Myelodysplastic syndromes (MDS) are a heterogeneous group of myeloid neoplasms, characterized by ineffective hematopoiesis leading to cytopenias, dysplastic features in bone marrow and by progression to acute myeloid leukemia in about a third of patients.1 The latest World Health Organization (WHO) classification and International Consensus Classification (ICC) have confirmed the importance of molecular biology in characterizing these diseases, particularly MDS with SF3B1 and TP53 mutations.2,3 As the course of these diseases is very diverse, it is essential to stratify patients, and the International Prognostic Scoring System-Revised (IPSS-R) classification was used for more than a decade.4 However, this classification only took into account hematological and cytogenetic data without formally integrating molecular biology, which have since shown its importance in understanding the pathophysiology of MDS but also in assessing prognosis.5,6 This oversight was rectified and in 2022 the IPSS-molecular (IPSS-M) was published, based on an initial cohort of 2,957 patients and then validated on an external cohort of 754 patients.7 Briefly, this score uses hematological parameters, cytogenetic abnormalities and somatic mutations in 31 genes to classify patients into six risk categories. However, this score does not include flow cytometry data. The role of multiparametric flow cytometry (MFC) in the diagnosis of MDS is not clearly defined. Several scores have been previously defined like the Ogata score focusing on progenitor cells or the RED score analyzing nucleated red blood cells8,9 but the use of these different tools is limited by the lack of standardization. Previously, we published an article in Haematologica on the use of artificial intelligence (AI) on flow cytometry data to improve the diagnosis of MDS.10 After selecting the most relevant parameters using a Boruta algorithm on a cohort of 191 patients, we developed a score, thanks to an Elasticnet model, that greatly improved the sensitivity of the Ogata score, enabling MDS to be diagnosed with a sensitivity of 91.8% and a specificity of 92.5%. Briefly, this score uses several parameters from the Ogata score like the granulocyte/lymphocyte side scatter (SSC) peak channel ratio, the percentage of B-cell and CD34 myeloid progenitors. This score has also been validated on an external cohort of 89 patients but only its diagnostic value had been evaluated.
We thought it might be interesting to compare our MFC score based on AI with the IPSS-M score, which is the current gold standard for prognostic classification of MDS, particularly regarding the diagnosis of low- and high-risk forms of progression.
We obtained data from 119 patients with complete molecular and MFC characteristics distributed over three different centers, 49 from Cochin Hospital (Assistance Publique des Hôpitaux de Paris, APHP), 16 from Ambroise Paré Hospital (APHP), and 54 from Amiens Hospital. This study was approved by the local institutional review board and was conducted according to the Declaration of Helsinki. We obtained in this cohort an average age of 77.1 years, with a standard deviation (SD) of 11.3 years. Regarding biological parameters at diagnosis, the white blood cell (WBC) count was evaluated at 5.72x109/L on average (SD: 4.9x109/L), absolute neutrophil count (ANC) at 3.11x109/L (SD: 3.3x109/L), hemoglobin at 10.45 g/dL (SD: 1.85 g/dL), and platelets at 165x109/L (SD: 129.9x109/L). Among the 119 patients, according to the World Health Organization (WHO) 2022 classification, we obtained 55 MDS with low blast (46%), 21 with increased blasts 1 (18%), 18 with SF3B1 mutation (15%), 12 with increased blasts 2 (10%), eight with a biallelic TP53 inactivation (7%), and five with a 5q deletion (4%). We could also describe the cohort by the ICC of myeloid neoplasms, with 29 MDS with multilineage dysplasia (24%), 21 excess blasts (18%), still the 18 patients with mutated SF3B1 (15%), 15 not otherwise specified (13%), 11 with single-lineage dysplasia (9%), 11 (9%) with an excess of blasts greater than 10% in bone marrow representing the MDS/AML category, nine with multi-hit TP53 (8%), and still the five patients with a 5q deletion (4%). Concerning the IPSS-M, we obtained a majority of low-risk score with 44 patients (37%), 17 moderate high (14%), 16 moderate low (13.5%), 16 very low (13.5%), 14 high (12%), and 12 very high (10%).
Table 1.Global cohort characteristics.
