BACKGROUND: Increasingly large and complex biomedical data sets challenge conventional hypothesis-driven analytical approaches, however, data-driven unsupervised learning can detect inherent patterns in such data sets. METHODS: While unsupervised analysis in the medical literature commonly only utilizes a single clustering algorithm for a given data set, we developed a large-scale model with 605 different combinations of target dimensionalities as well as transformation and clustering algorithms and subsequent meta-clustering of individual results. With this model, we investigated a large cohort of 1383 patients from 59 centers in Germany with newly diagnosed acute myeloid leukemia for whom 212 clinical, laboratory, cytogenetic and molecular genetic parameters were available. RESULTS: Unsupervised learning identifies four distinct patient clusters, and statistical analysis shows significant differences in rate of complete remissions, event-free, relapse-free and overall survival between the four clusters. In comparison to the standard-of-care hypothesis-driven European Leukemia Net (ELN2017) risk stratification model, we find all three ELN2017 risk categories being represented in all four clusters in varying proportions indicating unappreciated complexity of AML biology in current established risk stratification models. Further, by using assigned clusters as labels we subsequently train a supervised model to validate cluster assignments on a large external multicenter cohort of 664 intensively treated AML patients. CONCLUSIONS: Dynamic data-driven models are likely more suitable for risk stratification in the context of increasingly complex medical data than rigid hypothesis-driven models to allow for a more personalized treatment allocation and gain novel insights into disease biology.

There are various ways in which clinicians can predict the risk of disease progression in patients with leukemia, helping them to treat the patients accordingly. However, these approaches are usually designed by human experts and might not fully capture the complexity of a patient’s disease. Here, with a large cohort of patients with acute myeloid leukemia, we design an unsupervised machine learning model – a type of computer model that learns from patterns in data without human input—to separate these patients into subgroups according to risk. We identify four distinct groups which differ with regards to patient genetics, laboratory values, and clinical characteristics. These groups have differences in response to treatment and patient survival, and we validate our findings in another dataset. Our approach might help clinicians to better predict outcomes in patients with leukemia and make decisions on treatment.

Department of Hematology and Oncology University Hospital Schleswig Holstein Kiel Germany

Department of Hematology and Stem Cell Transplantation University Hospital Essen Essen Germany

Department of Hematology Oncology and Palliative Care Robert Bosch Hospital Stuttgart Germany

Department of Hematology Oncology and Tumor Immunology Charité Berlin Germany

Department of Internal Medicine 1 University Hospital Carl Gustav Carus Dresden Germany

Department of Internal Medicine 3 Klinikum Chemnitz GmbH Chemnitz Germany

Department of Internal Medicine 5 University Hospital Erlangen Erlangen Germany

Department of Internal Medicine 5 University Hospital Nuremberg Nuremberg Germany

Department of Internal Medicine A University Hospital Muenster Muenster Germany

Department of Internal Medicine Hematology and Oncology Masaryk University Hospital Brno Czech Republic

Department of Medicine 2 Hematology and Oncology Goethe University Frankfurt Frankfurt Germany

Department of Medicine 3 Hospital Leverkusen Leverkusen Germany

Department of Medicine 5 University Hospital Heidelberg Heidelberg Germany

Department of Software and Multimedia Technology Technical University Dresden Dresden Germany

Else Kröner Fresenius Center for Digital Health Technical University Dresden Dresden Germany

German Consortium for Translational Cancer Research DKFZ Heidelberg Germany

Hospital Barmherzige Brueder Regensburg Regensburg Germany

Institute for Biostatistics and Clinical Research University Muenster Muenster Germany

Laboratory for Leukemia Diagnostics Department of Medicine 3 University Hospital LMU Munich Munich Germany

Medical Clinic and Policlinic 1 Hematology and Cell Therapy University Hospital Leipzig Germany

National Center for Tumor Diseases Dresden Germany

Zobrazit více v PubMed

Papaemmanuil E, et al. Genomic classification and prognosis in acute myeloid leukemia. N. Eng. J. Med. 2016;374:2209–2221. doi: 10.1056/NEJMoa1516192. PubMed DOI PMC

Bullinger L, Döhner K, Döhner H. Genomics of acute myeloid leukemia diagnosis and pathways. JCO. 2017;35:934–946. doi: 10.1200/JCO.2016.71.2208. PubMed DOI

