Abstract
Background
Acute erythroid leukemia (AEL) is a rare and highly aggressive subtype of acute myeloid leukemia with a poor prognosis and no standardized treatment strategy. Its rarity and molecular complexity mean that conventional staging systems fail to predict its outcomes accurately. This study aimed to develop and validate nomogram models for predicting overall survival (OS) and cancer-specific survival (CSS) in patients with AEL.
Methods
We retrospectively analyzed data from the Surveillance, Epidemiology, and End Results (SEER) database (2000–2021) and identified 778 patients with AEL (ICD-O-3 code 9840/3). Patients were divided randomly into training (n = 544) and validation (n = 234) cohorts. Independent prognostic factors for OS and CSS were identified using Cox regression. Nomograms were developed and validated using concordance index (C-index), receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). Risk stratification was performed based on nomogram-derived scores.
Results
Multivariate Cox regression identified age, chemotherapy, marital status, and first primary tumor status as independent prognostic factors for both OS and CSS. The nomograms demonstrated good discrimination, with C-index values of 0.669 (OS) and 0.665 (CSS) in the training cohort, and 0.654 (OS) and 0.661 (CSS) in the validation cohort. ROC analysis confirmed good predictive accuracy, and calibration plots showed good agreement between predicted and observed survival. DCA confirmed the clinical utility of the nomograms. Risk stratification based on median nomogram scores effectively distinguished between high- and low-risk patients (P < 0.001).
Conclusions
We developed and validated novel SEER-based nomograms for predicting OS and CSS in patients with AEL, which demonstrated reliable performance in internal validation. However, the lack of molecular data (e.g., TP53 mutations) limits the biological interpretability of the models. External validation is required before clinical implementation.
Supplementary Information
The online version contains supplementary material available at 10.1007/s12672-026-04465-8.
Keywords: Acute erythroid leukemia, Prognosis, Nomogram, SEER
Introduction
Acute erythroid leukemia (AEL) is a rare and highly aggressive subtype of acute myeloid leukemia (AML), accounting for approximately 2%–5% of all cases of AML, and characterized by abnormal proliferation and differentiation arrest of erythroid precursors in the bone marrow [1, 2]. The French–American–British Cooperative Group first proposed a classification for AEL in 1976 [3], since when the diagnostic criteria have undergone multiple revisions. Early classifications divided AEL into acute erythroid/myeloid leukemia (M6a) and pure erythroid leukemia (M6b); however, the 2016 World Health Organization (WHO) classification system reclassified most previous cases of M6a as myelodysplastic syndromes or other AML subtypes, whereas M6b was retained as a subtype of AML not otherwise specified (AML-NOS) [4]. The latest International Consensus Classification further reclassified most AEL cases as AML with TP53 mutation, highlighting the central role of molecular pathological features in the classification of AEL [5]. Notably, AEL exhibits significant clinical and genetic heterogeneity and is frequently associated with complex karyotypes and biallelic TP53 mutations. These genetic abnormalities serve not only as key molecular markers of AEL, but also correlate directly with poor patient prognosis [6, 7].
Patients with AEL generally have a poor prognosis, with a median overall survival (OS) ranging from 3 to 14 months [2]. There is currently no standardized treatment strategy for AEL, and its management relies primarily on intensive chemotherapy or hypomethylating agents [7–9]; however, these approaches have limited efficacy, high relapse rates, and are frequently associated with severe treatment-related toxicities, such as cardiovascular complications, neurocognitive dysfunction, secondary malignancies, and endocrine disorders, further reducing the patient’s quality of life [10–13]. Despite progress in allogeneic hematopoietic stem cell transplantation and targeted therapies in other AML subtypes, their application in AEL remains challenging, and effective disease-specific therapies are currently lacking [9, 14]. In addition, its rarity and molecular complexity mean that conventional staging systems cannot accurately predict patient survival outcomes, highlighting the urgent need for the development of individualized prognostic assessment tools.
Nomogram models based on large-scale databases have recently been widely used for survival prediction in various cancers, by integrating multiple variables and providing intuitive, individualized risk assessments [15–17]; however, validated prognostic nomogram models specifically for AEL are currently lacking. This study therefore aimed to develop and validate nomogram models for predicting OS and cancer-specific survival (CSS) in patients with AEL using data from the Surveillance, Epidemiology, and End Results (SEER) database, with the goal of providing more accurate prognostic tools to guide individualized treatment decisions in clinical practice.
Materials and methods
Patient selection and data collection
This retrospective cohort study extracted data from the SEER database (2000–2021) using SEER*Stat (version 8.4.4, https://seer.cancer.gov/data-software/), identifying patients with AEL defined by ICD-O-3 code 9840/3. The collected variables included sex, age, race, marital status, year of diagnosis, chemotherapy, radiation, median household income, first primary malignancy indicator, OS, and CSS.
