Related Experiment Videos
Comparison between two partial likelihood approaches for the competing risks model with missing cause of failure.
Kaifeng Lu1, Anastasios A Tsiatis
1Merck Research Laboratories, Merck & Co., Inc., Rahway, NJ 07065, USA. kaifeng_lu@merck.com
Lifetime Data Analysis
|March 8, 2005
Summary
This study compares two statistical methods for analyzing clinical trial data with competing risks and missing cause of death. The Dewanji partial likelihood method is shown to be more efficient and robust for regression coefficient estimation.
Area of Science:
- Biostatistics
- Clinical Trials
- Survival Analysis
Background:
- Clinical studies often involve time-to-event data with multiple failure causes.
- Right censoring and missing cause of death are common challenges in survival analysis.
- Handling competing risks and missing data is crucial for accurate clinical outcome assessment.
Purpose of the Study:
- To compare the Goetgbebeur and Ryan (1995) and Dewanji (1992) partial likelihood approaches for analyzing time-to-failure data with missing causes of death.
- To evaluate the statistical properties, including consistency, asymptotic normality, and semiparametric efficiency, of these estimators.
- To assess the robustness of each method to potential misspecifications in the statistical models.
Main Methods:
- Utilized partial likelihood approaches for survival data analysis.
- Assumed cause of death is missing at random.
- Compared estimators for regression coefficients under competing risks scenarios.
- Investigated robustness against misspecification of proportional baseline hazards.
Main Results:
- The Dewanji partial likelihood estimator for regression coefficients is consistent, asymptotically normal, and semiparametric efficient.
- The Goetghebeur and Ryan estimator offers greater robustness to misspecified proportional baseline hazards.
- The Dewanji method accommodates missing failure causes dependent on covariates without explicit missingness mechanism modeling.
Conclusions:
- The Dewanji partial likelihood approach provides a statistically efficient and robust method for analyzing clinical trial data with competing risks and missing cause of death.
- While the Goetghebeur and Ryan method is more robust to certain model misspecifications, the Dewanji method offers advantages in handling missing data mechanisms.
- The study suggests tests for proportional baseline hazards and derives a robust variance estimator for improved analysis.