Deciphering elevated risk of life-threatening hemorrhage and early death in acute myeloid leukemia with APL-like phenotype: independent role of endothelial injury and implications of EASIX-based risk stratification in a multicenter study

Abstract

Background

Early hemorrhagic mortality remains the major cause of elevated early death (ED) incidence in acute promyelocytic leukemia (APL)-like subgroup of acute myeloid leukemia (AML). The underlying mechanisms remain unknown and risk-adapted management strategies are not well established. We aimed to investigate the significance of endothelial injury in major bleeding event (MBE) and ED of APL-like AML (APLL) using endothelial activation and stress index (EASIX), and to develop an EASIX-based risk stratification with translational implications on APLL in venetoclax (VEN) era.

Methods

This multicenter study investigated newly diagnosed AML patients from three Chinese institutions between January 2017 and September 2025. Propensity score matching (PSM) and prospective thrombomodulin (TM) measurements were applied to validate EASIX utility in APLL. The APLL cohort was divided into development and validation cohorts based on diagnostic institutions to establish an MBE risk stratification using logistic regression.

Results

Of 344 APLL, 212 APL, and 1596 ‘Other AML’ in the analysis, log2-EASIX of APLL was comparable to APL but significantly higher than ‘Other AML’ (3.777 vs. 2.692, P < 0.001), which was sustained after PSM (4.022 vs. 2.904, P < 0.001). Significant correlation was further validated between log2-EASIX and TM (r = 0.80, P < 0.001). Within APLL, high log2-EASIX (OR: 5.611, 95% CI: 2.212–14.233, P < 0.001), overt DIC at diagnosis (OR: 2.929, 95% CI: 1.127–7.616, P = 0.027) and CD56 + blasts (OR: 3.697, 95% CI: 1.691–8.082, P = 0.001) emerged as independent MBE risk factors, setting foundation for a novel risk stratification. High-risk APLL exhibited significantly higher cumulative ED incidence in both development and validation cohorts, particularly when treated with standard VEN combining hypomethylating agent (HMA) regimen (P < 0.05 in both cohorts). However, modified VEN-HMA regimen with reduced VEN initiation dose specifically improved early outcomes of high-risk APLL (P = 0.036) in an independent, real-world implementation cohort.

Conclusions

APLL is characterized by significant endothelial damage, which is quantifiable by EASIX to independently contribute to the elevated MBE risk. The EASIX-based stratification can be utilized to rapidly identify high-risk APLL, who may benefit from an optimized VEN-HMA induction strategy to reduce ED risk.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12967-026-08439-w.

Introduction

Emerging studies have revealed a distinct acute promyelocytic leukemia (APL)-like subset among newly diagnosed acute myeloid leukemia (AML), defined as AML with immunophenotypic feature of CD34 and HLA-DR double negativity but lacked evidence of retinoic acid receptor rearrangement (RAR-r) [1–3]. One of the greatest concerns of APL-like AML (APLL) was highlighted by its disproportionately high early death (ED) rate, which was mainly associated with common coagulation dysfunction and, in particular, the aggressive clinical manifestations of major bleeding events (MBE) [3]. The unestablished understanding of bleeding diathesis results in the unmet clinical need among APLL, calling for more robust tools for guidance of early assessment and risk-adapted clinical strategies on these patients.

MBE is a life-threatening complication of AML, especially in APL, an acknowledged entity characterized by excessive bleeding [4, 5]. Despite complicated mechanisms of hemorrhage, growing evidence suggested that endothelial dysfunction contributed to risks of bleeding in APL [6, 7]. In addition, endothelial damage has been reported to be involved in organ failure in non-promyelocytic leukemia, including coagulation disorders [8]. However, its role in APLL remains poorly defined, indicating the importance of further exploration.

Despite the abundance of direct endothelial function markers [9], the widespread adoption of endothelial function assessment in routine clinical practice remains severely constrained, due to technical complexities and considerable costs. Nevertheless, Endothelial Activation and Stress Index (EASIX) is a marker of endothelial injury that has been increasingly applied in clinical practice [10]. The index, which combines lactate dehydrogenase (LDH), creatinine and platelet (PLT) count, has emerged as an easily accessible tool to predict acute complications and prognosis of patients undergoing allogeneic hematopoietic stem cell transplantation or chimeric antigen receptor T-cell therapy [11–13]. However, its utility value in de novo AML remains largely unexplored, enabling EASIX as a candidate of decision-making tool in APLL.

Anti-leukemic treatment may influence early complications of AML as well. In APL, the introduction of all-trans retinoic acid (ATRA) markedly reduced severe coagulopathy and ED risk by promoting differentiation of abnormal promyelocytes, demonstrating the translational value of pathophysiology-directed therapy on alleviation of early fatal complications [14]. Despite the resembling biological alterations between APLL and APL, ATRA could even lead to unsatisfactory outcomes in NPM1-mutated (NPM1c) AML [15], limiting its application in APLL. On the other hand, the traditional anthracycline-based “7 + 3” regimen, a cornerstone of induction therapy for newly diagnosed AML, can induce endothelial injury and exacerbate coagulation dysfunction, with studies highlighting MBE in intensive chemotherapy (IC) context [16–18]. In recent years, however, venetoclax (VEN) combining hypomethylating agent (HMA) has become a standard low-intensity induction regimen for elderly or unfit AML patients who are ineligible for IC [19]. This substantially enhances the therapeutic potential of VEN-HMA in APLL. Given the uncertain impact of VEN-HMA on APLL, the optimization of this epochal regimen in this unique entity awaits investigation.

Based on a multicenter cohort study, we investigated the association between endothelial injury and critical clinical challenges of APLL by introducing EASIX as a crucial biological marker. Additionally, we explored how EASIX could be utilized for risk stratification and therapeutic implications of APLL in VEN era.

Methods

Study population

In this study, newly diagnosed adult AML patients from 3 centers in China between January 2017 and September 2025 were screened. Patients who were exposed to previous anti-leukemic treatment or loss of follow-up within the first 60 days of diagnosis were excluded from the study. Variant APL, which referred to AML carrying RAR-r except PML::RARA, was also excluded from the study due to its controversial classification [20]. All the patients met the diagnostic criteria of the 5th edition of World Health Organization (WHO) classification [21]. This study was approved by the Ethics Committee of The First Affiliated Hospital of Soochow University and was conducted in accordance with the principles of the Declaration of Helsinki.

Definition

APLL patients were defined according to double negativity of both CD34 and HLA-DR, with no evidence of any RAR-r by polymerase chain reaction or RNA-sequencing examination. APL was diagnosed with the proof of PML::RARA. The rest of the AML patients were defined as ‘Other AML’.

EASIX score was calculated by the formula of [lactate dehydrogenase (LDH, U/L) x creatinine (mg/dL)/ PLT (10⁹/L)] based on the laboratory data on the day of initial diagnosis and was converted to a log2 form (log2-EASIX) for analysis.