Patel JP, et al. Prognostic relevance of integrated genetic profiling in acute myeloid leukemia. N. Eng. J. Med. 2012;366:1079–1089. doi: 10.1056/NEJMoa1112304. PubMed DOI PMC

Döhner H, et al. Diagnosis and management of AML in adults: 2017 ELN recommendations from an international expert panel. Blood. 2017;129:424–447. doi: 10.1182/blood-2016-08-733196. PubMed DOI PMC

Breiman L. Statistical Modeling: The two cultures (with comments and a rejoinder by the author) Statist. Sci. 2001;16:199–231. doi: 10.1214/ss/1009213726. DOI

Shouval R, et al. Application of machine learning algorithms for clinical predictive modeling: a data-mining approach in SCT. Bone Marrow Transplant. 2014;49:332–337. doi: 10.1038/bmt.2013.146. PubMed DOI

Raghupathi W, Raghupathi V. Big data analytics in healthcare: promise and potential. Health Inf. Sci. Syst. 2014;2:3. doi: 10.1186/2047-2501-2-3. PubMed DOI PMC

Obermeyer Z, Emanuel EJ. Predicting the future—big data, machine learning, and clinical medicine. N. Engl. J. Med. 2016;375:1216–1219. doi: 10.1056/NEJMp1606181. PubMed DOI PMC

Eckardt J-N, Bornhäuser M, Wendt K, Middeke JM. Application of machine learning in the management of acute myeloid leukemia: current practice and future prospects. Blood Adv. 2020;4:6077–6085. doi: 10.1182/bloodadvances.2020002997. PubMed DOI PMC

Libbrecht MW, Noble WS. Machine learning applications in genetics and genomics. Nat. Rev. Genet. 2015;16:321–332. doi: 10.1038/nrg3920. PubMed DOI PMC

Barlow HB. Unsupervised learning. Neural Comput. 1989;1:295–311. doi: 10.1162/neco.1989.1.3.295. DOI

Cancer Genome Atlas Research Network, Ley, T. J. et al. Genomic and Epigenomic landscapes of adult de novo acute myeloid leukemia. N. Eng. J. Med.368, 2059–2074 (2013). PubMed PMC

Way GP, et al. Machine learning detects pan-cancer Ras pathway activation in the Cancer Genome Atlas. Cell Rep. 2018;23:172–180.e3. doi: 10.1016/j.celrep.2018.03.046. PubMed DOI PMC

Röllig C, et al. A novel prognostic model in elderly patients with acute myeloid leukemia: results of 909 patients entered into the prospective AML96 trial. Blood. 2010;116:971–978. doi: 10.1182/blood-2010-01-267302. PubMed DOI

Schaich M, et al. High-dose cytarabine consolidation with or without additional amsacrine and mitoxantrone in acute myeloid leukemia: results of the prospective randomized AML2003 trial. J. Clin. Oncol. 2013;31:2094–2102. doi: 10.1200/JCO.2012.46.4743. PubMed DOI

Röllig C, et al. Intermediate-dose cytarabine plus mitoxantrone versus standard-dose cytarabine plus daunorubicin for acute myeloid leukemia in elderly patients. Ann. Oncol. 2018;29:973–978. doi: 10.1093/annonc/mdy030. PubMed DOI

Röllig C, et al. Addition of sorafenib versus placebo to standard therapy in patients aged 60 years or younger with newly diagnosed acute myeloid leukaemia (SORAML): a multicentre, phase 2, randomised controlled trial. Lancet Oncol. 2015;16:1691–1699. doi: 10.1016/S1470-2045(15)00362-9. PubMed DOI

Arber DA, et al. The 2016 revision to the World Health Organization classification of myeloid neoplasms and acute leukemia. Blood. 2016;127:2391–2405. doi: 10.1182/blood-2016-03-643544. PubMed DOI

Krug U, et al. Increasing intensity of therapies assigned at diagnosis does not improve survival of adults with acute myeloid leukemia. Leukemia. 2016;30:1230–1236. doi: 10.1038/leu.2016.25. PubMed DOI

Braess J, et al. Sequential high-dose cytarabine and mitoxantrone (S-HAM) versus standard double induction in acute myeloid leukemia—a phase 3 study. Leukemia. 2018;32:2558–2571. doi: 10.1038/s41375-018-0268-9. PubMed DOI PMC

