Developing, translating and evaluating risk growth charts for chronic diseases and multimorbidities using population-wide electronic health records

多病是指存在两种或两种以上的慢性健康状况。多病在老年人中尤为重要：据估计，80岁以上的成年人中有80%患有两种或两种以上的慢性病，通常包括一种或多种心血管疾病（CVD）。拟议的研究项目将旨在开发风险增长图表，模拟发展疾病的动态风险，以使用全民电子健康记录（EHRs）预测多种疾病。通过利用数百万参与者的完整病史，我们将能够监测和计算他们在整个患者生命过程中患上一种或多种慢性疾病的风险。电子病历是一种具有独特优势的数据类型。随着初级保健的现代化和数字化，电子病历变得更加容易获得，并定期添加最新的新数据。此外，电子病历包括对临床特征的重复测量，如血压、低密度脂蛋白和高密度脂蛋白胆固醇的测量，这可以用来预测“估计的当前值”，这被认为比单一的危险因素测量更不容易出错。此外，有了完整的病史，我们就能观察到病人健康的总体趋势。通过使用人口健康数据和电子健康记录，包括流行病学危险因素的重复测量，我们旨在建立多病风险预测模型。不同的慢性疾病相互联系的方式将通过观察多种疾病的重叠及其在人群中的严重程度来调查。数据将来自英国生物银行和NHS数字。CVD-COVID- uk联盟还可用于探索COVID严重程度作为CVD危险因素的相关性，以及它与已知CVD危险因素的关系。心血管疾病将在个体尺度（例如仅调查心肌梗死）和疾病组尺度上同时调查多个心血管疾病（例如调查心肌梗死和中风）。各种统计方法将用于开发风险预测模型，包括回归，聚类和机器学习技术。博士培训将包括参加剑桥大学的机器学习和人工智能讲座，以及参加基尔大学为期三天的课程：“风险预测和预后模型的统计方法”。在项目的后期阶段，我们计划探索将电子病历和基因组数据以多基因风险评分的形式结合起来预测心血管疾病风险的未来前景。未来，基于电子病历的多病风险预测模型可用于临床实践，通过考虑患者的全部病史来监测患者及其慢性病风险。能够同时确认多种慢性疾病将使初级保健从以疾病为中心的方法转变为将患者及其健康作为一个整体来考虑。风险预测模型将帮助临床医生发现慢性病风险增加的早期迹象，确定哪些患者应该接受进一步的风险评估或筛查，并鼓励高风险人群改变生活方式。

Multimorbidity describes the presence of two or more chronic health conditions. Multimorbidities are particularly relevant in the context of older adults: it is estimated that 80% of adults over the age of 80 have two or more chronic conditions, often including one or more cardiovascular diseases (CVD). The proposed research project will aim to develop risk growth charts modelling dynamic risk of developing diseases to predict multimorbidities using population-wide electronic health records (EHRs). By leveraging full medical histories of millions of participants we will be able monitor and calculate their risk of developing one or multiple chronic diseases across the patient life course. EHRs are a type of data with unique advantages. With the modernisation and digitalisation of primary care, EHRs have become more readily available, with regular up-to-date addition of new data.Furthermore, EHRs include repeated measurements of clinical features, such as blood pressure and LDL and HDL cholesterol measurements, which can be used to predict an "estimated current value" which is thought to be a less error-prone than single measures of risk factors. Moreover, the availability of a full medical history allows us to observe the general trend of a patient's health. By using population health data and electronic health records, including the repeated measurements of epidemiological risk factors, we aim to develop multimorbidities risk prediction models. The way different chronic conditions are interconnected will be investigated by observing at the overlap of multiple diseases and their severity in the population. Data will be sourced from the UK Biobank and NHS Digital. The CVD-COVID-UK Consortium can also be used to explore the relevance of COVID severity as a risk factor for CVD and how it is associated with known risk factors of CVD. CVD will be considered on both an individual scale (for example just investigating myocardial infarction) and on a disease group scale investigating multiple CVDs at a time (for example investigating myocardial infarction and strokes).Various statistical approaches will be used to develop risk prediction models including regression, clustering and machine learning techniques. PhD training will include attending Machine Learning and Artificial Intelligence lectures from the University of Cambridge and attending a three-day course from Keele University: "Statistical Methods for Risk Prediction and Prognostic Models". Towards later stages of the project, we plan to explore future prospects of combining EHRs and genomic data, in the form of polygenic risk scores, to predict cardiovascular disease risk.In the future, multimorbidities risk prediction models based on EHRs can be used in clinical practices to monitor patients and their chronic diseases risk by considering their full medical history. Being able to acknowledge multiple chronic conditions at once will shift primary care from a disease-focused approach to considering the patient and their health as a whole. Risk prediction models will assist clinicians in detecting early signs of increased chronic diseases risk, prioritise which patients should undergo further risk assessments or screening and encourage lifestyle changes in populations at higher risk.