For coagulation function evaluation, overt disseminated intravascular coagulation (DIC) was retrospectively diagnosed according to the updated scoring of 2025 International Society on Thrombosis and Haemostasis guideline [22]. MBE and major thrombotic events (MTE) were defined as > Grade 2 events of the WHO bleeding assessment scale [23] and Common Terminology Criteria for Adverse Events version 5.0, respectively.

For clinical outcomes, ED was defined as death from any cause within the first 60 days of initial diagnosis.

Treatment

Therapeutic regimens in this study were categorized as follows: IC regimen of standard “7 + 3” regimen (idarubicin/daunorubicin plus cytarabine) and low-intensity regimen of VEN-HMA (venetoclax plus azacytidine/decitabine). Standard dose of VEN was the ramp-up dose of 100 mg at Day1, 200 mg at Day2 and 400 mg/d from Day3. Reduced VEN intiation dose was 10 mg at Day1, followed by ramp-up to 400 mg daily [24]. PLT was transfused to maintain a level greater than 20 × 10⁹/L. Fresh-frozen plasma, cryoprecipitate, and fibrinogen were also transfused as supportive therapy of active bleeding events.

Model development and validation

An MBE risk stratification of APLL was developed and validated in development (DC) and validation cohort (VC), respectively (Details of study cohorts are provided in supplementary method).

In DC, multivariate logistic regression analysis was firstly performed to select variables for risk stratification. Their integer weights were derived from the coefficients of the independent predictors to constitute the score of stratification in APLL. Optimal cutoff was determined by receiver operating characteristic (ROC) analysis. Sensitivity, specificity, accuracy, positive likelihood ratio (PLR), negative likelihood ratio (NLR) and area under curve (AUC) were applied for evaluation of performance in DC and VC. Net reclassification improvement (NRI) was also applied when comparing the risk stratification with the previously published bleeding risk score in AML [25].

Statistical analysis

The Mann–Whitney U test was used to compare patient characteristics between different groups for continuous variables. Analysis of categorical data was performed using the Chi-square or Fisher’s exact. Correlation between log2-EASIX and thrombomodulin (TM) level was assessed using Spearman rank correlation analysis. In risk stratification performance comparison, McNemar test was applied in all the metrics except AUC, where Delong test was used for evaluation comparison.

Cumulative incidence of MBE was performed by the Kaplan-Meier method and the log-rank test. Cumulative incidence of ED was estimated with Fine and Gray method, with ED of non-MBE causes being analyzed as competing events. Statistical significance was set at P < 0.05 in this study. Analyses were performed with R 4.3.3 (R Foundation for Statistical Computing, Vienna, Austria) and MedCalc software (MedCalc Software Ltd., Ostend, Belgium).

Results

Patient characteristics

Firstly, we screened newly diagnosed AML patients between January 2017 and June 2024. 1940 non-APL AML and 212 APL patients were included for our analysis. Of 1940 non-APL AML patients, 344 were defined as APLL and the remaining 1596 were defined as ‘Other AML’. The patient characteristics of APLL, APL and ‘Other AML’ are summarized in Table 1 and Table S1. Comparing ‘Other AML’, APLL demonstrated hyperleukocytosis (46.01 vs. 20.62, P < 0.001) and significantly higher LDH burden (795.9 vs. 404.0, P < 0.001). Overt DIC and MBE occurred in 41.3% and 9.0% of APLL patients at diagnosis, significantly higher than that of ‘Other AML’ (P < 0.001, respectively). Additionally, APLL was presented with significantly higher bone marrow (BM) blast burden (73.0% vs. 59.0%, P < 0.001) and NPM1 mutation incidence (84.6% vs. 31.7%, P < 0.001).

Table 1.

Baseline characteristics of APLL and ‘Other AML’

Parameters	Before PSM (N = 1940)			After PSM (N = 618)
	APLL (N = 344)	Other AML (N = 1596)	P-value	APLL (N = 309)	Other AML (N = 309)	P-value
Age (years), median (range)	53 (18, 80)	44 (18, 83)	< 0.001	53 (18, 80)	51 (18, 83)	0.012
Sex (Male), n (%)	158 (45.9)	803 (50.3)	0.140	144 (46.6)	174 (56.3)	0.016
ECOG PS, n (%)			0.082			0.247
0–2	320 (93.0)	1521 (95.3)		286 (92.6)	293 (94.8)
3–4	24 (7.0)	75 (4.7)		23 (7.4)	16 (5.2)
Complete blood count, median (IQR)
WBC, x 109/L	46.01 (16.12, 106.23)	20.62 (4.82, 58.01)	< 0.001	45.53 (14.17, 102.08)	41.35 (12.37, 97.93)	0.495
Hemoglobin, g/L	82 (70, 97)	80 (65, 99)	0.690	81 (70, 97)	82 (65, 97)	0.729
Platelets, x 109/L	46 (27, 85)	47 (26, 89)	0.904	41 (20, 78)	50 (28, 93)	0.006
Biochemistry, median (IQR)
LDH, U/L	795.9 (485.9, 1085.6)	404.0 (252.7, 625.7)	< 0.001	825.5 (495.7, 1115.9)	487.5 (277.4, 832.9)	< 0.001
Albumin, g/L	36.1 (32.9, 39.6)	37.1 (33.9, 40.0)	0.001	36.1 (32.9, 39.5)	35.6 (32.5, 38.7)	0.434
Creatinine, mg/dL	0.73 (0.58, 0.89)	0.70 (0.57, 0.86)	0.162	0.76 (0.62, 0.96)	0.73 (0.59, 0.89)	0.053
Bilirubin, µmol/L	9.8 (7.4, 13.9)	10.3 (7.8, 14.9)	0.051	9.7 (7.4, 13.7)	10.1 (8.0, 14.4)	0.246
Coagulation evaluation, n (%)
Overt DIC at onset	142 (41.3)	140 (8.8)	< 0.001	107 (34.6)	99 (32.0)	0.495
MBE at onset	31 (9.0)	6 (0.4)	< 0.001	26 (8.4)	2 (0.6)	< 0.001
MTE at onset	12 (3.5)	4 (0.3)	< 0.001	10 (3.2)	4 (1.3)	0.105
BM blast (%), median (IQR)	73.0 (50.1, 86.0)	59.0 (40.0, 76.0)	< 0.001	70.0 (49.5, 84.3)	68.5 (48.0, 86.5)	0.312
Immunophenotype, n (%)
CD11b positive	63 (18.3)	407 (25.5)	0.005	58 (18.8)	119 (38.5)	< 0.001
CD7 positive	33 (9.6)	498 (31.2)	< 0.001	30 (9.7)	96 (31.1)	< 0.001
CD56 positive	116 (33.7)	269 (16.9)	< 0.001	104 (33.7)	48 (15.5)	< 0.001
CD4 positive	83 (24.1)	322 (20.2)	0.102	78 (25.2)	86 (27.8)	0.466
Molecular features, n (%)
KMT2A-r	10 (2.9)	87 (5.7)	0.050	10 (3.2)	12 (3.9)	0.664
CBF-AML	7 (2.0)	276 (17.3)	< 0.001	7 (2.3)	5 (1.6)	0.560
NUP98-r	6 (1.7)	18 (1.1)	0.348	6 (1.9)	3 (1.0)	0.505
Other fusion	6 (1.7)	36 (2.3)	0.554	6 (1.9)	8 (2.6)	0.589
NPM1	291 (84.6)	506 (31.7)	< 0.001	256 (82.8)	265 (85.8)	0.320
DNMT3A	88 (25.6)	363 (22.7)	0.259	79 (25.6)	178 (57.6)	< 0.001
FLT3-ITD	164 (47.7)	392 (24.6)	< 0.001	140 (45.3)	184 (59.5)	< 0.001
IDH1	54 (15.7)	136 (8.5)	< 0.001	48 (15.5)	52 16.8)	0.662
IDH2	85 (24.7)	155 (9.7)	< 0.001	68 (22.0)	68 (22.0)	1.000
TET2	94 (27.3)	152 (9.5)	< 0.001	83 (26.9)	70 (22.7)	0.226
NRAS	53 (15.4)	286 (17.9)	0.266	52 (16.8)	116 (37.5)	< 0.001
KRAS	19 (5.5)	175 (11.0)	0.002	19 (6.1)	66 (21.4)	< 0.001
ELN-risk 2022, n (%)			< 0.001			< 0.001
Favorable	159 (46.2)	615 (38.5)		148 (47.9)	89 (28.8)
Intermediate	165 (48.0)	659 (41.3)		141 (45.6)	201 (65.0)
Adverse	20 (5.8)	322 (20.2)		20 (6.5)	19 (6.1)
Log2-EASIX, median (IQR)	3.777 (1.893, 4.315)	2.692 (1.570, 3.860)	< 0.001	4.022 (2.183, 4.694)	2.904 (1.582, 4.250)	< 0.001