Age groups were determined using X-tile software (https://medicine.yale.edu/lab/rimm/research/software/) to identify the optimal cutoff values based on OS and CSS in the entire cohort [18]. The software tests all possible cut-off points and selects the value with the minimum p-value based on the log-rank test. Patients were categorized into three groups: 0–62 years, 63–71 years, and ≥ 72 years (Fig S1). Marital status was classified as married (including common-law marriage, divorced, and separated), unmarried (including never married, unmarried, or domestic partner), and widowed. Patients were stratified into two cohorts by year of diagnosis: 2000–2010 and 2011–2021. Race was categorized as White, Black, or other. Patients were divided into two groups according to median household income: <$75,000 and ≥$75,000. Chemotherapy and radiation were categorized as yes or no/unknown. The study selection criteria are summarized in Fig. 1. The inclusion criteria were: (1) a diagnosis of AEL with ICD-O-3 code 9840/3; (2) diagnosis between 2000 and 2021; (3) complete data for key clinical variables, including sex, age, race, year of diagnosis, treatment information, survival status, and household income; and (4) documented survival time and cause of death, with follow-up information. The exclusion criteria were: (1) unknown or < 1 month survival time; (2) diagnosis based solely on clinical findings or missing diagnostic information; and (3) missing data for any key variable. The final sample was split randomly into a training set and a validation set at a ratio of 7:3 for model development and evaluation. The SEER database is a publicly available, de-identified dataset. All personal identifiers have been removed, and this study therefore did not require institutional review board approval.
Fig. 1.
Study flow chart
Development, validation, and assessment of the nomogram
Univariate and multivariate Cox regression analyses were performed to identify independent prognostic factors for OS and CSS using the training cohort. Key variables for inclusion in the nomogram were selected using stepwise regression based on the Akaike Information Criterion (AIC). The nomogram was constructed using the rms package to visually represent the contribution of each variable to survival prediction. Model performance was evaluated in the validation cohort. Discrimination was assessed using the concordance index (C-index) and the area under the receiver operating characteristic (ROC) curve (AUC). Calibration was carried out using calibration curves with bootstrapping. Clinical utility was assessed using decision curve analysis (DCA) to determine net clinical benefit across different threshold probabilities. Individualized risk scores were calculated using the nomogram, and patients were then stratified into high- and low-risk groups based on the median score, and survival differences between the groups were analyzed.
Statistical analysis
Statistical analyses were performed using R software (version 4.4.1) packages, including survival, rms, ggplot2, survivalROC, ggDCA, and survminer. Categorical variables were presented as frequencies (%) and group comparisons were performed using the χ2 test. Survival analysis was performed using the Kaplan–Meier method, and differences between groups were assessed using the log-rank test. Cox proportional hazards regression models were used to calculate hazard ratios (HR) and 95% confidence intervals (CI). A two-tailed P-value < 0.05 was considered statistically significant.
Results
Patient characteristics
A total of 778 patients who met the inclusion criteria were enrolled in this study and were divided randomly into a training cohort (n = 544) and a validation cohort (n = 234) at a ratio of 7:3. The demographic and clinical characteristics of the patients are summarized in Table 1. In the overall cohort, 63.2% were male (n = 492) and 36.8% were female (n = 286). Patient age was distributed across the following groups: 0–62 years (40.1%), 63–71 years (24.8%), and ≥ 72 years (35.1%). Most patients were White (83.0%), 8.0% were Black, and 9.0% were other races. Patients diagnosed between 2000 and 2010 accounted for 55.5%, while those diagnosed between 2011 and 2021 accounted for 44.5%. Regarding treatment, 74.6% of patients received chemotherapy, whereas 25.4% did not receive chemotherapy or had missing information, and only 4.8% of patients received radiotherapy. An annual median household income ≥$75,000 was reported in 57.5% of patients. At diagnosis, 70.6% of patients were married, 18.8% were unmarried, and 10.7% were widowed. In addition, 74.7% of patients had primary malignant tumors, whereas 25.3% had non-primary tumors. There was no significant difference in the distribution of any of the variables between the training and validation cohorts (P > 0.05).
Table 1.
Baseline characteristics of patients diagnosed with AEL from the SEER database
