A Tool to Improve the Diagnostic Accuracy of Myelodysplastic Syndromes

Histological and cytomorphological examinations are vital to diagnosis of myelodysplastic syndromes (MDS). But accurate diagnosis can still be challenging and variable, especially in patients with pancytopenia or in cases when the diagnostician has limited experience.

In an effort to improve diagnostic accuracy, a group of researchers has developed and validated a model that uses mutational data, peripheral blood counts, and clinical variables to differentiate MDS from other myeloid malignancies. They presented their work at the 62nd ASH Annual Meeting & Exposition.

The team collected data on 652 patients treated at Cleveland Clinic in Ohio, 1,509 patients from at Munich Leukemia Laboratory in Germany, and 536 patients from the University of Pavia in Italy.

All of the patients had been diagnosed with one of the following based on bone marrow aspiration according to World Health Organization 2017 criteria:

• MDS
• Chronic myelomonocytic leukemia (CMML)
• MDS/myeloproliferative neoplasm (MPN) overlap (MDS-MPN)
• MPN
• Clonal cytopenia of undetermined significance (CCUS)
• Idiopathic cytopenia of undetermined significance (ICUS)

The study collected clinical and genomic data, including commercially available next-generation sequencing panels. Clinical characteristics included:

• Median age was 70 years
• Median hemoglobin was 10.4g/dL
• Median platelet count was 132 k/dL
• Median white blood cell count was 5.3 k/dL
• Median absolute neutrophil count was 2.8 k/dL
• Median monocyte count was 0.3 k/dL
• Median lymphocyte count was 1.1 k/dL
• Median peripheral blast percentage was 0%

Then, the team used a machine learning algorithm to build a model based on the genomic and clinical variables. They trained and tested the model on the Cleveland Clinic and University of Pavia patients, then independently validated the final model with the Munich cohort.

They found that the most commonly mutated genes in all patients and in the patients with MDS were SF3B1 (27%), TET2 (25%), ASXL1 (19%), SRSF2 (16%), and DNMT3A (11%). In CMML and MDS-MPN, the most commonly mutated genes were TET2 (46%), ASXL1 (34%), SRSF2 (29%), RUNX1 (13%), and CBL (12%). In patients with MPN, the study identified JAK2 (64%), ASXL1 (27%), TET2 (14%), DNMT3A (8%), and U2AF1 (7%). Finally, among patients with CCUS, the most commonly mutated genes were TET2 (41%), DNMT3A (27%), ASXL1 (19%), (SRSF2 17%), and ZRSR2 (10%). The results on ICUS were not reported.

The most important features for model predictions, in order of importance, were:

• Number of mutations detected
• Peripheral blast percentage
• Absolute monocyte count
• JAK2status
• Hemoglobin
• Basophil count
• Age
• Eosinophil count
• Absolute lymphocyte count
• White blood cell count
• EZH2mutation status
• Absolute neutrophil count
• Mutation status of KRAS and SF3B1
• Platelets
• Gender

“The model can provide personalized interpretations of its outcome and can aid physicians and hematopathologists in recognizing MDS with high accuracy when encountering patients with pancytopenia and with a suspected diagnosis of MDS,” said lead presenter Nathan Radakovich, BA, of the Cleveland Clinic Lerner College of Medicine of Case Western Reserve University in Ohio. “The model also provides individual-level explanations for predictions, providing top differential diagnoses and individual-level explanations of how features influence a putative diagnosis.”