We observed a mean of 2.02 (SD: 2.98) for the MDS score and a mean of -0.30 (SD: 1.27) for the IPSS-M score in the overall cohort. Taking into account the IPSS-M classification, we obtained for the AI-based MDS score for patients in the very-low group a mean of 0.8 (SD: 1.4, max: 2.58, min: -1.62), for the low group a mean of 0.82 (SD: 1.9, max: 5.2, min: -2.1), for the moderate low group a mean of 1.69 (SD: 1.93, max: 6.2, min: -1.3), for the moderate high group a mean of 2.41 (SD: 2.91, max: 11, min: -2.9), for the high group a mean of 4.6 (SD: 3.5, max: 12, min: -0.6), very high group at 4.9 (SD: 4.39, max: 15.5, min: 1.55).
We then conducted a mean comparison test between the IPSS-M groups and the AI-based MDS score obtained for each patient, resulting in a P value <0.001. Post hoc Tukey tests corrected by the Bonferroni method were performed (Table 2) between the different classes and found a significant difference between the low and high (difference: 3.8; adjusted P< 0.001), the low and very high (difference: 4.1; adjusted P<0.001), the moderate low and high (difference: 2.9; adjusted P=0.02), the moderate low and very high (difference: 3.2; adjusted P=0.01), the high and very low (difference: -3.8; adjusted P=0.001), and very high and very low groups (difference: -4.1; adjusted P<0.001). The results are presented in Figure 1, with the mean and 95% confidence interval for each IPSS-M group and clearly show a significant positive association between the AI-based MDS score and the IPSS-M score.
We then conducted another comparison between the WHO 2022 groups and the AI-based MDS score, which show a significant difference (P<0.001). The post hoc tests found significant differences between the MDS increased blasts 1 and low blast groups (3.39 vs. 0.82, adjusted P=0.001), the increased blasts 2 and low blast (4.1 vs. 0.82; adjusted P<0.001), the biallelic TP53 inactivation and low blast (6.18 vs. 0.82; adjusted P<0.001), the increased blasts 2 and the SF3B1 mutated (4.1 vs. 1.25; adjusted P=0.03), the biallelic TP53 inactivation and SF3B1 mutated (6.18 vs. 1.25; adjusted P<0.001) and finally between the biallelic TP53 inactivation and the 5q deletion groups (6.18 vs. 0.69; adjusted P=0.002). Finally, we compared ICC 2022 and the MFC score, with another significant difference (P<0.001). The most significant ones were between MDS SLD and MDS/AML TP53 (1 .7 vs. 9.5; P<0.001), the 5q deletion and MDS/AML TP53 (0 .7 vs. 9.5; P<0.001), MDS NOS and MDS/AML (0.1 vs. 4.3; P=0.001), MDS-EB and MDS/AML TP53 (3.4 vs. 9.5; P=0.003) and between MDS MLD and MDS/AML TP53 groups (0.84 vs. 9.5; P=0.003).
Table 2.Post hoc tests characteristics.
Figure 1.Distribution of Myelodysplastic Syndromes Scores from the elasticnet algorithm across the different Molecular International Prognostic Scoring System groups and the 2022 World Health Organization classification. The range corresponds to the 95% confidence interval. It shows a significant positive association between the Artifical Intelligence-based Myelodysplastic Syndromes score and the Molecular International Prognostic Scoring System Score. MDS: myelodysplastic syndromes; LB: low blast; IB1: increased blast between 5 and 9%; IB2: increased blast >9%; bi-TP53: TP53 bi-allelic abnormalities.
