We retrospectively identified consecutive children with de novo RUNX1::RUNX1T1-positive AML diagnosed between September 2009 and May 2021 and treated on AML-CAMS serial trials at the Pediatric Blood Disease Center, Blood Diseases Hospital, Chinese Academy of Medical Sciences (CAMS). In total, 146 patients received treatment under AML-CAMS-2009 (NCT03165851) or AML-CAMS-2016 (NCT03173612) protocols. After excluding induction deaths (n = 3), loss to follow-up before the initial assessment (n = 1), and induction refractory disease (n = 6), 136 patients achieving morphological CR after one induction course were included (Figure 1).

Both protocols used induction followed by high-dose cytarabine-based consolidation. Regimen assignment was primarily determined by protocol availability during the corresponding trial period based on the calendar time of diagnosis and enrollment, rather than concurrent allocation within the same time period. AML-CAMS-2009 regimen comprised five consolidation cycles; AML-CAMS-2016 used four cycles with reduced-intensity chemotherapy plus dasatinib. Regimen details were reported previously (14, 17). This study was approved by the Institutional Ethics Committee of the Blood Diseases Hospital, CAMS, and performed in compliance with the Declaration of Helsinki. Written informed consent was obtained from guardians. Data were abstracted from existing records and anonymized before analysis.

Candidate covariates were prespecified a priori based on published evidence and biological plausibility in pediatric RUNX1::RUNX1T1-positive AML, with an emphasis on clinical relevance and routine availability at diagnosis and early treatment response. Variables were additionally constrained by data completeness and quality in our cohort. We reviewed baseline demographics (age, sex), peripheral blood counts at diagnosis (WBC, hemoglobin, and platelets), presence of extramedullary infiltration (EMI), treatment regimen, and transplantation status. Laboratory and molecular data included RUNX1::RUNX1T1 transcript quantification by real-time quantitative polymerase chain reaction (RT-qPCR) at diagnosis and subsequent time points, cytogenetics, and KIT mutations (14). “High-Risk MRD” was defined as <3-log transcript reduction in RUNX1::RUNX1T1 after induction and MRD ≥0.01% after two consolidations (14). This combined criterion was prespecified based on our published AML-CAMS MRD dynamics study, in which it was independently associated with inferior RFS and OS (14), and it was also included as a covariate in the prognostic modeling. To more comprehensively capture baseline disease burden and biologically relevant factors potentially associated with clinical outcomes, we additionally incorporated baseline bone marrow (BM) blasts by morphology (BM blasts [Morph]), BM blasts by flow cytometry (BM blasts [FCM]), lactate dehydrogenase (LDH) levels, and RAS-pathway mutations.

Conventional cytogenetic karyotyping and fluorescence in situ hybridization (FISH) were routinely performed to screen for and rapidly confirm the t (8,21)(q22;q22) translocation at diagnosis. RT-qPCR was used for baseline quantification of the RUNX1::RUNX1T1 fusion transcript and for subsequent response assessment and molecular MRD monitoring. A standardized RT-qPCR protocol was implemented. RNA was prepared from BM mononuclear cells, using fresh specimens from AML-CAML-2016 and cryopreserved specimens from AML-CAML-2009, and then reverse-transcribed into cDNA. Quantification of RUNX1::RUNX1T1 was performed with a TaqMan probe system, with ABL serving as the internal reference. Assays were accepted only when ABL exceeded 1 × 10^5 copies. The assay sensitivity reached 10 copies per reaction, and transcript levels were presented as the RUNX1::RUNX1T1-to-ABL copy-number ratio. All molecular analyses were performed in an accredited reference laboratory following standardized protocols. A complex karyotype indicated ≥2 additional clonal chromosomal abnormalities alongside t (8,21). KIT and RAS pathway mutations were identified by first- or next-generation sequencing in BM.

Patients were followed by visits, phone contact, and electronic records. BM aspiration was performed before each chemotherapy course to evaluate treatment response. CR was characterized by <5% blasts in bone marrow smears, no Auer rod, no circulating or extramedullary disease, absolute neutrophil ≥1.0 × 10⁹/L, and platelet count ≥100 × 10⁹/L.

