Demonstrating the efficacy at scale of a novel data-protecting AI system for surfacing clinically-eligible patients for clinical trials

**问题****临床试验参与者的识别是极度手工的，耗时且经常失败。因此，这是众所周知的医疗保健研发瓶颈。****临床招募是一个双重过程：患者识别，其次是筛选**。为了识别具有特定临床表现和属性的患者，制药公司通常将其外包给合同研究组织（CRO）或参与者识别中心（PICs）。cro广泛做广告，手动搜索记录，并聘请临床专家手动评估潜在参与者的适合程度。PICs和医院有效地等待符合条件的患者参加他们的咨询，因此纳入试验往往被遗忘！因此，使用这些低效的方法进行患者识别需要几个月或几年的时间。**新技术**过去十年，机器学习和人工智能技术在全社会的应用呈爆炸式增长。在医疗保健领域，这些技术具有显著提高患者护理质量和成本效益的潜力。**这一潜力有限，因为医疗数据隐私在传统机器学习中受到损害**。人工智能的一个新发展是联邦机器执行。算法被发送到数据中，而不是将数据移动到机器学习模型的中心位置进行应用或训练。对于医院来说，这确保了数据的隐私性，并提供了对其使用情况的控制和完整审计跟踪。**医院的数据永远不会离开其安全环境**。**该项目**该项目使用**Bitfount有限公司的联合机器执行平台**，使**Moorfields眼科医院的** **尖端软件模型** （ai -生物标志物）能够扫描数百万张医学图像，并从这一主要、客观的来源**快速、准确地识别潜在的试验参与者**，将患者数据和临床合格试验候选人的综合列表留在医院数据库中。我们的方法有望通过快速、准确、廉价的识别，在数据不离开医院的情况下，大规模增加适合筛查和试验招募的临床合格参与者的数量。它的效率和功效将通过人工智能临床试验专家萨里CTU**的一项试验内研究（SWAT）来衡量。该项目将展示干性年龄相关性黄斑变性（AMD）的方法，该方法影响了大约5%的65岁及以上的欧洲人（39）。眼科是国民医疗服务体系中门诊出勤率最高的，黄斑变性是目前英国致盲的主要原因，随着人口老龄化而增加。该项目有助于**NHSX的**眼科数字化转型，**BEIS的人工智能和数据大挑战**(8)，**DCMS的官方技术优先事项**(9)，**英国国家人工智能战略**(10)和**生命科学产业战略**(11)。

**Issue****Clinical trial participant identification is desperately manual, time-consuming and frequently fails. As such, it is a well-known bottle-neck to healthcare R&D.****Clinical recruitment is a two-fold process: patient identification, followed by screening**. In order to identify patients with specific clinical presentation and attributes, pharmaceutical companies typically outsource to Contract Research Organisations (CRO) or Participant Identification Centres (PICs). CROs advertise widely, trawl through records manually and pay clinical experts to manually assess potential participant's fit. PICs and hospitals effectively wait for eligible patients to attend their consultations, whereupon inclusion in the trial is often forgotten! Consequently, patient identification using these inefficient methods takes months or years.**New Technologies**The last decade has seen an explosion in the use of machine learning and AI technologies across society. In the healthcare sector, these techniques have the potential to dramatically improve the quality and cost effectiveness of patient care. **This potential has been limited because healthcare data privacy is compromised in traditional machine learning**.A new development in AI is **federated-machine-execution**. Instead of moving data to a central location for machine learning models to be applied or trained, algorithms are sent to the data. For a hospital, this ensures the privacy of the data and provides control and a complete audit trail of its usage. **The hospital's data never leaves its secure environment**.**This project**This project uses **Bitfount Ltd's federated-machine-execution platform** to enable **Moorfields Eye Hospital's** **cutting-edge software models** (AI-biomarkers) to scan millions of medical images and identify potential trial participants from this primary, objective source **swiftly and accurately**, leaving patient data and comprehensive lists of clinically-eligible trial candidates inside the hospital database. Our approach promises to massively increase the number of clinically-eligible participants suitable for screening and trial recruitment through **fast, accurate, cheap identification, without data leaving the hospital**. Its efficiency and efficacy will be measured via a Study Within a Trial (SWAT) by **Surrey CTU**, experts in AI Clinical Trials.This project will demonstrate the approach in dry Age-related Macular Degeneration(AMD), which affects approximately 5% of Europeans aged 65 and over(39). Ophthalmology has the NHS' highest outpatient attendance rate(6) and AMD is by far the UK's leading cause of blindness(7), increasing with an ageing population. This project contributes to **NHSX's** digital transformation of ophthalmology, **BEIS' Grand Challenge for AI and data**(8), **DCMS' official Tech Priorities**(9), **UK's National AI Strateg**y(10) and **Life Sciences Industrial Strategy**(11).