PSM, propensity score matching. ECOG, Eastern Cooperative Oncology Group. PS, performance status. IQR, interquartile range. WBC, white blood cell. LDH, lactate dehydrogenase. DIC, disseminated intravascular coagulation. MBE, major bleeding event. MTE, major thrombotic event. BM, bone marrow. CBF, core binding factor. ELN, European LeukemiaNet. EASIX, endothelial activation and stress index

EASIX indicates elevated endothelial damage of APLL

Next, EASIX distribution characteristics were investigated among our multicenter cohort. No significant difference of log2-EASIX was observed between APLL and APL (3.777 vs. 3.894, P = 0.215, Fig. 1A), while APLL exhibited elevated log2-EASIX level in comparison with ‘Other AML’ (3.777 vs. 2.692, P < 0.001, Fig. 1A; Table 1). Considering the possible confounding factors of endothelial injury in non-promyelocytic leukemia, we further performed propensity score matching (PSM) between APLL and ‘Other AML’ in a 1:1 ratio, selecting white blood cell (WBC) count, BM blast burden, overt DIC incidence and NPM1 mutation incidence as the PSM variables [8, 26]. 309 of 344 APLL and 309 of 1596 ‘Other AML’ were paired. After PSM, APLL retained significantly higher log2-EASIX level than ‘Other AML’ (4.022 vs. 2.944, P < 0.001, Fig. 1B; Table 1). Notably, significantly higher MBE incidence rate was observed in APLL (8.4% vs. 0.6%, P < 0.001, Table 1) after PSM. To further validate endothelial dysfunction of APLL, we prospectively measured TM level in a cohort containing 25 APLL and 25 ‘Other AML’. With well-matched baseline characteristics (Table S2), we found that APLL showed significantly higher soluble TM level (18.1 vs. 11.8, P = 0.001, Fig. 1C and Table S2) and log2-EASIX level (4.760 vs. 3.124, P = 0.006, Table S2) at diagnosis. Correlation analysis identified that log2-EASIX showed significant correlation with TM in the entire prospective cohort (r = 0.80, P < 0.001, Fig. 1D) and the subgroup of APLL (r = 0.83, P < 0.001, Figure S1A) and ‘Other AML’ (r = 0.77, P < 0.001, Figure S1B). These results collectively indicated that EASIX served as a valid marker of endothelial injury and APLL was presented with severe endothelial damage by significantly higher log2-EASIX level.

Fig. 1.

Endothelial function assessment of newly diagnosed APLL, APL and ‘Other AML’. (A) Box plot of log2-EASIX in newly diagnosed APLL, APL and ‘Other AML’. (B) Box plot of log2-EASIX in newly diagnosed APLL and ‘Other AML’ after propensity score matching. (C) Scatter plot of soluble thrombomodulin in newly diagnosed APLL and ‘Other AML’. (D) Correlation scatter plot of log2-EASIX and soluble thrombomodulin in newly diagnosed APLL and ‘Other AML’. APL, acute promyelocytic leukemia. AML, acute myeloid leukemia. APLL, APL-like AML. EASIX, endothelial activation and stress index. TM, thrombomodulin

Given the specific clinical significance of EASIX in APLL, we further explored the predictors of EASIX in this subset. Beyond WBC count, we found that mutations of NPM1 and IDH1 were associated with reduced log2-EASIX, while NRAS mutation was correlated with a high log2-EASIX level (Figure S2A). Due to the central role of NPM1 mutation in APLL, we also conducted investigations in NPM1c APLL (Figure S2B). One of the findings was that the type of NPM1 mutation was linked to the level of log2-EASIX. In detail, type A of NPM1 mutations was an independent factor in association with the decreased log2-EASIX level.

EASIX is associated with coagulation disorder of APLL

Given the common coagulation dysfunction of APLL, we investigated the association between EASIX and coagulation abnormalities in the development cohort (DC) and validation cohort (VC; Table S3). 75 of 106 in DC and 42 of 53 in VC who underwent IC regimen, along with 77 of 109 in DC and 40 of 59 in VC who received VEN-HMA regimen, were matched after PSM (Table S4). These patients, combining those who were dead before induction treatment (11 from DC, 6 from VC), comprised the cohorts for subsequent analysis.

It was revealed (Tables S5) that for APLL with overt DIC at diagnosis, their log2-EASIX level was significantly higher than those without overt DIC in both DC (3.904 vs. 2.601, P < 0.001) and VC (4.019 vs. 2.162, P < 0.001). As for MBE/MTE, APLL patients with MBE at diagnosis (5.592 vs. 3.068, P < 0.001 in DC; 5.746 vs. 2.721, P < 0.001 in VC) or MBE within 60 days of diagnosis (4.638 vs. 2.998, P < 0.001 in DC; 5.066 vs. 2.478, P = 0.001 in VC) showed significantly higher log2-EASIX level. By contrast, the log2-EASIX level demonstrated no association with the incidence of MTE within APLL. Furthermore, on the basis of the median of log2-EASIX, high log2-EASIX APLL showed significantly higher cumulative incidence of MBE in DC (25.6% vs. 7.4%, P = 0.002, Figure S3A) and VC (36.4% vs. 2.3%, P < 0.001, Figure S3B) within first 60 days of diagnosis.