Thiede C, et al. Analysis of FLT3-activating mutations in 979 patients with acute myelogenous leukemia: association with FAB subtypes and identification of subgroups with poor prognosis. Blood. 2002;99:4326–4335. doi: 10.1182/blood.V99.12.4326. PubMed DOI

Thiede C, et al. Prevalence and prognostic impact of NPM1 mutations in 1485 adult patients with acute myeloid leukemia (AML) Blood. 2006;107:4011–4020. doi: 10.1182/blood-2005-08-3167. PubMed DOI

Taube F, et al. CEBPA mutations in 4708 patients with acute myeloid leukemia—differential impact of bZIP and TAD mutations on outcome. Blood. 2021 doi: 10.1182/blood.2020009680. PubMed DOI

Gebhard C, et al. Profiling of aberrant DNA methylation in acute myeloid leukemia reveals subclasses of CG-rich regions with epigenetic or genetic association. Leukemia. 2019;33:26–36. doi: 10.1038/s41375-018-0165-2. PubMed DOI

Stasik S, et al. An optimized targeted next-generation sequencing approach for sensitive detection of single nucleotide variants. Biomol. Detect. Quantif. 2018;15:6–12. doi: 10.1016/j.bdq.2017.12.001. PubMed DOI PMC

Marimont RB, Shapiro MB. Nearest neighbour searches and the curse of dimensionality. IMA J. Appl. Math. 1979;24:59–70. doi: 10.1093/imamat/24.1.59. DOI

Rousseeuw PJ. Silhouettes: A graphical aid to the interpretation and validation of cluster analysis. J. Comput. Appl. Math. 1987;20:53–65. doi: 10.1016/0377-0427(87)90125-7. DOI

Caliński T, Harabasz J. A dendrite method for cluster analysis. Commun. Stat. 1974;3:1–27.

Davies DL, Bouldin DW. A cluster separation measure. IEEE Trans. Pattern Anal.Mach. Intell. 1979;PAMI-1:224–227. doi: 10.1109/TPAMI.1979.4766909. PubMed DOI

Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J. R. Stat. Soc. Series B (Methodol.) 1995;57:289–300.

Bennett JM, et al. Proposals for the classification of the acute leukaemias. French-American-British (FAB) co-operative group. Br. J. Haematol. 1976;33:451–458. doi: 10.1111/j.1365-2141.1976.tb03563.x. PubMed DOI

Gale RE, et al. The impact of FLT3 internal tandem duplication mutant level, number, size, and interaction with NPM1 mutations in a large cohort of young adult patients with acute myeloid leukemia. Blood. 2008;111:2776–2784. doi: 10.1182/blood-2007-08-109090. PubMed DOI

Ley TJ, et al. DNMT3A mutations in acute myeloid leukemia. N. Engl. J. Med. 2010;363:2424–2433. doi: 10.1056/NEJMoa1005143. PubMed DOI PMC

Yang L, Rau R, Goodell MA. DNMT3A in haematological malignancies. Nat. Rev. Cancer. 2015;15:152–165. doi: 10.1038/nrc3895. PubMed DOI PMC

Gaidzik VI, et al. RUNX1 mutations in acute myeloid leukemia are associated with distinct clinico-pathologic and genetic features. Leukemia. 2016;30:2160–2168. doi: 10.1038/leu.2016.126. PubMed DOI

Pratcorona M, et al. Acquired mutations in ASXL1 in acute myeloid leukemia: prevalence and prognostic value. Haematologica. 2012;97:388–392. doi: 10.3324/haematol.2011.051532. PubMed DOI PMC

Bowen D, et al. TP53 gene mutation is frequent in patients with acute myeloid leukemia and complex karyotype, and is associated with very poor prognosis. Leukemia. 2009;23:203–206. doi: 10.1038/leu.2008.173. PubMed DOI

Haferlach C, et al. Mutations of the TP53 gene in acute myeloid leukemia are strongly associated with a complex aberrant karyotype. Leukemia. 2008;22:1539–1541. doi: 10.1038/leu.2008.143. PubMed DOI

Middeke JM, et al. TP53 mutation in patients with high-risk acute myeloid leukaemia treated with allogeneic haematopoietic stem cell transplantation. Br. J. Haematol. 2016;172:914–922. doi: 10.1111/bjh.13912. PubMed DOI

Jerez A, et al. Topography, clinical, and genomic correlates of 5q myeloid malignancies revisited. JCO. 2012;30:1343–1349. doi: 10.1200/JCO.2011.36.1824. PubMed DOI PMC

