Big data analysis of electronic hospital records: inpatient trajectories and pharmacological exposures associated with mortality in older adults.

英国国家医疗服务体系(NHS)医院的入院人数在过去十年里增加了28%，近一半入院的成年人年龄为65岁。一些老年人衰老得很厉害，但另一些人则变得虚弱，这种情况的特点是抵御疾病等压力因素的能力降低。此外，大约三分之一的成年住院患者有一种或多种慢性健康状况，许多患者被开出多种长期药物，即所谓的多药联用。我们的目标是了解老年人如何使用紧急医院服务，以及从入院到出院的医院旅程。我们还将探索与住院死亡等不良住院结果相关的因素，特别是与处方药有关的因素。世界卫生组织在2017年宣布了第三次全球患者安全挑战，并由卫生政策研究计划部委托发布的一份报告估计，用药失误每年给NHS造成9850万GB的损失。虽然药物相关的危害得到了广泛的认识，但需要更多的信息来说明哪些药物具有最高的风险，哪些患者需要告知更安全的处方实践。我们将使用四年来NHS一家三级医院(剑桥阿登布鲁克医院)急诊住院的8万例老年人(65岁)的数据。在2014年引入电子病历系统后，数据可用于大规模回顾分析。提供了描述入院各方面的信息，从患者的特征，如年龄组，到与床边观察、处方药和血液测试有关的信息。这些数据在入院期间将有许多重复测量，例如每天测量几次血压，导致非常大和复杂的数据集。因此，匿名的患者记录将被转移到欧洲生物信息学研究所(EBI；剑桥欣克斯顿的惠康基因组校园)。EBI是一家领先的研究机构，专注于开发尖端技术来处理和管理“大数据”。我们将使用机器学习(ML)，这是一种能够可视化复杂数据中的模式的人工智能，来探索我们数据集中的数千个患者实例。我们将首先定义多少种不同的住院类型，描述大多数老年人的住院情况，并描述这些住院患者的轨迹。例如，入院期间的特点可能是它们的长度(短与长)或医院操作因素，如病房搬家的数量。其次，我们将使用ML来研究不同的处方药物或药物组合如何代表一种与住院患者死亡持续相关的模式。我们可以利用已知的关联，例如血液稀释剂的使用和较高的出血死亡可能性，来教育ML过程。然后，ML可以确定与住院死亡相关的其他处方模式，并探索某些患者特征或入院类型是否会使患者更容易受到伤害。这将建立一个影响最终住院结果的患者、治疗和医院因素的综合图景。住院死亡是我们的主要结果，但也可以考虑其他结果，如出院后重新进入疗养院。最后，ML可以模拟如果采用假设的替代治疗计划，医院结果可能会如何改变。例如，可以用另一种治疗方法取代药物，看看这将如何改变死亡发生的可能性。这项研究将描述老年人使用急性医院服务的情况，并确定潜在的不适当药物供进一步研究。东北-纽卡斯尔和北泰恩赛德研究伦理服务委员会批准了这项研究。

Admissions to National Health Service (NHS) hospitals in England have increased by 28% over the last decade and nearly half of all adults admitted are >=65 years old. Some older adults age robustly but others develop frailty, a condition characterised by reduced ability to withstand stressors such as illness. In addition, approximately a third of adult inpatients have one or more chronic health conditions, and many patients are prescribed multiple long term medications, so called 'polypharmacy'. We aim to understand how older adults use emergency hospital services and journey through the hospital from admission to discharge. We will also explore factors, particularly related to prescription medications, associated with poor hospital outcomes such as inpatient death. The World Health Organisation declared 'Medication without harm' its 3rd global patient safety challenge in 2017 and a report commissioned by the Department of Health Policy Research Programme estimated medication errors cost the NHS £98.5 million per year. Whilst medication related harm is widely recognised, more information is needed about which drugs confer the highest risk in which patients to inform safer prescribing practices. We will use data from 80 000 inpatient episodes of older adults (>/=65 years old) admitted as an emergency to one tertiary NHS hospital (Addenbrooke's Hospital, Cambridge) over four years. Data is available for large scale retrospective analysis after an electronic patient record system was introduced in 2014. Information describing all aspects of admission from patient characteristics such as age group to information pertaining to bedside observations, prescription medications and blood tests are available. These data will have many repeated measurements over the admission duration, for example blood pressure measurements taken several times each day, leading to a very large and complex dataset. Therefore, anonymised patient records will be transferred to the European Bioinformatics Institute (EBI; Wellcome Genome Campus, Hinxton, Cambridge). EBI is a leading research institution focused on developing cutting-edge technologies to process and manage 'big' data. We will employ machine learning (ML), a type of artificial intelligence capable of visualising patterns within complex data, to explore the thousands of patient examples in our dataset. We will firstly define how many different types of hospital admission describe the majority of admissions in older adults and characterise these inpatient trajectories. For example, admission episodes may be characterised by their length (short versus prolonged) or hospital operational factors such as number of ward moves. Secondly, we will use ML to study how different prescribed medications, or combinations of medications, represent a pattern that is consistently associated with inpatient death. We can use known associations, such as the use of blood thinning medications and higher likelihood of death from bleeding, to educate the ML process. ML can then identify other prescribing patterns associated with inpatient death and explore whether certain patient characteristics or types of admission make patients more vulnerable. This will build a comprehensive picture of patient, treatment and hospital factors that impact on the eventual hospital outcome. Inpatient death is our primary outcome but other outcomes such as new admission to a care home following discharge can be considered. Finally, ML can simulate how the hospital outcome might change if a hypothetical alternative treatment plan was employed. For example, medications can be substituted with an alternative treatment to see how this would change the likelihood of death occurring. This research will describe use of acute hospital services by older adults and identify potentially inappropriate medications for further study. The Northeast-Newcastle & North Tyneside research ethics service committee approved the study.