Abstract: SA-OR37
Prediction of Atrial Fibrillation Using Clinical and Cardiac Biomarker Data: The CRIC Study
Session Information
- Hypertension and Vascular Disease: From the Lab to Trials
October 24, 2020 | Location: Simulive
Abstract Time: 05:00 PM - 07:00 PM
Category: Hypertension and CVD
- 1401 Hypertension and CVD: Epidemiology, Risk Factors, and Prevention
Authors
- Zelnick, Leila R., University of Washington Department of Medicine, Seattle, Washington, United States
- Shlipak, Michael, University of California San Francisco, San Francisco, California, United States
- Soliman, Elsayed Z., Wake Forest University, Winston-Salem, North Carolina, United States
- Anderson, Amanda Hyre, Tulane University, New Orleans, Louisiana, United States
- Christenson, Robert, University of Maryland School of Medicine, Baltimore, Maryland, United States
- Lash, James P., University of Illinois at Chicago, Chicago, Illinois, United States
- Deo, Rajat, University of Pennsylvania, Philadelphia, Pennsylvania, United States
- Rao, Panduranga S., University of Michigan, Ann Arbor, Michigan, United States
- Afshinnia, Farsad, University of Michigan, Ann Arbor, Michigan, United States
- Chen, Jing, Tulane University, New Orleans, Louisiana, United States
- He, Jiang, Tulane University, New Orleans, Louisiana, United States
- Seliger, Stephen L., University of Maryland School of Medicine, Baltimore, Maryland, United States
- Townsend, Raymond R., University of Pennsylvania, Philadelphia, Pennsylvania, United States
- Cohen, Debbie L., University of Pennsylvania, Philadelphia, Pennsylvania, United States
- Go, Alan S., Kaiser Permanente Northern California, Oakland, California, United States
- Bansal, Nisha, University of Washington Department of Medicine, Seattle, Washington, United States
Background
Clinically available biomarkers of myocardial injury (high sensitivity troponin T, hsTnT) and hemodynamic stress (N-terminal brain natriuretic peptide, NT-proBNP) are strongly associated with atrial fibrillation (AF) in chronic kidney disease (CKD), and have been included in AF prediction models in community-based populations. We investigated the incremental prognostic value of NT-proBNP and hsTnT for AF prediction compared to standard clinical variables in CKD patients enrolled in the Chronic Renal Insufficiency Cohort (CRIC) using machine learning methods.
Methods
Among 2690 CRIC participants without prior AF with complete cardiac biomarker, demographics, medical history/lifestyle, medications, physical characteristics, and laboratory data, we evaluated the utility of Cox regression, LASSO, ridge regression, elastic net, and boosting methods, as well as a previously validated clinical prediction model (CHARGE-AF, using both original and re-estimated coefficients) to predict incident AF. Discriminatory ability of each model was assessed using 10-fold cross-validation; calibration was evaluated graphically.
Results
Mean (SD) age of participants was 57 (11) years, 55% men, 38% black, and mean (SD) eGFR 45 (15) mL/min/1.73m2; 251 incident AF events occurred during 7.3 (SD 2.8) years of follow-up. CHARGE-AF prediction equations using original and re-estimated coefficients each had a cross-validated C-index of 0.69 (Figure 1a). A LASSO model using only clinical data had a C-index of 0.69, while adding NT-proBNP, hsTnT, or both biomarkers improved the C-index to 0.75, 0.73, and 0.76, respectively (p for difference compared to clinical only model <0.0001 for all). Calibration of top biomarker models was generally adequate (Figure 1b).
Conclusion
Cardiac biomarkers NT-proBNP and hsTnT can improve AF prediction in CKD, particularly when paired with machine learning algorithms.
Funding
- NIDDK Support