Establishment of EASIX-based MBE risk stratification of APLL

Integrating comprehensive clinical variables, it was found in DC that high log2-EASIX level (OR: 5.611, 95%CI: 2.212–14.233, P < 0.001), overt DIC at diagnosis (OR: 2.929, 95%CI: 1.127–7.616, P = 0.027) and CD56 expression of leukemic blasts (OR: 3.697, 95%CI: 1.691–8.082, P = 0.001) served as independent risk factors of MBE (Table S6), with no evidence of significant multicollinearity found between EASIX and DIC (VIF = 1.130). Utilizing coefficents, a weighted score was assigned to each variable for establishment of MBE risk stratification: 2 points to high log2-EASIX, 1 point to overt DIC at diagnosis and CD56 expression of blasts, respectively (Fig. 2A). Based on the ROC curve analysis, a scoring of 3 was determined as the optimal cutoff to discriminate between low-risk and high-risk APLL of MBE. Under the stratification, 64 of 163 (39.3%) and 28 of 88 (31.8%) patients were categorized as high-risk subgroup in DC and VC, respectively.

Fig. 2.

MBE risk stratification of APLL. (A) The MBE risk stratification of APLL according to the initial diagnositic data. MBE, major bleeding event. EASIX, endothelial activation and stress index. DIC, disseminated intravascular coagulation. (B-C) Cumulative incidence of MBE by group of risk stratification in the (B) development cohort (DC) and (C) validation cohort (VC). (D-E) Pie charts illustrating the MBE sites of low- and high-risk APLL in the (D) DC and (E) VC. CNS, central nervous system

The sensitivity and specificity of the stratification were 0.963 (95%CI: 0.810–0.999) and 0.721 (95%CI: 0.637–0.794) in DC, respectively. The PLR was 3.446 (95%CI: 2.605–4.559) and NLR was 0.051 (95%CI: 0.007–0.353). In the VC, the model performed a sensitivity of 0.882 (95%CI: 0.636–0.985) and specificity of 0.817 (95%CI: 0.707–0.899) in VC. In addition, it was calculated that PLR was 4.819 (95%CI: 2.862–8.115) and NLR was 0.144 (95%CI: 0.039–0.532).

We subsequently compared our EASIX-based risk stratificaton with a previously published SiAML model for bleeding prediction in AML [25]. The comparison results are summarized in Table S7. Our EASIX-based risk stratification showed superior specificity (0.721 vs. 0.515, P < 0.001 in DC; 0.817 vs. 0.592, P < 0.001 in VC), accuracy (0.761 vs. 0.564, P < 0.001 in DC; 0.830 vs. 0.636, P < 0.001 in VC) and PLR (3.446 vs. 1.679, P = 0.013 in DC; 4.819 vs. 2.016, P = 0.031 in VC) to SiAML. Besides, ROC analysis of the two scoring systems revealed that the AUC of our stratification was significantly higher than that of SiAML model (0.842 vs. 0.665, P < 0.001 in DC; 0.850 vs. 0.708, P < 0.001 in VC). Moreover, the NRI was 35.4% (95%CI: 16.5%-54.3%) in DC, indicating that 5 of 27 (18.5%) MBE cases were reclassified as high-risk APLL and 34 of 136 (25.0%) non-MBE cases were corrected as low-risk APLL comparing SiAML, while there were merely 3.7% (1/27) of MBE cases and 4.4% (6/136) of non-MBE cases who were misclassified by EASIX-based stratification. Similarly, the NRI in VC was 28.4% (95%CI: 12.6%-44.2%), as 1 MBE case (5.9%) was correctly reclassified as high-risk subgroup and 18 non-MBE cases were reclassified as low-risk subgroup (25.4%), while only 2 non-MBE cases (2.8%) was incorrectly categorized as high-risk APLL. The NRI calculation showed significant improvement of our EASIX-based risk stratification in APLL (P < 0.001 in DC and VC, repectively). Together, these examination results indicated that our EASIX-based risk stratification demonstrated satisfactory performance that could be better utilized in MBE evaluation of APLL.

Characteristics of high-risk APLL

On the basis of our EASIX-based risk stratification, the characteristics of low- and high-risk APLL are displayed in Table 2. High-risk APLL was presented with elevated WBC count (94.56 vs. 42.15, P < 0.001 in DC; 94.51 vs. 34.43, P = 0.002 in VC) and higher BM blast (84.5% vs. 71.5%, P < 0.001 in DC; 81.3% vs. 69.8%, P = 0.014 in VC). At a median of 2 days in DC and 1 day in VC, the cumulative MBE incidence was significantly higher in high-risk subgroup (40.6% vs. 1.0% in DC, P < 0.001, Figs. 2B and 53.6% vs. 3.3% in VC, P < 0.001, Fig. 2C). The sites of MBE in low- and high-risk APLL also showed distinct patterns (Fig. 2D-E), as central nervous system ranked top among MBE sites of high-risk APLL (54.5% in DC; 60.0% in VC), while MBE in low-risk APLL was rarely observed in gastrointestinal tract, urinary tract and lung.

Table 2.

Patient characteristics of low- and high-risk APLL in the development and validation cohort

