A Novel Probabilistic Methodology for Prediction of Emerging Diseases in Patients with Multiple Chronic Conditions

 描述(由申请人提供)：治疗患有多种慢性病(MCC)的人目前估计占全国医疗保健费用的66%，并将继续增长。这一日益严峻的挑战已经成为一个重大的公共卫生问题，因为MCC与不理想的健康结果和不断上涨的医疗保健成本有关。然而，它现在知道了MCC是如何在个人或普通人群中出现的。传统的流行病学方法导致了疾病联系和共病关联的重要发现。然而，它们在描述患者如何获得新的慢性病以及预测/个性化个别患者出现MCC方面的能力有限。目的：这项研究将开发可用于确定人群中最可能的共病组合的方法。这些方法可以量身定做，以检查特定亚群或单个患者水平的MCC模式。我们还将研究大量风险因素对MCC组合出现的影响。此外，我们将使用数据挖掘方法在人口和个人层面预测和监测MCC组合的发展。假设：我们假设MCC患者的合并症的出现和进展形成了可以预测的模式，并与既往的医疗条件、人口统计学和社会经济特征有关。我们进一步假设，这些方法可以通过个性化个别患者的记录，更准确地预测新的慢性病的时间和出现。目标和方法：在Aim1中，我们将描述MCC是如何以不同的模式出现和发展的，以及这些模式如何随着时间的推移在不同的疾病组合之间转移。然后，我们将使用马尔可夫聚类(MCL)算法识别主要的MCC转换并对其进行分组，该算法是一种处理大数据的新颖、高效的图形方法。在目标2中，我们将确定哪一个 风险因素与MCC的出现有关，使用机器学习方法可以处理共病模式的复杂异质性。危险因素包括年龄、性别、种族/民族、教育程度、经济状况、婚姻状况和既往的医疗状况。在目标3中，我们将使用相似学习方法来开发能够预测MCC是否会出现在个体患者或人群中的模型，并且我们将能够使用这些模型来监测随着时间的推移MCC的出现。结论：我们的发现将为未来的研究提供基础，该研究将评估与MCC模式进展相关的特定治疗模式，并最终确定对患有或有多种慢性病风险的人进行干预的最佳时间点。这些调查结果还将提供可在社区一级使用的信息，以管理卫生保健资源，以改善护理的连续性和可及性。

 DESCRIPTION (provided by applicant): Treatment for people living with multiple chronic conditions (MCC) currently accounts for an estimated 66 percent of the Nation's health care costs and will continue to grow. This mounting challenge has become a major public health issue since MCC is linked to suboptimal health outcomes and rising health care costs. However, it now known how MCC emerge among individuals or in the general population. Traditional epidemiological approaches have led to important findings of disease links and comorbidity associations. However, they are limited in their ability to characterize how patients acquire new chronic conditions and predict/personalize the emergence of MCC for individual patients. Objectives: This study will develop approaches that can be used to identify the most likely combinations of comorbidity within a population. These approaches can be tailored to examine MCC patterns in specific sub- populations or at the level of the individual patient. We will also study the effect of a large set of risk factors on MCC combinations emergence. Furthermore, we will use data mining approaches to predict and monitor the development of MCC combinations in at the population and individual levels. Hypotheses: We hypothesize that the emergence and progression of comorbidities in MCC patients form patterns that can be predicted and which are associated with prior medical conditions, demographic, and socio-economic characteristics. We further hypothesize that these methods can predict the timing and emergence of new chronic diseases more accurately by personalizing the records for individual patients. Aims and methodology: in Aim1, we will characterize how MCC emerge and progress in distinct patterns and how these patterns transition between different combinations of diseases over time. We will then identify and group major MCC transitions using the Markov clustering (MCL) algorithm, which is a novel, efficient graphical approach to handle big data. In Aim 2, we will identify which risk factors are associated with MCC emergence using a machine learning approach that can handle the complex heterogeneity of comorbidity patterns. Risk factors include age, sex, race/ethnicity, education, economic status, marital status, and prior medical conditions. In Aim 3, we will use a similarity learning approach to develop models that can predict if MCC will emerge in individual patients or among populations and we will be able to use these models to monitor MCC emergence over time. Conclusion: Our findings will provide a foundation for future research that will evaluate specific treatment patterns associated with progression in MCC patterns and ultimately identify optimal time points of intervention for those with, or at risk for multiple chronic conditions. These findings will also provide information that can be used at the community level to manage healthcare resources to improve continuity and accessibility of care.