Abstract:We propose a novel deep learning approach to nonparametric statistical inference for the conditional hazard function of survival time with right-censored data. We use a deep neural network (DNN) to approximate the logarithm of a conditional hazard function given covariates and obtain a DNN likelihood-based estimator of the conditional hazard function. Such an estimation approach renders model flexibility and hence relaxes structural and functional assumptions on conditional hazard or survival functions. We establish the nonasymptotic error bound and functional asymptotic normality of the proposed estimator. Subsequently, we develop new one-sample tests for goodness-of-fit evaluation and two-sample tests for treatment comparison. Both simulation studies and real application analysis show superior performances of the proposed estimators and tests in comparison with existing methods.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is how to use deep - learning methods for non - parametric statistical inference of the conditional hazard function in survival analysis, especially in the presence of right - censored data. Traditional methods usually rely on structural and functional assumptions of the conditional hazard or survival function, which may not hold in practical applications, leading to unreliable or even wrong inference and prediction results. Therefore, this paper proposes a new method based on deep neural networks (DNN) to approximate the logarithm of the conditional hazard function and obtain the DNN likelihood estimator of the conditional hazard function. This method improves the flexibility of the model and reduces the dependence on the structural and functional assumptions of the conditional hazard or survival function, thereby reducing the bias caused by improper model specification.
Specifically, the main contributions of the paper include:
1. **Proposing a new model - free method**: Use DNN to estimate the logarithm of the conditional hazard function without making any assumptions about its form, thereby reducing the bias that may be caused by improper model specification.
2. **Establishing non - asymptotic error bounds and asymptotic properties**: Study the non - asymptotic error bounds of the proposed estimator and explore its asymptotic properties, including consistency and functional asymptotic normality.
3. **Developing new test statistics**: Construct one - sample and two - sample test statistics, as well as goodness - of - fit test statistics, and further derive the asymptotic normality of these test statistics. In particular, develop a new goodness - of - fit test method for testing non - parametric Cox models.
4. **Demonstrating the superior performance of the method through simulation studies and real - data analysis**: Compared with existing methods, the proposed method shows better performance in simulation studies and practical applications.
In summary, this paper aims to improve the accuracy and reliability of conditional hazard function estimation in survival analysis by introducing non - parametric statistical inference methods based on deep learning, while providing a complete set of statistical test tools to evaluate the model fitting effect and compare the conditional hazard functions between different groups.