Machine learning approaches for the detection of emergency department patients with opioid misuse

用于检测阿片类药物滥用的急诊科患者的机器学习方法

• 批准号: 10350200
• 负责人: Neeraj Chhabra
• 金额: $ 19.8万
• 依托单位: COOK COUNTY HEALTH AND HOSPITAL SYSTEM
• 依托单位国家: 美国
• 财政年份: 2022
• 资助国家: 美国
• 起止时间: 2022-04-15 至 2023-02-15
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/10350200
• 关键词: Academic Medical Centers; Accident and Emergency department; Affect; Algorithms; American; Area; Biometry; COVID-19 pandemic; Cause of Death; Cessation of life; Characteristics; Clinical; Clinical Ethics; Cluster Analysis; Code; Cost Savings; County; Data; Data Science; Data Sources; Databases; Decision Trees; Detection; Development Plans; Diagnosis; Discrimination; Disease; Electronic Health Record; Emergency Care; Emergency Department patient; Emergency Department-based Intervention; Emergency Health Services; Ensure; Ethics; Evidence based intervention; Foundations; Fright; Future; Goals; Grant; Harm Reduction; Health; Hospitalization; Hour; Human; Hybrids; Individual; Institution; Intervention; Investigation; K-Series Research Career Programs; Link; Logistic Regressions; Machine Learning; Manuals; Measures; Medical; Mentored Patient-Oriented Research Career Development Award; Mentors; Mentorship; Methods; Modeling; Morbidity - disease rate; Natural Language Processing; Opioid; Outcome; Patient Care; Patient Self-Report; Patients; Performance; Process; Provider; Research; Research Personnel; Research Training; Resources; Risk; Role; Stigmatization; Systematic Bias; Techniques; Testing; Time; Toxicology; Training; United States; advanced analytics; base; career development; clinically relevant; cohort; cooking; deep neural network; improved; individual patient; innovation; machine learning algorithm; machine learning model; model building; mortality; multidisciplinary; multiple data sources; opioid misuse; opioid mortality; opioid overdose; opioid use disorder; patient engagement; patient oriented; patient-level barriers; predictive modeling; premature; prescription monitoring program; prevent; programs; prospective; research and development; screening; skills; social bias; substance use; targeted treatment

Project Summary/Abstract Patients with opioid misuse disproportionately utilize emergency health services and are at increased risk for premature death. The timely and accurate identification of patients with opioid misuse in the Emergency Department (ED) is critical to provide evidence-based interventions to decrease mortality. Challenges to opioid misuse detection in the ED include provider time constraints, inconsistent screening approaches, and patient barriers to self-reporting. Advanced analytic techniques such as machine learning and cluster analyses offer promise in efficiently characterizing and identifying patients with opioid misuse during their ED encounter by leveraging data within the electronic health record (EHR) and the prescription drug monitoring program (PDMP). The role of machine learning approaches utilizing multiple data sources to identify ED patients with opioid misuse has yet to be fully explored. In aim 1, multiple machine learning algorithms using ED encounter data will be developed for the identification of opioid misuse. Models will be systematically assessed for social biases and mitigation strategies implemented to ensure equity in model performance. In aim 2, the inclusion of longitudinal PDMP data for the identification of ED patients with opioid misuse will be evaluated by building models from both data sources utilizing ensemble stacking methods. Finally, in aim 3, an unsupervised latent class analysis model will be built to identify clinically relevant subphenotypes of ED patients with opioid misuse, describe their characteristics, and determine patient-oriented outcomes. An innovative approach to the detection of ED patients with opioid misuse will be pursued by rigorously testing machine learning models utilizing multiple data sources, conducting social bias assessments prior to clinical deployment, and characterizing latent groups of patients with opioid misuse. The candidate for this Mentored Patient-Oriented Career Development Award (Dr. Neeraj Chhabra) possesses a strong foundation in emergency care, medical toxicology, substance use research, and biostatistics. Through this K23, he will further develop skills in data science to build comprehensive and scalable models spanning multiple data domains for the identification of patients with opioid misuse. The multidisciplinary mentorship team led by his primary mentor (Dr. Niranjan Karnik) and co-mentors (Dr. Majid Afshar, Dr. Harold Pollack, and Dr. Gail D’Onofrio) consists of nationally renowned experts in the fields of substance use research, machine learning, natural language processing, and clinical ethics. Through an integrated program of formal coursework, ethics training, mentorship, and research, Dr. Chhabra will develop the skillset necessary to complete these aims and transition to independent investigation. His proposal takes full advantage of the combined resources provided by the affiliated institutions of Cook County Health and Rush University Medical Center. Dr. Chhabra’s long-term goal is to utilize machine learning techniques to focus treatments and resources towards patients with opioid misuse within the ED setting. This K23 award provides the necessary foundation to pursue this goal and will form the basis for future R01 proposals evaluating the clinical impact of these models.

项目总结/摘要 滥用阿片类药物的患者不成比例地使用紧急卫生服务， 过早死亡紧急情况下阿片类药物滥用患者的及时准确识别 艾德部门是提供循证干预措施以降低死亡率的关键。阿片类药物的挑战 艾德中的误用检测包括提供者时间限制、不一致的筛查方法和患者 自我报告的障碍。先进的分析技术，如机器学习和聚类分析， 有望通过以下方式有效表征和识别艾德治疗期间阿片类药物滥用的患者 利用电子健康记录（EHR）和处方药监测计划（PIMs）中的数据。 利用多个数据源的机器学习方法在识别阿片类药物滥用的艾德患者中的作用 还有待充分探索。在目标1中，将使用艾德就诊数据的多种机器学习算法 用于识别阿片类药物滥用。将系统地评估模型的社会偏见， 实施缓解战略，以确保模型性能的公平性。在目标2中，列入纵向 将通过从两个模型中构建模型，评价用于识别阿片类药物滥用艾德患者的PADER数据。 数据源利用集合叠加方法。最后，在aim 3中， 将建立识别阿片类药物滥用的艾德患者的临床相关亚表型，描述其 特征，并确定以患者为导向的结果。一种检测艾德患者的创新方法 阿片类药物滥用将通过利用多个数据源严格测试机器学习模型， 在临床部署之前进行社会偏见评估，并描述潜在患者群体， 阿片类药物滥用该指导患者导向职业发展奖的候选人（Neeraj博士 查布拉）在急救护理、医学毒理学、物质使用研究方面拥有坚实的基础， 生物统计学通过这个K23，他将进一步发展数据科学技能，以建立全面和可扩展的 跨越多个数据域的模型，用于识别阿片类药物滥用患者。多学科 导师团队由他的主要导师（Niranjan Karnik博士）和共同导师（Majid Afshar博士，Harold博士 Pollack和Gail D 'Onofrio博士）由物质使用研究领域的全国知名专家组成， 机器学习、自然语言处理和临床伦理。通过一个正式的综合方案， 课程，道德培训，指导和研究，博士。查布拉将开发必要的技能， 完成这些目标并过渡到独立调查。他的建议充分利用了 由库克县卫生和拉什大学医学附属机构提供的综合资源 中心查布拉博士的长期目标是利用机器学习技术集中治疗和资源 针对在艾德环境中滥用阿片类药物的患者。该K23奖项为以下方面提供了必要的基础 追求这一目标，并将成为未来R 01提案的基础，评估这些模型的临床影响。