Parameters	Development cohort (N = 163)			Validation cohort (N = 88)
	Low-risk group (N = 99)	High-risk group (N = 64)	P-value	Low-risk group (N = 60)	High-risk group (N = 28)	P-value
Age (years), median (range)	54 (22, 74)	54 (20, 80)	0.466	54 (18, 76)	50 (18, 69)	0.169
Sex (Male), n (%)	44 (44.4)	34 (53.1)	0.279	32 (53.3)	12 (42.9)	0.360
ECOG PS, n (%)			0.001			0.378
0–2	98 (99.0)	55 (85.9)		57 (95.0)	25 (89.3)
3–4	1 (1.0)	9 (14.1)		3 (5.0)	3 (10.7)
Complete blood count, median (IQR)
WBC, x 109/L	42.15 (11.45, 83.00)	94.56 (37.01,167.10)	< 0.001	34.43 (9.73, 86.69)	94.51 (43.83, 179.80)	0.002
Hemoglobin, g/L	82 (71, 100)	77 (67, 98)	0.301	81 (69, 97)	84 (63, 98)	0.932
Platelets, x 109/L	59 (40, 101)	25 (14, 42)	< 0.001	64 (48, 103)	22 (13, 36)	< 0.001
Biochemistry, median (IQR)
LDH, U/L	396.8 (261.0, 602.0)	787.0 (540.0, 1474.3)	< 0.001	432.3 (262.8, 614.2)	823.7 (478.7, 2001.5)	< 0.001
Albumin, g/L	36.5 (33.2, 40.0)	36.6 (33.2, 40.2)	0.817	35.3 (32.8, 39.0)	34.9 (28.8, 38.8)	0.415
Creatinine, mg/dL	0.69 (0.54, 0.84)	0.82 (0.65, 1.02)	< 0.001	0.77 (0.60, 0.89)	0.73 (0.53, 0.89)	0.641
Bilirubin, µmol/L	10.2 (7.9, 13.4)	11.2 (7.8, 15.5)	0.221	9.2 (7.1, 13.7)	11.7 (8.8, 17.4)	0.016
Coagulation evaluation, n (%)
Overt DIC	23 (23.2)	51 (79.7)	< 0.001	7 (11.7)	26 (92.3)	< 0.001
MBE at onset	0 (0)	9 (14.1)	< 0.001	0 (0)	10 (35.7)	< 0.001
MTE at onset	1 (1.0)	5 (7.8)	0.035	3 (5.0)	2 (7.1)	0.651
MBE within 60d	1 (1.0)	26 (40.6)	< 0.001	2 (3.3)	15 (53.6)	< 0.001
MTE within 60d	1 (1.0)	6 (9.4)	0.015	4 (6.7)	3 (10.7)	0.675
BM blast (%), median (IQR)	71.5 (45.5, 85.0)	84.5 (63.6, 91.0)	< 0.001	69.8 (48.0, 82.1)	81.3 (63.3, 88.3)	0.014
Immunophenotype, n (%)
CD11b positive	17 (17.2)	18 (28.1)	0.096	5 (8.3)	7 (25.0)	0.047
CD7 positive	12 (12.1)	6 (9.4)	0.585	8 (13.3)	2 (7.1)	0.493
CD56 positive	24 (24.2)	31 (48.4)	0.001	13 (21.7)	17 (60.7)	< 0.001
CD4 positive	24 (24.2)	23 (35.9)	0.107	13 (21.7)	8 (28.6)	0.479
Molecular features, n (%)
KMT2A-r	2 (2.0)	3 (4.7)	0.382	3 (5.0)	0 (0)	0.548
CBF-AML	1 (1.0)	4 (6.3)	0.078	0 (0)	1 (3.6)	0.318
NUP98-r	2 (2.0)	2 (3.1)	0.646	0 (0)	1 (3.6)	0.318
Other fusion	1 (1.0)	1 (1.6)	1.000	0 (0)	3 (10.7)	0.030
NPM1	88 (88.9)	47 (73.4)	0.011	52 (86.7)	22 (78.6)	0.360
DNMT3A	32 (32.3)	12 (18.8)	0.057	20 (33.3)	2 (7.1)	0.008
FLT3-ITD	48 (48.5)	32 (50.0)	0.850	25 (41.7)	17 (60.7)	0.096
IDH1	20 (20.2)	6 (9.4)	0.065	7 (11.7)	5 (17.9)	0.509
IDH2	25 (25.3)	14 (21.9)	0.622	14 (23.3)	8 (28.6)	0.597
TET2	24 (24.2)	21 (32.8)	0.232	17 (28.3)	7 (25.0)	0.744
NRAS	18 (18.2)	10 (15.6)	0.673	9 (15.0)	4 (14.3)	1.000
KRAS	5 (5.1)	4 (6.3)	0.739	3 (5.0)	1 (3.6)	1.000
Induction Regimen			0.008			0.165
Death prior to induction therapy	2 (2.0)	9 (14.1)		2 (3.3)	4 (14.3)
IC	45 (45.5)	30 (46.9)		29 (48.3)	13 (46.4)
VEN-HMA	52 (52.5)	25 (39.1)		29 (48.3)	11 (39.3)

ECOG, Eastern Cooperative Oncology Group. PS, performance status. IQR, interquartile range. WBC, white blood cell. LDH, lactate dehydrogenase. DIC, disseminated intravascular coagulation. MBE, major bleeding event. MTE, major thrombotic event. BM, bone marrow. CBF, core binding factor. ELN, European LeukemiaNet. IC, intensive chemotherapy. VEN, venetoclax. HMA, hypomethylating agent

EASIX-based risk stratification guides ED prediction of APLL

The specific role of MBE in early outcomes of APLL prompted investigation of the association between our EASIX-based risk stratification and ED. In our DC and VC, the ED rate was 19.6% (32/163) and 15.9% (14/88), respectively (Table 3). Among the ED cases, high-risk APLL accounted for 68.8% (22/32) in DC and 85.7% (12/14) in VC. MBE was associated with 81.8% (18/22) ED of high-risk APLL in DC and 91.7% (11/12) in VC, which was significantly different from that in low-risk APLL [40.0% (4/10) in low-risk subgroup of DC, P = 0.037; 0 (0/2) in low-risk subgroup of VC, P = 0.033]. With the median occurrence of 12 days in high-risk APLL and 18.5 days in low-risk APLL, significant difference was identified in DC regarding the cumulative ED incidence by MBE (sHR: 9.97, 95%CI: 3.47–28.70, P < 0.001, Fig. 3A). Also, at the median of 8 days, high-risk APLL demonstrated significantly higher cumulative ED incidence by MBE in VC (sHR: 28.40, 95༅CI: 3.67–220.00, P < 0.001, Fig. 3B).

Table 3.

