Performance Evaluation of A Machine Learning Model For Systematic Identification of Wild-type Transthyretin Amyloid Cardiomyopathy At Two Academic Medical Centers

Abstract

Introduction Wild-type transthyretin amyloid cardiomyopathy (wtATTR-CM) is a progressive, life-threatening disease that is a recognized but underdiagnosed cause of heart failure (HF). A previously developed and validated machine learning (ML) model trained on US medical claims data delivered a robust performance predicting wtATTR-CM in HF patients with sensitivity/specificity/accuracy of 90/79/84% and an ROC AUC of 0.95.1 To observe the model performance in real-world datasets, the ML algorithm was tested at two academic medical centers—Oregon Health & Science University (OHSU) and New York University (NYU). Methods A retrospective, case-control study was conducted using electronic health records from patients with wtATTR-CM (cases) and non-amyloid HF (controls [random sample]; 1:1) at OHSU (Jul 2005-Nov 2019) and NYU (Oct 2015-Jan 2020). Inclusion criteria were age ≥50 years; HF diagnosis (based on ICD-10 codes/SNOMED CT); and ≥1 of the following: ≥12 months of medical history, ≥5 overall clinical visits, or ≥10 documented diagnosis codes. Results Of 25,233 and 25,174 patients meeting study criteria in the OHSU and NYU datasets, respectively, 41 (0.2%) and 27 (0.1%) patients had wtATTR-CM. In both datasets, the model performed well in predicting wtATTR-CM HF vs non-amyloid HF (Table), with sensitivity/specificity/accuracy of 83/81/82% and ROC AUC 0.91 (OHSU) and 93/74/83% and 0.95 (NYU). Conclusions Using retrospective data from OHSU and NYU, we confirmed the validated ML model for predicting wtATTR-CM.1 Consistency in model performance suggests that it may be an effective tool for screening patients at-risk for wtATTR-CM in the clinical setting. Reference: 1. Huda A, et al. Presented at 2019 HFSA 23rd Annual Scientific Meeting. (Available at: https://www.eventscribe.com/2019/HFSA/fsPopup.asp?efp=Tk9DWk9ZUFUxMDAyNA&PosterID=235694&rnd=0.3349664&mode=posterinfo ).