A case-cohort study presents an economical advantage since it builds a sub-cohort and tracks events of interest only with a subset of the full-cohort. However, applying the existing survival analysis methodology using the partial likelihood function t...
A case-cohort study presents an economical advantage since it builds a sub-cohort and tracks events of interest only with a subset of the full-cohort. However, applying the existing survival analysis methodology using the partial likelihood function to the case-cohort design directly can yield biased results. Therefore, the case-cohort design should estimate the survival rate through the pseudo-likelihood function. The typical methods are Prentice, Self & Prentice, and Barlow weights. However, prior literatures on cohort study design have only focused on comparisons in weighted Cox regression analysis, and has not yet examined the comparison of the weighted Kaplan-Meier survival curves.
Thus, the current study compares the performance of the survival curves drawn based on 'weighted Kaplan-Meier estimator' through simulation. The results reveal no significant difference between the weights in the general situation except for Prentice method, however, Barlow's weight reflects the full-cohort most appropriately as the sub-cohort size is larger and event occurrence rate is lower. Moreover, in the paper, we suggest ‘a modified Barlow’ that simplifies the original Barlow method. In comparison with the original Barlow method, the modified Barlow method appropriately estimated the survival curve of the full-cohort when the sub-cohort size is small or the event rate is low as 3% or less. It is recommended that Barlow's weight is the most accurate method when drawing survival curves in case-cohort data.