Challenges of studying and predicting chronic kidney disease progression and its complications using routinely collected electronic healthcare records

背景：慢性肾脏疾病是一个世界性的公共卫生问题，在英国大约有5%-10%的成年人受到影响。这是一种进行性疾病，早期没有症状，但随着疾病的进展，发病率会增加，导致患者预后和生活质量较差。在无症状的患者中，可以通过血液和尿液测试来检测到它。慢性肾脏病的并发症包括心血管风险增加、死亡率、急性肾损伤和感染易感性增加。此外，少数慢性肾脏病患者进展为终末期肾脏疾病，需要常规透析或移植，这对公共卫生服务构成沉重的经济负担。如果治疗成功，慢性肾脏疾病可能会在多年内保持稳定。然而，在某些情况下，持续一段时间的肾功能迅速恶化可能会发生，称为进行性肾病。进行性肾病显然是一个重要的健康问题，对患者的福利构成高风险，并给公共卫生服务带来负担。需要进一步的研究来提高对进行性肾病的性质、负担和后果的认识。及早识别进展性肾病的高危患者并识别进展性CKD的亚型将允许更好的个性化护理，并将有助于资源管理，并可能对政策产生影响。在英国，在初级保健中管理的慢性肾脏疾病患者可以使用电子医疗记录(EHR)的大型数据集。除了捕获详细的患者特征外，还可以通过常规测试定期捕获有关肾功能的纵向数据，从而能够调查单个患者随时间推移的肾功能轨迹。然而，常规检查的频率因个人的全科医生而异，而检查往往发生在医疗团队最关心的人身上，这意味着对进展性疾病的检测是不同的。这就是所谓的测试偏差，即接受测试的人往往代表病情较重的人群，而没有接受测试的人代表没有参与医疗服务的病人和健康状况良好的人的混合。此外，那些患有快速进展的肾脏疾病的人被转介到二级专家护理，这意味着这些患者将从信息上从数据中消失。目的：这项研究项目将使用常规收集的电子医疗记录来研究慢性肾脏疾病的进展和并发症，重点是数据质量和适当的研究设计和方法来解决这些问题。主要目标：1)评估英国初级保健(国家慢性肾脏疾病审计数据库)常规收集的EHR中肾功能测试的数据完整性，以及数据质量问题对估计肾功能衰减率的准确性的影响2)全面回顾以前使用EHR研究慢性肾脏疾病进展的研究，特别是评估：报告数据完整性和/或丢失数据；用于克服丢失数据导致的问题的统计方法；以及作者如何在存在缺失数据的情况下解释研究结果3)调查全科医生(GP)实践行为的可变性(尤其是CKD在患者健康记录上的电子记录/编码)对已知与CKD进展相关的不良事件的影响(国家慢性肾脏病审计数据库)。[这项分析通过研究实践层面的暴露来处理患者层面的混淆。]4)利用CKD患者群体的EHR中常规可用的数据，开发不受肾功能测试行为影响的CKD进展预测模型。(此分析将使用Scream数据库，这是我们的Rese拥有的外部数据库

Background: Chronic kidney disease is a worldwide public health problem affecting approximately 5-10% of adults in the UK. It is a progressive disease that is asymptomatic in the early stages but causes increasing morbidity as the disease progresses, leading to poor patient outcomes and quality of life. It can be detected in asymptomatic patients using blood and urine tests. Complications of CKD include increased cardiovascular risk, mortality, acute kidney injury and increased susceptibility to infection. Further, a minority of CKD patients progress to end stage renal disease requiring routine dialysis or a transplant which presents a heavy economic burden to public health services.Chronic kidney disease may remain stable over many years, if managed successfully. However, in some cases, a rapid deterioration in kidney function sustained over a period of time can occur, termed progressive kidney disease. Progressive kidney disease is clearly an important health problem, posing a high risk to patient welfare and a burden to public health services. Further research is needed to improve knowledge of the nature, burden and consequences of progressive kidney disease. Early identification of patients at high risk of progressive kidney disease and identification of sub-phenotypes of progressive CKD would allow better individualised care and would be useful for resource management with potential implications for policy.Large datasets of electronic healthcare records (EHRs) are available for chronic kidney disease patients managed in primary care in the UK. As well as capturing detailed patient characteristics, longitudinal data on kidney function are captured regularly over time through routine testing, allowing investigations in to the trajectory of kidney function over time in individual patients. However, the frequency of routine testing varies by individual general practitioners, and testing tends to occur in those the medical team is most concerned about, which means that detection of progressive disease is varied. This is what is termed testing bias where those who are tested tend to represent a sicker group of people and those who do not have a test represent a mix of sick people not engaging with the health service and people who are in good health. Moreover, those who have fast progressive kidney disease are referred to secondary specialist care, which means that these patients will be informatively missing from the data. Aims: This research project will study chronic kidney disease progression and complications using routinely collected electronic healthcare records, focussing on data quality and appropriate study designs and methods to address these issues. Key objectives:1) Evaluate data completeness for renal function tests in routinely collected EHRs in UK primary care (National chronic kidney disease audit database) and the impact of data quality issues on accuracy of estimation of rate of decline in renal function2) Conduct a comprehensive review of previous research using EHRs to study the progression of chronic kidney disease, in particular evaluating: reporting of data completeness and/or missing data; statistical methods used to overcome issues caused by missing data; and how authors interpreted study results in the presence of missing data issues 3) Investigate the impact of variability in general practitioner (GP) practice behaviour (in particular, electronic recording/coding of CKD on the patient health record) on adverse events known to be associated with CKD progression (National chronic kidney disease audit database). [This analysis deals with patient-level confounding by studying exposure at the practice level.]4) Develop predictive models for CKD progression that is not biased by renal function testing behaviours, utilising data which is routinely available in EHRs in the population of CKD patients. (This analysis will use the SCREAM ddatabase, an external database owned by our rese