Dynamic and Personalized Risk Forecast in Step-Down Units. Implications for Monitoring Paradigms

Area of Science:

• Critical Care Medicine
• Biomedical Engineering
• Machine Learning in Healthcare

Background:

• Cardiorespiratory insufficiency (CRI) signifies a critical loss of cardiorespiratory reserve, often leading to poor patient outcomes.
• While common in hospitalized individuals, the specific patterns of CRI risk escalation remain largely unexamined.
• Continuous vital sign monitoring offers a potential avenue for understanding dynamic risk changes.

Purpose of the Study:

• To characterize the dynamic and individualized nature of cardiorespiratory insufficiency risk evolution.
• To analyze risk trends using continuous vital sign data from patients in a step-down unit.
• To compare personalized versus non-personalized risk escalation patterns preceding CRI events.

Main Methods:

• A machine learning model was employed to estimate CRI risk trends for 1,971 admissions using continuous vital signs.
• Risk trends were analyzed during the initial 4 hours post-admission and the 4 hours preceding a CRI event.
• Comparison was made between patients who experienced CRI (cases) and those who did not (controls).

Main Results:

• Estimated CRI risk was significantly higher in cases compared to controls during initial monitoring periods (P ≤ 0.001).
• Personalized risk trends diverged from controls 90 minutes before CRI, while non-personalized trends diverged 2 hours prior.
• Distinct phenotypes of risk escalation were identified for both personalized and non-personalized risk assessments.

Conclusions:

• Continuous vital sign monitoring and machine learning can reveal dynamic, personalized CRI risk patterns.
• These insights can inform the development of advanced monitoring systems for predicting CRI.
• Such systems may optimize resource allocation and clinical interventions in acute care settings.