Abstract: 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.

Abstract: About 20-25% of patients with Acute Myeloid Leukemia (AML) have primary drug resistant disease and fail to achieve complete remission after induction therapy. These patients have an extremely poor prognosis and cannot reliably be identified prior to therapy with current methods. The aim of this work was to develop a predictive tool that can identify therapy resistant patients with high accuracy at the time of diagnosis. We used two independent Affymetrix gene expression (GE) data sets and standard molecular and clinical variables to develop a predictive score for response to cytarabine/anthracycline-based induction chemotherapy. The "training set 1" consisted of 407 adult AML patients enrolled in the AMLCG-1999 trial (GSE37642). Training set 2 included 449 adults treated in various HOVON trials (GSE6891). GE-based classifiers for primary treatment resistance were developed in training set 1 using a penalized logistic regression approach (Lasso). A cut off with a specificity of 90% was predefined in training set 1. Training set 2 was used to select the best classifier. The predictive score and cut off were then validated in a third, fully independent data set, comprising 260 patients enrolled in AMLCG-1999 and 2008 trials studied by RNA sequencing. Additionally, targeted amplicon sequencing data for 68 recurrently mutated genes in AML was available for training set 1 and the validation set. The final classifier (Predictive score 29 MRC - PS29MRC) consisted of 29 gene expression values and the cytogenetic risk group (defined according to the United Kingdom Medical Research Council (MRC) classification) and was calculated as a weighted sum of Lasso coefficients and predictor values. PS29MRC was a highly significant predictor of resistant disease in the validation set with an odds ratio of 2.32 (p=1.53x10-8, AUC: 0.75). We tested the signature in a multivariable model including all variables with univariate p-value & lt;0.05. TP53 mutations, age and PS29MRC (OR: 1.70; p=0.0020) were left significant in the validation set. In comparison to published predictive classifiers like the model by Walter et al. (integrating information on age, performance status, white blood cell count, platelet count, bone marrow blasts, gender, type of AML, cytogenetics and NPM1 and FLT3-ITD status; OR: 1.27; p=0.00083; AUC: 0.70) or the modified molecular version of this score (OR: 1.37; p=0.0027; AUC: 0.63) PS29MRC reached superior predictive accuracy. (Walter et al.; Leukemia 2015) Since we aimed to develop a clinically useful score, we categorized PS29MRC to distinguish between patients who have a high probability of refractory disease and those who are likely to benefit from induction therapy (complete remission or complete remission with incomplete hematologic recovery). By applying the predefined cut off, we were able to reach a specificity of 90% and sensitivity of 46% in the validation set (OR: 7.83; p=6.06x10-9). The accuracy of PS29MRC was 77%. In the multivariable model the categorized classifier was highly significant (OR: 4.45; p=0.00040) and only age and TP53 mutations were left as significant variables again. Within the cytogenetic subgroups favorable (n=14; refractory: n=0; responders: n=13), intermediate (n=189; refractory: n=43; responders: n=136) and adverse (n=49; refractory: n=29; responders: n=15) the classifier showed an accuracy of 100%, 78% and 66%, respectively. Furthermore, the classifier predicted survival and was able to unravel the intermediate MRC subgroup (Figure). Additionally, genes included in our predictive signature seem to be involved in AML pathogenesis and potentially actively contribute to mechanisms responsible for primary therapeutic resistance. For example MIR-155HG, an already known parameter of inferior outcome in AML, contributed significantly to PS29MRC. There are currently ongoing trials with the novel inhibitor Pevonedistat that aim to modulate this target in AML. In summary we were able to develop a predictive risk classifier summarizing 29 gene expression values and the MRC classification that outperformed all currently used methods to predict refractory disease in intensively treated adult AML patients. PS29MRC demonstrates that it is possible to identify patients at risk of treatment failure in AML at diagnosis with high specificity. Figure 1. Kaplan-Meier estimates showing overall survival of AML patients in the validation set according to PS29MRC Figure 1. Kaplan-Meier estimates showing overall survival of AML patients in the validation set according to PS29MRC Figure 2. Figure 2. Disclosures No relevant conflicts of interest to declare.