| Characteristic | Overall (n = 778) |
Training (n = 544) |
Validation (n = 234) |
χ² | P value |
|---|---|---|---|---|---|
| Age, n(%) | |||||
| 0–62 | 312 (40.1) | 214 (39.3) | 98 (41.9) | 0.558 | 0.757 |
| 63–71 | 193 (24.8) | 135 (24.8) | 58 (24.8) | ||
| ≥ 72 | 273 (35.1) | 195 (35.8) | 78 (33.3) | ||
| Sex, n(%) | |||||
| Female | 286 (36.8) | 198 (36.4) | 88 (37.6) | 0.103 | 0.748 |
| Male | 492 (63.2) | 346 (63.6) | 146 (62.4) | ||
| Race, n(%) | |||||
| White | 646 (83.0) | 453 (83.3) | 193 (82.5) | 1.608 | 0.448 |
| Black | 62 (8.0) | 46 (8.5) | 16 (6.8) | ||
| Other | 70 (9.0) | 45 (8.3) | 25 (10.7) | ||
| Chemotherapy, n(%) | |||||
| No/Unknown | 198 (25.4) | 143 (26.3) | 55 (23.5) | 0.668 | 0.414 |
| Yes | 580 (74.6) | 401 (73.7) | 179 (76.5) | ||
| Radiation, n(%) | |||||
| No/Unknown | 741 (95.2) | 518 (95.2) | 223 (95.3) | 0.002 | 0.962 |
| Yes | 37 (4.8) | 26 (4.8) | 11 (4.7) | ||
| Marital status, n (%) | |||||
| Married | 549 (70.6) | 379 (69.7) | 170 (72.7) | 0.758 | 0.685 |
| Unmarried | 146 (18.8) | 106 (19.5) | 40 (17.1) | ||
| Widowed | 83 (10.7) | 59 (10.8) | 24 (10.3) | ||
| Median household income, n(%) | |||||
| <$75,000 | 331 (42.5) | 241 (44.3) | 90 (38.5) | 2.283 | 0.131 |
| ≥$75,000 | 447 (57.5) | 303 (55.7) | 144 (61.5) | ||
| First primary tumor, n(%) | |||||
| No | 197 (25.3) | 141 (25.9) | 56 (23.9) | 0.342 | 0.559 |
| Yes | 581 (74.7) | 403 (74.1) | 178 (76.1) | ||
| Year of diagnosis, n(%) | |||||
| 2000–2010 | 432 (55.5) | 303 (55.7) | 129 (55.1) | 0.022 | 0.883 |
| 2011–2021 | 346 (44.5) | 241 (44.3) | 105 (44.9) | ||
Survival analysis
Among the 778 patients with AEL included in this study, 675 (86.8%) died, of whom 589 (75.7%) died from AEL. OS and CSS stratified by key clinical characteristics were evaluated by Kaplan–Meier survival analysis and log-rank tests. The median OS of patients with AEL was 7 months, with a 3-year OS of 17.1%, and the median CSS was 8 months, with a 3-year CSS of 20.7%. In the analysis of treatment factors, patients receiving radiotherapy had a significantly longer median CSS of 23 months, compared with 8 months in those without radiotherapy (P = 0.00026). Similarly, patients receiving chemotherapy achieved a median CSS of 10 months, which was significantly longer than the 4 months in those without chemotherapy (P < 0.0001). Female (P = 0.025), unmarried status (P < 0.0001), age 0–62 years (P < 0.0001), and first primary tumor (P < 0.0001) were associated with significantly better survival. In contrast, median household income (P = 0.8), year of diagnosis (P = 0.47), and race (P = 0.17) showed no significant association with survival (Figs. 2 and 3; Table 2).
Fig. 2.
Kaplan–Meier analysis of OS in patients with AEL. Kaplan–Meier survival curves of OS for patients stratified by age (A), sex (B), race (C), chemotherapy (D), radiation (E), marital status (F), median household income (G), first primary tumor (H), and year of diagnosis (I)
Fig. 3.
Kaplan–Meier analysis of CSS in patients with AEL. Kaplan–Meier survival curves of CSS for patients stratified by age (A), sex (B), race (C), chemotherapy (D), radiation (E), marital status (F), median household income (G), first primary tumor (H), and year of diagnosis (I)
Table 2.
Median OS and CSS across different demographic groups in patients with AEL
| Groups | Median OS (in Months) | Median CSS (in Months) |
|---|---|---|
| Overall | ||
| All | 7 (7–8) | 8 (7–9) |
| Age | ||
| 0–62 | 15 (13–18) | 17 (14–20) |
| 63–71 | 6 (6–8) | 7 (6–9) |
| ≥ 72 | 4 (3–5) | 5 (4–6) |
| Sex | ||
| Male | 7 (6–8) | 8 (7–9) |
| Female | 8 (6–10) | 9 (7–11) |
| Race | ||
| Other | 8 (6–14) | 9 (7–17) |
| White | 7 (6–8) | 8 (7–10) |
| Black | 7 (6–8) | 7 (6–9) |
| Chemotherapy | ||
| No/Unknown | 3 (3–5) | 4 (3–5) |
| Yes | 9 (8–10) | 10 (9–12) |
| Radiation | ||
| No/Unknown | 7 (6–8) | 8 (7–9) |
| Yes | 15 (10–#) | 23 (11-#) |
| Marital status | ||
| Married | 7 (6–8) | 8 (7–9) |
| Unmarried | 17 (11–26) | 17 (13–36) |
| Widowed | 4 (3–6) | 5 (4–8) |
| Median household income | ||
| <$75,000 | 8 (6–9) | 8 (7–10) |
| ≥$75,000 | 7 (6–8) | 8 (7–10) |
| First primary tumor | ||
| No | 5 (3–6) | 6 (4–7) |
| Yes | 8 (8–10) | 10 (8–11) |
| Year of diagnosis | ||
| 2000–2010 | 8 (7–9) | 9 (8–10) |
| 2011–2021 | 7 (6–8) | 8 (7–10) |
OS, overall survival; CSS, cancer-specific survival; #survival probability did not reach the corresponding threshold
Prognostic factor analysis and development of nomograms
Prognostic factors in patients with AEL were evaluated systematically using univariate and multivariate Cox proportional hazards models. Univariate analysis identified age, sex, chemotherapy, radiotherapy, marital status, and first primary tumor status as significant factors influencing OS and CSS. Age, chemotherapy, marital status, and first primary tumor status remained independent prognostic factors for both OS and CSS in multivariate analysis (Tables 3 and 4).
Table 3.