Details of APLL with early death in the development and validation cohort

Cohort	Age	Sex	ECOG	WBC	PLT	LDH	Cr	Log2- EASIX	Overt DIC	BM Blast	CD56	Risk of MBE	Induction treatment	ED from diagnosis (days)	ED cause
D	30	F	2	10.90	14	11297.8	0.70	9.152	Yes	84.0	No	High	VEN-HMA	14	MBE
D	51	F	2	180.45	19	6653.6	0.89	8.284	Yes	90.5	No	High	VEN-HMA	14	MBE
D	76	F	2	187.38	14	1415.0	2.21	7.804	Yes	92.0	Yes	High	DPI	3	MBE
D	50	F	3	38.84	23	3705.4	0.73	6.884	No	58.0	Yes	High	DPI	9	MBE
D	76	F	2	69.69	8	624.0	1.36	6.729	Yes	86.0	No	High	IC	28	MBE
D	54	M	2	156.55	29	3744.0	0.78	6.655	Yes	86.0	No	High	VEN-HMA	7	MBE
D	75	M	2	136.92	18	1894.0	0.94	6.626	Yes	97.0	Yes	High	VEN-HMA	16	MBE
D	34	F	3	48.93	42	4868.9	0.83	6.585	Yes	89.5	Yes	High	IC	56	MBE
D	42	F	2	54.01	36	5221.1	0.51	6.206	Yes	60.0	Yes	High	DPI	3	MBE
D	80	M	3	185.53	28	1493.8	1.14	5.921	Yes	98.0	Yes	High	DPI	1	MBE
D	75	M	2	15.31	24	1177.5	1.13	5.787	Yes	70.0	Yes	High	DPI	6	MBE
D	49	F	2	97.75	28	2291.3	0.64	5.722	Yes	90.5	No	High	IC	40	Infection
D	44	M	2	35.26	37	1076.2	1.64	5.577	Yes	86.5	Yes	High	VEN-HMA	11	MBE
D	55	M	3	91.37	13	540.0	0.85	5.139	Yes	80.5	No	High	DPI	4	MBE
D	55	F	2	97.99	8	446.0	0.63	5.127	Yes	87.5	No	High	IC	22	MBE
D	68	M	2	192.84	22	636.0	1.02	4.882	Yes	85.5	Yes	High	VEN-HMA	12	Myocardial Infarction
D	62	M	3	220.96	46	1205.0	0.95	4.638	Yes	20.5	Yes	High	DPI	4	MBE
D	77	M	3	83.54	112	1325.3	1.77	4.386	No	90.0	Yes	High	DPI	6	Myocardial Infarction
D	46	F	2	61.07	31	678.2	0.77	4.077	Yes	40.0	No	High	IC	17	MBE
D	60	M	2	242.82	45	876.0	0.86	4.065	Yes	93.0	No	High	VEN-HMA	32	Infection
D	73	M	2	74.93	78	1177.2	1.04	3.879	No	65.5	Yes	High	IC	15	MBE
D	20	M	2	28.96	8	366.0	0.30	3.772	Yes	62.0	No	High	VEN-HMA	12	MBE
D	65	F	1	230.75	55	996.5	0.78	3.826	No	86.0	No	Low	IC	21	MBE
D	70	M	2	42.15	87	1060.1	1.07	3.703	No	92.5	No	Low	VEN-HMA	18	Infection
D	52	F	2	137.50	34	396.8	0.66	2.939	Yes	87.0	Yes	Low	IC	15	MBE
D	57	F	3	20.14	20	364.4	0.41	2.901	Yes	88.5	Yes	Low	DPI	4	MBE
D	72	M	2	25.39	36	628.7	0.43	2.893	Yes	86.5	Yes	Low	DPI	2	Myocardial Infarction
D	66	F	1	30.02	42	453.5	0.53	2.515	Yes	53.0	Yes	Low	VEN-HMA	23	MBE
D	56	M	1	114.88	105	758.0	0.74	2.417	No	90.5	No	Low	IC	51	Infection
D	66	F	1	1.73	46	235.3	0.60	1.609	Yes	86.0	Yes	Low	IC	8	Cerebral Infarction
D	52	F	1	83.00	111	643.0	0.46	1.426	No	42.0	No	Low	IC	24	Infection
D	65	M	1	11.45	122	326.9	0.92	1.302	No	49.0	No	Low	IC	19	Infection
V	38	F	3	183.76	18	6538.8	0.66	7.858	Yes	36.5	No	High	VEN-HMA	8	MBE
V	61	F	1	213.56	31	2581.8	1.63	7.083	Yes	82.5	No	High	DPI	1	MBE
V	20	F	2	104.14	28	2187.6	1.28	6.642	Yes	63.0	Yes	High	IC	16	MBE
V	30	F	3	41.35	17	1451.4	0.89	6.250	Yes	22.0	No	High	VEN-HMA	46	MBE
V	71	F	2	72.72	38	2866.4	0.87	6.044	Yes	53.0	No	High	DPI	2	MBE
V	50	F	2	84.88	13	853.0	0.69	5.501	Yes	82.0	Yes	High	DPI	2	MBE
V	69	F	2	34.28	40	2521.0	0.53	5.066	Yes	81.5	Yes	High	VEN-HMA	8	MBE
V	61	M	2	151.84	85	2184.9	1.05	4.751	Yes	55.5	Yes	High	DPI	1	MBE
V	18	F	2	45.70	6	233.1	0.42	4.019	Yes	79.0	Yes	High	IC	8	MBE
V	50	M	2	145.37	38	989.0	0.59	3.936	Yes	83.0	Yes	High	VEN-HMA	49	MBE
V	57	F	2	142.66	40	1120.3	0.53	3.893	Yes	93.0	No	High	IC	51	Infection
V	51	F	2	252.53	54	867.0	0.53	3.094	Yes	90.5	No	High	VEN-HMA	14	MBE
V	59	F	3	248.97	66	2423.4	0.74	4.755	No	57.0	No	Low	VEN-HMA	7	TLS- MODS
V	67	F	2	53.26	68	867.9	1.12	3.831	No	74.0	No	Low	IC	8	MBE
Summary
Cohort	Median	F (%)	> 2 (%)	Median	Median	Median	Median	Median	Overt DIC (%)	Median	Yes (%)	High (%)	DPI (%)	Median	MBE (%)
D	59	50.0	18.8	78.97	33	1028.3	0.81	4.760	75.0	86.0	53.1	68.8	31.3	14	68.8
V	54	85.7	21.4	123.40	38	1818.2	0.72	4.911	85.7	76.5	42.9	85.7	28.6	8	92.9

ECOG, Eastern Cooperative Oncology Group. WBC, white blood cell. PLT, platelet. LDH, lactate dehydrogenase. Cr, creatinine. EASIX, endothelial activation and stress index. DIC, disseminated intravascular coagulation. BM, bone marrow. MBE, major bleeding event. ED, early death. D, development. V, validation. DPI, death prior to induction. IC, intensive chemotherapy. VEN, venetoclax. HMA, hypomethylating agent. TLS, tumor lysis syndrome. MODS, multiple organ dysfunction syndrome

Fig. 3.

Association between EASIX-based risk stratification and 60-day outcomes of APLL. (A-B) Cumulative incidence of MBE-associated early death (ED) by group of risk stratification in the (A) development cohort (DC) and (B) validation cohort (VC), with ED of non-MBE causes considered as a competing event. (C-D) Cumulative incidence of MBE-associated ED in VEN-HMA context by group of risk stratification in the (C) DC and (D) VC, with ED of non-MBE causes considered as a competing event. VEN, venetoclax. HMA, hypomethylating agent

For high-risk APLL, no significance of 60-day outcomes was found between IC and VEN-HMA subgroup in DC (sHR: 1.38, 95%CI: 0.41–4.63, P = 0.598, Figure S4A) and VC (sHR: 2.58, 95༅CI: 0.51–13.01, P = 0.252, Figure S4B). However, it was surprisingly identified that for those undergoing VEN-HMA regimen, high-risk APLL exhibited increased cumulative ED incidence by MBE (sHR: 14.20, 95༅CI: 1.81-110.86, P = 0.011 in DC, Fig. 3C; sHR: 11.60, 95༅CI: 1.30-103.41, P = 0.028 in VC, Fig. 3D), while no such significant difference was observed between low- and high-risk APLL who underwent IC regimen (sHR: 3.88, 95༅CI: 0.76–19.92, P = 0.103 in DC, Figure S4C; sHR: 4.53, 95༅CI: 0.43–48.08, P = 0.211 in VC, Figure S4D).

