Beyond the Routine CBC: Research CBC Parameters Associated with Myelodysplastic Syndromes and Underlying Mutations

Background: Myeloid neoplasms and reactive conditions cannot be readily distinguished by conventional CBC parameters. Given the time, expense and clinical expertise associated with the diagnostic workup of myeloid neoplasms, there is a clear need for a cost-effective screening laboratory test that can rapidly and accurately select for patients with cytopenias or cytoses potentially related to myelodysplastic syndrome (MDS) or myeloproliferative neoplasms (MPN), respectively. Recent studies showed that incorporating NE-WX, a research CBC parameter related to neutrophil complexity, and IPF, an immature platelet fraction parameter, into a conventional CBC analysis helps identify patients with high likelihood of MDS (PMID 30406952) or JAK2-positive polycythemia vera (PMID 30601597), respectively. Our exploratory analysis was focused on using all research CBC parameters in combination with standard CBC parameters and mutational data to compare control patients to those with active myeloid disease, and MDS in particular. Methods: We measured standard and research CBC parameters (92 parameters overall) using the Sysmex XN series hematology analyzer in a cohort of 592 consecutive patients (pts), out of which 221 had abnormal CBC due to an underlying myeloid neoplasm (MDS 83 pts, MPN 54 pts, MDS/MPN 24 pts, acute myeloid leukemia 21 pts, clonal or non-clonal cytopenia or cytoses 33 and 6 pts, respectively) and 371 age-matched control patients with normal CBC. The former group was tested for the presence of somatic mutations in 95 genes associated with hematologic malignancies and included patients with active disease (n=127) and in remission (n=94). The MDS cohort included 59 patients with active disease and 24 patients in remission. We developed a variety of models (Linear and Logistic Regression, Decision Trees, Random Forests, Neural Networks, Support Vector Machines, Gradient Boosting) to predict the presence of an active myeloid neoplasm or active MDS using both CBC and molecular data, as well as CBC data alone. These models were compared using a leave one patient out cross-validation method and a champion model was chosen. Correlation matrix analysis was used for establishing relationships between MDS-associated mutations and CBC parameters, where a relevant association was considered at the levels of above 0.25 or below -0.25. Results: Table 1 lists the leave one out performance of the champion predictive models for the active myeloid and MDS cohorts based o