Univariate and multivariate Cox regression analyses for AEL OS (training cohort)
| Characteristics | Univariate analysis | Multivariate analysis | ||
|---|---|---|---|---|
| HR (95% CI) | P value | HR (95% CI) | P value | |
| Age | ||||
| 0–62 | Reference | – | Reference | – |
| 63–71 | 2.09 (1.65, 2.65) | < 0.001 | 1.68 (1.30, 2.16) | < 0.001 |
| ≥ 72 | 2.77 (2.23, 3.45) | < 0.001 | 1.88 (1.46, 2.42) | < 0.001 |
| Sex | ||||
| Male | Reference | – | Reference | – |
| Female | 0.82 (0.68, 0.99) | 0.041 | 0.81 (0.66, 1.00) | 0.052 |
| Race | ||||
| Other | Reference | – | – | – |
| White | 1.17 (0.83, 1.65) | 0.369 | – | – |
| Black | 1.26 (0.80, 1.97) | 0.321 | – | – |
| Chemotherapy | ||||
| No/Unknown | Reference | – | Reference | – |
| Yes | 0.45 (0.37, 0.55) | < 0.001 | 0.55 (0.44, 0.68) | < 0.001 |
| Radiation | ||||
| No/Unknown | Reference | – | Reference | – |
| Yes | 0.49 (0.31, 0.77) | 0.002 | 0.77 (0.48, 1.23) | 0.269 |
| Marital status | ||||
| Married | Reference | – | Reference | – |
| Unmarried | 0.48 (0.37, 0.62) | < 0.001 | 0.67 (0.51, 0.88) | 0.005 |
| Widowed | 1.59 (1.20, 2.11) | 0.001 | 1.12 (0.81, 1.56) | 0.485 |
| Median household income | ||||
| <$75,000 | Reference | – | – | – |
| ≥$75,000 | 0.97 (0.81, 1.16) | 0.752 | – | – |
| First primary tumor | ||||
| No | Reference | – | Reference | – |
| Yes | 0.61 (0.50, 0.74) | < 0.001 | 0.78 (0.64, 0.97) | 0.022 |
| Year of diagnosis | ||||
| 2000–2010 | Reference | – | – | – |
| 2011–2021 | 0.99 (0.82, 1.19) | 0.889 | – | – |
OS, overall survival; HR, hazard ratio; CI, confidence interval
Table 4.
Univariate and multivariate Cox regression analyses for AEL CSS (training cohort)
| Characteristics | Univariate analysis | Multivariate analysis | ||
|---|---|---|---|---|
| HR (95% CI) | P value | HR (95% CI) | P value | |
| Age | ||||
| 0–62 | Reference | – | Reference | – |
| 63–71 | 2.02 (1.57, 2.60) | < 0.001 | 1.71 (1.31, 2.23) | < 0.001 |
| ≥ 72 | 2.79 (2.22, 3.51) | < 0.001 | 1.98 (1.52, 2.57) | < 0.001 |
| Sex | ||||
| Male | Reference | – | Reference | – |
| Female | 0.83 (0.68, 1.02) | 0.071 | – | – |
| Race | ||||
| Other | Reference | – | – | – |
| White | 1.15 (0.80, 1.66) | 0.442 | – | – |
| Black | 1.36 (0.85, 2.17) | 0.202 | – | – |
| Chemotherapy | ||||
| No/Unknown | Reference | – | Reference | – |
| Yes | 0.46 (0.37, 0.57) | < 0.001 | 0.57 (0.46, 0.72) | < 0.001 |
| Radiation | ||||
| No/Unknown | Reference | – | Reference | – |
| Yes | 0.46 (0.28, 0.76) | 0.003 | 0.73 (0.44, 1.22) | 0.233 |
| Marital status | ||||
| Married | Reference | – | Reference | – |
| Unmarried | 0.50 (0.38, 0.66) | < 0.001 | 0.69 (0.52, 0.93) | 0.013 |
| Widowed | 1.56 (1.16, 2.10) | 0.003 | 0.99 (0.71, 1.37) | 0.932 |
| Median household income | ||||
| <$75,000 | Reference | – | – | – |
| ≥$75,000 | 1.03 (0.85, 1.25) | 0.758 | – | – |
| First primary tumor | ||||
| No | Reference | – | Reference | – |
| Yes | 0.59 (0.48, 0.73) | < 0.001 | 0.76 (0.61, 0.95) | 0.016 |
| Year of diagnosis | ||||
| 2000–2010 | Reference | – | – | – |
| 2011–2021 | 0.97 (0.80, 1.18) | 0.787 | – | – |
CSS, cancer-specific survival; HR, hazard ratio; CI, confidence interval
Based on the Cox regression and stepwise regression results using the AIC, all independent prognostic factors associated with OS or CSS were subsequently used to develop the prognostic nomograms. The OS nomogram was constructed using five variables: age, sex, chemotherapy, marital status, and first primary tumor, and the CSS nomogram was developed using four variables: age, chemotherapy, marital status, and first primary tumor (Fig. 4).
Fig. 4.
