Exploring tooth survival using Bayesian spatial models

描述（由申请人提供）：使用贝叶斯空间模型探索牙齿存活率龋齿和严重的牙周病最终导致牙齿脱落，这仍然是美国主要的公共卫生负担。未来的牙科治疗计划将受益于先进的统计方法的发展，以整合有效的风险评估和牙齿脱落的短期预测。牙科数据集带来了许多有趣的统计挑战，这些挑战严重限制了现有方法的潜力。除了口内牙齿聚类，事件的时间是空间依赖性的，非平稳的（随牙齿位置而变化），并经历沉重的审查。这些因素也使临床结果的解释复杂化，这需要在条件（受试者水平）和边缘（人群）水平上进行解释。目前可用的统计方法可能会处理一些，但不是所有这些在一个统一的范式。目标：使用贝叶斯框架，拟议的研究将评估和监测感兴趣的人群的牙科疾病状态，并确定与牙齿脱落相关的协变量，从而实现有效的短期预测。主题：统计方法最初将在来自McGuire和纳恩数据的约100名牙齿受试者的数据集上进行评估，这些受试者在休斯顿地区的私人牙科诊所接受了约16年的监测。为了具有普遍性，这些方法将在一个为期4年的纵向数据库中进行测试，该数据库由Creighton大学收集的约16，500名患者组成。研究设计：一项具有事件发生时间终点的聚类纵向研究设计包括记录年龄、性别、种族、完整的修复和牙周记录以及随访、吸烟状况、糖尿病状况、口腔卫生和其他基本参数的数据库。重要性：目前的项目将提供新的知识来解开决定牙齿脱落的复杂的协变量-响应关系，并且可以很容易地推广到其他牙科数据集。长期目标是能够实现对牙齿存活的准确预测推断，使牙科医生能够制定具有成本效益的牙科治疗计划。

DESCRIPTION (provided by applicant): Exploring tooth survival using Bayesian spatial models Caries and severe periodontal disease eventually lead to tooth loss, and this remains a major public health burden in the US. Future dental treatment plans will benefit from development of advanced statistical methods to integrate efficient risk assessment and short-term prediction of tooth loss. Dental datasets come with many interesting statistical challenges which severely limit the potential of currently available methods. In addition to tooth-within-mouth clustering, the times to events are spatially dependent, non-stationary (varying with tooth-locations), and experience heavy censoring. These factors also complicate the interpretation of clinical findings, which are needed at the conditional (subject-level) and the marginal (population) levels. Currently available statistical methods might handle some, but not all of these within an unified paradigm. Goals: Using a Bayesian framework, the proposed study will assess and monitor dental disease status of a population of interest and identify covariates associated with tooth- loss leading to efficient short-term prediction. Subjects: The statistical methods will be initially evaluated on a dataset of about 100 dentate subjects from the McGuire and Nunn data who were monitored at a private dental practice in the Houston area for about 16 years. For generalizability, the methods will be tested on a 4-year longitudinal database consisting of about 16,500 patients collected at Creighton University. Study design: A clustered-longitudinal study design with time to event endpoint comprises the databases that recorded age, gender, race, complete restorative and periodontal records with follow-up, smoking status, diabetes status, oral hygiene, and other essential parameters. Significance: The current project will provide new knowledge to unravel the complex covariate-response relationship that determines tooth loss, and can be easily generalized to other dental datasets. The long-term goal is to be able to achieve accurate predictive inference on tooth survival enabling dental practitioners to develop cost-effective dental treatment plans.