A new machine learning model could help doctors spot an associated blood cancer that may be masked by advanced systemic mastocytosis (SM), according to data presented at the European Hematology Association (EHA) 2026 Congress, held June 11-14, 2026, in Stockholm, Sweden.
The tool was designed to address a common diagnostic problem: When a patient has both SM and an associated hematologic neoplasm (AHN), the AHN can change their treatment and prognosis, but it may be hard to detect the AHN when features of SM dominate a patient’s clinical picture.
Read more about the SM with an associated hematological neoplasm (SM-AHN) subtype
The National Organization for Rare Disorders estimates that one in five patients with SM has the SM-AHN subtype. Still, there is no single test for SM-AHN. Doctors instead have to piece together blood counts, bone marrow biopsyBone marrow biopsy A procedure to collect a sample of bone marrow using a needle. Often used to examine the characteristics of mast cells and diagnose SM. results and genetic data to confirm the diagnosis — a process that can be lengthy and involved for patients.
Using baseline information from 374 patients, the researchers built several versions of the model to reflect how much information doctors might have available in practice. The model was trained to look for patterns among patients who were already classified as having or not having AHN. It then used those patterns (including blood test results, clinical features, bone marrow findings and genetic data) to estimate which patients were more likely to have SM-AHN.
The full model used the broadest set of inputs. But even a simpler version based only on peripheral blood laboratory results correctly identified 96.6% of patients in the test cohort who had AHN. Across all versions, the figure ranged from 95.4% to 97.7%.
One notable finding involved patients who had not been diagnosed with SM-AHN. The full model predicted the neoplasm in 28 patients who were clinically classified as AHN-negative. Although the model’s prediction did not amount to a confirmed diagnosis, 21 of those patients had at least one finding that could point to a possible obscured AHN.
The abstract suggests the model was good at catching AHN cases, but does not show how often it might mistakenly flag patients who do not have AHN.
The results point to a possible role for machine learning as a triage tool in advanced SMAdvanced SM In these subtypes of SM, mast cells begin to damage organs. Advanced SM includes the subtypes aggressive SM, mast cell leukemia and SM with an associated hematological neoplasm., helping clinicians identify patients who may warrant a closer search for AHN. For now, though, the findings are preliminary.
Sign up here to get the latest news, perspectives, and information about SM sent directly to your inbox. Registration is free and only takes a minute.
