Identification of Previously Unrecognized Multiple Myeloma Risk Subgroups with a Novel Biological Disease Stratifier

• Published in: Blood 2021-11, Vol.138 (Supplement 1), p.4718-4718
• Main Authors: M. Shariatpanahi, Afsaneh, Grasedieck, Sarah, Jarvis, Matthew C., Borzooee, Faezeh, Harris, Reuben S., Larijani, Mani, Song, Kevin, Rouhi, Arefeh, Kuchenbauer, Florian
• Format: Article
• Language: English
• ISSN: 0006-4971; 1528-0020
• DOI: 10.1182/blood-2021-148718

Background: The prognosis of MM is determined by affected organs, tumor burden as measured by e.g., the international staging system (ISS), disease biology such as cytogenetic abnormalities, and response to therapy. The outcome of high-risk MM patients classified by ISS or adverse risk cytogenetics is not uniform and patients show heterogeneous survival. Recent insights into the pathogenesis of MM highlighted genome/transcriptome editing as well as inflammation as drivers for the onset and progression of MM. We hypothesized that inclusion of molecular features into risk stratification could potentially resolve the challenge of accurately distinguishing between high-risk and low-risk MM patients at initial diagnosis and improve outcome. Aim: We aimed to create a simple molecular risk score to identify unrecognized patient subgroups, who have been previously misclassified by current risk stratifiers. Method: The Multiple Myeloma Research Foundation CoMMpass study genomics dataset, combining mRNA Seq and clinical data from more than 700 MM patients, allowed us to evaluate the prognostic value of demographic and clinical parameters, cytogenetics, and gene expression levels of APOBEC genes as well as inflammation-modulating cytokines in MM patients. We calculated hazard ratios and Kaplan-Meier survival estimates for all extracted features. Combining clinical variables that were significantly associated with PFS and OS, we then applied machine learning approaches to identify the most accurate classification model to define a new risk score that is easy to compute and able to stratify NDMM patients more accurately than cytogenetics-based classifiers. Based on a Kaplan-Meier survival curve analysis, we then evaluated the performance of our newly built EI score in sub-classifying of current multiple myeloma risk stratifiers. Results: Based on machine learning models, we defined a weighted OS/PFS risk score (Editor-Inflammation (EI) score) based on mRNA expression of APOBEC2, APOBEC3B, IL11, TGFB1, TGFB3, as well as ß2-microglobulin and LDH serum levels. We showed that the EI score subclassified patients into high-risk, intermediate-risk, and low-risk prognostic groups and demonstrated superior performance (C-index: 0.76) compared to ISS (C-index: 0.66) and R-ISS (C-index: 0.64). We further showed that EI low-risk patients do not benefit from autograft and maintenance therapy. Re-classification of ISS (Figure 1a, b, c) and R-ISS risk groups further confirmed the sup