Nomograms for predicting 1-, 3- and 5-year OS rates (A) and CSS rates (B) for patients with AEL
Nomogram validation and evaluation
The performances of the OS and CSS nomogram models were assessed by calculating the C-index and using ROC curves, calibration curves, and DCA. The C-index values in the training cohort were 0.669 (95% CI: 0.667–0.672) for the OS nomogram and 0.665 (95% CI: 0.662–0.667) for the CSS nomogram, and the equivalent values in the validation cohort were 0.654 (95% CI: 0.648–0.659) and 0.661 (95% CI: 0.655–0.667). For 1-, 3-, and 5-year predictions, the AUC values in the training cohort were 0.750 (95% CI: 0.706–0.794), 0.764 (95% CI: 0.711–0.816), and 0.794 (95% CI: 0.739–0.847), respectively, for OS, and 0.745 (95% CI: 0.699–0.791), 0.755 (95% CI: 0.698–0.806), and 0.767 (95% CI: 0.710–0.827), respectively, for CSS. The corresponding AUCs in the validation cohort were 0.694 (95% CI: 0.624–0.767), 0.761 (95% CI: 0.657–0.850), and 0.814 (95% CI: 0.719–0.893) for OS, and 0.692 (95% CI: 0.616–0.773), 0.723 (95% CI: 0.619–0.825), and 0.759 (95% CI: 0.646–0.868) for CSS, respectively. Calibration curves demonstrated good agreement between the actual and nomogram-predicted survival rates at 1-, 3-, and 5- years in both the training and validation cohorts (Fig. 5). DCA indicated that the OS and CSS nomograms provided significant net benefits for predicting 1-, 3-, and 5-year survival, with good predictive accuracy and clinical utility (Fig. 6). These results indicate that both nomogram models possessed good discriminative ability and stable predictive performance.
Fig. 5.
ROC curves and calibration plots of nomograms for predicting OS and CSS in patients with AEL. In the training cohort, the AUCs for 1-, 3-, and 5-year OS were 0.750, 0.764, and 0.794, respectively, (A), and for CSS were 0.745, 0.755, and 0.767, respectively (E). In the validation cohort, the AUCs for 1-, 3-, and 5-year OS were 0.694, 0.761, and 0.814, respectively, (C), and for CSS were 0.692, 0.723, and 0.759, respectively (G). Calibration plots for predicting 1-, 3-, and 5-year OS in the training cohort (B) and validation cohort (D), as well as for predicting 1-, 3-, and 5-year CSS in the training cohort (F) and validation cohort (H), showed good agreement between predicted and observed outcomes. The x-axis represents predicted survival, y-axis represents observed survival, 45° line indicates perfect calibration, and pink, dark green, and orange lines represent predictions for 1-, 3-, and 5-year survival, respectively. AUC, area under the ROC curve; ROC, receiver operating characteristic curve
Fig. 6.
DCA for OS and CSS in patients with AEL. A–C DCA curves for predicting 1-, 3-, and 5-year OS in the training and validation cohorts. D–F DCA curves for predicting 1-, 3-, and 5-year CSS in the training and validation cohorts. The x-axis represents threshold probability, y-axis represents net benefit, horizontal line along the x-axis indicates the assumption that no patients survive (“None”), and green line indicates the assumption that all patients survive (“All”)
Risk stratification based on nomogram points
To establish an effective prognostic stratification system, patients were divided into low- and high-risk groups according to the optimal cut-off values based on the median risk scores in the training cohort (OS: 0.086; CSS: 0.018). Kaplan–Meier survival analysis showed that this risk stratification method significantly distinguished survival differences between the risk groups. Patients in the low-risk group had significantly better OS (P < 0.001) and CSS (P < 0.001) than patients in the high-risk group (Fig. 7). In summary, the median risk score-based stratification model demonstrated good discriminative ability for predicting prognosis in patients with AEL, and could effectively distinguish between low- and high-risk populations.
Fig. 7.
Kaplan–Meier survival analysis of risk stratification in patients with AEL for OS and CSS. Kaplan–Meier curves for OS (A, B) and CSS (C, D) stratified by total risk scores in the training cohort (A, C) and validation cohort (B, D)
Subgroup analysis of diagnostic periods
To evaluate the influence of temporal changes in AEL diagnostic criteria on model performance, we stratified patients into two subgroups by diagnostic period: 2000–2010 (n = 432) and 2011–2021 (n = 346). The predictive accuracy of the OS and CSS nomograms was assessed in each subgroup using C-index and AUC values.
The results showed that the nomograms maintained consistent predictive performance across the two subgroups. For the OS nomogram, the C-index was 0.680 (95% CI: 0.621–0.739) in the 2000–2010 subgroup and 0.655 (95% CI: 0.576–0.733) in the 2011–2021 subgroup. For the CSS nomogram, the C-index was 0.674 (95% CI: 0.611–0.736) in the 2000–2010 subgroup and 0.662 (95% CI: 0.582–0.741) in the 2011–2021 subgroup. In addition, the ROC curves, calibration plots, and DCA demonstrated similarly favorable discriminative ability, calibration accuracy, and clinical utility in both diagnostic period subgroups (Table S1, Fig. S2–S4).
Discussion
In this study, we systematically analyzed the clinical characteristics and prognostic factors for 778 patients with AEL from the SEER database between 2000 and 2021. Univariate and multivariate Cox regression analyses identified age, chemotherapy, marital status, and primary tumor status as independent prognostic factors for OS and CSS. We then established novel OS and CSS predictive nomograms for AEL.
Previous studies reported a median survival time of 8–17 months for patients with AEL [19]. In contrast, the median OS in the present study was only 7 months, with a 3-year OS rate of 17.1%, while the median CSS was 8 months with a 3-year CSS rate of 20.7%, indicating a poor overall prognosis. Similarly, a Mayo Clinic analysis of 41 patients with pure AEL reported a median survival of only 1.8 months (range, 0.2–9.3 months) [20], further confirming the highly aggressive nature and poor prognosis of this disease.
