Flexible statistical models for dynamic prediction in survival analysis with time-varying prognostic factors

我研究的长期目标是开发新的、灵活的统计方法，以提高关于许多疾病的发生、进展、治疗和结果的临床研究的科学有效性。人类健康研究面临许多概念和方法上的挑战，而生物统计学的使命是阐述复杂的方法来应对这些挑战。 在接下来的5年里，我的研究将集中于改进现有的统计方法，以处理与a)风险和预后因素随时间的变化以及它们对健康结果的影响相关的复杂性，以及b)个别患者在其疾病演变过程中可能经历的大量间歇性和最终的临床结果。在临床实践中，患者和他们的医生必须了解随着时间的推移，生活方式、饮食和其他可改变的危险因素以及实验室测试和疾病进展的生物标志物的变化如何可能影响住院、受伤、疾病复发或死亡的风险。要弄清各种患者特征、生物标志物和治疗方法的影响，需要分析动态、纵向过程的几个方面，这一点很复杂。首先，虽然同时影响给定临床终点风险的许多因素是相互关联的。另一项挑战涉及风险因素和治疗的变化之间的复杂时间关系，包括过去治疗或风险因素值的延迟或累积影响，这也可能相互影响。例如，尽管抗高血压治疗会影响患者目前的血压，但决定是否开出、停止或改变这种治疗的剂量，可能反过来又取决于患者以前的血压及其最近的变化。此外，最终结果(如中风)可能会受到血压病史以及治疗的其他影响的影响，包括治疗的间接益处和可能的意外“副作用”(不良反应)。 为了应对这些挑战，我的研究将建立在统计理论和计算技术的最新进展以及我过去开发新方法分析健康结果纵向研究的经验的基础上。 通过我的大量临床合作，拟议的研究结果将有助于改善许多疾病的临床预后和治疗。

The long-term objective of my research is to develop new, flexible statistical methods that will enhance the scientific validity of clinical studies on the occurrence, progression, treatment and outcomes of many diseases. Studies of human health face many conceptual and methodological challenges, and the mission of biostatistics is to elaborate sophisticated methods to address these challenges. In the next 5 years, my research will focus on improving the existing statistical methods to handle complexities related to a) changes over time in risk and prognostic factors, as well as in their impact on the health outcomes, and b) multitude of intermittent and final clinical outcomes individual patients may experience during the evolution of their disease. In clinical practice, patients and their physicians have to understand how changes over time, in lifestyle, diet and other modifiable risk factors, as well as in the laboratory tests and biomarkers of disease progression, may affect the risk of hospitalizations, injuries, disease recurrence, or death. Disentangling the impact of various patient characteristics, biomarkers and treatments is complicated by several aspects of dynamic, longitudinal processes being analyzed. First, while many factors that affect simultaneously the risk of a given clinical endpoint, are correlated with each other. Another challenge is related to complex temporal relationships between changes in risk factors and treatments, including delayed or cumulative effects of past treatments or risk factor values, which may also affect each other. For example, whereas an anti-hypertensive treatment will affect the patient's current blood pressure, the decision to prescribe, discontinue or change the dose of such treatment may, in turn, depend on the patient's previous blood pressure and its recent changes. In addition, the final outcome (such as a stroke) may be affected by blood pressure history, as well as by the other effects of the treatment, including both indirect benefits and possible unintended “side effects” (adverse reactions) of the treatment. To address these challenges, my research will build on the recent progresses in both statistical theory and computational techniques, as well as on my past experience in developing new methods for analyzing longitudinal studies of health outcomes. Through my numerous clinical collaborations, the results of the proposed research will help improve clinical prognosis and treatment of many diseases.