Measuring Infant Pain Objectively using Sensor Fusion and Machine Learning Algorithms

Autonomous Healthcare, Inc., Santa Clara CA

Investigators

Linked publications & trials

Abstract

Newborns are routinely and frequently exposed to pain during Neonatal ICU (NICU) care. Pain assessments in neonates are difficult, labor intensive, subjective and unreliable ? often resulting in excessive or inadequate analgesia. Our overall objective is to measure infant pain objectively, reliably, and in real-time. We will extract pain-related information from multiple non-invasive sensors, develop a sensor fusion framework to integrate multi-modal sensor data into a single pain score, and assess the validity of this approach by comparing with validated clinical pain scores. Specific aims: 1) To differentiate acute pain from baseline or non-painful events, we will study 60 newborns using: facial electromyography (EMG) to record facial expressions specific for infant pain, electrocardiography (ECG) to measure heart rate changes and heart rate variability, skin conductance to measure catecholamine-dependent palmar sweating, electroencephalography (EEG) using 32 ?active? electrodes to assess pain-related brain activity, and pulse oximetry (SpO2) to record pain-induced changes in oxygenation and peripheral perfusion. We will study acute painful procedures associated with mild, moderate, or severe pain in 30 late preterm (34-36 weeks) and 30 term newborns (37-42 weeks). Bedside nurses will use validated pain scoring methods to concurrently assess these infants for pain. A pain expert will independently assess 50% of subjects, to establish inter-rater reliability and to authenticate the bedside nurses? pain scores. From each sensor, we will extract pain-related data that correlate strongly with the clinically relevant pain scores. 2) To develop sensor fusion frameworks integrating data from multiple sensors. Proprietary machine learning algorithms will fuse pain-related data from all 5 sensors, ?calibrate? itself for each newborn by using data from prior pain events, and compensate for missing or unreliable data. Sensor fusion frameworks including combinations of these sensors will help to identify infant pain with far greater specificity and sensitivity than the subjective pain scales used clinically. Procedures will be included to assess the scaling properties of this objective approach and to refine the principal algorithms. Data analyses will assess inter-rater reliability and internal consistency, verify content, concurrent and construct validity, and include multivariable modeling for optimal selection and weighting of the sensor variables that will compute the final objective pain score. This approach will eventually lead to a bedside ICU monitor (compatible with the ECG, SpO2, EEG, EMG, and skin conductivity sensors), which displays the current pain intensity and trends within the time periods of clinical interest. An objective, automated pain detection device developed for newborns (and adapted for other nonverbal patients) will reduce the subjectivity and variability of pain assessments, improve the safety and efficacy of various analgesics used for treating neonatal pain, avoid the acute side effects and long-term effects of both unrelieved pain or excessive analgesia in newborns, prevent iatrogenic tolerance and neonatal abstinence syndrome, reduce the workload of bedside NICU nurses and improve clinical outcomes. !