Comparison of the prognostic predictive value of Molecular International Prognostic Scoring System and Revised International Prognostic Scoring System in patients undergoing allogeneic hematopoietic stem cell transplantation for myelodysplastic neoplasms

Myelodysplastic neoplasms (MDS) are heterogeneous clonal hematopoietic neoplasms with a high risk of evolution into acute myeloid leukemia.1 Given the highly variable clinical courses of MDS, the importance of risk stratification for distinguishing high-risk patients has been underscored.2 Currently, the most widely used survival prediction models are the International Prognostic Scoring System (IPSS), revised IPSS (IPSS-R), and World Health Organization Classification-based Prognostic Scoring System, which considers hematological parameters and cytogenetic abnormalities to identify patients at different risk stages.3 A major limitation of these scores is that they fail to consider the effect of the genetic mutation topography. Thus, in 2022, a new prognostication tool, the Molecular IPSS (IPSS-M), which employed hematologic data, cytogenetic abnormalities, and somatic mutations of 31 genes for survival estimation, has emerged as a powerful tool for therapeutic decision-making.4 Thus, the superior prognostic power of IPSS-M compared with IPSS-R has been demonstrated. However, whether IPSS-M is applicable for estimating the outcomes in the setting of allogeneic hematopoietic stem cell transplantation (allo-HSCT) has not yet been fully investigated. Here, we aimed to evaluate the predictive utility of IPSS-M in a large multicenter real-life cohort of Chinese MDS patients undergoing allo-HSCT. We recruited a total of 341 adult patients diagnosed with MDS who underwent allo-HSCT at 11 bone marrow centers across the Zhejiang Province between January 2017 and October 2022. Full information on genomic screening at diagnosis, HSCT procedures, and statistical analysis are provided in the Supporting Information (section Method). The study protocol was approved by the institutional review boards and written informed consent was obtained from each participant or their legal guardians in accordance with the Declaration of Helsinki. The follow-up period was updated in April 2023. Clinical features and transplant-specific characteristics of the enrolled 341 patients are summarized in Table S1. The median age of patients at transplantation was 48.6 years (range, 18.1–67.4 years), with 41.3% of them being female. The majority of patients (67.7%) underwent haploidentical HSCT (haplo-HSCT). Median follow-up time among survivors was 29.2 (range, 4.7–76.6) months. The molecular profiles of MDS patients at diagnosis and the interrelationships among the various mutations are presented in Figure S1a, b. In total, 231 subjects (67.7%) had at least one genomic alteration. When considering 31 IPSS-M-associated genes, we identified one or more molecular abnormalities in 61.9% of subjects (n = 211). Specific gene mutations in each diagnostic group are presented in Figure S1c. The IPSS-R and IPSS-M scores were calculated in the entire cohort at diagnosis, and the distribution of risk subtypes showed the prevalence of IPSS-R intermediated risk and IPSS-M high-risk subgroups, respectively (Figure S2). According to IPSS-M, patients are clustered in very low risk (n = 4, 1.2%), low risk (n = 39, 11.4%), moderate low risk (n = 38, 11.1%), moderate high risk (n = 61, 17.9%), high risk (n = 115, 33.7%), and very high risk (n = 84, 24.6%). We then evaluated the restratification of patients from IPSS-R to IPSS-M (through merging moderate low and moderate high into moderate). Approximately 49.0% (n = 167) of patients were reclassified: 53.3% (n = 89) were upstaged and 46.7% (n = 78) were downstaged. Specifically for patients in the intermediate IPSS-R group, 21 (17.1%) were downstaged with the majority (76.2%) having no mutations, whereas 48 (39.0%) were reclassified into the higher-risk category with 66.7% of them harboring two or more mutated IPSS-M genes. Table S2 displays the full clinical characteristics