Kidney Week

Abstract: FR-PO0589

Machine Learning-Based Fluid Trajectories and Associated Outcomes in Critically Ill Children

Session Information

• Fluid, Electrolyte, and Acid-Base Disorders: Clinical - 2
  November 07, 2025 | Location: Exhibit Hall, Convention Center
  Abstract Time: 10:00 AM - 12:00 PM

Category: Fluid, Electrolytes, and Acid-Base Disorders

• 1102 Fluid, Electrolyte, and Acid-Base Disorders: Clinical

Authors

• Thadani, Sameer, Baylor College of Medicine, Houston, Texas, United States

Sameer Thadani, MD

• Silos, Christin N., Baylor College of Medicine, Houston, Texas, United States

Christin N. Silos

• Maffei, Salvador R, Baylor College of Medicine, Houston, Texas, United States

Salvador R Maffei, MD

• Kennedy, Curtis E., Baylor College of Medicine, Houston, Texas, United States

Curtis E. Kennedy, MD, PhD

• Chen, Jin, The University of Alabama at Birmingham, Birmingham, Alabama, United States

Jin Chen, PhD

• Neyra, Javier A., The University of Alabama at Birmingham, Birmingham, Alabama, United States

Javier A. Neyra, MD, MS, FASN

• Akcan Arikan, Ayse, Baylor College of Medicine, Houston, Texas, United States

Ayse Akcan Arikan, MD, FASN

Background

Fluids are central to managing critically ill children; however, fluid overload (FO) is increasingly recognized to be associated with poor outcomes. Prior studies have measured FO as a static value, failing to capture how daily changes in fluid balance affect outcomes. We aimed to use machine learning to identify fluid balance trajectories and assess their impact on outcomes.

Methods

We conducted a single-center retrospective study of children admitted to the PICU at Texas Children’s Hospital from 1/1/2021 to 5/30/2022. Patients were excluded if ICU stay was <7 days. We split the cohort into 80% training and 20% testing sets. We used finite mixture models to identify phenotypes based on delta net fluid balance (NFB) over the first 7 days. Delta NFB was calculated as: NFB (day N) minus NFB (day N–1). The primary outcome for model development was hospital length of stay; in-hospital mortality was the outcome for final logistic regression.

Results

We analyzed 2,371 encounters from 1,954 patients, comprising 16,597 ICU days. Median age was 7.4 years (IQR 1.6–14.8); median weight was 23 kg (IQR 11–50). In total, 1,009 encounters required mechanical ventilation and 75 needed renal replacement therapy. Median PICU stay was 13 days (IQR 8–24); 93 patients (3.9%) died. Five fluid trajectory groups were identified (Figure 1), with differing demographic characteristics and prescribed therapies. Mean posterior class membership probability was >98% and model performance showed R² = 0.68. In adjusted logistic regression, groups 1 [aOR 3.18 (2.14–4.72)], 3 [3.07 (2.09–4.52)], and 5 [6.90 (4.50–10.36)] were associated with mortality.

Conclusion

We identified five fluid trajectory phenotypes with distinct profiles and outcomes. These findings support trajectory-based approaches for risk stratification. Future work should explore how fluid type, diuretics, extracorporeal fluid removal, and vasoactives influence trajectories and outcomes.

Digital Object Identifier (DOI)

• doi: 10.1681/ASN.2025hybay66t