Given the rarity of relapse beyond three years, the observation window was set at three years (15). OS was measured from diagnosis to death due to any cause or last follow-up within this period. RFS was measured from the date of CR to relapse, death, or last follow-up within the same timeframe. Relapse was defined as ≥5% BM blasts (excluding post-chemotherapy regeneration), recurrence of leukemic cells in peripheral blood, or extramedullary infiltration confirmed by morphological or cytogenetic examination.

Continuous variables were presented as medians with interquartile ranges (IQRs), and categorical variables were reported as frequencies and proportions. Variables with <10% missingness (excluding outcomes/follow-up) were imputed using the random forest algorithm (mice package, R). Clinical, MICM (morphology, immunophenotyping, cytogenetics, and molecular biology), and treatment variables were analyzed by the least absolute shrinkage and selection operator (LASSO) regression, with optimal penalty parameter (λ) determined by ten-fold cross-validation. Selected variables entered multivariate Cox regression to identify independent prognostic factors, based on which nomograms predicting 1-, 2-, and 3-year OS and RFS were constructed. Model performance was evaluated using receiver operating characteristic (ROC) curves, calibration plots, decision curve analysis (DCA), and concordance index (C-index). Internal validation was performed using 500 bootstrap resamples. Risk scores derived from the nomogram were calculated, and optimal cutoffs identified via X-tile (v3.6.1) to classify patients into high- and low-risk groups. Survival differences were compared using Kaplan-Meier and log-rank tests. In the survival analyses, HSCT was not treated as a censoring event. Patients were followed from diagnosis to the predefined endpoints regardless of HSCT, and relapse or death occurring after HSCT was counted as an event for RFS or OS, respectively. All statistical analyses were conducted using R software (version 4.3.0), and statistical significance was defined as a two-tailed P value below 0.05.

A total of 136 pediatric patients diagnosed with RUNX1::RUNX1T1-positive AML who achieved CR following one course of induction chemotherapy were enrolled in the study. In this cohort, the earliest death occurred 8 months after diagnosis; no induction-related deaths were observed within the first 30 or 60 days after diagnosis. The median diagnostic age was 8 years (IQR, 6–11), and 57.35% were male. During follow-up, 22 patients (16.2%) died and 28 patients (20.6%) relapsed (Table 1).

At diagnosis, the median WBC count was 13.51×10⁹/L (IQR, 7.43–23.11). Mortality was higher in patients with WBC ≥20×10⁹/L than in those with WBC <20×10⁹/L (54.55% vs 25.44%; P = 0.006). Consistently, a greater proportion of relapsed patients had WBC ≥20×10⁹/L at diagnosis compared with non-relapsed patients (46.43% vs 25.93%; P = 0.035). When stratified by diagnostic WBC, relapse occurred more frequently in the WBC ≥20×10⁹/L group (13/41, 31.71%) than in the WBC <20×10⁹/L group (15/95, 15.79%) (Supplementary Table 1). Moreover, among patients who relapsed, post-relapse mortality was higher in the WBC ≥20×10⁹/L group than in the WBC <20×10⁹/L group (Supplementary Figure 1). No significant intergroup differences were observed for hemoglobin, platelet count, LDH levels, or BM blasts (Morph). Cytogenetic abnormalities included complex karyotype (2.94%), monosomy X (12.50%), monosomy Y (32.35%), and any monosomy (loss of X, Y, or chromosome 9, 46.32%) without significant differences between survival or relapse groups. KIT, NRAS, and KRAS mutations were detected in 39.71%, 19.85%, and 8.82% of patients, respectively, with no prognostic significance (all P>0.05). Similarly, no marked difference was observed in the frequencies of KIT exon 17 or KIT p.D816 mutations among groups. Of the entire cohort, 18 patients (13.2%) underwent allogeneic hematopoietic stem cell transplantation (allo-HSCT), with comparable frequencies between survivors and non-survivors (13.16% vs. 13.64%; P>0.05). Additional clinical characteristics of HSCT recipients and HSCT-related details, including indications and timing, are summarized to document real-world HSCT practice in our cohort (Supplementary Tables 2, 3). Treatment regimen and MRD status were key prognostic indicators. The AML-CAMS-2016 regimen was used significantly more often among survivors than non-survivors (53.51% vs. 22.73%; P<0.008), but the difference was not significant between relapse-free and relapsed patients (51.85% vs. 35.71%; P = 0.128). High-Risk MRD was markedly more prevalent in non-survivors vs survivors (90.91% vs. 37.72%; P<0.001) and in relapsed versus relapse-free patients (82.14% vs. 37.04%; P<0.001) (Table 1).