Age has long been recognized as an important prognostic factor in AML, and several studies have demonstrated that OS decreases significantly with increasing age [21, 22]. In the current study, patients aged ≥ 72 years had a significantly increased risk of death (OS: HR = 1.88; CSS: HR = 1.98), and compared with patients aged ≤ 62 years, those aged 63–71 and ≥ 72 years exhibited progressively worse OS and CSS, with patients aged ≥ 72 years having the poorest survival. This finding is highly consistent with previous multicenter studies of AML, which showed that advanced age significantly reduced remission rates and survival [23, 24]. Reichard et al. and Reinig et al. reported similar trends, with almost no long-term survival in older patients [20, 25]. Notably, although our study confirmed that increasing age was associated with worse OS and CSS, the underlying mechanism remains uncertain. The poor prognosis in older AEL patients is likely multifactorial: on the one hand, age-related decline in physiological reserve, impaired bone marrow function, and a higher burden of comorbidities may limit tolerance to intensive treatment, resulting in reduced therapeutic efficacy; on the other hand, older patients may be more likely to harbor adverse biological features, such as complex karyotypes or molecular abnormalities (e.g., TP53 mutations), which are not recorded in the SEER database.
In terms of treatment, chemotherapy was associated with a significantly reduced risk of death, indicating that it remains a key approach to improving the prognosis in patients with AEL. Even in patients treated with chemotherapy, however, the median OS in this study was < 1 year, consistent with previous reports. For example, a multinational retrospective study by Almeida et al. reported a median OS of only 10.5–13.7 months in patients with AEL, with no significant survival differences between induction regimens, such as intensive chemotherapy or hypomethylating agents [26]. Similarly, a single-center study by Gabbard et al. found that patients with AEL, defined according to the WHO 2022 criteria, had a median OS of only 4.5 months, regardless of treatment strategy, highlighting the limitations of current therapies [27]. This poor prognosis may be closely related to the high heterogeneity and distinct molecular features of AEL. Fang et al. noted that pure erythroid leukemia is frequently characterized by biallelic inactivation of TP53 and complex karyotypes, contributing to a poor treatment response and short survival [28]. This highlights the need to improve existing chemotherapy regimens or explore novel therapeutic strategies. Additionally, although radiotherapy showed a trend towards improved survival benefit in univariate analysis, it did not demonstrate an independent protective effect in multivariate analysis, possibly due to the limited sample size and lack of detailed treatment data.
Marital status was also associated with prognosis in this study, with unmarried patients demonstrating better OS and CSS compared with married patients. This finding contrasts with the typical trend for several solid tumors, where married patients showed better survival [29–31]. This apparent discrepancy may be related to the age distribution in our cohort: the median age of unmarried patients was 40 years, compared with 66 years overall and 67 and 79 years for married and widowed patients, respectively. A recent study of 90 patients with AML with t(3;3) (q21;q26.2) or inv(3) (q21;q26.2) reported a similar finding, with unmarried patients showing better OS and CSS than married, divorced, or widowed patients [32].
In addition, patients with non-primary tumors had significantly worse prognoses than those with primary tumors (HR = 0.78, P = 0.022), consistent with the generally poorer outcomes observed in secondary or therapy-related AML subtypes [33, 34]. For example, Sedeta et al. analyzed 47,704 patients with AML from the SEER database between 2001 and 2018, and found that patients with secondary AML had an 8% higher risk of death than those with de novo AML (HR = 1.08, 95% CI: 1.05–1.11) [35]. In previous studies, secondary or therapy-related cases accounted for a substantial proportion of cases of AEL and were associated with significantly shorter median OS than primary cases. For example, an analysis of 41 AEL cases reported a median OS of only 2.3 months for therapy-related cases, 2.6 months for cases secondary to myelodysplastic syndrome, and 3.9 months for de novo AEL [28]. Another multicenter retrospective analysis similarly confirmed that secondary AEL generally had a worse prognosis than de novo AEL [25].
The model fit was assessed using the AIC, and the model with the lowest AIC was selected as the final prognostic model. By integrating these prognostic factors, we developed nomograms to predict 1-, 3-, and 5-year OS and CSS in patients with AEL. To the best of our knowledge, no previous studies have provided risk assessment models specific to AEL. The proposed nomograms demonstrated good discrimination and calibration in both the training and validation cohorts, and DCA confirmed their potential clinical utility. Moreover, the risk stratification system based on the nomogram effectively separated patients into low- and high-risk groups, with significant survival differences (P < 0.001), providing a preliminary tool for identifying high-risk patients and thus guiding individualized treatment strategies.
To evaluate the potential impact of evolving diagnostic criteria on model performance, we conducted subgroup analyses stratified by year of diagnosis (2000–2010 vs. 2011–2021). The predictive accuracy of the nomograms, as assessed by C-index and calibration, remained consistent across these periods, suggesting that temporal changes in classification practices had minimal influence on our results.
This study had several limitations. First, the C-index of the training and validation cohorts were relatively modest, possibly because of the limited sample size. Second, although age was identified as a major prognostic factor, the mechanism underlying the age-related survival disadvantage remains unclear. Third, the SEER database lacks key molecular data and measurable residual disease data, such as TP53 mutations, which limits our ability to explore survival differences and optimize prognostic assessments from a biological perspective. Fourth, detailed treatment information, including specific chemotherapy regimens, dosages, and stem cell transplantation status, is not recorded, limiting any in-depth analysis of the relationship between treatment patterns and clinical outcomes. Finally, this study was limited to internal validation and lacked external validation, and potential biases were inevitable due to the retrospective nature of the data. Future studies incorporating prospective, multicenter clinical data are needed to validate and refine the model.
