Computational alternative to cardiac biopsy in multiple myeloma: utilizing the shape of routine laboratory data for the classification of AL cardiac amyloidosis

SVM models using longitudinal EMR “DescFits” features achieved an AUC of 1.0 for identifying AL cardiac amyloidosis in a retrospective cohort at Moffit Cancer Center (2014–2019).

• Why it matters: Cardiac amyloidosis in multiple myeloma is typically diagnosed with biopsy and imaging, which are challenging; a computational alternative could reduce reliance on these tests.
• What they did: The study trained MISTIC support vector machine classifiers on EMR data from patients with multiple myeloma or MGUS, using descriptive statistics, regression-fit parameters, or both (DescFits).
• The result: DescFits (AUC=1.0) substantially outperformed the EchoAdj timepoint model (AUC=0.36) and other conventional approaches, selecting features tied to temporal hematologic, cardiac, and renal function.