EHR-based machine learning model predicts cardiac amyloidosis on bone scintigraphy, outperforming clinical routine

Amylo-Detect, a machine learning model using 50 EHR variables, predicted cardiac amyloidosis on bone scintigraphy with an AUC of 0.91 in external validation, outperforming clinical routine and detecting 29% of missed cases.

• Of 42 patients with cardiac amyloidosis missed by clinical routine, Amylo-Detect additionally identified 12 (29%).
• Performance was consistent across subgroups even when crucial predictors were missing.
• This retrospective study requires prospective evaluation before clinical use.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.