Rebooting Infant Pain Assessment: Using Machine Learning to Exponentially Improve Neonatal Intensive Care Unit Practice.

住院婴儿的疼痛无法控制，有严重的长期并发症。然而，要控制疼痛，必须有准确的婴儿疼痛评估。婴儿不能自我报告他们的疼痛，目前卫生专业人员使用的婴儿疼痛评估工具存在主要问题，因为：1）不同的疼痛指标（心率，氧气水平，面部活动）在测量中被相同地处理，没有尝试优化它们以改善疼痛评估（例如，对它们进行不同的加权）。2)没有一个“黄金标准”工具可以区分创伤性手术（如脚跟穿刺）和非创伤性手术（如脏尿布）造成的痛苦。3)照顾者可能会有偏见，影响他们的疼痛判断。我们的知识用户和健康/自然科学/工程/社会科学研究人员的国际团队已经走到一起，建立一个机器学习算法，将学习如何区分侵入性和非侵入性的痛苦。将对300名早产儿及其母亲的样本进行大约间隔1周的2次痛苦手术（脚跟切开术）的随访。第一次疼痛手术期间的疼痛指标（面部表情、心率、脑电活动、氧气水平）将用于训练算法，以区分第二次手术期间的不同类型的痛苦。我们的团队还将在痛苦的手术之前更长时间测量婴儿心率和氧气水平模式，以更好地了解婴儿在痛苦的手术期间和之后的反应。此外，这将是第一次在临床工具中考虑大脑活动。我们断言，这增加了我们区分侵入性和非侵入性痛苦的能力。疼痛的复杂性需要一种机器学习解决方案，能够对疼痛期间大脑、行为和生理的个体模式进行建模。从本质上讲，这将最终允许早产儿为自己“自我报告”疼痛。

Unmanaged pain in hospitalized infants has serious long term complications. However, to manage pain, one must have accurate infant pain assessment. Infants cannot self-report their pain and current infant pain assessment tools used by health professionals have major problems because: 1) Different pain indicators (heart rate, oxygen levels, facial activity) are treated the same inmeasures, with no attempts to optimize them to improve pain assessment (e.g. weightingthem differently). 2) None of the "gold standard" tools can discriminate distress from invasive procedures (like aheel lance) from distress due to non-invasive procedures (like a soiled diaper). 3) Caregivers can have biases that impact their pain judgments. Our international team of knowledge users and health/natural science/engineering/socialscience researchers have come together to build a machine learning algorithm that will learn how to discriminate invasive and non-invasive distress. A sample of 300 preterm infants and their mothers will be followed during 2 painful procedures (heel lance) approximately 1-week apart. Pain indicators (facial grimacing, heart rate, brain electrical activity, oxygen levels)during the first painful procedure will be used to train the algorithm to discriminate between the different types of distress during the second procedure. Our team will also measure the infant heart rate and oxygen level patterns longer before the painful procedure to better contextualize infant responses during and after the painful procedure. Also, this will be the first time brain activity is being considered in a clinical tool. We assert this increases our ability to discriminate between invasive and non-invasive distress. The complexity of pain requires a machine learning solution that is capable of modelling individual patterns of brain, behaviour, and physiology during pain. In essence, this will finally allow preterm infants to'self-report' pain for themselves.