Dynamic prediction of type 1 diabetes risk and autoantibody status by a joint model of longitudinal and multistate models

项目总结/摘要 1型糖尿病是一种慢性自身免疫性疾病，其特征是胰腺β细胞的破坏 导致胰岛素缺乏和每天注射胰岛素以维持生存。1型糖尿病的早期诊断 通过连续监测胰岛自身抗体状态和测量胰岛自身抗体水平的纵向标志物来实现。 免疫和代谢功能。该提案的目标是开发一个统计模型， 根据自身抗体状态和糖尿病患者的历史数据， 单独的.描述时变风险因素的纵向模型，预测 自身抗体状态和预测疾病进展的生存模型将在联合模型中组合 来实现这个目标。该模型将被应用到一个数据集来自环境的决定因素， 年轻人糖尿病（TEDDY）研究开发具有如此复杂结构的模型可能具有挑战性。 然而，统计方法和计算技术的进步为我们提供了机会， 来解决问题。在目标1中，我们将制定拟议的联合模型，并将其应用于TEDDY数据。 可以进行统计推断，以研究糖尿病相关自身抗体和其他 纵向危险因素与1型糖尿病诊断的风险相关。在目标2中，根据 提出的联合模型，一个动态的预测算法，预测自身抗体的发展 以及在给定个体的历史数据的情况下的随后的1型糖尿病的风险。在目标3中，我们将 使用各种诊断测量来评估所提出的动态预测算法的准确性。 我们预计，拟议的联合模型将表现出更好的性能比传统的静态 使用基线特征或最后可用测量的生存模型。拟议的研究可以 回答关于1型糖尿病自然史的关键研究问题，以及 纵向风险因素。

Project Summary/Abstract Type 1 diabetes is a chronic autoimmune disease that features the destruction of pancreatic beta-cells resulting in insulin deficiency and daily insulin injections for survival. Early identification of type 1 diabetes can be achieved by continuously monitoring islet autoantibody status and longitudinal markers that measure the immunological and metabolic functions. The goal of this proposal is to develop a statistical model that can give dynamic predictions about type 1 diabetes risk based on autoantibody status and the historical data of an individual. A longitudinal model for characterizing time-varying risk factors, a multistate model for predicting autoantibody status, and a survival model for predicting disease progression will be combined in a joint model to achieve the goal. The model will be applied to a dataset derived from The Environmental Determinants of Diabetes in the Young (TEDDY) study. It may be challenging to develop a model with such a complex structure. However, the advances in statistical methodology and computational technology have opened up opportunities to resolve the problems. In Aim 1, we will formulate the proposed joint model and apply it to the TEDDY data. Statistical inferences can be made to investigate how the changes in diabetes-related antoantibodies and other longitudinal risk factors are associated with the risk for type 1 diabetes diagnosis. In Aim 2, based on the proposed joint model, a dynamic prediction algorithm will be derived that predicts autoantibody development and the subsequent risk of type 1 diabetes given the historical data of an individual. Lastly, in Aim 3, we will evaluate the accuracy of the proposed dynamic prediction algorithm using a variety of diagnostic measures. We expect that the proposed joint model will demonstrate better performance than the conventional static survival models that use baseline characteristics or last available measurements. The proposed research can answer critical research questions about the natural history of type 1 diabetes and the relationship between longitudinal risk factors.