Use of Machine Learning Classifiers to Forecast Severe Acute Postoperative Pain F

描述（由申请人提供）：高达40%的手术患者报告术后中度至重度疼痛。临床决策支持系统的发展，允许对这部分患者进行术前干预，可能会对他们的康复产生深远的影响，并可能对他们的长期预后产生影响。为了准确预测严重的术后疼痛，我们建议使用机器学习分类器（MLC），这是一种采用一系列新颖搜索和分类方法的分类算法，随着新信息的可用性不断更新其性能。该奖项将允许申请人完成严格的教学课程，强调分类理论、算法评估和临床决策支持系统的开发。这些研究的性质使它们远远超出了传统医学教育的范围。通过抽出时间接受疼痛生物学和心理学、机器学习和临床区域麻醉专家的持续指导，候选人将很好地成为围手术期疼痛预测领域的独立资助研究人员。在本研究的Specific Aim 1中，我们将检验机器学习分类器可以准确预测癌症手术患者术后严重疼痛的假设。这部分研究将回顾性测试MLC预测术后第1天严重疼痛的能力。一系列MLC将在彼此之间进行测试，无论是否执行文本分析。此外，所有MLC将与更传统的多变量回归技术（如逻辑回归）进行比较。在具体目标2中，

DESCRIPTION (provided by applicant): Up to 40% of patients undergoing surgery report moderate to severe pain in the postoperative period. The development of a clinical decision support system to allow preoperative intervention for this subset of patients may have a profound impact on their recovery, and potentially their long-term outcome. To accurately forecast severe postoperative pain, we propose the use of machine learning classifiers (MLC's), which are classification algorithms employing a range of novel search and classification methodologies that continually update their performance as new information becomes available. This award will permit the applicant to complete a rigorous didactic curriculum emphasizing classification theory, algorithm evaluation, and development of clinical decision support systems. The nature of these studies place them far outside the realm of traditional medical education. By protecting time for continued mentorship from experts in pain biology and psychology, machine learning, and clinical regional anesthesia, the candidate is well-positioned to become an independently-funded researcher in the field of perioperative pain prediction. In Specific Aim 1 of this study, we will test the hypothesis that Machine Learning Classifiers can accurately predict severe post-operative pain in patients undergoing cancer surgery. This portion of the study will retrospectively test MLC's ability to predict severe pain on post-operatie day 1. An array of MLC's will be tested amongst each other, both with and without the implementation of text analytics. Additionally, all MLC's will be compared against more traditional multiple variable regression techniques such as logistic regression. In Specific Aim 2, we will test the hypothesis that the addition of prospectively obtained attributes and instances will permit continued improvement in MLC performance. This prospective portion of the study will examine the role of prospectively-obtained psychometric attributes, as well as the ability of MLC's to learn and adapt their accuracy during continued refinements to surgical and anesthetic care.