Improving antibiotic treatment decisions through machine learning

项目总结/摘要 抗生素耐药性细菌的感染是现代医疗面临的最大挑战之一。 次优抗生素使用是抗生素耐药性的主要驱动因素之一，但临床医生缺乏强大的工具 帮助他们在患者层面做出合理的治疗决定。这个为期3年的指导临床的目标， 科学家研究职业发展计划是应用最先进的机器学习模型， 收集电子健康记录中的数据，以预测之前的抗菌素耐药性（AMR）风险，以及 抗生素治疗后。候选人Sanjat Kanjilal博士确定了两个重要的临床问题， AMR风险预测的改善可能对护理质量产生重大影响。第一个是过度使用 广谱抗生素治疗社区败血症患者。为了解决这个问题，候选人 将开发一套机器学习预测模型，这些模型是根据电子数据库中定期收集的数据进行训练的。 健康记录，以帮助临床医生确定哪种抗生素将有效地治疗患者的病原体， 最大的可能频谱。第二个问题是无法评估患者发展的风险 在用抗生素治疗后产生抗生素耐药性的感染。候选人建议建立一个强大的 使用目标最大似然估计结合机器学习的因果推理模型， 估计服用各种常用的门诊抗生素对药物开发风险的影响 在治疗后12个月内耐药感染。这项工作的结果将形成一个精确的基础， 抗生素管理和治疗的医学方法。 候选人是一名执业传染病临床医生和临床副医学主任 布里格姆妇女医院的微生物实验室他以前有制造机器的经验 学习算法，提供强大的抗菌剂管理。他独特的背景加上 哈佛医学院和哈佛人口医学系的支持环境 朝圣者卫生保健研究所，定位他很好地过渡到成为一个独立资助的临床医生- 研究传染病和机器学习的科学家。他已经召集了一个 多学科导师团队由败血症流行病学，抗菌药物管理， 实现科学，机器学习和因果推理，以帮助他实现他的目标，并已确定 一个全面的培训计划，为他提供必要的技能，成为一个领导者在他的领域。他短暂 一个学期的目标是成为机器学习算法开发方面的专家， 导致抗生素耐药性感染。他的中期目标是大规模部署这些模型， 他们在随机试验中的实际效用。候选人的长期目标是使用这些算法， 基础设施，以维持它们作为学习健康系统的技术基础， 提供者和公共卫生层面的个性化决策支持。

PROJECT SUMMARY / ABSTRACT Infections from antibiotic resistant bacteria represent one of the biggest challenges facing modern medical care. Suboptimal antibiotic use is one of the major drivers for antibiotic resistance, however clinicians lack robust tools to help them make rational treatment decisions at the patient-level. The goal of this 3-year mentored clinical scientist research career development program is to apply state-of-the-art machine learning models to routinely collected data in the electronic health record to predict the risk of antimicrobial resistance (AMR) prior to, and after antibiotic treatment. The candidate, Dr. Sanjat Kanjilal, has identified two important clinical problems where improved risk prediction for AMR could have a significant impact on quality of care. The first is the overuse of broad-spectrum antibiotics in patients presenting with community-onset sepsis. To address this, the candidate will develop a set of machine learning prediction models trained on routinely collected data in the electronic health record to help clinicians identify which antibiotic(s) will effectively treat the patient's pathogen while being of the narrowest possible spectrum. The second problem is the inability to assess the risk of a patient developing an antibiotic resistant infection after being treated with an antibiotic. The candidate proposes to build a robust causal inference model using targeted maximum likelihood estimation combined with machine learning to estimate the impact of taking various commonly used outpatient antibiotics on the risk of developing a drug resistant infection in the 12 month period after treatment. The results of this work will form the basis of a precision medicine approach to antibiotic stewardship and treatment. The candidate is a practicing infectious diseases clinician and the Associate Medical Director of the clinical microbiology laboratory at the Brigham & Women's Hospital. He has prior experience in building machine learning algorithms that provide robust antimicrobial stewardship. His unique background combined with the rich supporting environment of the Department of Population Medicine at Harvard Medical School and Harvard Pilgrim Health Care Institute, position him well for the transition to becoming an independently funded clinician- scientist working at the interface of infectious diseases and machine learning. He has assembled a multidisciplinary mentorship team consisting of experts in sepsis epidemiology, antimicrobial stewardship, implementation science, machine learning and causal inference to help him achieve his goals and has identified a comprehensive training plan that provides him the skills necessary to become a leader in his field. His short term goal is to become an expert in the development of machine learning algorithms that improve decision making for antibiotic resistant infections. His medium term goal is to deploy these models at scale and evaluate their real-world utility with randomized trials. The candidate's long term goal is to use these algorithms and the infrastructure necessary to maintain them as the technologic basis, of a learning health system that provides personalized decision support at the provider and public health level.