Haem.io

The Complexity Crisis in Genomic Diagnosis

With genomic-era diagnosis, myeloid disease classification is no longer a small checklist. It is a high-dimensional state space of genetic findings, cytogenetics, blast patterns, prior-treatment context, and clinical qualifiers, all evaluated through increasingly complex guideline logic. In practice, this can produce hundreds of thousands of possible pathway combinations before arriving at a final subtype label.

More broadly, this is becoming a system-wide challenge across oncology as diagnostic practice shifts from morphology-first classification toward genomics-first classification.

Expecting clinicians to execute that full combinatorial logic by hand, repeatedly and consistently, is not realistic. We need algorithmic execution for reliability, and formal verification layers so those algorithms can be checked for drift, contradiction, and internal coherence as they run in the real world. Diagnosticians in the west — and increasingly in LMICs — are expected to do this hundreds of times per year, and these diagnoses are clinically critical. It can be the difference between being put on a chemotherapy regimen or not.