Using NLP to Target Patient Led and Scalable Primary Care Interactions

全科医生士气低落，全科医生人数减少，需求不断增加，工作量不断增加，导致许多人对全科医生的“危机”发出警告\[1\]。在线咨询（OC）系统是解决方案的一部分。它们帮助全科医生更有效地管理病人的请求和工作量，同时让病人无需等待电话就能联系他们的全科医生。使用自由文本界面的系统允许患者用自己的话描述他们的需求，从而提高患者对\[2\]的理解和体验。然而，他们要求临床工作人员手动对请求进行分类和编码，这在英格兰每年带来了大约2.4亿英镑的时间和资源负担\[3\]。这意味着临床工作人员花在行政管理上的时间可以更好地花在其他地方，特别是如果由于教育或语言障碍而要求不充分描述。此外，OC捕获当前患者的症状，这意味着全科医生必须参考患者的纵向临床病史，以获取背景信息，这在已经很短的会诊中可能会耗费时间。自然语言处理（NLP）建立在人工智能的基础上，以实现文本理解和摘要。自然语言处理的进步使得训练计算机模型能够以超越人类能力的速度和数量处理文本。在这个项目中，我们将利用我们内部的NLP学术专家，通过开发一个名为ASPIRE的工具来显著改善自由文本OC，该工具可以自动对\[4\]患者的自由文本请求进行分类和编码，并向全科医生提供临床总结和建议。重要的是，该工具将以病人为导向，个性化。这意味着ASPIRE将利用患者的自由文本请求和他们的个人健康记录（PHR）的数据；这是最全面的健康记录，包括病史（全科医生记录）数据和患者报告的环境数据（例如非处方药物、过敏、自我报告的健康结果）。一般实践将受益于准确性的提高和临床编码负担的减轻。全科医生将受益于相关数据的自动编码和标记，从而做出更明智的决定。患者将受益于更个性化的护理决定——在他们的PHR背景下做出的决定。\[1\] ＿General＿Practice＿In＿England＿[https://journals.lww.com/ambulatorycaremanagement/Abstract/2022/04000/General＿Practice＿in＿England＿＿The＿Current＿Crisis,.7.aspx][0] \[2\] ＿access＿to＿and＿delivery＿of＿general＿practice＿services＿ https://www.health.org.uk/publications/access-to-and-delivery-of-general-practice-services \[3\] ＿Calculation＿based＿on＿ 30seconds＿GP＿coding＿time＿required＿per＿consultation,＿at＿an＿equivalent＿cost＿of＿£80/hr。3.67亿英镑，相当于2.4亿英镑。\[4\] ＿临床“编码”在这种情况下描述了在电子健康记录中使用结构化临床词汇的使用，以确保护理信息被清晰记录、一致和全面。[0]: https://journals.lww.com/ambulatorycaremanagement/Abstract/2022/04000/General＿Practice＿in＿England＿＿The＿Current＿Crisis,.7.aspx

Collapsing morale among general practitioners (GPs), a shrinking GP workforce, relentless demands, and increasing workload have caused many to sound alarm on a general practice in "crisis" \[1\].Online Consultation (OC) systems are part of the solution. They help GPs to manage patient requests and workload more effectively while allowing patients to contact their GP without waiting on the phone. Systems that use a free-text interface improve patient uptake\[2\] and experience by allowing patients to describe their needs in their own words.They, however, require clinical staff to manually sort and code requests introducing a time and resource burden of around £240Million/year in England\[3\]. This means clinical staff spend time on administration that could be better spent elsewhere, especially if requests are insufficiently descriptive due to education or language barriers. In addition, OC captures current patient symptoms meaning the GP must refer to the patient's longitudinal clinical history, for context, which can be time consuming in an already short consultation.Natural language processing (NLP) builds on artificial intelligence to enable text understanding and summarisation. Advances in NLP allow training computer models that process text at a speed and volume way beyond the capability of humans.In this project we will utilise our in-house academic expert in NLP to significantly improve free-text OC by developing a tool named ASPIRE, that can automatically categorise and code\[4\] free-text requests from patients and provide a clinical summary and recommendation to GPs.Importantly, the tool will be patient-led and personalised. Meaning ASPIRE will utilise data from both the patient's free-text request and their personal health record (PHR); this is the most comprehensive health record, including both medical history (GP record) data and patient-reported environmental data (e.g. over-the-counter medication, allergies, self-reported health outcomes).General practice will benefit from improved accuracy and reduced burden of clinical coding. GPs will benefit from support to make better informed decisions from the automatic coding and flagging of relevant data. Patients will benefit from more personalised care-decisions made within the context of their PHR.\[1\]\_General\_Practice\_In\_England\_[https://journals.lww.com/ambulatorycaremanagement/Abstract/2022/04000/General\_Practice\_in\_England\_\_The\_Current\_Crisis,.7.aspx][0]\[2\]\_ Access\_to\_and\_delivery\_of\_general\_practice\_services\_https://www.health.org.uk/publications/access-to-and-delivery-of-general-practice-services\[3\]\_Calculation\_based\_on\_~30seconds\_GP\_coding\_time\_required\_per\_consultation,\_at\_an\_equivalent\_cost\_of\_£80/hr.~367\_million\_GP\_appointments\_in\_2021\_equates\_to\_~£240million.\[4\]\_clinical 'coding' in this context describes the use of structured clinical vocabulary for use in an electronic health record to ensure care information is clearly recorded, consistent, and comprehensive.[0]: https://journals.lww.com/ambulatorycaremanagement/Abstract/2022/04000/General_Practice_in_England__The_Current_Crisis,.7.aspx