Background:

Real-world data (RWD) provide essential insights into diagnostic and treatment patterns for myeloid malignancies. Comprehensive genomic profiling (CGP), like the Neo Comprehensive Myeloid Disorders panel, is crucial for diagnosis, risk assessment, and personalized treatment of myeloid malignancies such as acute myeloid leukemia (AML), myelodysplastic neoplasms (MDS), myeloproliferative neoplasms (MPN), and Myeloid/Lymphoid Neoplasms with Eosinophilia and tyrosine kinase gene fusion. Understanding the real-world impact of CGP as a diagnostic tool across myeloid diseases provides insights into critical genomic variants that may be missed by targeted panels.

Methods:

Molecular profiles of 533 patients were analyzed by integrating Neogenomics sequencing data (03/2024 – 07/2025) with clinical data aggregated from national health information exchanges and processed via the xCures platform. Cohort 1 was a random cohort of 477 patients with confirmed diagnoses of AML, MDS, and MPN. Cohort 2 was a selected group of 56 patients whose Neo Comprehensive Myeloid Disorders or Heme Cancers panels revealed rare fusion events. Data were harmonized from a median of 156 medical records per patient from a median of 24 distinct locations per patient. Structured data from electronic medical records (EMRs) were supplemented large language models (LLMs) trained to perform natural language processing (NLP) and extraction, which standardized information from pathology reports, progress notes, and imaging records. Diagnosis dates, prior disease history, and treatment patterns were derived from this enhanced clinicogenomic dataset.

Results:

In cohort 1 (n=377 patients), 357 (75%) had AML, 278 (58%) had MDS, and 46 (10%) had MPN. Forty AML patients (11%) had a prior MDS diagnosis, with a median interval from MDS to AML transformation of 320 days (IQR, 105–710). In AML, the most common first-line therapy involved cytotoxic chemotherapy in combination with targeted therapy (n=132), followed by cytotoxic agents alone (n=48). The median time from AML diagnosis to treatment initiation was 7 days (IQR, 2–36). 288 patients (60%) received Neo Comprehensive Heme or Myeloid panels (resulted Sept 2024–June 2025). The most common alterations were missense SNVs (n=257, 87%), frameshifts (n=121, 42%), stop gained (n=77, 27%), splice donor events (n=20, 7%), and inframe insertions, deletions, and fusions (n=14 each, 5%). IDH2 mutations were observed in 21 patients (7%), while pathogenic alterations in the DNA damage response pathway (BRCA1/2, ATM, CHEK2) were noted in 16 patients (5.5%), at a higher frequency than some known targets such as FLT3 (n=14) and IDH1 (n=8). Median time from diagnosis to testing was 310 days (MDS) and 373 days (AML). 95 tests (33%) were resulted within 90 days of diagnosis, and 17 (6%) before any formal diagnosis had been made. Three patients initially diagnosed with MDS were reclassified as AML shortly after testing. Among patients tested early in their diagnoses (within 90 days), 32 samples (34%) revealed pathogenic alterations across 39 genes. In cohort 2 (n=56), 34 patients (61%) harbored pathogenic or likely pathogenic fusions involving PDGFRA, PDGFRB, FGFR1, JAK2, FLT3, ETV6, ALK, or NTRK3. Six patients (18%) received therapies targeting these pathogenic fusions. Twenty-one patients (37.5%) presented with cytopenia, eosinophilia, or an unspecified diagnosis, but within days after CGP, EMR data confirmed diagnoses were made for 13 of these patients (62%).

Conclusions:

CGP identified pathogenic alterations in one-third of patients tested early in their disease course and led to diagnostic reclassification in some cases. The identification of rare fusion events by CGP was instrumental in providing diagnostic insights. These findings underscore CGP's critical role in refining diagnoses and enabling timely, individualized treatment decisions in myeloid malignancies, particularly when integrated early in the care trajectory. Future analysis could be extended to include tens of thousands of myeloid patients tested for critical genomic variants using CGP, which may be missed by targeted, disease-specific NGS panels.

*Author 1 and Author 2 contributed equally to this work

This content is only available as a PDF.
2025
Sign in via your Institution