Score and deviance residuals based on the full likelihood approach in survival analysis

Area of Science:

• Statistics
• Biostatistics
• Survival Analysis

Background:

• The proportional hazards model is a cornerstone of survival data analysis.
• Identifying outliers is crucial for robust statistical modeling.
• Existing methods like partial likelihood residuals have limitations, particularly with high censoring rates.

Purpose of the Study:

• To develop and evaluate novel residuals for survival analysis using a full likelihood approach.
• To compare the performance of these new residuals against traditional partial likelihood residuals.
• To assess the utility of these residuals in identifying outliers in the presence of high censoring.

Main Methods:

• The study employed a full likelihood approach under the proportional hazards model with non-informative censoring.
• Two new types of residuals were derived: score-type and deviance-type residuals.
• Extensive simulation studies were conducted to compare the efficiency of the proposed residuals with partial likelihood residuals.

Main Results:

• The newly developed full likelihood-based residuals demonstrated superior efficiency in identifying potential outliers compared to partial likelihood residuals.
• This improved performance was particularly evident in simulation studies with a high proportion of censored data.
• Graphical techniques were utilized to showcase the practical application of these residuals.

Conclusions:

• Full likelihood-based residuals offer a more efficient and reliable method for outlier detection in survival analysis.
• These new residuals are especially valuable when dealing with datasets characterized by high censoring proportions.
• The findings suggest a significant advancement in diagnostic tools for survival models.