Sudheesh,K. K.,Anjana
Abstract:The analysis left truncated and right censored data is very common in survival and reliability analysis. In lifetime studies patients often subject to left truncation in addition to right censoring. For example, in bone marrow transplant studies based on International Bone Marrow Transplant Registry (IBMTR), the patients who die while waiting for the transplants will not be reported to the IBMTR. In this paper, we develop novel U-statistics under left truncation and right censoring. We prove the $\sqrt{n}$-consistency of the proposed U-statistics. We derive the asymptotic distribution of the U-statistics using counting process technique. As an application of the U-statistics, we develop a simple non-parametric test for testing the independence between time to failure and cause of failure in competing risks when the observations are subject to left truncation and right censoring. The finite sample performance of the proposed test is evaluated through Monte Carlo simulation study. Finally we illustrate our test procedure using lifetime data of transformers.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the statistical method problem of dealing with left - truncated and right - censored data in survival analysis and reliability analysis. Specifically, the authors are concerned with how to effectively estimate and test the parameters in the data in the presence of left - truncated and right - censored data. The main contribution of the paper lies in developing new U - statistics to handle this type of data and proving the $\sqrt{n}$ - consistency and asymptotic normal distribution of these U - statistics. In addition, based on the proposed U - statistics, the author also developed a simple non - parametric test method for testing the independence between failure time and failure cause in the competing risk model.
### Background and Motivation
In survival analysis and reliability research, left - truncated and right - censored data are very common. For example, in bone marrow transplantation research, patients who die while waiting for transplantation will not be reported to the International Bone Marrow Transplant Registry (IBMTR), which leads to left - truncation of the data; and the survival time of some patients may not be fully observed due to the end of the study or loss of follow - up, which leads to right - censoring. The existence of these problems makes traditional statistical methods no longer applicable, and new methods need to be developed to accurately analyze this type of data.
### Main Contributions
1. **Development of U - Statistics**: The author proposed new U - statistics applicable to left - truncated and right - censored data and proved its $\sqrt{n}$ - consistency and asymptotic normal distribution.
2. **Asymptotic Properties**: Through the counting process technique, the asymptotic distribution of U - statistics was derived, and a consistent estimate of the asymptotic variance was given.
3. **Non - parametric Test**: Based on U - statistics, a non - parametric test method was developed to test the independence between failure time and failure cause in the competing risk model.
4. **Empirical Research**: The finite - sample performance of the proposed test method was evaluated through Monte Carlo simulation research, and practical applications were carried out using transformer life data.
### Application Example
The paper uses transformer life data as an example of practical application. These data include transformers installed before 1980 but failed after 1980, and transformers installed after 1980 and failed before 2008. For transformers still in use in 2008, their lifetimes are regarded as right - censored. In addition, there is no information about transformers installed before 1980 and failed, so these data are left - truncated. By analyzing these data, the author found that there is a dependence relationship between the transformer's lifetime and its failure cause.
### Conclusion
The paper successfully developed a U - statistic applicable to left - truncated and right - censored data and proved its good statistical properties. The proposed non - parametric test method performs well in actual data and can effectively detect the dependence relationship between failure time and failure cause. Future research can consider extending this method to other types of censoring mechanisms, such as double - censoring and interval - censoring.