Identifying opioid response phenotypes in low back pain electronic health data

疼痛是成人门诊和急诊科就诊的首要原因，影响了100多人 百万美国人，每年花费超过6000亿美元。腰背痛(LBP)占28% 这是医疗保健问题，也是美国和全世界残疾的主要原因。阿片类药物 是美国最常见的处方药类别，这些处方中的大多数 是给伦敦金融服务公司的。尽管阿片类药物治疗在LBP中得到了广泛应用，但 阿片类药物治疗的疼痛缓解经验尚未确定，这使得提供者没有明确 指导安全有效的治疗。鉴于这种令人震惊的疾病负担和医疗保健利用， 有关LBP的临床信息广泛存在于电子健康记录(EHR)中，提供了有价值的 数据源。然而，这些信息目前除了个别患者的经历之外几乎没有什么意义。 因为来自电子病历的大部分疼痛相关数据都嵌入了自由文本中。使用EHR数据可能会 为更好地理解LBP提供了重要的桥梁。因此，这一提议的中心假设是 将临床经验转化为离散和可分析的数据，特别是对阿片类药物的反应进行建模 LBP患者的表型，将确定临床相关的表型治疗反应。为了测试这一点 假设，这个有指导的职业发展项目将适应和应用自然语言处理 (NLP)、数据标准化、挖掘和分析工具，以对阿片类药物反应表型进行专门建模 评估LBP患者的疼痛强度、功能状态以及疼痛对活动的干扰。 通过整合的AIMS，这项提议将：1)支持LBP和阿片类音符语料库的注释，以及 与疼痛强度、功能状态和疼痛干扰相关的临床概念图 活动；2)使用NLP识别和关联LBP患者的相关阿片反应表型 在EHR中；以及3)表征与阿片类药物剂量增加相关的LBP表型。NLP的临床应用 用于提取和转换高维临床数据的统计建模，当使用 PI的领域知识创造了一个了解LBP管理、结果和 治疗效果。最终，这一基础可用于预测临床结果和对 治疗性干预。我们的长期目标是超越识别疾病表型特征，而是 建立一个识别治疗反应表型的系统。根据疼痛强度对患者进行分层， 功能状态、疼痛干扰和其他因素，我们计划确定潜在的队列 从遗传学的角度进一步研究。这项受指导的职业发展补助金(K08)将支持临床 专家对工具的调整和系统方法的培训，以使增长朝着规划方向发展 对NIH疼痛研究议程做出令人难以置信的反应，并可以过渡到独立的研究 R01级资金。

Pain is the leading reason for adult outpatient and emergency department medical visits, impacting over 100 million Americans at a cost of over $600 billion dollars annually. Low back pain (LBP) represents 28% of this health-care problem and is the leading cause of disability, both in the United States and worldwide. Opioids are the most commonly prescribed drug class in the United States, and the majority of these prescriptions are for LBP. Despite the broad application of opioid therapy in LBP, the phenotypes of individuals who experience pain relief from opioid treatment have not been identified, leaving providers without clear guidance for safe and effective therapy. Given this staggering burden of disease and health-care utilization, clinical information regarding LBP widely populates the electronic health record (EHR), providing a valuable data source. However, this information presently has little meaning beyond the individual patient experience because the majority of pain-related data from the EHR is embedded in free text. Using EHR data may provide the crucial bridge to a better understanding of LBP. Thus, the central hypothesis of this proposal is that translating clinical experiences into discrete and analyzable data, specifically modeling opioid response phenotypes for patients with LBP, will identify clinically relevant phenotypic treatment responses. To test this hypothesis, this mentored career development project will adapt and apply natural language processing (NLP), data standardization, mining, and analysis tools to specifically model opioid response phenotypes for patients with LBP to characterize pain intensity, functional status, and pain interference with activity. Through integrated aims, this proposal will, 1) support the annotation of LBP and opioid note corpus, and the mapping of clinical concepts related to pain intensity, functional status, and pain interference with activities; 2) use NLP to identify and relate relevant opioid response phenotypes in patients with LBP in the EHR; and 3) characterize LBP phenotypes associated with opioid dose escalation. Clinical NLP uses statistical modeling to extract and transform high dimensional clinical data, which, when developed with the PI’s domain knowledge, creates a unique opportunity to understand LBP management, outcomes, and therapeutic efficacy. Ultimately this foundation may be used to predict clinical outcomes and responses to therapeutic interventions. Our long-term goal is to move beyond identifying disease phenotype profiles to create a system to identify treatment response phenotypes. Stratifying patients based on pain intensity, functional status, pain interference and other factors, we plan to identify potential cohorts that warrant further study from a genetic focus. This mentored career development grant (K08) will support a clinical expert’s adaptation of tools and training in a systematic method to allow growth toward a programmatic line of research that is incredibly responsive to the NIH pain research agenda and can transition to independent R01-level funding.