Cancer Emulation Analysis with Deep Neural Network

项目摘要 客观地量化治疗癌症的药物、设备和治疗程序的相对有效性 预后、严格设计和执行的随机临床试验(RCT)仍然是金标准。 然而，正如本申请和许多已发表的研究所例证的那样，随机对照试验并不总是可行的。 幸运的是，电子病历和保险索赔数据库的快速发展使其成为 有可能挖掘大量观测数据，并有效补充随机对照试验。这一战略一直是 得到多个国家组织的热情支持。在现有的观测数据分析中 旨在得出RCT类型结论的技术，仿真已经变得特别吸引人，它的试验-- 比如架构、可解释性和可伸缩性。它已被应用于多种癌症和其他复杂疾病 并导致了具有临床意义的发现。 这项研究有两个同样重要的目标。第一个目标是开发基于深度神经网络的 仿真分析方法和软件。现有的大多数仿真分析都是基于经典的 回归技术。与回归相比，DNN具有更好的模型拟合性和更高的灵活性。 最近，我们团队率先开发了一种基于DNN的仿真分析方法，并将其应用于 心血管疾病。在最近的成功基础上，我们将开发更多可解释的和更多的 为RCT分析量身定做的稳定DNN。然后我们将进一步扩大分析范围，并进行基于DNN的 对一系列模拟试验的分析。对于单个模拟试验和一系列试验，我们都将 开发有效的推理，这是RCT分析的关键，但在大多数DNN研究中被忽视。用户- 将开发友好的软件。这种方法论的发展将极大地扩大 仿真分析、深度学习、因果推理、观察数据分析和医疗记录/保险 索赔数据分析。第二个目标是开发和分析两个模拟试验。我们将解决 SEER-Medicare(A)肺叶切除术和有限切除术对肺癌生存率的比较 老年人群；(B)根治性前列腺癌切除术和局部前列腺癌生存观察 弗吉尼亚州人口。调查结果将得到全面和严格的评估。以提供更全面的 图片，我们还将使用多种替代方法进行分析，并与现有的RCT和 观察性研究。随着方法的重大进步和强大的数据，我们的分析将 导致更明确的发现，直接为临床实践提供信息，并作为未来应用的原型。

Project Summary To objectively quantify the relative effectiveness of drugs, devices, and treatment procedures on cancer prognosis, rigorously designed and executed randomized clinical trials (RCTs) remain the gold standard. However, as exemplified in this application and many published studies, RCTs are not always feasible. Fortunately, the fast development of electronic medical records and insurance claims databases has made it possible to mine a large amount of observational data and efficiently complement RCTs. This strategy has been enthusiastically endorsed by multiple national organizations. Among the available observational data analysis techniques that aim to draw RCT-type conclusions, emulation has emerged as especially appealing, with its trial- like architecture, interpretability, and scalability. It has been applied to multiple cancers and other complex diseases and led to clinically significant findings. This study has two equally important aims. The first aim is to develop deep neural network (DNN)-based emulation analysis methods and software. Most of the existing emulation analyses are based on classic regression techniques. Compared to regression, DNN excels with superior model fitting and higher flexibility. Recently, our group was the first to develop a DNN-based emulation analysis approach and applied it to cardiovascular diseases. Advancing from this recent success, we will develop more interpretable and more stable DNNs tailored to RCT analysis. We will then further expand the analysis scope and conduct DNN-based analysis of a sequence of emulated trials. For both a single emulated trial and a sequence of trials, we will develop valid inference, which is essential for RCT analysis but has been neglected in most DNN studies. User- friendly software will be developed. This methodological development will substantially expand the scope of emulation analysis, deep learning, causal inference, observation data analysis, and medical record/insurance claims data analysis. The second aim is to develop and analyze two emulated trials. We will address the comparative effectiveness of (a) lobectomy and limited resection on lung cancer survival for the SEER-Medicare elderly population, and (b) radical prostatectomy and observation on localized prostate cancer survival for the VA population. The findings will be comprehensively and rigorously evaluated. To provide a more comprehensive picture, we will also analyze using multiple alternative methods and compare against existing RCTs and observational studies. With the significant methodological advancements and powerful data, our analysis will lead to more definitive findings, directly inform clinical practice, and serve as the prototype for future applications.