Optimization of VEN-HMA strategies on high-risk APLL

The specific unsatisfactory outcomes of high-risk APLL in VEN-HMA context suggested the importance of optimized induction strategies of VEN-HMA in APLL, and we explored the effect of modified VEN doses on APLL in an independent, real-world implementation cohort (Table S8) since 2024. 24 of 64 (37.5%) patients were high-risk APLL, among whom 12 (50.0%) were initiated with reduced VEN dose, while 12 (50.0%) underwent standard ramp-up VEN-HMA regimen. 10 of 12 (83.3%) who received reduced VEN dose were alive at + 60 days of diagnosis, significantly alleviating 60-day survival rate compared with standard ramp-up dose (4/12, 33.3%, P = 0.036, Fig. 4A). Nevertheless, no significant benefit was obtained in low-risk subgroup between reduced VEN initiatidose and standard ramp-up dose [90.0% (9/10) vs. 96.7% (29/30), P = 0.442, Fig. 4B]. Detailed analysis of high-risk APLL revealed that of 12 patients who underwent reduced VEN initiation doses, 5 patients experienced MBE before initiation of induction treatment and 4 out of 5 (80.0%) obtained 60-day survival (Fig. 4C). However, within 5 of 12 high-risk APLL who received standard ramp-up VEN-HMA regimen with early MBE occurrence, 4 (80.0%) advanced to ED and MBE contributed to mortality of all these cases. On the other hand, for the rest of 7 patients undergoing reduced VEN initiation doses, 6 (85.4%) managed to encounter no MBE occurrence during induction treatment and only 1 (14.6%) of these patients ended in ED due to MBE. However, in standard ramp-up VEM-HMA context, 4 of 7 (57.1%) were observed as ED cases, among whom MBE occurred in 3 cases (75.0%) on the first day of induction treatment and rapidly progressed to ED. These results indicated that our modified VEN-HMA regimen highly mitigated early outcomes of high-risk APLL.

Fig. 4.

Association between VEN initiation doses and 60-day outcomes of APLL. (A-B) Bar plot indicating 60-day outcomes of APLL from implementation cohort by different VEN initiation doses within (A) high-risk APLL and (B) low-risk APLL. ED, early death. MBE, major bleeding event. (C) Swimmer plot of high-risk APLL from implementation cohort within the first 60 days of diagnosis. ED, early death. MBE, major bleeding event. VEN, venetoclax

Discussion

APLL is a distinct subset of AML that has attractted increasing attention recently. While previous studies focused on elucidation of its unique biological profiles, it has been revealed that APLL is also highlighted with prominent coagulation dysfunction, which poses great threat to its clinical outcomes [3]. Here, we attempted to explore the novel perspective of MBE in APLL and how it influenced the understanding and management of APLL. We firstly revealed that EASIX, a promising marker of the endothelial function evaluation, indicated elevated endothelial injury of APLL and was crucial to translational implications of APLL clinical strategies in the VEN era.

By investigating our muleticenter cohort of newly diagnosed AML, we identified that APLL exhibited no significant difference of log2-EASIX level with classic APL, and was statistically higher than that of ‘Other AML’. However, it is noteworthy that some baseline characteristics of APLL, including hyperleukocytosis, heavy leukemic burden and abnormal coagulation function, could contribute to the endothelial injury in non-M3 AML [27, 28]. More importantly, the endothelial dysfunction could massively amplify the procogulant drive from the leukemic cells, accelerating thrombin generation, microvascular thrombosis, and consumption of platelets and coagulation factors. This could potentiate and sustain the DIC process, indicating a “double-hit” model between DIC and endothelial damage [8, 29, 30]. The shared role of PLT count in the definition of both indexes also gave a hint on their mutual impact. Therefore, given the distinct clinical characterization of APLL, we included variables of WBC count, BM blast burden, DIC and NPM1 mutation incidence for PSM analysis, aiming to minimize their possible confounding effect and to investigate whether endothelial dysfunction played a specific role in pathogenesis of APLL. After PSM, we still observed significantly higher log2-EASIX level of APLL than ‘Other AML’. Moreover, we also measured the level of soluble TM, a solid biomarker of endothelial damage [31], in a well-matched cohort of newly diagnosed APLL and ‘Other AML’. We revealed that APLL showed increased TM level than ‘Other AML’ and log2-EASIX demonstrated significant correlation with TM, strengthening the proof of severe endothelial injury as an inherent biological alteration in de novo APLL patients. Together, these results suggested that APLL was a unique subgroup of newly diagnosed AML characterized by specifically elevated endothelial dysfunction, which could be quantifiable by EASIX, a biomarker of endothelial function evaluation that was easily accessible in clinical practice and had not been previously reported in de novo AML.

The specific role of EASIX in APLL inspired us to investigate the possible influencing factors of the biomarker in this subset. Apart from the well-known hyperleukocytosis [26], there were some unexpected findings. NPM1 mutation, the core biological alteration of APLL, was associated with the EASIX level in APLL. There was no evidence of blast burden or proliferative mutation load difference between NPM1c and NPM1 wild-type (NPM1wt) APLL except for a different LDH level in two subgroups [3], suggesting an alternative explanation to the result. One of the prominent characteristics of NPM1c APLL was its distinct genomic profile of NPM1 mutation and common co-mutations of IDH1/2 and TET2. These mutations collectively defined a stable, differentiation-blocked cellular state through epigenetic reprogramming and transcriptional dysregulation, which could carry a corresponding adaptive metabolic program of restrained glycolytic flux, resulting in a lower LDH level in NPM1c APLL [32]. A supplementary evidence of the explanation was that IDH1 mutation was also identified as an independent protective factor of endothelial damage in APLL. In contrast, the elevated LDH activity in NPM1wt APLL might activate and damage the vascular endothelium as well, ultimately leading to a progressive deterioration of endothelial function. These findings suggested the possible impact of APLL biological alterations on the endothelial injury and careful management was also highlighted to NPM1wt APLL for EASIX evaluation in clinical practice.

In this study, we managed to develop an MBE predictive scoring system of APLL, aiming to pave a way for risk stratification of the entity that was characterized by increased risk of severe coagulopathy. Under our EASIX-based stratification, 39.3% (64/163) from DC and 31.8% (28/88) from VC were defined as high-risk APLL of MBE. Moreover, the comparison with the previously published SiAML scoring [25] showed a better efficacy. The SiAML model, containing fibrinogen and D-dimer as the key components of bleeding prediction, reflects the primary role of DIC in the occurrence of bleeding events in AML patients. However, independent predictors of MBE in APLL consisted of high EASIX level, overt DIC at diagnosis and CD56 expression of leukemic cells. No collinearity was observed between EASIX and DIC in our analysis. Furthermore, the scoring indicated that the effect size of EASIX on MBE was as approximately 2 times as that of overt DIC. These findings suggests that while DIC remains a critical and well-established risk factor of MBE in AML, diffuse endothelial damage may exert an even stronger influence on hemostatic failure, underscoring the endothelial injury as a predominant and distinct driver of hemorrhage in APLL. This challenges the traditional paradigm where DIC is considered as the primary hematological culprit for bleeding in AML [30], indicating the possible reasons for ineffectiveness of relying solely on blood transfusion to reduce ED risk of APLL. It further implies that the bleeding risk conferred by endothelial injury may not merely be a surrogate for concurrent DIC but represent a unique pathophysiological entity in APLL. In the studies of APL, novel mechanisms of endothelial damage was proposed [6, 7]. Despite the importance of further research, our findings still refine the understanding of hemorrhagic pathophysiology in APLL and highlight the importance of evaluation and management of endothelial dysfunction in APLL beyond the DIC disturbance.

