Abstract: TH-PO230
Predicting Liberation from Continuous Kidney Replacement Therapy in Critically Ill Patients Using a Machine Learning Model
Session Information
- Hemodialysis, Hemodiafiltration, and Frequent Dialysis
October 24, 2024 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Dialysis
- 801 Dialysis: Hemodialysis and Frequent Dialysis
Authors
- Kashani, Kianoush, Mayo Clinic Minnesota, Rochester, Minnesota, United States
- Alfieri, Francesca, U-Care Medical s.r.l, Torino, Italy
- Ancona, Andrea, U-Care Medical s.r.l, Torino, Italy
- Zappalà, Simone, U-Care Medical s.r.l, Torino, Italy
Background
Continuous renal replacement therapy (CRRT) is utilized in nearly 14% of critically ill patients. Currently, there are no standardized practice guidelines to wean CRRT, apart from clinicians' discretion, as per the KDIGO guidelines. This study aims to develop and validate a model to predict successful CRRT liberation.
Methods
For this single-center, retrospective cohort study, we used data from 661 adult patients connected to 668 disinct ICU stays from MIMIC-IV dataset who received CRRT between 2008 to 2019.
CRRT liberation was defined as renal replacement therapy (RRT)-free survival within seven days after the liberation. We randomly divided the cohort into derivation (70%) and validation sets [KK1] (30%). The outcomes were successful CRRT liberation vs. unsuccessful CRRT liberation and/or death. A multiclass decision tree model was developed and internally validated.
Results
The final cohort included 668 ICU stays requiring CRRT, among them 266 (39.8%) were successfully liberated from CRRT, 265(39.7%) had unsuccessful CRRT liberation, and 137(20.5%) died while on CRRT.
The auROC of the model reflects its ability to discriminate between successful and unsuccessful liberation, varying from 0.797(CI 95% 0.743, 0.864) to 0.739(CI 95% 0.700, 0.810) when the model is activated at the time of CRRT liberation up to 12 hours later. Moreover, the model outputs a probability of mortality within seven days from the CRRT discontinuation, achieving auROCs from 0.746 (CI 95% 0.682, 0.861) to 0.861(CI 95% 0.795, 0.944) moving forward over time.
Conclusion
The model is designed to be used by clinicians at the time of CRRT liberation and can be consulted up to 12 hours after the discontinuation.
This validated model could be integrated into the clinical practice with the aim of assisting the decision-making related to the CRRT liberation in critically ill patients with AKI hospitalized in intensive care units, facilitating a timely reinstitution of CRRT where needed and ensuring better patient outcomes.
Funding
- Commercial Support – U-Care Medical s.r.l.