Icon Legend

This session is not in your schedule.

This session is in your schedule. Click again to remove it.

Device Technologies and Biomedical Robotics

Device Technologies and Biomedical Robotics - Poster Session B

Poster I17 - Development of sedation state assessment models using machine learning and accelerometry for critically ill children

Thursday, October 24, 2024

2:45 PM – 3:45 PM EST

Location: Exhibit Hall E, F & G

Presenting Author(s)

Sidharth Raghavan

Student
Department of Biomedical Engineering, Johns Hopkins University
Baltimore, Maryland, United States

Co-Author(s)

JH

Jake Hoffmann

Student
Department of Computer Engineering, Johns Hopkins University, United States
Sayantika Roy, BS

Medical Student
University of Rochester School of Medicine and Dentistry
Getzville, New York, United States
SP

Sara Barbara Pejic

Student
Department of Biomedical Engineering, Johns Hopkins University, United States
SW

Sarah Wu

Student
Department of Computer Engineering, Johns Hopkins University, United States
Feiyang Huang, BS

Graduate Student
Weill Cornell Medicine, Tri-Institutional Training Program in Computational Biology and Medicine, United States
KB

Kristen Brown

Associate Professor
Johns Hopkins School of Nursing, United States
ND

Nicholas Durr

Assistant Professor
Johns Hopkins University, United States
Sapna Ravi Kudchadkar, MD, PhD

Professor
Johns Hopkins University School of Medicine, Anesthesiology and Critical Care Medicine, United States
James Fackler, MD (he/him/his)

Professor
Johns Hopkins University School of Medicine, Anesthesiology and Critical Care Medicine
BALTIMORE, Maryland, United States

Primary Investigator(s)

Jessica LaRosa, MD

Assistant Professor
Johns Hopkins University School of Medicine, Anesthesiology and Critical Care Medicine, United States

Introduction: Standardized evaluation of sedation levels in critically ill children is essential in sedation management. 250,000 children receive critical care in the US annually, over 30% of whom receive sedative medications at some point during their stay. However, only 58% of those are optimally sedated, with 32% oversedated and 10% undersedated. Improper sedation increases risk of adverse effects, with over-sedation leading to prolonged respiratory support and withdrawal syndrome, while under-sedation increases patients' physical and physiological stress, often leading to self-extubation and the need for restraints. The State Behavioral Scale (SBS) is one validated strategy for evaluating sedation levels; however, it does not provide real-time assessment and is subject to user variability. We hypothesize that (1) the incorporation of a computational model incorporating heart rate (HR), respiratory rate (RR) and oxygen saturation (SpO2) changes will accurately predict nurse-reported SBS scores and (2) that the addition of accelerometry to routinely collected clinical data can be used as a strong indicator of sedation level.

Materials and

Methods: (Study 1) A retrospective electronic medical record (EMR) dataset from 188 pediatric ICU patients receiving continuous sedation at Johns Hopkins was captured. Time-series features were extracted for heart rate, respiratory rate, and oxygen saturation around each documented SBS score. Ten repeats of ten-fold cross validation were performed on two ordinal regression models: three-class (SBS: {-3, -2, -1}, {0}, {+1, +2}), and five-class (SBS: {-3, -2}, {-1}, {0}, {+1}, {+2}). Accuracy, F1-score, sensitivity, specificity and Cohen’s Kappa were computed.

(Study 2) Time-series accelerometry, vitals waveform data, and nurse-reported SBS scores were collected from thirteen pediatric patients on continuous sedatives in the Johns Hopkins Pediatric Intensive Care Unit through an ongoing clinical study. Patients receiving continuous complete neuromuscular blockade were excluded. Mean-amplitude deviation (MAD) data, calculated from the raw time-series accelerometry data, and vitals waveform data were analyzed in 15-minute intervals surrounding nurse-reported SBS scores. Paired t-tests were performed to compare scores and correlation analysis was performed.

Results, Conclusions, and Discussions: In our retrospective analysis, our three-class model predicted SBS with a mean of 40.09 ± 3.26% accuracy, 44.18 ± 11.11% F1-score, 40.97 ± 3.25% sensitivity, 80.67 ± 2.66% specificity, and 11.78 ± 5.71% Cohen’s Kappa. Our five-class model had a mean of 25.45 ± 3.53% accuracy, 34.45 ± 10.04% F1-score, 25.45 ± 3.35% sensitivity, 79.30 ± 2.94% specificity, and 6.78 ± 4.53% Cohen’s Kappa. However, our ongoing analysis found that while most SBS scores didn't show significant differences in vitals waveform data and accelerometry, there were positive, moderate correlations at the extremes.

Our retrospective study revealed that both machine learning models, particularly the three-class, indicate a positive correlation between vitals and SBS, but the low resolution of EMR data is insufficient in prediction. Our ongoing study revealed that the low frequency of nurse-reported SBS scores, in conjunction with our preliminary data sample size, resulted in low-levels of correlation for vitals waveform data and accelerometry. Ongoing efforts to improve model performance and data analysis includes standardized frequent SBS scoring, the combined analysis of accelerometry and vitals data to model complex patient states, and the addition of the high-frequency data from our clinical study to our machine learning models. Though further investigation is necessary, our future efforts aim to create a robust sedation state prediction algorithm with the goal of improving real-time sedation assessment.

Acknowledgements (Optional): We would like to thank the Johns Hopkins Discovery Grant and Undergraduate Catalyst Award for supporting our work.