The impact of different censoring methods for analyzing survival using real-world data with linked mortality...

Area of Science:

• Real-world evidence research
• Biostatistics
• Health outcomes research

Background:

• Reliable outcome evaluation is essential in real-world evidence (RWE) studies.
• Overall survival (OS) is a common RWE outcome, but its data capture is often incomplete.
• External mortality data linkage is frequently used, yet censoring recommendations vary.

Purpose of the Study:

• To investigate the impact of different censoring methods on median survival and log hazard ratio estimation.
• To assess these impacts under conditions of partially captured external mortality information.
• To understand how censoring affects real-world data (RWD) analyses.

Main Methods:

• Monte Carlo simulation of a comparative effectiveness study using RWD and linked mortality data.
• Evaluation of two censoring schemes: last activity date vs. data cutoff.
• Assessment of bias, coverage, variance, and rejection rate under varying data completeness and study parameters.

Main Results:

• Censoring at data cutoff yielded unbiased median survival when mortality data were complete; censoring at last activity underestimated it.
• With incomplete mortality data, censoring at last activity underestimated survival, while censoring at data cutoff overestimated it.
• Bias in median survival estimates changed inversely with the amount of missing mortality data depending on the censoring method.

Conclusions:

• The completeness of linked external mortality data must guide the choice of censoring strategy for OS in RWD studies.
• Inappropriate censoring can introduce substantial bias into median survival estimates.
• RWD providers should validate and publish their mortality data to support methodological decisions in RWE research.