Li H-Y, et al. Favorable prognosis of biallelic CEBPA gene mutations in acute myeloid leukemia patients: a meta-analysis. Eur. J. Haematol. 2015;94:439–448. doi: 10.1111/ejh.12450. PubMed DOI

Mannelli F, et al. CEBPA-double-mutated acute myeloid leukemia displays a unique phenotypic profile: a reliable screening method and insight into biological features. Haematologica. 2017;102:529–540. doi: 10.3324/haematol.2016.151910. PubMed DOI PMC

Hou H-A, et al. GATA2 mutations in patients with acute myeloid leukemia-paired samples analyses show that the mutation is unstable during disease evolution. Ann. Hematol. 2015;94:211–221. doi: 10.1007/s00277-014-2208-8. PubMed DOI

Fasan A, et al. GATA2 mutations are frequent in intermediate-risk karyotype AML with biallelic CEBPA mutations and are associated with favorable prognosis. Leukemia. 2013;27:482–485. doi: 10.1038/leu.2012.174. PubMed DOI

Bowman RL, Levine RL. TET2 in normal and malignant hematopoiesis. Cold Spring Harb. Perspect. Med. 2017;7:a026518. doi: 10.1101/cshperspect.a026518. PubMed DOI PMC

Weissmann S, et al. Landscape of TET2 mutations in acute myeloid leukemia. Leukemia. 2012;26:934–942. doi: 10.1038/leu.2011.326. PubMed DOI

Wang R, Gao X, Yu L. The prognostic impact of tet oncogene family member 2 mutations in patients with acute myeloid leukemia: a systematic-review and meta-analysis. BMC Cancer. 2019;19:389. doi: 10.1186/s12885-019-5602-8. PubMed DOI PMC

Dastugue N, et al. Prognostic significance of karyotype in de novo adult acute myeloid leukemia. The BGMT group. Leukemia. 1995;9:1491–1498. PubMed

Schiffer CA, Lee EJ, Tomiyasu T, Wiernik PH, Testa JR. Prognostic impact of cytogenetic abnormalities in patients with de novo acute nonlymphocytic leukemia. Blood. 1989;73:263–270. doi: 10.1182/blood.V73.1.263.263. PubMed DOI

Gerstung M, et al. Precision oncology for acute myeloid leukemia using a knowledge bank approach. Nat. Genet. 2017;49:332–340. doi: 10.1038/ng.3756. PubMed DOI PMC

Bullinger L, et al. Gene-expression profiling identifies distinct subclasses of core binding factor acute myeloid leukemia. Blood. 2007;110:1291–1300. doi: 10.1182/blood-2006-10-049783. PubMed DOI

Bullinger L, et al. Use of gene-expression profiling to identify prognostic subclasses in adult acute myeloid leukemia. N. Engl. J. Med. 2004;350:1605–1616. doi: 10.1056/NEJMoa031046. PubMed DOI

Awada H, et al. Machine learning integrates genomic signatures for subclassification beyond primary and secondary acute myeloid Leukemia. Blood. 2021 doi: 10.1182/blood.2020010603. PubMed DOI PMC

Lang KM, et al. Core outcome set measurement for future clinical trials in acute myeloid leukemia: the HARMONY study protocol using a multi-stakeholder consensus-based Delphi process and a final consensus meeting. Trials. 2020;21:437. doi: 10.1186/s13063-020-04384-1. PubMed DOI PMC

LeCun Y, Bengio Y, Hinton G. Deep learning. Nature. 2015;521:436–444. doi: 10.1038/nature14539. PubMed DOI

Röllig C, et al. Sorafenib or placebo in patients with newly diagnosed acute myeloid leukaemia: long-term follow-up of the randomized controlled SORAML trial. Leukemia. 2021;35:2517–2525. doi: 10.1038/s41375-021-01148-x. PubMed DOI PMC

Kayser S, Levis MJ. Advances in targeted therapy for acute myeloid leukaemia. Br. J. Haematol. 2018;180:484–500. doi: 10.1111/bjh.15032. PubMed DOI PMC

Perl AE. The role of targeted therapy in the management of patients with AML. Blood Adv. 2017;1:2281–2294. doi: 10.1182/bloodadvances.2017009829. PubMed DOI PMC

Wendt, K. KarstenWendtTUD/sal-metaclustering. Zenodo10.5281/zenodo.7841798 (2023).