Abstract: Background: AML is a disease affecting predominantly older patients (pts), but does occur across the entire age spectrum; younger adults [age & lt;60 years (y)] have better outcomes. Using 2 large datasets (from Germany and the US), we sought to identify whether mutational frequencies, cytogenetic aberrations or outcome measures would demonstrate unique patterns to independently assort populations by age. Methods: We analyzed the molecular profiles of 2,823 adult AML pts enrolled onto clinical frontline protocols of 2 large cooperative study groups from 2 continents [US, Cancer and Leukemia Group B (CALGB)/Alliance for Clinical Trials in Oncology (Alliance), n=1743; Germany, AML Cooperative Group [AMLCG], n=1080] between 1986 and 2016. Treatment of all pts included intensive induction therapy, whereas pts enrolled on CALGB/Alliance protocols precluded allogeneic transplantation in 1 st complete remission. Pts in both cohorts were profiled for molecular features via targeted sequencing platforms. Frequencies of mutations genes and selected cytogenetic findings were then calculated in both datasets for the group of pts aged 18-24 y and for older pts by 5-year intervals until the age of 74 y and for pts older than 75 y. We also analyzed survival outcomes of 1,669 AML pts younger than 60 y using the same age intervals up to age 59 y. Results: Our side-by-side analysis shows remarkable congruence of results between German and US pt populations. Selected AML-associated gene mutations (mutation frequency ≥4%) and recurrent cytogenetic abnormalities followed 3 basic distribution patterns across the age spectrum (Fig. 1A): group 1 with increasing frequency with increasing age [ASXL1, BCOR, IDH1/2, RUNX1, SRSF2, TET2, TP53; complex karyotype and cytogenetically normal AML (CN-AML)]; group 2 with decreasing frequency with increasing age (CEBPA, EZH2, FLT3-TKD, GATA2, KIT, KRAS, PTPN11, NRAS, WT1; inv(16), t(8;21) and 11q23/KMT2A rearrangements) and group 3 with non-linear frequency distribution, which included the 3 most common AML-associated gene mutations (NPM1, DNMT3A, FLT3-ITD), SF3B1 and mutations in the cohesin complex genes (RAD21, SMC1A, SMC3, STAG2) (Fig. 1A). Notably, within the first 2 distribution groups, there seem to be no obvious age that could serve as a cut point separating age groups that are markedly different with regard to their molecular patterns. Particularly, this includes an age group that is commonly used for pt cohort definitions such as pts aged 18-39 y referred to as adolescent and young adults (AYA) or even treatment decisions and eligibility (eg, ages 60 or 65 and older for consideration as elderly AML). With respect to pt outcomes, expectedly, there was almost linear shortening of overall survival (OS) as age increased (p & lt;.001; Fig. 1B, Table 1). Within the European LeukemiaNet (ELN) genetic-risk groups, there was also an age-associated shortening in OS rates (Table 1): in favorable and intermediate risk pts the 5y-OS declined over the age range (favorable risk pts; US, p=.002; GER, p & lt;.001; intermediate risk pts, US, p & lt;.001; GER, p=.009, Table 1). Adverse risk pts had less variability in survival outcome across age (US, p=.004; GER, p=.22). Thus, while ELN criteria risk stratifies each age group, age itself is an important qualifier with regards to OS within each ELN group given the wide survival range among the age group. Again, there were no distinct outcome changes at certain age groups, further supporting the consideration of age as a continuum in AML for both biology and risk stratification. Conclusions: To our knowledge, this is the first large scale depiction of mutational patterns in AML inclusive of the entire adult age spectrum. Our international study demonstrates that patterns of individual mutations based on age are remarkably consistent between countries, and defy assortment based on typical age conventions. Given the continuous distribution of either increasing or decreasing frequency of many mutations, there are distinctly different mutational profiles for the youngest pts compared with older pts, however choosing a precise cut-off, such as age 39 for AYA pts or 59 for consideration as "younger AML", does not seem to be supported by our analyses. This observation supports a more personalized approach that also considers molecular subgroups in clinical practice instead of the age rigidity set in many clinical trials. *shared first: M.C.,K.L.; #last: T.H.,AK.E. Figure 1 Figure 1. Disclosures Berdel: Philogen S.p.A.: Consultancy, Current equity holder in publicly-traded company, Honoraria, Membership on an entity's Board of Directors or advisory committees. Hiddemann: F. Hoffmann-La Roche: Membership on an entity's Board of Directors or advisory committees, Research Funding; Janssen: Research