Kidney Week

Abstract: PUB084

Cardiovascular Risk Stratification Using Predicted Coronary Artery Calcium Score from a Machine-Learning Model

Session Information

• Publication Only

Category: Augmented Intelligence, Digital Health, and Data Science

• 300 Augmented Intelligence, Digital Health, and Data Science

Authors

• Takkavatakarn, Kullaya, Icahn School of Medicine at Mount Sinai, New York, New York, United States

Kullaya Takkavatakarn, MD, PhD

• Gownivaripally, Pooja Anand, Icahn School of Medicine at Mount Sinai, New York, New York, United States

Pooja Anand Gownivaripally, MS

• Nadkarni, Girish N., Icahn School of Medicine at Mount Sinai, New York, New York, United States

Girish N. Nadkarni, MD, MPH

• Chan, Lili, Icahn School of Medicine at Mount Sinai, New York, New York, United States

Lili Chan, MD, MS

Background

Cardiovascular disease (CVD) is a common complication in patients with chronic kidney disease (CKD). Coronary artery calcium (CAC) score is a clinically validated marker of CVD risk but not readily accessible for many patients. We aimed to develop and validate machine learning (ML) models to predict CAC groups and assessed the use for risk stratification of CVD events.

Methods

We obtained electronic health record (EHR) data from patients with non-dialysis CKD stages 3-5, defined by an eGFR of less than 60 mL/min/1.73 m2 for at least 3 months, and without a history of CVD at the Mount Sinai Health System who underwent a CAC scan. CAC scores were categorized into 4 groups (0, 1-100, 101-400, and > 400). We divided the cohort into a training and test dataset (70/30) then developed and validated four models, decision tree (DT), random forest (RF), XGBoost, and deep neural network (DNN), to predict CAC score groups. Area under the receiver operating characteristic curve (AUROC) was used to evaluate model performances. We utilized Cox regression to assess the association between CAC group and the best-performing ML model-predicted CAC group with CVD events within 5 years. Risk discrimination was measured using C-statistics.

Results

We included 648 patients; the median age was 70 (IQR 63-77) and the median CAC score was 64 (IQR 0-306). XGBoost yielded the best AUROC (0.72; 95%CI 0.63 to 0.79), followed by RF (0.71; 95%CI 0.63 to 0.78), DNN (0.64; 95%CI 0.53 to 0.71), and DT (0.61; 95%CI 0.52 to 0.68). 210 (32%) had CVD events during the 5-year follow-up. CT scan-measured CAC and the XGBoost-predicted CAC risk stratification were significantly associated with CVD events (Figure1). The C-statistics were 0.67 and 0.66 respectively.

Conclusion

ML-derived CAC risk stratification is comparable to CT scan-measured CAC risk stratification in predicting CVD events. Implementing ML techniques on EHR data can serve as an alternative approach for assessing CAC and improving CVD risk prediction, particularly in settings with limited resources.