Statistical issues in survival analysis (Part XVIII)
November 22, 2023
In this article that appeared in Journal of Probability and Statistics, the authors described a new test as an alternative to the log-rank test to compare late differences between survival curves. Normally, in these approaches, a weighted version or the log-rank test had been proposed to handle differences in survival (Usha: is it late or early?). It appears that the authors have simply modified the weighted log-rank test with a different weight that they proposed. They varied the number at risk at each risk time. AS they said, their new weight would be more adaptive as it would take into account more differences across time points due to its varying nature instead of being fixed. By design it was monotonically increasing.
The authors then went on to test their method in simulations. What was not discussed by the authors was how they coded up their revised weighted log-rank test for use in coding. However they looked at different rates of censoring and different sample sizes and compared them to several other tests which have been used as alternatives to the standard log-rank test over the years, especially the Fleming-Harrington tests used in the Max Combo to adjust for early or late separation of curves. According to all their comparisons, the new weight was powerful in cases of heavy censoring and small sample sizes. They also tested their method in two real dataset applications. Again, their method worked out. The authors noted no issues with their methods.
Keywords: survival, log-rank test, weights, censoring
Ildephonse Nizeyimana, Samuel Mwalili, George Orwa, “New Test for the Comparison of Survival Curves to Detect Late Differences”, Journal of Probability and Statistics, vol. 2023, Article ID 9945446, 10 pages, 2023. https://doi.org/10.1155/2023/9945446