Research on Multiple Comparisons of Non-Parametric Tests In Survival Analysis  

Oleh:

Shimamura, Fumiya ; Sano, Masataka ; Hamada, Chikuma

Jenis: Article from Proceeding
Dalam koleksi: 12th ANQ Congress in Singapore, 5-8 Agustus 2014, page 1-6.
Topik: multiplicity; survival analysis; joint-ranking method; separate-ranking methods; multiple comparison methods
Fulltext: IC2-2.3-P0260.pdf (306.33KB)

Isi artikel

There is a problem of multiplicity occurs when performing multiple tests in one trial. Matsuda and Nagata (1991) solves this problem for parametric tests. On the other hand, it has not been studied the problem of multiplicity for non-parametric tests. This study focuses on the problem of multiplicity of inter-level comparison in survival analysis. Non-parametric tests are frequently used when considering the difference in survival times between groups. It is necessary to consider the following two points to compare three or more groups by using the multiple comparison method. The first one is the selection of ranking method for the observation which is necessary to calculate the test statistics. There are Separate-ranking method and Joint-ranking method. In general, Joint-ranking method is better for detecting large differences, and Separate-ranking method is for detecting small differences, although the performance evaluation is not sufficient yet. The second is the determination of the adjusting method of the multiplicity. It is common to use Tukey method, or Bonferroni method which adjustment method is simple in clinical trials. However, the performance evaluation of the multiple comparison methods in survival analysis is not sufficient yet. This study evaluates the performance of the Separate-ranking method and Joint-ranking method in addition to the plurality of multiple comparison method assuming pairwise comparisons tests of all groups in three or four groups. Each method is evaluated based on the power and a error. As a result, Separate-ranking is better than Joint-ranking method. In particular, Holm method is higher than the other methods in almost all conditions; around 2-20% higher

Opini Anda

Klik untuk menuliskan opini Anda tentang koleksi ini!

Kembali