|The following article features coverage from the American Society of Hematology 2020 meeting. Click here to read more of Hematology Advisor‘s conference coverage.|
A recently developed and validated differential diagnosis model has been reported to accurately distinguish myelodysplastic syndromes (MDS) from other myeloid malignancies using clinical and mutational data alone, without bone marrow biopsy data, according to study results presented by Nathan Radakovich, BA, a medical student at Case Western Reserve University in Cleveland, Ohio, at the virtual 62nd American Society of Hematology (ASH) Annual Meeting and Exposition.
“Despite having varied prognoses and treatments, myeloid malignancies have many phenotypic similarities, which may complicate their differential diagnosis. This is further complicated by interobserver variability in tissue diagnosis, as well as the fact that tissue is not always readily available,” said Mr Radakovich, “Because of its ability to handle complex variables, machine learning is an appealing tool to handle this kind of challenge.”
Using next-generation sequencing, 3 institutions collected clinical and genomic data from 2697 patients with diagnoses of MDS, chronic myelomonocytic leukemia, MDS/myeloproliferative neoplasm overlap (MDS/MPN), myeloproliferative neoplasm (MPN; either polycythemia vera, essential thrombocythemia, or myelofibrosis), clonal cytopenia of undetermined significance (CCUS), or idiopathic cytopenia of undetermined significance (ICUS); 24 commonly mutated genes were shared across all cohorts. All diagnoses were confirmed with bone marrow aspiration (World Health Organization 2017 criteria).
Data from 2 of the 3 institutions (1188 patients) were used to train the machine learning algorithm of the model, while data from the third institution (1509 patients) were used to independently validate the final model, which was intentionally designed to be used without the need for bone marrow biopsy in practice.
Median age of patients in the entire cohort was 70 years (range, 36-86), and median hemoglobin level was 10.4 g/dL (range, 6.9-15.7). As expected, compared with all other disease states, patients with ICUS were generally younger, patients with MDS/MPN overlap syndromes and MPNs had elevated white blood cell (WBC) count and absolute neutrophil count, patients with MPN and CCUS had milder anemia, patients with MPNs had elevated platelets, and patients with MDS and MDS/MPN overlap syndromes had higher percentages of peripheral blood blasts (all P <.001).
Among all patients, the 5 most commonly mutated genes were SF3B1 (27%), TET2 (25%), ASXL1 (19%), SRSF2 (16%), and DNMT3A (11%); these were also the most commonly mutated genes in patients with MDS. The mutations for each subtype were consistent with previous reports.
Listed in order of importance, the most important features for model predictions were the number of mutations detected per sample, percentage of blasts in peripheral blood, absolute monocyte count, JAK2 status, hemoglobin level, basophil count, age, eosinophil count, absolute lymphocyte count, WBC count, EZH2 mutation status, absolute neutrophil count, mutation status of KRAS and SF3B1, platelets, and sex.
The final model yielded an average area under the receiver operating curve (AUROC) of 0.95 (95% CI, 0.93-0.96) for the test cohort and 0.93 (95% CI, 0.91-0.94) for the validation cohort. The model is able to provide personalized predictions for top differential diagnoses accompanied by explanations of how features influence the potential diagnosis.
“A tool such as this may be useful for clinicians when faced with ambiguous diagnoses or in cases where the confirmation of a suspected diagnosis is desired,” Mr Radakovich noted. “Importantly, the predictions generated by the model are explainable, which allows them to be weighed against other evidence and, therefore, to be integrated into existing workflows used by practitioners.”
Disclosure: Some authors have declared affiliations with or received funding from the pharmaceutical industry. Please refer to the original study for a full list of disclosures.
Read more of Hematology Advisor’s coverage of the ASH 2020 meeting by visiting the conference page.
Radakovich N, Meggendorfer M, Malcovati L, et al. A personalized clinical-decision tool to improve the diagnostic accuracy of myelodysplastic syndromes. Presented at: American Society of Hematology (ASH) 62nd Annual Meeting and Exposition; December 5-8, 2020. Abstract 541.