Pervasive Sensing and AI in Intelligent ICU

Official title

Pervasive Sensing and Artificial Intelligence in Intelligent ICU Subtitles: -Intelligent Intensive Care Unit (I2CU): Pervasive Sensing and Artificial Intelligence for Augmented Clinical Decision-making -ADAPT: Autonomous Delirium Monitoring and Adaptive Prevention

Overview

Important information related to the visual assessment of patients, such as facial expressions, head and extremity movements, posture, and mobility are captured sporadically by overburdened nurses, or are not captured at all. Consequently, these important visual cues, although associated with critical indices such as physical functioning, pain, delirious state, and impending clinical deterioration, often cannot be incorporated into clinical status. The overall objectives of this project are to sense, quantify, and communicate patients' clinical conditions in an autonomous and precise manner, and develop a pervasive intelligent sensing system that combines deep learning algorithms with continuous data from inertial, color, and depth image sensors for autonomous visual assessment of critically ill patients. The central hypothesis is that deep learning models will be superior to existing acuity clinical scores by predicting acuity in a dynamic, precise, and interpretable manner, using autonomous assessment of pain, emotional distress, and physical function, together with clinical and physiologic data.

Detailed description

The under-assessment of pain is one of the primary barriers to the adequate treatment of pain in critically ill patients, and is associated with many negative outcomes such as chronic pain after discharge, prolonged mechanical ventilation, longer ICU stay, and increased mortality risk. Many ICU patients cannot self-report their pain intensity due to their clinical condition, ventilation devices, and altered consciousness. The monitoring of patients' pain status is yet another task for over-worked nurses, and due to pain's subjective nature, those assessments may vary among care staff. These challenges point to a critical need for developing objective and autonomous pain recognition systems. Delirium is another common complication of patient hospitalization, which is characterized by changes in cognition, activity level, consciousness, and alertness and has rates of up to 80% in surgical patients. The risk factors that have been associated with delirium include age, preexisting cognitive dysfunction, vision and hearing impairment, severe illness, dehydration, electrolyte abnormalities, overmedication, alcohol abuse, and disruptions in sleep patterns. Estimates show that about one third of delirium cases can benefit from drug and non-drug prevention and intervention. However, detecting and predicting pain and delirium is still very limited in practice.

The aim of this study is to evaluate the ability of the investigators' proposed model to leverage accelerometer, environmental, circadian rhythm biomarkers, and video data in autonomously quantifying pain, characterizing functional activities, and delirium status. The Autonomous Delirium Monitoring and Adaptive Prevention (ADAPT) system will use novel pervasive sensing and deep learning techniques to autonomously quantify patients' mobility and circadian dyssynchrony in terms of nightly disruptions, light intensity, and sound pressure level. This will allow for the integration of these risk factors into a dynamic model for predicting delirium trajectories. Commercially available cameras will be used to monitor patients' facial expressions and contextualize patients' actions by providing imaging data to provide additional patient movement information. Commercially available environmental sensors will be used to provide data on illumination, decibel level, and air quality. Patient blood samples will help determine their circadian rhythm and compare and validate the pervasive sensing system's capabilities of autonomously monitoring circadian dyssynchrony. Electronic health record data will also be collected.

Interventions

• Other Video Monitoring
  continuous video monitoring
• Other Accelerometer Monitoring
  continuous accelerometer monitoring of patient movements
• Other Noise Level Monitoring
  continuous environmental noise monitoring
• Other Light Level Monitoring
  continuous environmental light monitoring
• Other Air Quality Monitoring
  continuous environmental air quality monitoring
• Other EKG Monitoring
  continuous EKG monitoring
• Other Vitals Monitoring
  continuous vitals monitoring (heart rate, oxygen saturation)
• Other Biosample Collection
  blood and urine samples collected once on Day 1 and once on Day 2
• Other Delirium Motor Subtyping Scale 4 (DMSS-4)
  done daily on delirious patients to subtype delirium

Primary outcome measures

• Algorithmic Activity Labeling [Time frame: Image frames collected continuously for up to 7 days maximum.]
• Algorithmic Pain Labeling [Time frame: Image frames collected continuously for up to 7 days maximum.]
• Decibel Levels [Time frame: Noise sensor data collected continuously for up to 7 days maximum.]
• Lux Levels [Time frame: Light sensor data collected continuously for up to 7 days maximum.]
• Air Quality [Time frame: Air quality sensor data collected continuously for up to 7 days maximum.]
• Circadian Dyssynchrony Index [Time frame: Change in internal circadian profile from Day 1 to Day 2.]
• Algorithmic Delirium Recognition Profile [Time frame: Data collected for up to 7 days maximum.]
• Delirium Motor Subtyping Scale 4 (DMSS-4) [Time frame: Changes from baseline up to a maximum of 7 days]

Secondary outcome measures (1)

• Mortality [Time frame: From baseline (study enrollment) up to a maximum of 7 days]

Eligibility criteria

Inclusion criteria

• aged 18 or older
• admitted to UF Health Shands Gainesville ICU ward
• expected to remain in ICU ward for at least 24 hours at time of screening

Exclusion criteria

• under the age of 18
• on any contact/isolation precautions
• expected to transfer or discharge from the ICU in 24 hours or less
• unable to provide self-consent or has no available proxy/LAR

Criteria are shown verbatim from the registry (in English). Final eligibility is always assessed by the study center.

Healthy volunteers: No

Study design

Observational model

Case-only

Study locations

United States · 1 center

• University of Florida Health Shands Hospital — Gainesville

Publications

• Davoudi A, Malhotra KR, Shickel B, Siegel S, Williams S, Ruppert M, Bihorac E, Ozrazgat-Baslanti T, Tighe PJ, Bihorac A, Rashidi P. Intelligent ICU for Autonomous Patient Monitoring Using Pervasive Sensing and Deep Learning. Sci Rep. 2019 May 29;9(1):8020. doi: 10.1038/s41598-019-44004-w. PMID 31142754

Identifiers

NCT: NCT05127265 · IRB-202101013 · R01NS120924 · R01EB029699

Primary sources (government registries)

• ClinicalTrials.gov (US NIH)

View this study on ClinicalTrials.gov ↗