INFORMING INTENSIVE CARE UNIT DIGITAL TWINS: DYNAMIC ASSESSMENT OF CARDIORESPIRATORY FAILURE TRAJECTORIES IN PATIENTS WITH SEPSIS.

Summary

Using an unsupervised two-stage clustering of 19,177 ICU sepsis patients over 8 years, the authors identified four robust cardiorespiratory support trajectories with dramatically different mortality (fast/slow recovery vs fast/delayed decline). Clusters were associated with comorbidity and severity scores, providing a basis for predictive analytics and ICU digital twin decision support.

Key Findings

• Identified four distinct ICU trajectories: fast recovery (27%, mortality 3.5%), slow recovery (62%, mortality 3.6%), fast decline (4%, mortality 99.7%), and delayed decline (7%, mortality 97.9%).
• Trajectories were robust and separable by Charlson Comorbidity Index, APACHE III, and day 1/3 SOFA scores (ANOVA P<0.001).
• Modeling used unsupervised two-stage clustering of dynamic cardiorespiratory support and hospital discharge status over the first 14 ICU days.
• Large retrospective cohort from Mayo Clinic hospitals (N=19,177) spanning 8 years.

Clinical Implications

Trajectory-aware tools could enhance communication with families, guide levels of support, and target interventions to those on decline trajectories early.

Limitations

• Retrospective single health system; external validation not reported.
• Clusters are associative; causal inferences and intervention effects cannot be drawn.

Future Directions

Prospective validation across diverse ICUs; integration into real-time dashboards; testing trajectory-informed interventions and digital twin simulations.