Abstract: TH-PO0737
Predicting Bicarbonate Dynamics During CKRT: Model Validation and Detection of Complications in a US Cohort
Session Information
- AKI: Critical Care, CKRT, Cardiorenal Syndromes, and Clinical Outcomes
October 22, 2026 | Location: Exhibit Hall A, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Acute Kidney Injury
- 102 AKI: Clinical, Outcomes, and Trials
Authors
- Casper, Sabrina, Renal Research Institute, New York, New York, United States
- Fuertinger, Doris H., Renal Research Institute, New York, New York, United States
- Pohlmeier, Robert, Fresenius Medical Care Deutschland GmbH, Bad Homburg, HE, Germany
- Kneppel, Angelika, Fresenius Medical Care Deutschland GmbH, Bad Homburg, HE, Germany
- Derlet-Savoia, Anja, Fresenius Medical Care Deutschland GmbH, Bad Homburg, HE, Germany
- Thompson, David, Fresenius Medical Care Holdings Inc, Waltham, Massachusetts, United States
Background
Regional citrate anticoagulation (RCA) during continuous kidney replacement therapy (CKRT) affects acid-base balance and electrolyte dynamics. We refined a previously published model (Mueller et al., 2025) to predict bicarbonate (HCO3) trajectories and evaluated its performance in a U.S. cohort. We examined whether deviations between model predictions and clinical data could identify complications, such as metabolic disturbances or filter clogging.
Methods
We conducted a retrospective analysis of the SARCA study (NCT05339139). Patients with missing or out-of-range inputs were excluded, and analyses were truncated at the first invalid time point. Model inputs included demographics, CKRT prescription parameters (time-varying fluid flows, citrate and calcium dosing), sodium bicarbonate infusions, hematocrit, and initial HCO3 levels. Simulations were restarted with each filter change. Model performance was assessed using mean absolute error (MAE), with deviations analyzed in relation to clinical events and metabolic markers.
Results
Twenty-seven patients (20 CVVHD, 7 CVVHDF; age 66±12 years, 48% female) were analyzed (median CKRT duration 73 h, IQR 37–128). Predicted and observed HCO3 levels closely matched, with a low per-patient MAE (median 2.07 mmol/l, IQR 1.22-2.70; Figure 1A). Higher errors occurred mainly in patients with elevated lactate (associated with citrate accumulation), where HCO3 was overestimated, and in one case of filter clogging (associated with diminished filter performance), where HCO3 was underestimated. Here, MAE increased sharply after onset, reflecting divergence between predicted and observed values (Figure 1B).
Conclusion
The model captures HCO3 dynamics generally well. Relevant deviations of the model were associated with complications such as elevated lactate and filter clogging, suggesting model-data disagreement may signal treatment issues requiring user action. With further validation, this approach could enable real-time monitoring and alerts.
Funding
- Commercial Support – Fresenius Medical Care