Application of Machine Learning to Personalised Medicine in Children

随着电子健康记录的发展，可以捕获更广泛、更完整的数据，例如连续的监测信号，计算方法分析患者并揭示更深层次的共性和差异的能力也在不断提高。该项目将研究机器学习方法，这些方法超出了传统患者特征的分析，例如既往诊断、体重、性别或已知风险因素的历史。大奥蒙德街医院--实际上是更广泛的医学界--一直在将曾经是分布式和非正式的数据存储系统集成到高度结构化和集中化的数据库中。这些具有明显的安全益处，但重要的是允许有效地访问数据以进行研究和部署辅助工具。这些现代化的电子健康记录（EHR）不仅包括疾病和治疗史，还包括手术室、主治医生以及值得注意的住院患者信号记录（如心率和血氧饱和度）等精细信息。与金标准随机对照试验（RCT）中的广泛队列相比，每个患者的这种连贯性图片，以及HER中包含的患者结局的整个历史，允许识别细粒度的患者队列。该项目将研究使用代表患者完整纵向病史的特征，包括诊断，干预和治疗。在创建了这些患者模型之后，该项目将进一步研究这些新的患者表示如何与机器学习方法和纵向GOSH EPR数据一起使用，以（1）发现患者群体中潜在的共性和差异，可以支持患者水平的预测并与临床结果相关联，以及（2）分析，了解并建议修改现有的治疗途径。成功的结果将是使用这些方法的辅助工具的内部部署和临床验证。该项目符合EPSRC的两个一般研究主题，即“工程”和“医疗保健技术”，也福尔斯人工智能战略研究领域的职权范围。该研究还将利用EPSRC命名的战略优先事项“工程和技术前沿”和“转变健康和医疗保健”。

As electronic health records progress to capture a wider and more complete range of data, such as continuous monitoring signals, so too does the ability of computational methods to analyse patients and uncover deeper reaching commonalities and differences.This project will research machine learning methods that extend beyond the analysis of traditional patient features, such as prior diagnoses, weight, sex, or their history of known risk-factors. Great Ormond Street Hospital - and indeed the broader medical community - has been integrating what were once distributed and informal data storage system into highly structured and centralised databases. These have obvious security benefits, but importantly allow efficient access to data for research and for the deployment of assistive tools. These modernised Electronic Health Records (EHRs) extend beyond disease and treatment history, including information as granular as operating theatres, presiding surgeons, and - of note - records of inpatient signals such as heart rate and oximetry. Such a cohesive picture of each patient, in tandem with the entire history of patient outcomes contained in the HER, permits the identification of fine-grained patient cohorts, in contrast to the broad cohorts in gold-standard randomised controlled trials (RCTs). This project will research the use of characteristic features representing a patient's full longitudinal history, including diagnoses, interventions and treatments. This will allow for a more comprehensive computational model of a patient and their needs.Having created these patient models, this project will further research how these novel patient representations can be used with machine learning methods and longitudinal GOSH EPR data to (1) find latent commonalities and differences in groups of patients, that can support patient-level predictions and associate with clinical outcomes, and (2) analyse, understand and suggest modifications to existing treatment pathways. A successful outcome would be the internal deployment and clinical validation of an assistive tool using these methods.This project aligns with two EPSRC general research themes, namely "Engineering" and "Healthcare technologies" and also falls within the remit of the strategic research area of Artificial Intelligence. The research will also tap into EPSRC named strategic priorities "Frontiers in Engineering and Technology" and "Transforming Health and Healthcare".