Based on LASSO regression analysis, key clinical variables associated with OS and RFS in pediatric RUNX1::RUNX1T1-positive AML cases were identified. LASSO algorithm with an L1 penalty was employed for variable selection, effectively shrinking the coefficients of non-significant variables to zero, thus eliminating those with minimal contribution to survival prediction. For OS, BM blasts (FCM) %, High-Risk MRD (No/Yes), KRAS mutation (No/Yes), treatment regimen (AML-CAMS-2009 vs. AML-CAMS-2016), and WBC group (<20×10⁹/L vs. ≥20×10⁹/L) were selected at λmin (Figure 2A). Multivariate Cox analysis confirmed High-Risk MRD, treatment regimen, and WBC group as independent prognostic indicators for OS (Table 2), forming the basis of a 1-, 2-, and 3-year OS nomogram (Figure 2C). Similarly, the λ min for RFS via LASSO regression selected the following key variables: sex (male vs. female), BM blasts (FCM) %, EMI (No/Yes), High-Risk MRD (No/Yes), KRAS mutation (No/Yes), treatment regimen (AML-CAMS-2009 vs. AML-CAMS-2016), and WBC group (<20×10⁹/L vs. ≥20×10⁹/L) (Figure 2B). Multivariate Cox regression analysis indicated that BM blasts (FCM), EMI, High-Risk MRD, treatment regimen, and WBC group were independent prognostic factors for RFS in pediatric t (8,21) AML patients (Table 2). A nomogram model for 1-year, 2-year, and 3-year RFS was developed and visualized (Figure 2D).

In the established nomogram, each category of the prognostic variables was assigned a weighted score proportional to its relative contribution to survival outcomes. The total score, obtained by summing individual points, corresponded to the estimated 1-, 2-, or 3-year OS or RFS probability. A higher total score corresponds to an increased risk and poorer prognosis.

Nomogram performance was systematically evaluated for discrimination, calibration, and clinical applicability. ROC curves showed high area under curve (AUC) values for OS (0.878, 0.911, 0.864 at 1-, 2-, and 3-years) and RFS (0.823, 0.813, 0.793 at 1-, 2-, and 3-years), confirming strong discriminatory ability across different time points (Figures 3A–F). Calibration plots showed excellent consistency between predicted and observed outcomes, with calibration curves for 1-, 2-, and 3-year OS and RFS closely aligning with the reference line (Figures 3G–L). DCA revealed favorable net clinical benefit across a wide range of threshold probabilities (Supplementary Figure 1). Compared with single predictors (OS: High-Risk MRD, treatment regimen, WBC group; RFS: BM blasts (FCM), EMI, High-Risk MRD, treatment regimen, WBC group), the nomograms exhibited consistently higher AUCs, C-index values, and greater clinical net benefit (Figures 4A–G; Supplementary Figure 2). These findings highlighted the models’ strong predictive performance and practical value for individualized survival estimation in pediatric t (8,21) AML. Bootstrap validation (500 resamples) produced C-indices of 0.842 (OS) and 0.789 (RFS), supporting model robustness (Figure 4H).

Drawing on nomogram-based scoring results, the optimal cutoff values were determined using X-Tile software to establish a risk stratification system. All participants were stratified into low- and high-risk groups. For OS, scores of 0–160 and 161–194 represented low- and high-risk groups, respectively; for RFS, low-risk group corresponded to scores of 0–240 and high-risk group to scores of 241-350. The Kaplan-Meier survival curves demonstrated that this risk stratification system exhibited excellent discriminatory and stratification abilities (Figure 5).