of each IPSS-M category. The allocation of patients into the six IPSS-M risk groups had a significantly distinct overall survival (OS) compared with the five IPSS-R risk groups (p = .032 vs. .2) (Figure 1a, b). IPSS-M and IPSS-R categories showed significant differences in leukemia-free survival (LFS) (p = .0029 vs. .0085) (Figure 1c, d) and the cumulative incidence of relapse (CIR) (p = .001 vs. p < .001), but did not show differences in non-relapse mortality (NRM) (Figure S3). Then we compared the two prognostic scoring systems through time-dependent area under curve. With respect to OS and LFS, IPSS-M classification resulted in refined discrimination relative to IPSS-R. Concordance indices (C-indices) derived from IPSS-M versus IPSS-R for OS, LFS, and CIR at 3 years were 0.684 (95% confidence interval [CI], 0.612–0.757) versus 0.612 (95% CI, 0.535–0.688), 0.699 (95% CI, 0.629–0.769) versus 0.641 (95% CI, 0.568–0.714), and 0.708 versus 0.708, respectively (Figure 1e). In the multivariable analyses, we fitted two Cox models separately including IPSS-R and IPSS-M and found IPSS-M and IPSS-R were independent risk factors for OS and LFS (Figure S4). Akaike Information Criterion (AIC) for the IPSS-M versus IPSS-R model was 1168.83 versus 1172.53 for OS, and 1311.3 versus 1312.37 for LFS; C-index was 0.651 (SE = 0.028) versus 0.643 (SE = 0.028) for OS, and 0.640 (SE = 0.027) versus 0.632 (SE = 0.027) for LFS. Further, given the IPSS-M score is represented by its best performance mainly assessed in reduced intensity conditioning-based human leukocyte antigen-matched donor HSCT cases,5 we investigated the prognostic value of IPSS-M in patients undergoing myeloablative conditioning-based haplo-HSCT (n = 231, 67.7%), to refine the applicability of IPSS-M in this setting. Both IPSS-R and IPSS-M showed significant separation of LFS (p = .036 vs. .012) and CIR (p = .040 vs. .017), but did not affect OS and NRM (Figure S5). Slightly improved performance in predicting LFS and CIR was observed for IPSS-M relative to IPSS-R, as indicated by increase in C-index (LFS, 0.711 [95% CI, 0.627–0.794] vs. 0.605 [95% CI, 0.515–0.696]; CIR, 0.711 vs. 0.664) (Figure 1F). By comparing two multivariable models including IPSS-M versus IPSS-R for probability of LFS, only IPSS-M remained an independent prognostic risk factor for patient survival, and AIC was 898.43 versus 901.08, and concordance was 0.632 (SE = 0.03) versus 0.623 (SE = 0.03) (Table S3). Our results further validated the improved predictive performance of IPSS-M for LFS and CIR in the context of haplo-HSCT cohort. We subsequently stratified patients into mutation group (n = 231) and no mutation group (n = 110). Figure S6 shows the restratification of patients from IPSS-R to IPSS-M. In the mutation cohort, IPSS-M categories still maintained a significant prognostic value for the probability of OS (p = .024), LFS (p = .0032), and CIR (p = .022), whereas IPSS-R system only showed significantly different probabilities of LFS (p = .047) and CIR (p = .005) (Figure S7). Concordances derived from IPSS-M versus IPSS-R for OS, LFS at 3 years were 0.704 (95% CI, 0.605–0.803) versus 0.626 (95% CI, 0.523–0.729), and 0.707 (95% CI, 0.610–0.804) versus 0.660 (95% CI, 0.565–0.755), respectively, favoring the predictive value of the former (Figure 1e). In the multivariate analyses regarding LFS, IPSS-M was considered as an independent factor with a lower AIC (738.37 vs. 740.95) and higher C-index (0.699 vs. 0.675) than IPSS-R-based model (Table S4). As for OS, IPSS-M very high risk group had adverse effect in multivariate analysis (HR, 3.02 [95% CI, 1.05–8.66], p = .004), while IPSS-R had lost its significance in univariate analysis (p = .196). Besides, both IPSS-M and IPSS-R failed to stratify the different probabilities of OS, LFS, NRM, and CIR in the non-mutated group (Figure S8). This is in good accordance with the mutation architecture of IPSS-M model. Together, the clinical implementation of IPSS-M is expected to improve the risk identification of mutated patients, but patients without detectable