The combined use of natural language processing and electronic health records data to identify historical tolerances of β-lactams and promote clinician confidence in future use

We used natural language processing (NLP) to improve the utility of clinical decision support (CDS) β-lactam allergy alerts and promote informed allergy evaluation. NLP was performed on a corpus of clinical notes from hospital-based encounters to identify previous tolerance of β-lactam products using a rule-based approach. Historical tolerance of β-lactams was then combined with structured electronic health records data to produce improved CDS alerts. A survey was used to evaluate the utility of the improved alerts compared to standard allergy alerts. The rule-based pipeline identified previous β-lactam tolerance in between 3% and 28.4% of clinical notes and performed with high positive predictive value (83.6–97.6%) and recall (71.2−79.4%). The surveyed clinicians (<i>N</i> = 9) reported increased confidence in using β-lactam products despite the presence of a documented β-lactam allergy when using the information presented by the NLP-enriched CDS alerts, and all surveyed clinicians indicated the alerts would improve the care of their patients. NLP of clinical notes shows potential to improve the utility of CDS allergy alerts. Clinicians were receptive to allergy alerts containing NLP-derived information. Allergy-related CDS alerts should be improved to provide additional information such as historical tolerance of relevant products to empower providers to make informed decisions regarding patient allergies.

我们采用自然语言处理（Natural Language Processing，NLP）技术，以提升临床决策支持（Clinical Decision Support，CDS）β-内酰胺类过敏警报的实用性，并促进知情过敏评估工作。本研究针对医院就诊产生的临床笔记语料库，采用基于规则的方法，识别其中提及的β-内酰胺类药物既往耐受情况。随后将β-内酰胺类药物既往耐受信息与结构化电子健康记录数据相结合，生成优化后的临床决策支持警报。本研究通过问卷调查，对比评估了优化后警报与标准过敏警报的实用性。该基于规则的流程可在3%至28.4%的临床笔记中识别出β-内酰胺类药物既往耐受记录，且具备较高的阳性预测值（83.6%~97.6%）与召回率（71.2%~79.4%）。受访的9名临床医师报告称，在使用经NLP增强的临床决策支持警报所提供的信息时，即便患者存在已记录的β-内酰胺类过敏史，他们在使用该类药物时的信心也有所提升；所有受访医师均表示，此类警报可改善患者的诊疗工作。临床笔记的自然语言处理技术展现出提升过敏相关临床决策支持警报实用性的潜力。临床医师对包含NLP衍生信息的过敏警报接受度良好。应优化过敏相关的临床决策支持警报，补充如相关药物既往耐受情况等额外信息，以帮助临床医师针对患者过敏情况做出知情决策。