Apart from high EASIX level, another finding was that CD56 expression was identified as an independent risk factor of MBE in APLL. In previous studies, CD56 was also confirmed as a unique immunophenotypic marker that showed a prevalance in APLL [1, 3]. However, no evidence has been proposed to date with regard to its impact on hemorrhage of AML. The adhesive properties of the CD56 molecule might contribute to the leukemic cells for migration through the blood endothelial barrier [33, 34]. Nonetheless, according to the analysis results, CD56 expression did not serve as an independent predictor of the EASIX level in APLL. This might indicate other biological alterations of leukemic cells by CD56 expression. Therefore, in addition to supportive therapy such as blood transfusions to remedy the coagulation function, appropriate induction strategies to supress leukemic cells may be crucial to these patients as well.

Taking the novel components of the risk stratification into account, our EASIX-based system not only demonstrated the crucial role of endothelial enjury in MBE incidence of APLL, but also represented a more robust tool to reveal the prominent MBE risk of this unique entity. This would strengthen the utility of our EASIX-based scoring, as all the indicators saw extensive application in routine clinical setting.

It was identified that the EASIX-based MBE risk stratification also well-guided ED prediction of APLL. To the best of our knowledge, this was the first time to reveal that EASIX guided early outcomes prediction of newly diagnosed AML. Importantly, we here noticed that standard ramp-up of VEN-HMA could be specifically detrimental to induction of high-risk APLL, even though we have performed PSM to ensure a balanced baseline background between our VEN-HMA and IC subgroup. A possible explanation could be that VEN, as a BCL-2 inhibitor, induced rapid and synchronous apoptosis, resulting in an abrupt, avalanche-like release of damage-associated molecular patterns and cytokines [35]. On the contrary, as a cell cycle-specific agent, the cell death triggered by conventional chemotherapy was more asynchronous and delayed, allowing for a relatively graduated systemic response. In high-risk APLL characterized by pre-existing endothelial injury and inherent poor coagulation function, the VEN-driven tumor lysis syndrome (TLS) might overwhelm endogenous buffering systems, resulting in a vicious cycle of exacerbated endothelial disruption and consumptive coagulopathy [36]. On the other hand, given the established role of BCL-2 family in maintaining endothelial cell survival and homeostasis [37, 38], its pharmacological inhibition by VEN at standard ramp-up doses might inadvertently compromise the integrity and function of already dysfunctional endothelial cells. This rendered the vasculature more susceptible to the cytotoxic and inflammatory milieu of concomitant rapid TLS. Therefore, the heightened early hemorrhage-related ED risk with VEN-HMA in high-risk APLL might not only be a result of TLS severity but also a consequence of TLS velocity coupled with a potential direct vascular fuction interference, which was superimposed on a profoundly vulnerable homeostatic function.

It has long been emphasized regarding the importance of optimizing VEN starting dose in clinical setting [39], and emerging evidence has been proposed of benefit obtained from VEN modified doses, particularly in those ineligible for IC [24, 40]. Thus, the EASIX-based MBE stratification may prompt modified VEN-HMA induction strategies on high-risk APLL. It was identified that redeuced VEN initiation dose improved early outcomes in high-risk APLL. This constructed a biomarker-guided, translational strategy in clinical decision-making of APLL. The underlying interpretation of alleviation from our modified VEN-HMA regimen might be that the optimized VEN initiation dose could moderate the TLS effect, which decreased the intensity of inflammatory storm, providing a critical time window for the clearance of apoptotic debris and endothelial repair. Also, the lower dose of VEN might mitigate the “off-target” inhibition of BCL-2 in vascular endothelial cells, preserving their homeostatic function. This modification controlled the disease while maintaining the fragile balance of the coagulation-endothelial barrier, allowing supportive care to intervene effectively and thus reducing early hemorrhage-related mortality incidence.

While our study yields some crucial findings regarding the translational value of EASIX-based stratification on a unique APLL subset, we also acknowledge the limitations of the study, including the general retrospective design of the research. In addition, the efficacy of our modified VEN-HMA regimen warrants further investigation in larger population of high-risk APLL. These limitations highlight the importance of future studies to extend our findings, particularly the exploration on high-risk APLL patients.

Conclusions

This multicenter study demonstrates the aggressive clinical manifestations of severe coagulopathy and high ED risk of APLL by introducing EASIX evaluation for the first time. EASIX is indicated as a robust quantifiable marker of specifically elevated endothelial dysfunction in APLL to contribute to MBE and ED risk of the distinct subgroup. The EASIX-based scoring not only presents a novel, potent risk stratification of APLL from the pathogenesis perspective, but also guides the risk-adapted therapeutic strategies of this unique entity, optimizing its clinical management in the era of VEN.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

AML

Acute myeloid leukemia

APL

Acute promyelocytic leukemia

APLL

Acute promyelocytic leukemia-like acute myeloid leukemia

AUC

Area under curve

BM

Bone marrow

CI

Confidence interval

DC

Development cohort

DIC

Disseminated intravascular coagulation

EASIX

Endothelial activation and stress index

ED

Early death

HMA

Hypomethylating agent

IC

Intensive chemotherapy

LDH

Lactate dehydrogenase

MBE

Major bleeding event

MTE

Major thrombotic event

NLR

Negative likelihood ratio

NPM1c

NPM1-mutated

NPM1wt

NPM1 wild-type

NRI

Net reclassification improvement

OR

Odds ratio

PLR

Positive likelihood ratio

PLT

Platelet

PSM

Propensity score matching

RAR-r

Retinoic acid receptor rearrangement

ROC

Receiver operating characteristic

sHR

Subdistribution hazard ratio

TLS

Tumor lysis syndrome

TM

Thrombomodulin

VC

Validation cohort

VEN

Venetoclax

WBC

White blood cell

WHO

World Health Organization

Author contributions

MS and SC contributed to the design of the study. MS, XL, YW and WD conducted data collection of the patients. MS performed statistical analysis and prepared the manuscript. SC and WD were responsible for the supervision of the study and helped revise the manuscript. DW also assisted the revision of the manuscript. All authors read and approved the final manuscript.

Funding

This study was supported by grant from the National Natural Science Foundation of China (82530011), the Natural Science Foundation of Jiangsu Province (BK20231195, BK20210087), the NHC Key Laboratory of Thrombosis and Hemostasis, The First Affiliated Hospital of Soochow University (KJS2420), Shandong Provincial Natural Science Foundation (ZR2025QC2033Z), Research and development fund project of Affiliated Hospital of Shandong Second Medical University (2025FYQ024).

Data availability

The dataset supporting the findings of the study is available from the corresponding authors on reasonable request.

Declarations

Ethics approval and consent to participate

The study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of The First Affiliated Hospital of Soochow University (ethic approval protocal number 2024-607). Informed consent was obtained from individual patient to participate in this study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.