Clinical Pathways and Digital Twins in MIMIC-IV: A Proof-of-Concept Study.

Area of Science:

• Clinical Informatics
• Health Data Science
• Machine Learning in Healthcare

Background:

• Electronic health records (EHRs) contain rich temporal data.
• Clinical pathway analysis is crucial for understanding patient journeys.
• Digital twins offer a novel approach to personalized medicine.

Purpose of the Study:

• To assess the utility of the MIMIC-IV database for constructing clinical pathways.
• To explore the feasibility of creating proof-of-concept digital twins from MIMIC-IV data.
• To investigate the potential of machine learning for automated pathway updates.

Main Methods:

• Reconstruction of patient-level event logs from MIMIC-IV using admissions, diagnoses, and procedures.
• Utilization of directed graphs for visualizing patient trajectories.
• Development of a baseline machine learning pipeline for regenerating predictive features.

Main Results:

• MIMIC-IV provides sufficient temporal detail for event ordering and pathway-outcome linkage.
• The database supports the reconstruction of heterogeneous clinical pathways.
• A machine learning pipeline demonstrated the capability to update pathways with new data.

Conclusions:

• MIMIC-IV is a suitable resource for clinical pathway and digital twin construction.
• Automated pathway updates using machine learning are feasible.
• Findings are based on a US-centric database and proof-of-concept design, requiring cautious interpretation for generalizability.