Kidney Week

Abstract: SA-PO403

Proteomics-Based Machine-Learning Approach for Predicting Cardiac Dysfunction in Patients on Hemodialysis

Session Information

• Hemodialysis and Frequent Dialysis - 2
  October 26, 2024 | Location: Exhibit Hall, Convention Center
  Abstract Time: 10:00 AM - 12:00 PM

Category: Dialysis

• 801 Dialysis: Hemodialysis and Frequent Dialysis

Authors

• Wu, Ping-Hsun, Kaohsiung Medical University Chung Ho Memorial Hospital, Kaohsiung, Taiwan

Ping-Hsun Wu, MD, MS

• Lin, Yi-Ting, Kaohsiung Medical University Chung Ho Memorial Hospital, Kaohsiung, Taiwan

Yi-Ting Lin, MD, PhD

• Chiu, Yi-Wen, Kaohsiung Medical University Chung Ho Memorial Hospital, Kaohsiung, Taiwan

Yi-Wen Chiu, MD

• Kuo, Mei-Chuan, Kaohsiung Medical University Chung Ho Memorial Hospital, Kaohsiung, Taiwan

Mei-Chuan Kuo, MD

• Hwang, Shang-Jyh, Kaohsiung Medical University Chung Ho Memorial Hospital, Kaohsiung, Taiwan

Shang-Jyh Hwang, MD

Background

Evaluating cardiac function is crucial for hemodialysis patients due to its association with cardiovascular mortality. Evaluating cardiac function in hemodialysis patients is crucial, and blood-based biomarker testing represents a potentially convenient and insightful approach for assessing cardiac function in this population. This study aims to explore using cardiovascular proteomics and machine learning (ML) to predict cardiac dysfunction in hemodialysis patients.

Methods

The study enrolled 347 hemodialysis patients who underwent cardiac ultrasonography to assess cardiac dysfunction, which was defined as an ejection fraction < 50% (primary analysis) or < 45% (sensitivity analysis). The proteomic analysis measured 184 proteins using proximity extension assays. ML techniques (classification and regression tree [CART], Least Absolute Shrinkage and Selection Operator [LASSO], random forest, Ranger, eXtreme Gradient Boosting [XgBoost]) were applied to develop predictive models using the proteomic and clinical data. Model performance was evaluated by area under the curve (AUC). The Significance of the Hierarchical Averaging of Shapley Values (SHAP) values identified key predictive features.

Results

The proteomic biomarkers outperformed routine clinical/laboratory variables in predicting cardiac dysfunction across ML models. LASSO and XgBoost models with feature selection highlighted N terminal pro B type natriuretic peptide (NT-ProBNP) as the top predictor, followed by Chitotriosidase 1 (CHIT1), Angiotensin-Converting Enzyme 2 (ACE2), and Matrix Metalloproteinase-2 (MMP-2). SHAP analysis confirmed these findings.

Conclusion

Cardiovascular proteomics combined with ML enables superior prediction of cardiac dysfunction compared to clinical variables alone in hemodialysis patients. The NT-proBNP and CHIT1 emerged as important protein biomarkers, potentially facilitating early interventions for preventing cardiovascular complications.