Analysing and recommending personalised care pathways for multimorbid patients with the use of artificial intelligence (AI) techniques

背景医疗保健系统无法科普管理多病患者的多种健康状况的复杂需求。他们的病情由专业医疗保健专业人员单独治疗，没有目标和治疗计划的协调，这往往导致护理和医源性的重复。数字化工具被认为是实现以人为本、主动和综合护理的基础，但大多数实施都是孤立研究的，缺乏工作流程优化。与此同时，人们对在医疗保健中使用流程挖掘（PM）（一种流程分析）的兴趣越来越大。这是因为人们对如何提供护理知之甚少，特别是对患有并存疾病的患者，他们的护理往往是分散的。事实上，有越来越多的文献支持使用PM在改善患者的治疗效果，降低成本和等待时间guidelines.In为了处理的问题，护理碎片和相互冲突的指导方针，在multimorphies研究人员开发的方法来识别和解决不利的相互作用，如约束逻辑编程和分析计算机可解释的临床指南。然而，这些几乎不能说明护理是如何提供的，也不能确定新的、新兴的最佳实践模式。另一方面，尽管PM成为研究现有护理途径的常用方法，但只有13%的论文提出了一种新的解决方案，据我们所知，还没有关于PM的研究发表在多媒体上。以人为本的综合护理模式，使用数据科学和人工智能技术为多病患者分析和推荐个性化护理途径。将使用一个决策支持工具来传播调查结果，该项目分为三个主要部分，重点领域日益扩大：糖尿病作为单一疾病，糖尿病及其相关并发症，如肾脏疾病，最后是糖尿病伴随不相关的合并症，如癌症。在每个范围部分，我们的目标是发现和分析护理途径，这将提供对如何提供护理以及与已发布的指南的一致性的深入了解。此外，它将使我们能够提出新的护理途径，重点关注患者进展，疾病结局和护理效果。这些建议将作为一种决策支持工具进行传播，这些工具将根据当前的医疗指南提出护理途径，或者当由于共存疾病的目标冲突而无法获得时，根据具有最佳结果的途径提出护理途径。这将包括事件发生时间分析以及预测模型，以估计相关的临床结果。随后，将通过识别瓶颈，发现和分析医疗保健流程是否符合已发布的指南和性能，同时正式验证模型的正确性。人工智能工具将用于对患者进行聚类和分类，从而识别和分析通路变异。为了为个性化护理计划提供建议，我们将使用人工智能方法对患者进行分组，以识别具有相似特征的亚群。在每个子集中具有最佳结果的护理路径以及临床指南将被用于推荐针对个人的新护理路径，然后这些路径将被正式验证。最后，研究结果将以交互式应用程序的形式作为决策支持工具进行传播。在输入患者特征后，它将以直观的格式向用户呈现不良事件预测风险的视觉表示。

BackgroundHealthcare systems are unable to cope with the complex need of managing multiple health conditions of multimorbid patients. Their conditions are treated individually by specialised healthcare professionals, without a coordination of goals and treatment plans, which often leads to duplications of care and iatrogenicity. Digital tools are being recognised as the foundation towards person-centered, proactive and integrated care, however most implementations have been studied in isolation with a lack of workflow optimisation.At the same time, there is a growing interest in the use of process mining (PM), a type of process analysis, in healthcare. This is because little is known about how care is being delivered, especially to patients with coexisting conditions whose care is often fragmented. In fact, there is a growing body of literature supporting the use of PM in improving patient outcomes and reducing cost and waiting times guidelines.In order to deal with the problem of care fragmentation and conflicting guidelines in multimorbidity researchers developed approaches to identify and address adverse interactions, such as constraint logic programming and analysis of computer interpretable clinical guidelines. These, however, can tell little about how care is being delivered and cannot identify new, emergent patterns of best practice. On the other hand, despite PM becoming a common approach to study existing care pathways, only 13% of papers propose a novel solution and, to the best of our knowledge, there has been no research published on PM in multimorbidity.AimsThe primary aim of the project is to contribute to the digitally supported, person centered and integrated model of care using data science and AI techniques to analyse and recommend personalised care pathways for multimorbid patients. A decision support tool will be used to disseminate the findings.The project is broken down into three main segments with increasingly broader area of focus: diabetes as single disease, diabetes and its related complications, such as renal disease, and finally diabetes accompanied by unrelated comorbidities, such as cancer.In each scope segment we aim to discover and analyse care pathways, which will provide insight into how care is delivered and into the conformance with published guidelines. Additionally, it will allow us to suggest novel care pathways with the focus on patient progression, disease outcome and care efficacy. The recommendations will be disseminated as a decision support tool that suggest care pathways based on current medical guidelines or, when these are not available due to conflicting targets of coexisting diseases, based on pathways with best outcomes.MethodologyDatasets will be explored in order to produce an overview of the population and the data. This would include time to event analysis as well as predictive modelling to estimate relevant clinical outcomes.Subsequently, healthcare processes will be discovered and analysed in terms of conformance with published guidelines and performance, by identification of bottlenecks, while the correctness of the model will be formally verified. AI tools will be used to cluster and classify patients, enabling the identification and analysis of pathway variants.In order to create recommendations for personalised care planning, we will use AI approaches to group patients to identify subpopulations with similar characteristics. Care pathways with best outcomes within each subset, as well as clinical guidelines, will be used to recommend novel care pathways, tailored to the individual, and these will then be formally verified.Finally, the findings will be disseminated as a decision support tool in the form of an interactive application. After inputting patient characteristics, it will present the user with a visual representation of predicted risk of adverse events in an intuitive format.