Funding. Blachly: KITE: Consultancy, Honoraria; INNATE: Consultancy, Honoraria; AbbVie: Consultancy, Honoraria; AstraZeneca: Consultancy, Honoraria. Mims: Glycomemetics: Research Funding; Daiichi Sankyo: Consultancy, Research Funding; Aptevo: Research Funding; Leukemia and Lymphoma Society's Beat AML clinical study: Consultancy, Research Funding; Xencor: Research Funding; Kartos Pharmaceuticals: Research Funding; Genentech: Consultancy; Abbvie: Consultancy; BMS: Consultancy; Kura Oncology: Consultancy; Syndax Pharmaceuticals: Consultancy; BMS: Consultancy; Jazz Pharmaceuticals: Consultancy; Aptevo: Research Funding. Walker: Karyopharm Therapeutics: Current Employment, Current holder of individual stocks in a privately-held company, Current holder of stock options in a privately-held company. Blum: Celyad Oncology: Research Funding; Forma Therapeutics: Research Funding; Xencor: Research Funding; Nkarta: Research Funding; Leukemia and Lymphoma Society: Research Funding; Abbvie: Honoraria; AmerisourceBergen: Honoraria; Syndax: Honoraria. Larson: Epizyme: Consultancy; Astellas: Consultancy, Research Funding; Gilead: Research Funding; CVS/Caremark: Consultancy; Takeda: Research Funding; Novartis: Research Funding; Rafael Pharmaceuticals: Research Funding; Cellectis: Research Funding. Stone: Syntrix/ACI: Membership on an entity's Board of Directors or advisory committees; Novartis: Consultancy, Research Funding; Astellas: Membership on an entity's Board of Directors or advisory committees; BerGen Bio: Membership on an entity's Board of Directors or advisory committees; Actinium: Membership on an entity's Board of Directors or advisory committees; Elevate Bio: Membership on an entity's Board of Directors or advisory committees; Syndax: Membership on an entity's Board of Directors or advisory committees; Onconova: Consultancy; Jazz: Consultancy; Janssen: Consultancy; Innate: Consultancy; GlaxoSmithKline: Consultancy; Gemoab: Membership on an entity's Board of Directors or advisory committees; Foghorn Therapeutics: Consultancy; Boston Pharmaceuticals: Consultancy; Bristol Myers Squibb: Consultancy; AbbVie: Consultancy; Arog: Consultancy, Research Funding; Aprea: Consultancy; Amgen: Membership on an entity's Board of Directors or advisory committees; Syros: Membership on an entity's Board of Directors or advisory committees; Takeda: Consultancy; Agios: Consultancy, Research Funding; Celgene: Consultancy; Macrogenics: Consultancy. Byrd: Vincerx Pharmaceuticals: Current equity holder in publicly-traded company, Membership on an entity's Board of Directors or advisory committees; Novartis, Trillium, Astellas, AstraZeneca, Pharmacyclics, Syndax: Consultancy, Honoraria; Newave: Membership on an entity's Board of Directors or advisory committees. Metzeler: Jazz Pharmaceuticals: Consultancy; Novartis: Consultancy; Daiichi Sankyo: Honoraria; Astellas: Honoraria; AbbVie: Honoraria; Pfizer: Consultancy; Celgene/BMS: Consultancy, Honoraria, Research Funding. Eisfeld: Karyopharm (spouse): Current Employment.

Abstract: Background: Mutations in the protein tyrosine phosphatase gene PTPN11 (also known as SHP2) are found in approximately 10% of adult patients with acute myeloid leukemia (AML). A recent study reported that mutated PTPN11 associates with inferior response rates and shorter survival among intensively treated AML patients, independently of the ELN prognostic groups (Alfayez et al., Leukemia 2020). Earlier analyses of the genomic landscape of AML did not uncover a similar prognostic relevance of PTPN11 mutations. Therefore, our aim was to clarify the prognostic relevance of mutated PTPN11 variants in AML patients receiving intensive front-line therapy. Patients and Methods: We studied 1116 AML patients enrolled on two subsequent multicenter phase III trials of the German AML Cooperative Group (AML-CG 1999, NCT00266136; and AML-CG 2008, NCT01382147) who were genetically characterized by amplicon-based targeted next-generation sequencing (Herold et al., Leukemia 2020). All patients had received induction chemotherapy containing cytarabine and daunorubicin or mitoxantrone. Results: We identified 146 PTPN11 mutations in 114 of 1116 patients (10%). Mutations clustered in two hotspot regions (5': codons 52-79; n=108 and 3': codons 491-512, n=38) as previously reported. Associations of PTPN11 mutations with baseline clinical and genetic patient characteristics are shown in Figure A. PTPN11 mutations were most frequent in the European LeukemiaNet (ELN) "favorable" genetic risk group, and associated with higher leukocyte counts. Patients with mutated PTPN11more commonly had mutated NPM1, IDH1 and DNMT3A, and less frequently had FLT3-ITD, IDH2 and TP53 mutations, compared to patients with wild-type PTPN11. With regard to treatment outcomes, the rate of complete remission was similar