Which method can address non-random informative censoring in survival analyses?

• A. Increasing sample size
• B. Stratified randomization
• C. Inverse probability weighting
• D. Ignoring censoring

Answer: (C)

Non-random informative censoring happens when the reason for censoring is related to the risk of the event, which can bias survival estimates if censoring is treated as random. Inverse probability weighting addresses this by weighting each individual's data by the inverse of the probability they remain uncensored up to each time, given their covariates. This reweights the data to create a pseudo-population in which censoring is independent of the outcome, enabling unbiased estimation of survival curves and hazard ratios under the censoring model is correctly specified. We estimate these weights from a model of the censoring mechanism and apply them in analyses such as weighted Kaplan-Meier or weighted Cox regression, with the key assumptions that, conditional on observed covariates, censoring is independent of the failure time and there is adequate positivity. Increasing sample size reduces random error but does not fix bias from informative censoring; stratified randomization addresses balance at baseline but does not inherently remove informative censoring in the analysis; ignoring censoring implies assuming non-informative censoring and leads to biased results.