Intradialytic hypotension and hemodynamic phenotypes in children following continuous renal replacement therapy initiation.

BACKGROUND

Intradialytic hypotension (IDH) leads to inadequate organ perfusion and occurs frequently after continuous renal replacement therapy (CRRT) connection. Unsupervised learning can enhance our understanding of how clinical trajectories impact outcomes. We aim to investigate the association between IDH during CRRT connection and outcomes, while also identifying hemodynamic trajectory-based phenotypes.

METHODS

A single center retrospective observational study of children (<18 years) undergoing CRRT from 9/2016 to 10/2018. IDH was defined as a sustained >20% decrease in mean arterial pressure (MAP) from baseline for ≥2 consecutive minutes. IDH burden was calculated by dividing connections with IDH by total observed connections. The primary outcome was major adverse kidney events at 30 days (MAKE30). K-means clustering was used to identify MAP trajectory-based phenotypes.

RESULTS

59 patients, 232 connections, and 13,920 minutes were included. Median age was 59 months (IQR 8-152). In multivariable analysis, higher IDH burden [β 4.35 (CI: 0.01-8.70)] was associated with MAKE30. Two distinct MAP trajectories phenotypes were identified, with differing incidence of MAKE30 [21 (100%) vs. 29 (76%), p < 0.01].

CONCLUSIONS

IDH within the first hour of CRRT connection is associated with poor outcomes, and time-series clustering is feasible and could improve our understanding of the impact of CRRT in children.

IMPACT

Repeated episodes of intradialytic hypotension within the first hour of continuous renal replacement therapy connection are associated with increased morbidity and mortality. Our findings suggest that intradialytic hypotension in the hour following CRRT connection in children is associated with poor outcomes. Unsupervised machine learning, an underutilized approach in pediatric research, identified two significantly different mean arterial pressure trajectory-based phenotypes with differing anthropometric features and outcomes. Leveraging unsupervised machine learning, we can identify trajectory-based subgroups that can provide insights into the impact of continuous renal replacement therapy in critically ill children.