among patients with mutated and wild-type PTPN11 (65% vs. 59%, P=.25). In univariate analyses, PTPN11-mutated patients had significantly longer relapse-free survival (RFS; 5-year estimate, 55% vs 33% for PTPN11-wild type patients; P=.001; Figure B) and tended to have longer overall survival (OS; 5-year estimate, 43% vs 32%; P=.06; Figure C). However, in multivariable models adjusting for age, sex, leukocyte count, AML type (de novo/sAML/tAML) and ELN-2017 genetic risk group, mutated PTPN11 no longer associated with RFS (hazard ratio [HR], 0.89, 95% confidence interval [CI] , 0.63 - 1.27; P=0.53) or OS (HR, 1.03; 95% CI, 0.80 - 1.33; P=.79). Moreover, PTPN11 mutations did not significantly associate with RFS or OS within any of the ELN genetic risk groups. Finally, we detected no significant differences in baseline characteristics or outcomes between patients with PTPN11 mutations affecting the 5' hotspot region (n=82), the 3' hotspot region (n=21), or mutations at both hotspots (n=11). Conclusion: In our cohort of newly diagnosed and intensively treated AML patients, mutations in PTPN11 occurred in 10% and associated with prognostically favorable genetic characteristics such as mutated NPM1 and absence of FLT3-ITD and TP53mutations. Consequently, PTPN11 mutations were most commonly found within the ELN-2017 favorable risk category. While patients with PTPN11 mutations had relatively favorable survival outcomes, multivariable models suggest this observation is confounded by the frequent co-occurrence of known favorable genetic markers. Our data are in disagreement with a recently published study on 880 newly diagnosed patients that found an unfavourable prognostic impact of mutated PTPN11, particularly among the 410 patients who received intensive treatment. Possible explanations for these discrepant results include differences in treatment regimens between the two cohorts, as well as the play of chance when studying a relatively rare gene mutation in medium-sized cohorts. In summary, our data do not support a role of PTPN11 mutations as an adverse prognostic biomarker in newly diagnosed, intensively treated adult AML patients. Figure Disclosures Metzeler: Daiichi Sankyo: Honoraria; Otsuka Pharma: Consultancy; Pfizer: Consultancy; Celgene: Consultancy, Honoraria, Research Funding; Novartis: Consultancy; Jazz Pharmaceuticals: Consultancy; Astellas: Honoraria. Subklewe:AMGEN: Consultancy, Honoraria, Research Funding; Celgene: Consultancy, Honoraria; Pfizer: Consultancy, Honoraria; Novartis: Consultancy, Research Funding; Janssen: Consultancy; Morphosys: Research Funding; Seattle Genetics: Research Funding; Roche AG: Consultancy, Research Funding; Gilead Sciences: Consultancy, Honoraria, Research Funding.

Abstract: Achievement of complete remission signifies a crucial milestone in the therapy of acute myeloid leukemia (AML) while refractory disease is associated with dismal outcomes. Hence, accurately identifying patients at risk is essential to tailor treatment concepts individually to disease biology. We used nine machine learning (ML) models to predict complete remission and 2-year overall survival in a large multicenter cohort of 1,383 AML patients who received intensive induction therapy. Clinical, laboratory, cytogenetic and molecular genetic data were incorporated and our results were validated on an external multicenter cohort. Our ML models autonomously selected predictive features including established markers of favorable or adverse risk as well as identifying markers of so-far controversial relevance. De novo AML, extramedullary AML, double-mutated CEBPA, mutations of CEBPA-bZIP, NPM1, FLT3-ITD, ASXL1, RUNX1, SF3B1, IKZF1, TP53, and U2AF1, t(8;21), inv(16)/t(16;16), del(5)/del(5q), del(17)/del(17p), normal or complex karyotypes, age and hemoglobin concentration at initial diagnosis were statistically significant markers predictive of complete remission, while t(8;21), del(5)/del(5q), inv(16)/t(16;16), del(17)/del(17p), double-mutated CEBPA, CEBPA-bZIP, NPM1, FLT3-ITD, DNMT3A, SF3B1, U2AF1, and TP53 mutations, age, white blood cell count, peripheral blast count, serum lactate dehydrogenase level and hemoglobin concentration at initial diagnosis as well as extramedullary manifestations were predictive for 2-year overall survival. For prediction of complete remission and 2-year overall survival areas under the receiver operating characteristic curves ranged between 0.77–0.86 and between 0.63–0.74, respectively in our test set, and between 0.71–0.80 and 0.65–0.75 in the external validation cohort. We demonstrated the feasibility of ML for risk stratification in AML as a model disease for hematologic neoplasms, using a scalable and reusable ML framework. Our study illustrates the clinical applicability of ML as a decision support system in hematology.