Patient-Specific Classification of ICU Sedation Levels From Heart Rate Variability*

Q: What is the clinical evidence from this study?

Study design: Observational. Population: Mechanically ventilated adults requiring sedation (n=70). Intervention: Patient-specific (personalized) HRV-based sedation assessment algorithm vs. Patient-independent HRV-based algorithm. Primary outcome: Overall accuracy in discriminating between four sedation levels (comatose, deeply sedated, lightly sedated, agitated) (8 percentage point increase in accuracy).

Sunil B. Nagaraj; Siddharth Biswal; Emily J. Boyle; David Zhou; Lauren McClain; Ednan K. Bajwa; Sadeq A. Quraishi; Oluwaseun Akeju; Riccardo Barbieri; Patrick L. Purdon; M. Brandon Westover

doi:10.1097/ccm.0000000000002364

April 25, 2017Critical Care MedicineOpen Access

Patient-Specific Classification of ICU Sedation Levels From Heart Rate Variability*

View Full Paper

Ask AI

Bookmark

Share

Key result

Personalized ECG-derived HRV algorithm improves sedation-level classification accuracy by ~8 percentage points.

8 percentage point increase in accuracy
n=70

Why the study?

Does a patient-specific algorithm using heart rate variability improve the accuracy of classifying sedation levels in mechanically ventilated ICU patients?

Study Design

Type

Observational (n=70)

Multicenter

Yes

Structured PICO

Does a patient-specific algorithm using heart rate variability improve the accuracy of classifying sedation levels in mechanically ventilated ICU patients?

P

Population

70 mechanically ventilated adult ICU patients receiving sedatives were monitored to develop a heart rate variability algorithm for sedation level assessment.

E

Exposure

Patient-specific automatic HRV-based sedation-level inference system (AHSISt), calibrated with the individual patient's labeled HRV epochs from the preceding 24 hours.

C

Comparator

Patient-independent version of the HRV-based classification system.

O

Outcome

Overall accuracy of discriminating between four grouped Richmond Agitation Sedation Scale (RASS) levels: comatose, deeply sedated, lightly sedated, and agitated.surrogate

Main Result

Effect estimate: 8 percentage point increase in accuracy

Absolute Event Rate: 67% vs 59%

A personalized machine learning algorithm utilizing continuous ECG-derived heart rate variability shows promise as an objective, automated tool for monitoring depth of sedation in the ICU.

Limitations

Did not account for infusion rates and types of sedative drugs (PKPD)
Did not consider associations between deeper levels of sedation and increasing organ failure severity
RASS scores were grouped into 4 categories rather than treated continuously
RASS scoring is subjective and prone to human error/annotation noise
Required initial 24 hours of monitoring and labeling as human input (semi-automatic)
Did not account for infusion rates and types of sedative drugs administered
Did not consider associations between deeper levels of sedation and increasing organ failure
Grouped RASS scores into 4 distinct groups instead of a continuous variable
RASS scores are subject to human error (annotation noise)
Misclassifications to nearby states were penalized equally to far states ('hard' instead of 'soft' classification)
System uses observations from the first 24 hours and is considered semi-automatic

Implication

May support objective sedation monitoring in ventilated ICU patients; hypothesis-generating pilot data requiring prospective validation.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Nagaraj et al. (2017) conducted an observational in Mechanically ventilated adults requiring sedation (n=70). Patient-specific (personalized) HRV-based sedation assessment algorithm vs. Patient-independent HRV-based algorithm was evaluated on Overall accuracy in discriminating between four sedation levels (comatose, deeply sedated, lightly sedated, agitated) (8 percentage point increase in accuracy). A personalized machine learning algorithm based on continuous ECG-derived heart rate variability improved the overall accuracy of discriminating between four sedation levels to 67%, compared to 59% with a patient-independent approach.

synapsesocial.com/papers/6a237305a2806805a29ff40e https://doi.org/10.1097/ccm.0000000000002364

View Full Paper

Ask AI

Bookmark

Share

Patient-Specific Classification of ICU Sedation Levels From Heart Rate Variability*

Study Design

Structured PICO

Main Result

Limitations

Implication

Discussion

Cite This Study

Patient-Specific Classification of ICU Sedation Levels From Heart Rate Variability*

Study Design

Structured PICO

Main Result

Limitations

Implication

Discussion

Cite This Study

Keep reading

Keep reading