A simple hybrid variance estimator for the Kaplan-Meier survival function.

Area of Science:

• Biostatistics
• Survival Analysis

Background:

• The Kaplan-Meier estimator is a standard for survival analysis but its variance estimation can be inaccurate, particularly with censored data.
• Traditional variance estimators like Greenwood and Peto may underestimate true variance in survival distribution tails.

Purpose of the Study:

• To propose a novel hybrid variance estimator for the Kaplan-Meier survival function.
• To introduce an adjusted hybrid estimator for small sample sizes.
• To compare the performance of these new estimators against established methods.

Main Methods:

• A hybrid variance estimator approximating true variance using a Binomial variance formula.
• Definition of the proportion parameter as a piecewise non-increasing function.
• Incorporation of effective sample size as subjects not censored.
• Simulation studies comparing hybrid estimators to Greenwood and Peto estimators.

Main Results:

• Hybrid variance estimators provide more accurate estimates of true variance on average.
• Confidence intervals constructed with hybrid estimators exhibit more nominal coverage rates.
• Greenwood and Peto estimators can significantly underestimate variance in survival distribution tails.

Conclusions:

• The proposed hybrid variance estimators offer superior performance compared to traditional methods.
• These estimators enhance the reliability of confidence intervals in survival analysis.
• The adjusted hybrid estimator is particularly useful for small sample sizes.