AKI-twinX: explainable organ structured digital twin for sepsis AKI trajectory forecasting

Area of Science:

• Critical Care Medicine
• Biomedical Informatics
• Digital Health

Background:

• Existing Intensive Care Unit (ICU) models for sepsis-induced acute kidney injury (AKI) primarily focus on onset, neglecting cardio-renal interactions and trajectory prediction.
• Understanding the dynamic interplay between cardiovascular and renal systems is crucial for managing sepsis patients.

Purpose of the Study:

• To develop AKI-twinX, an organ-structured, explainable digital twin for joint forecasting of AKI onset, AKI trajectory, and near-term mortality risk in sepsis.
• To capture and model cross-organ coupling between renal and cardiovascular systems.

Main Methods:

• Developed AKI-twinX, a digital twin incorporating sparse feature gating and attention mechanisms to learn latent states and cross-organ coupling.
• Trained and validated the model on large sepsis cohorts (MIMIC-IV and Indiana University Health) using 5-fold cross-validation.
• Evaluated forecasting accuracy for mortality, AKI onset, and AKI trajectory, and assessed blood pressure prediction and intervention sensitivity.

Main Results:

• Consistent discrimination across prediction tasks: AUC for mortality (0.86-0.88), AKI onset (0.78-0.82), and AKI trajectory (0.73-0.78).
• Accurate 12-hour systolic blood pressure forecasting (MAE: 8.5 mmHg) in vasopressor-treated patients.
• Demonstrated sensitivity to simulated interventions, such as vasopressor withdrawal, predicting increased risk.

Conclusions:

• AKI-twinX provides trajectory-aware forecasting for sepsis-induced AKI with bedside auditability.
• The digital twin enhances understanding of cardio-renal interactions and supports clinical decision-making for sepsis management.