospective nature of this study introduces potential confounding factors. Additionally, critical data were incomplete in the MP2PRT cohort, including baseline WBC counts, MLL rearrangement status, and others, limiting comprehensive risk assessment in this group. Although the external validation cohorts support the reproducibility of our findings across several Chinese centers, they do not fully resolve the generalisability of POD12 and POD24 across international treatment protocols, racial or ethnic groups, and treatment eras.

Our findings suggest that POD12 and POD24 may serve as clinically informative, data‐driven landmarks within the conventional relapse timing framework based on 18 months and 36 months, given their significantly stronger association with adverse outcomes in our HR analyses spanning 6 to 48 months. Earlier identification of POD12 or POD24 may help refine the clinical interpretation of relapse timing and support more individualized assessment after relapse. In addition, our results raise the possibility that 1‐year and 2‐year progression‐free survival could be explored as candidate early efficacy endpoints in frontline pediatric B‐ALL trials, potentially enabling earlier assessment of treatment activity. However, such exploratory, hypothesis‐generating use would require prospective validation before these measures can be considered surrogate endpoints for trial design or drug evaluation. Future studies in carefully characterized prospective cohorts are needed to confirm the clinical usefulness, generalizability, and relevance of POD12 and POD24 in the trial setting.

Author Contributions

B.X. performed most of the data analysis and drafted the manuscript. J.Z., X.Z., F.Z., and S.H. contributed to data collection and data curation. Y.S. and J.C. provided unique insights into recurrence. H.S. and Y.F. provided technical support. During the revision process, Y.F. made significant contributions to the manuscript update, primarily through his statistical insights and novel interpretations of the data. Q.L. and W.T. contributed to the discussion. H.Y. developed the conception, designed the project, supervised the study, and critically revised the manuscript. All authors read and approved the final manuscript.

Funding

This work was supported by National Natural Science Foundation of China, 81911530169, Chongqing Medical Scientific Research Project (Joint project of Chongqing Health Commission and Science and Technology Bureau), 2025ZDXM005, The Science and Technology Research Program of Chongqing Municipal Education Commission, KJZD‐K202300408, and the Innovation Support Program for Chongqing Overseas Returnees, cx2025115. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Ethics Statement

This study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committees of the participating institutions.

Consent

This retrospective study was approved by the Ethics Committees of the participating institutions, with a waiver of informed consent. The study strictly adhered to patient privacy protection and confidentiality principles.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Cohort‐specific univariate Cox regression analyses of overall survival in TARGET and MP2PRT.

Figure S2: Defining POD12 and POD24 as critical early progression thresholds and their association with survival outcomes in the TARGET B‐ALL cohorts.

Figure S3: Defining POD12 and POD24 as critical early progression thresholds and their association with survival outcomes in the MP2PRT B‐ALL cohorts.

Figure S4: Optimal time point identification for early relapse with adjusted non‐POD survival time.

Figure S5: Sensitivity analysis of 13/27‐month relapse thresholds for 12/24‐month clinical thresholds in TARGET and MP2PRT B‐ALL cohorts.

Figure S6: Overlap between conventional relapse timing categories and POD12/POD24‐based categories in combined CHSOU, CHZAU, UTH, and FAHU B‐ALL cohorts.

Table S1: Demographic and clinical characteristics of patients in two datasets.

Table S2: Univariate logistic regression analysis for POD12 and POD24 relapse in MP2PRT patients with B‐ALL.

Table S3: Univariate logistic regression analysis for POD12 and POD24 relapse in TARGET patients with B‐ALL (imputed data).

Table S4: Multinomial logistic regression analysis for POD12 and POD24 relapse in TARGET patients with B‐ALL (competing events).

Table S5: Univariate logistic regression analysis for POD12 and POD24 relapse in MP2PRT patients with B‐ALL.

Table S6: Multinomial logistic regression analysis for POD12 and POD24 Relapse in MP2PRT patients with B‐ALL (competing events).

Data Availability Statement

The data are not publicly available due to privacy or ethical restrictions.