mutations may not benefit from it. To the best of our knowledge, this is the first external validation of the newly published IPSS-M model in Chinese MDS cohort homogeneously treated with allogeneic HSCT from a multicenter cohort, which demonstrated its modestly improved prognostic predictive power over time relative to conventional IPSS-R scoring system. This is similar to that reported by Gurnari et al.5 The allocation to IPSS-M risk groups in the current study was distinct from real-word MDS cohorts without HSCT as previously reported, showing a skewing toward high- and very high-risk categories. A significant association between the allocation to risk groups and survival outcomes was observed, with the C-indices derived from IPSS-M versus IPSS-R for 3-year OS and LFS being 0.684 versus 0.612 and 0.699 versus 0.641, respectively, which were higher than the previously reported transplant-specific scores,5 but not as good as the results reported by Sauta et al.6 The heterogeneity of transplantation procedure may be part of the explanation. Similarly, when restricting our analyses to patients with haplo-HSCT and patients carrying detectable mutations, IPSS-M retained improved prognostic value with respect to OS and LFS relative to IPSS-R; while it failed to stratify individual probability of OS and LFS in those with no mutation. There are potential weaknesses in our work, mainly resulting from the retrospective nature of this registry-based, multicenter study. Not all laboratories are able to offer broad molecular genetic panels with all 31 genes of IPSS-M. There were missing data on GNB1, PPM1D, and PRPF8 status in some patients. However, this is closer to the real-world since not all institutions could support comprehensive genetic detection. Sauta et al. has confirmed that even in the presence of a small amount of missing data on molecular features, IPSS-M model was robust in achieving high accuracy in predicting the probability of LFS.6 Moreover, our cohort consisted mostly of haploidentical patients with myeloablative conditioning regimen, and therefore, generalization of our findings to other populations should be cautious. Additionally, we only concentrate on the mutational landscape at first diagnosis, and further studies at other time points in the course of disease, such as pre-transplantation, should be fully explored. Collectively, our results provide evidence that IPSS-M model significantly increases prognostic discrimination of MDS patients undergoing allo-HSCT, and may provide a basis for optimized clinical decision-making and early adoption of potential interventions in transplant procedure for high- or very high-risk patients. Jimin Shi, Tingting Yang, and Binqian Jiang designed the whole study. Guifang Ouyang, Jian Yu, Jianping Lan, Ying Lu, Xiaoyu Lai, Baodong Ye, Yi Chen, Lizhen Liu, Yang Xu, Pengfei Shi, Haowen Xiao, Huixian Hu, Qunyi Guo, Huarui Fu, Yishan Ye, Jie Sun, Weiyan Zheng, Jingsong He, Yi Zhao, Wu Wenjun, and Zhen Cai collected data; Tingting Yang, Binqian Jiang, and Xinyu Wang organized the data. Tingting Yang and Binqian Jiang analyzed and interpreted the data, and wrote the manuscript, which was proofreaded by Guoqing Wei, He Huang, and Jimin Shi. All of the authors have read and approved the final version of the manuscript. We are grateful to all the clinicians and patients who contributed to this study. The study was supported by the National Natural Science Foundation of China (grant numbers: 82070179, 82170205, 82270221 and 82170210). The authors declare no conflicts of interest. All data used and analyzed are available from the corresponding author on reasonable request. Figure S1. Gene mutation profile in patients with myelodysplastic neoplasms at diagnosis. (a) Frequency of gene mutations including IPSS-M-related genes in total cohort. Colors linked to the bars represent the disease diagnosis subtypes according to the 2022 WHO classification. The distribution of mutations <2% is magnified in the upper panel. (b) Circos diagram depicts the relative frequency and pairwise