Conclusions
This study identified age, chemotherapy, marital status, and first primary tumor as independent prognostic factors in patients with AEL. Based on these variables, we developed novel SEER-based nomograms to predict 1-, 3-, and 5-year OS and CSS. These nomograms may aid in risk stratification and individualized treatment planning. However, prospective validation in multicenter cohorts is required to confirm their robustness and clinical utility before widespread implementation.
Supplementary Information
Acknowledgements
The authors thank the staff of the National Cancer Institute (USA) for their contributions to the SEER Program.
Author contributions
Conceptualization, Xin Ma, Le Fu, Haiyan Gao; Data analysis, Xin Ma, Jiao Li; Methodology, Xin Ma, Le Fu, Cong Zhao, Jiao Li; Software, Xin Ma, Yuemei Feng; Validation, Cong Zhao, Xiaolan Yu; Visualization, Xin Ma, Ying Xi; Writing – original draft, Xin Ma, Le Fu; Writing – review & editing, Haiyan Gao; All authors have read and agreed to the published version of the manuscript.
Funding
This study was supported by Heilongjiang Provincial Health Commission scientific research project, China (No. YQJH2023051), and The project of “Basic Research Support Plan for Outstanding Young Teachers” in Heilongjiang Provincial universities, China (No. 20231111000188).
Data availability
The data used in this study are publicly available from the Surveillance, Epidemiology, and End Results (SEER) database (https://seer.cancer.gov/).
Declarations
Ethics approval and consent to participate
This study used the SEER public database, which contains de-identified, publicly available data. Therefore, ethics approval was not required. Not applicable.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Xin Ma and Le Fu contributed equally to this work.
References
- 1.Chen Q, Liu S, Wei X, Zhao P, Tian F, Yang K, et al. Dibromo-edaravone induces anti-erythroleukemia effects via the JAK2-STAT3 signaling pathway. Int J Mol Sci. 2025. 10.3390/ijms26094000. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Long W, Liu S, Li XX, Shen X, Zeng J, Luo JS, et al. Whole transcriptome sequencing and integrated network analysis elucidates the effects of 3,8-di-O-methylellagic acid 2-O-glucoside derived from Sanguisorba offcinalis L., a novel differentiation inducer on erythroleukemia cells. Pharmacol Res. 2021;166:105491. [DOI] [PubMed] [Google Scholar]
- 3.Bennett JM, Catovsky D, Daniel MT, Flandrin G, Galton DA, Gralnick HR, et al. Proposals for the classification of the acute leukaemias. French-American-British (FAB) co-operative group. Brit J Haematol. 1976;33(4):451–8. [DOI] [PubMed] [Google Scholar]
- 4.Arber DA, Orazi A, Hasserjian R, Thiele J, Borowitz MJ, Le Beau MM, et al. The 2016 revision to the World Health Organization classification of myeloid neoplasms and acute leukemia. Blood. 2016;127(20):2391–405. [DOI] [PubMed] [Google Scholar]
- 5.Arber DA, Orazi A, Hasserjian RP, Borowitz MJ, Calvo KR, Kvasnicka HM, et al. International Consensus Classification of Myeloid Neoplasms and Acute Leukemias: integrating morphologic, clinical, and genomic data. Blood. 2022;140(11):1200–28. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Zhou M, Yuan CM, He ZX, Long MT, He LT, Yang Y, et al. A natural flavagline derivative A2073 inhibits the proliferation of erythroleukemia cells by targeting the MAPK, PI3K, NF-kappaB, and cell cycle pathways. Bioorg Chem. 2025;163: 108612. [DOI] [PubMed] [Google Scholar]
- 7.Boddu P, Benton CB, Wang W, Borthakur G, Khoury JD, Pemmaraju N. Erythroleukemia-historical perspectives and recent advances in diagnosis and management. Blood Rev. 2018;32(2):96–105. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Zuo Z, Polski JM, Kasyan A, Medeiros LJ. Acute erythroid leukemia. Arch Pathol Lab Med. 2010;134(9):1261–70. [DOI] [PubMed] [Google Scholar]
- 9.Fernandes P, Waldron N, Chatzilygeroudi T, Naji NS, Karantanos T. Acute erythroid leukemia: from molecular biology to clinical outcomes. Int J Mol Sci. 2024. 10.3390/ijms25116256. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Thol F, Ganser A. Treatment of relapsed acute myeloid leukemia. Curr Treat Options Oncol. 2020;21(8):66. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Saultier P, Michel G. How i treat long-term survivors of childhood acute leukemia. Blood. 2024;143(18):1795–806. [DOI] [PubMed] [Google Scholar]
- 12.Wu SY, Short NJ, Nasr L, Dabaja BS, Fang PQ. Central nervous system prophylaxis and treatment in acute leukemias. Curr Treat Options Oncol. 2022;23(12):1829–44. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Granroth G, Khera N, Arana YC. Progress and challenges in survivorship after acute myeloid leukemia in adults. Curr Hematol Malig Rep. 2022;17(6):243–53. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Bhansali RS, Pratz KW, Lai C. Recent advances in targeted therapies in acute myeloid leukemia. J Hematol Oncol. 2023;16(1):29. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Balachandran VP, Gonen M, Smith JJ, DeMatteo RP. Nomograms in oncology: more than meets the eye. Lancet Oncol. 2015;16(4):e173–80. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Liu L, Xiao Y, Wei D, Wang Q, Zhang JK, Yuan L, et al. Development and validation of a nomogram for predicting suicide risk and prognostic factors in bladder cancer patients following diagnosis: a population-based retrospective study. J Affect Disord. 2024;347:124–33. [DOI] [PubMed] [Google Scholar]
