Rebooting Infant Pain Care: Using Machine Learning and Skin-to-Skin Contact to Exponentially Improve Neonatal Intensive Care Unit Practice

More detail

Unmanaged pain in hospitalized infants has serious long-term complications. Existing infant pain assessment approaches demonstrate several key flaws (e.g., dependent on human cognitive capacity to simply combine data from multiple indicators into a total pain score, none have passed a critical discriminant validity test, bias introduced from human caregivers). Thus, the complexity of preterm pain assessment necessitates a machine learning approach. Our international team of knowledge users and health/natural science/engineering/social science researchers have come together to build a machine learning algorithm that will learn how to discriminate invasive and non-invasive distress. Furthermore, better understanding how skin-to-skin contact works in caregiver-infant dyads and factors that influence the effectiveness of this pain management strategy is a critical step in improving infant pain care in NICUs. Relatedly, the design and sample of our current study (acute pain paradigm while infant is either in skin-to-skin contact with the birthing parent or in the cot) allows us to not only test the influence of skin-to-skin contact vs. cot on preterm newborn pain responding, but also interrogate potential mechanisms underlying the effectiveness of skin-to-skin contact (i.e., cardiac regulation attunement between caregivers and their infants during the procedure; influence of birthing parent perceived stress given the particularly elevated stress levels of NICU parents). A sample of 400 preterm infants (300 from Mount Sinai Hospital and 100 from University College London Hospital \[UCLH\]) and their birthing parents (if available) will be followed during a routine painful procedure (heel lance). Pain indicators (facial grimacing \[behavioural indicators\], heart rate, respiration rate, oxygen saturation levels \[physiologic indicators\], brain electrical activity) during the painful procedure will be used to train the algorithm to discriminate between different types of distress (pain-related and non-pain related). Heart rate and respiration rate, as well as maternal-reported perceived stress levels, will be collected from the birthing parent to examine factors impacting the effectiveness of skin-to-skin contact.