co-occurrence of mutations. (c) Frequency of gene mutations among MDS diagnosis subtypes according to 2022 WHO classification. IPSS-M, Molecular International Prognostic Scoring System; MDS, myelodysplastic neoplasms; MDS-5q, MDS with isolated deletion of long arm of chromosome 5; MDS-SF3B1, MDS with mutated SF3B1; MDS-biTP53, MDS with biallelic TP53 inactivation; MDS-LB, MDS with low blasts; MDS-h, hypoplastic MDS; MDS-IB, MDS with increased blasts (1, 2); MDS-f, MDS with fibrosis; WHO, World Health Organization. Figure S2. The redistribution of IPSS-R to IPSS-M for all patients with myelodysplastic neoplasms at diagnosis. (a) Sankey diagram comparing IPSS-R and IPSS-M risk categories. (b) The distribution of IPSS-M in different IPSS-R risk groups. Vertical axis represents IPSS-R categories, and horizontal axis denotes IPSS-M categories (indicated with different colors). IPSS-M, Molecular International Prognostic Scoring System; IPSS-R, Revised International Prognostic Scoring System. Figure S3. NRM and CIR assessments of IPSS-R and IPSS-M in the entire cohort. NRM (a and b). CIR (c and d). CIR, cumulative incidence of relapse; HSCT, hematopoietic stem cell transplantation; IPSS-M, Molecular International Prognostic Scoring System; IPSS-R, Revised International Prognostic Scoring System; NRM, non-relapse mortality. Figure S4. Multivariable analyses for survival outcomes separately included IPSS-R and IPSS-M. AIC, the Akaike Information Criterion; CI, confidence interval; HR, hazard ratio; HRD, haploidentical related donor; HSCT, hematopoietic stem cell transplantation; IPSS-M, Molecular International Prognostic Scoring System; IPSS-R, Revised International Prognostic Scoring System; LFS, leukemia-free survival; MSD, human leukocyte antigen-matched sibling donor; MUD, human leukocyte antigen-matched unrelated donor; OS, overall survival; se, standard error. Figure S5. Clinical assessment of IPSS-R and IPSS-M in myelodysplastic neoplasms patients undergoing haploidentical hematopoietic stem cell transplantation. OS (a and b). LFS (c and d). NRM (e and f). CIR (g and h). CIR, cumulative incidence of relapse; HSCT, hematopoietic stem cell transplantation; IPSS-M, Molecular International Prognostic Scoring System; IPSS-R, Revised International Prognostic Scoring System; LFS, leukemia-free survival; NRM, non-relapse mortality; OS, overall survival. Figure S6. The redistribution of IPSS-R to IPSS-M according to the presence of mutations. Abbreviations: IPSS-M, Molecular International Prognostic Scoring System; IPSS-R, Revised International Prognostic Scoring System. Figure S7. Clinical assessment of IPSS-R and IPSS-M in mutated patients. OS (a and b). LFS (c and d). NRM (e and f). CIR (g and h). CIR, cumulative incidence of relapse; HSCT, hematopoietic stem cell transplantation; IPSS-M, Molecular International Prognostic Scoring System; IPSS-R, Revised International Prognostic Scoring System; LFS, leukemia-free survival; NRM, non-relapse mortality; OS, overall survival. Figure S8. Clinical assessment of IPSS-R and IPSS-M in non-mutated patients. OS (a and b). LFS (c and d). NRM (e and f). CIR (g and h). CIR, cumulative incidence of relapse; HSCT, hematopoietic stem cell transplantation; IPSS-M, Molecular International Prognostic Scoring System; IPSS-R, Revised International Prognostic Scoring System; LFS, leukemia-free survival; NRM, non-relapse mortality; OS, overall survival. Table S1. Patient characteristics in the entire cohort. Table S2. Patient characteristics among each Molecular International Prognostic Scoring System risk category. Table S3. Molecular International Prognostic Scoring System versus Revised International Prognostic Scoring System in multivariable analyses for leukemia-free survival in patients undergoing haploidentical hematopoietic stem cell transplantation. Table S4. Molecular International Prognostic Scoring System versus Revised International Prognostic Scoring System in multivariable analyses for leukemia-free survival in patients with mutations. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.