Here, we compared the AI-based MDS score’s performance with the latest classifications of hemopathies and the IPSS-M prognostic score and found a perfect correlation between the score and these different entities. Furthermore, as shown in Figure 1, the score illustrates the linear progression between low- and high-risk categories. In this way, it could be used as a prognostic score in patients for whom molecular biology cannot be performed, for cost reasons in particular, in order to propose the most appropriate treatment for their MDS. Indeed, flow cytometers are available in almost all hospitals and the Ogata score, which is required to calculate the AI-based score, costs around 90 euros (compared with 2,000 euros for targeted next-generation sequencing). The main limitation of this study lies in the cohort size, which is distributed across only three hospitals. While this is sufficient for a proof of concept, increasing the cohort size, especially in the less common molecular groups, seems necessary in the future. The use of MFC instead of molecular biology can be useful in certain situations, such as determining IGHV mutational status in CLL.11 To our knowledge, the diagnostic and prognostic scores based on MFC have neither been compared with the latest classifications, nor with the IPSS-M score, unlike other prognostic classifications such as IPSS-R. In a real-world validation cohort, Sauta and colleagues showed that the IPSS-M provided a better prognostic classification than the IPSS-R, with 46% of patients falling within the risk group and a better selection of candidates for hematopoietic stem cell transplantation.12 Therefore, it would be interesting to carry out this correlation, in order to highlight the performance of MFC scores, like the iFS score which is known for its excellent balance between sensitivity and specificity.13 Other teams have developed fully automated systems using the FlowSOM algorithm with raw data preprocessing and then using a machine learning algorithm.14 This process has considerably improved scores such as iFS and Ogata and would therefore be an ideal candidate for evaluation in the face of new prognostic tools such as IPSS-M.
Footnotes
- Received July 25, 2024
- Accepted September 27, 2024
Correspondence
Disclosures
No conflicts of interest to disclose.
References
- Adès L, Itzykson R, Fenaux P. Myelodysplastic syndromes. Lancet. 2014; 383(9936):2239-2252. Google Scholar
- Khoury JD, Solary E, Abla O. The 5th edition of the World Health Organization Classification of Haematolymphoid Tumours: myeloid and histiocytic/dendritic neoplasms. Leukemia. 2022; 36(7):1703-1719. Google Scholar
- Arber DA, Orazi A, Hasserjian RP. International Consensus Classification of Myeloid Neoplasms and Acute Leukemias: integrating morphologic, clinical, and genomic data. Blood. 2022; 140(11):1200-1228. Google Scholar
- Greenberg PL, Tuechler H, Schanz J. Revised International Prognostic Scoring System for myelodysplastic syndromes. Blood. 2012; 120(12):2454-2465. Google Scholar
- Papaemmanuil E, Gerstung M, Malcovati L. Clinical and biological implications of driver mutations in myelodysplastic syndromes. Blood. 2013; 122(22):3616-3627. Google Scholar
- Nazha A, Al-Issa K, Hamilton BK. Adding molecular data to prognostic models can improve predictive power in treated patients with myelodysplastic syndromes. Leukemia. 2017; 31(12):2848-2850. Google Scholar
- Bernard E, Tuechler H, Greenberg PL. Molecular International Prognostic Scoring System for myelodysplastic syndromes. NEJM Evid. 2022; 1(7):EVIDoa2200008. Google Scholar
- Ogata K, Della Porta MG, Malcovati L. Diagnostic utility of flow cytometry in low-grade myelodysplastic syndromes: a prospective validation study. Haematologica. 2009; 94(8):1066-1074. Google Scholar
- Mathis S, Chapuis N, Debord C. Flow cytometric detection of dyserythropoiesis: a sensitive and powerful diagnostic tool for myelodysplastic syndromes. Leukemia. 2013; 27(10):1981-1987. Google Scholar
- Clichet V, Lebon D, Chapuis N. Artificial intelligence to empower diagnosis of myelodysplastic syndromes by multiparametric flow cytometry. Haematologica. 2023; 108(9):2435-2443. Google Scholar
- Couillez G, Morel P, Clichet V. Flow cytometry as a fast, cost-effective tool to assess IGHV mutational status in CLL. Blood Adv. 2023; 7(17):4701-4704. Google Scholar
- Sauta E, Robin M, Bersanelli M. Real-world validation of molecular international prognostic scoring system for myelodysplastic syndromes. J Clin Oncol. 2023; 41(15):2827-2842. Google Scholar
- Cremers EMP, Westers TM, Alhan C. Implementation of erythroid lineage analysis by flow cytometry in diagnostic models for myelodysplastic syndromes. Haematologica. 2017; 102(2):320-326. Google Scholar
- Duetz C, Van Gassen S, Westers TM. Computational flow cytometry as a diagnostic tool in suspected-myelodysplastic syndromes. Cytometry A. 2021; 99(8):814-824. Google Scholar
Figures & Tables
Article Information

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.