Modeling Therapy Sequence for Advanced Cancer: A Microsimulation Approach Using Real-World Data

项目摘要/摘要 在现代晚期癌症治疗中，许多患者使用不同的方法进行多轮治疗。 药剂。虽然治疗方法的转换在临床实践中很常见，但 治疗的顺序还在研究之中。多条治疗路线的有效和准确的建模仍然存在 一个悬而未决的问题。通常使用的马尔可夫模型是无记忆的，这在研究多个 治疗路线，因为该模型不能轻易地处理患者内部的依赖。因此，有一个未满足的 而且越来越需要开发专门的模型来评估治疗序列(即 是，指定了哪些代理以及按什么顺序提供代理)。从电子版中提取真实的患者信息 病历(EMR)为在实践中更好地了解治疗结果提供了一个新的机会。我们 可以使用电子病历数据创建改进的健康状态模型，该模型综合了发布的患者级别数据 来自临床试验的结果，以及来自文献的健康效用(生活质量)估计。在此，我们建议 通过在微观模拟模型中结合真实世界的数据来研究治疗顺序。这些状态转换 模型使用蒙特卡罗方法来模拟个体通过不同健康状态的路径。我们的模特会 尤其注重正确地解释患者体内随时间发生的事件的相关性。患者级别 变量，如肿瘤生物学、年龄和合并症，将影响所有系列的治疗结果。 治疗，从而引入统计依赖。我们将首先估计状态之间的转移概率 使用几个参数模型来拟合基于电子病历的数据。然后，这些转移概率将用于 微模拟模型，还包括医疗费用和生活质量等辅助信息 调整(实用程序)。我们还将开发一种完全非参数的方法，使用 患者群体，其中每个人通过模型状态的路径将被直接使用。同样，成本和 每个健康州的公用事业都将纳入其中。在模拟研究中，我们将评估每种方法的性能 模型与传统的马尔可夫模型方法进行了比较。最后，我们将说明这一点的应用 方法在晚期尿路上皮癌研究中使用由Flatiron Health创建的来自EMR的数据集。 该数据集包含关于治疗和结果的大量数据，包括进展事件和死亡， 然而，它缺乏我们进行全面成本效益分析所需的所有措施(即财政 成本和生活质量没有包括在内)。我们将演示如何使用我们的方法来估计 基于卡铂的一线治疗(与基于顺铂的一线治疗相比)的成本-效果 二线免疫检查点抑制剂，我们将比较不同策略的结果。

Project Summary/Abstract In modern treatment of advanced cancers, many patients go through multiple rounds of therapy using different pharmaceutical agents. Although therapy switching is common in clinical practice, the cost-effectiveness of sequences of therapies is under-studied. Effective and accurate modeling of multiple lines of therapy remains an open problem. The often-used Markov model is memoryless, which is problematic when studying multiple lines of therapy, as the model cannot easily deal with dependence within patients. Therefore, there is an unmet and growing need to develop specialized models to evaluate the cost-effectiveness of therapy sequence (that is, which agents are given and in what order). Real-world patient information extracted from the Electronic Medical Record (EMR) provides a novel opportunity to better understand treatment outcomes in practice. We can use EMR data to create improved health-state models which synthesize patient-level data, published results from clinical trials, and health utility (quality of life) estimates from the literature. Here, we propose to study therapy sequence by incorporating real-world data in microsimulation models. These state-transition models use Monte Carlo methods to simulate individual paths through different health states. Our models will particularly focus on correctly accounting for the dependence of events over time within patients. Patient-level variables, such as tumor biology, age, and comorbid conditions, will impact treatment outcomes on all lines of therapies, thus introducing statistical dependence. We will first estimate transition probabilities between states using several parametric models fit to the EMR-based data. These transition probabilities will then be used in a microsimulation model which also includes auxiliary information including healthcare costs and quality of life adjustments (utilities). We will also develop a fully non-parametric approach using bootstrap resampling of the patient population, where each individual’s path through model states will be used directly. Again, costs and utilities will be incorporated for each health state. In simulation studies, we will assess the performance of each model compared to a traditional Markov model approach. Finally, we will illustrate the application of this method in a study of advanced urothelial carcinoma using an EMR-derived dataset created by Flatiron Health. This dataset contains extensive data on treatments and outcomes including progression events and death, however, it lacks all the measures we would need to conduct a full cost-effectiveness analysis (i.e. financial costs and quality of life are not captured). We will demonstrate how our approach can be used to estimate the cost-effectiveness of first-line carboplatin based therapy (vs. first-line cisplatin based therapy) followed by second-line immune checkpoint inhibitors, and we will compare results from the different strategies.