- 17.Dong Z, Zhang Y, Geng H, Ni B, Xia X, Zhu C, et al. Development and validation of two nomograms for predicting overall survival and cancer-specific survival in gastric cancer patients with liver metastases: a retrospective cohort study from SEER database. Transl Oncol. 2022;24:101480. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Camp RL, Dolled-Filhart M, Rimm DL. X-tile: a new bio-informatics tool for biomarker assessment and outcome-based cut-point optimization. Clin Cancer Res. 2004;10(21):7252–9. [DOI] [PubMed] [Google Scholar]
- 19.Andrieu-Soler C, Soler E. Erythroid cell research: 3D chromatin, transcription factors and beyond. Int J Mol Sci. 2022. 10.3390/ijms23116149. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Reichard KK, Tefferi A, Abdelmagid M, Orazi A, Alexandres C, Haack J, et al. Pure (acute) erythroid leukemia: morphology, immunophenotype, cytogenetics, mutations, treatment details, and survival data among 41 Mayo Clinic cases. Blood Cancer J. 2022;12(11):147. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Gera K, Martir D, Xue W, Wingard JR. Survival after pure (acute) erythroid leukemia in the United States: a SEER-based study. Cancers. 2023. 10.3390/cancers15153941. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Hao Q, Foroutan F, Han MA, Devji T, Nampo FK, Mukherjee S, et al. Prognosis of older patients with newly diagnosed AML undergoing antileukemic therapy: A systematic review. PLoS One. 2022;17(12): e278578. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Thein MS, Ershler WB, Jemal A, Yates JW, Baer MR. Outcome of older patients with acute myeloid leukemia: an analysis of SEER data over 3 decades. Cancer. 2013;119(15):2720–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Oran B, Weisdorf DJ. Survival for older patients with acute myeloid leukemia: a population-based study. Haematologica. 2012;97(12):1916–24. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Reinig EF, Greipp PT, Chiu A, Howard MT, Reichard KK. De novo pure erythroid leukemia: refining the clinicopathologic and cytogenetic characteristics of a rare entity. Mod Pathol. 2018;31(5):705–17. [DOI] [PubMed] [Google Scholar]
- 26.Almeida AM, Prebet T, Itzykson R, Ramos F, Al-Ali H, Shammo J, et al. Clinical outcomes of 217 patients with acute erythroleukemia according to treatment type and line: a retrospective multinational study. Int J Mol Sci. 2017;18(4) [DOI] [PMC free article] [PubMed]
- 27.Gabbard T, Marini BL, Mirza KM, Benitez L, Fraga M, Pettit KM, et al. Therapeutic outcomes of patients with acute erythroid leukemia treated with hypomethylating agents. J Oncol Pharm Pract. 2025:1072610118 [DOI] [PubMed]
- 28.Fang H, Wang SA, Khoury JD, El HS, Kim DH, Tashakori M, et al. Pure erythroid leukemia is characterized by biallelic TP53 inactivation and abnormal p53 expression patterns in de novo and secondary cases. Haematologica. 2022;107(9):2232–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Zhu S, Lei C. Association between marital status and all-cause mortality of patients with metastatic breast cancer: a population-based study. Sci Rep. 2023;13(1):9067. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Gao C, Li H, Wu R, Chen W. Prognostic nomogram and treatment efficacy analysis for vaginal cancer: a SEER database and external validation study. Int J Gynaecol Obstet. 2023;162(2):541–51. [DOI] [PubMed] [Google Scholar]
- 31.Ding J, Wen Z. Survival improvement and prognosis for hepatocellular carcinoma: analysis of the SEER database. BMC Cancer. 2021;21(1):1157. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Ouyang X, Wu X, Hu S, Jiang W, Guo Y, Huang Y. The demographic characteristics and prognosis of acute myeloid leukemia with t(3;3)/inv(3) in the united states: a SEER-based study. Sci Rep. 2025;15(1):13477. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Dou X, Dan C, Zhang D, Zhou H, He R, Zhou G, et al. Genomic mutation patterns and prognostic value in de novo and secondary acute myeloid leukemia: a multicenter study from China. Int J Cancer. 2024;155(12):2253–64. [DOI] [PubMed] [Google Scholar]
- 34.Zeichner SB, Arellano ML. Secondary adult acute myeloid leukemia: a review of our evolving understanding of a complex disease process. Curr Treat Options Oncol. 2015;16(8):37. [DOI] [PubMed] [Google Scholar]
- 35.Sedeta E, Jemal A, Nisotel L, Sung H. Survival difference between secondary and de novo acute myeloid leukemia by age, antecedent cancer types, and chemotherapy receipt. Cancer. 2024;130(11):1952–63. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The data used in this study are publicly available from the Surveillance, Epidemiology, and End Results (SEER) database (https://seer.cancer.gov/).