A New Class of Mechanistic Risk Prediction Models for Cancer Treatment Outcomes

描述（由申请人提供）：本研究的总体目标是开发一个统计框架，为构建一个智能临床支持系统奠定基础，预测不同前列腺癌治疗方案下的潜在结局。在这个建议中，结合统计方法与癌症治疗机制，我们建议开发一类新的统计回归模型，通过有效地考虑治疗特征和治疗后个体生物标志物的信息来预测肿瘤复发的概率和时间。新模型源自迭代的细胞出生和死亡过程，模拟治疗后肿瘤细胞的生物学机制，从而在统计模型构建和治疗结果预测中调用生物学考虑。从统计学上讲，新模型允许比例和非比例风险结构，纳入治愈率，并适应短期癌症复发预防和长期生化无病生存的非同质治疗效果。所提出的模型扩展了治愈率模型，允许更一般的依赖于个别协变量，它是半参数性质：非参数成分涉及癌症进展时间分布和参数成分涉及治疗特征，治疗后的生物标志物，和其他显着的协变量。我们提出了非参数平滑技术估计的进展时间分布，参数估计的似然和贝叶斯方法。该方法将应用于前列腺癌患者在罗切斯特大学医学中心积累的临床随访数据。该项目的新奇在于，新模型基本上是基于肿瘤对治疗反应的生物学机制，并利用统计建模技术进行风险预测。在这个R21应用程序中，我们的目标是使用统计技术开发新模型，如果成功，将在未来提交R01提案，以充分开发预测模型，并应用于构建和验证预测计算机支持系统，以帮助医生为适应性癌症治疗策略做出明智的临床决策。 受治疗后肿瘤发展的随机建模的启发，该项目提出开发一类新的统计回归模型，用于预测前列腺癌复发的时间依赖性风险。

DESCRIPTION (provided by applicant): The general objective of this study is to develop a statistical framework to lay a foundation for building an intelligent clinical support system with predictions of potential outcomes under different scenarios of prostate cancer treatment. In this proposal, combining statistical methods with cancer treatment mechanism, we propose to develop a new class of statistical regression models for predicting the probability and the timing of tumor recurrence by effectively taking account of information on treatment characteristics and post-treatment individual biomarkers. The new model is derived from an iterated cell birth and death process, mimicking the biological mechanism of tumor cells after treatments, and thereby invokes biological considerations in statistical model building and treatment outcome prediction. Statistically speaking, the new model allows for both proportional and non-proportional hazards structures, incorporates a cure rate, and accommodates non-homogeneous treatment effects on short-term cancer recurrence prevention and long-term biochemical disease-free survival. The proposed model extends the cure rate models by allowing for a more general dependence on individual covariates, and it is of semiparametric nature: the nonparametric component involves the cancer progression time distribution and the parametric component involves treatment characteristics, post-treatment biomarkers, and other significant covariates. We propose nonparametric smoothing techniques for estimation of the progression time distribution, and likelihood and Bayesian methods for parametric estimation. The methodology will be applied to clinical follow-up data of prostate cancer patients amassed at The University of Rochester Medical Center. The novelty of this project is that the new model is essentially based on biological mechanism of tumor response to treatment and utilizes strength of statistical modeling techniques for risk prediction. In this R21 application, we aim to develop the new model using statistical techniques, and if successful, an R01 proposal will be submitted in the future to fully develop the prediction model with application to construct and validate a prediction computer support system to assist physicians in making informed clinical decisions for adaptive cancer treatment strategies. Motivated by stochastic modeling of post-treatment tumor development, the project proposes to develop a new class of statistical regression models for predicting the time- dependent risk of prostate cancer recurrence.