to request graphs that show whether your data are consistent with the hazard proportionality

assumption that we describe later in “Assessing the assumptions.”

Looking at the calculations

The log-rank test should not be done manually because it is an error-prone task. But we believe you’ll

have a better appreciation of the log-rank test if you understand how it works, so we describe how the

calculations could theoretically be carried out using a Microsoft Excel spreadsheet.

The log-rank test utilizes information from the life tables needed to produce the graph shown earlier in

Figure 22-1. Figure 22-2 shows a portion of the life tables that produced the curves shown in Figure

22-1, with the data for the two groups displayed side by side.

© John Wiley & Sons, Inc.

FIGURE 22-2: A portion of the life-table calculations for two groups of laboratory animals.

In Figure 22-2, the Drug group’s results are in columns B through E, and the Control group’s results are

in columns F through I. The only measurements needed from Figure 22-2 for the log-rank test are At

risk (columns E and I), meaning number at risk in each time slice for each group, and Died (columns C

and G), meaning the number of observed deaths in that time slice for each group. The log-rank test

calculations are in a second spreadsheet (shown in Figure 22-3).

© John Wiley & Sons, Inc.

FIGURE 22-3: Basic log-rank calculations done manually (but please use software instead!).

The spreadsheet shown in Figure 22-3 has the following columns:

Column A identifies the time slices, consistent with Figure 22-2.

Columns B and C pertain to the Drug group, and reprint the At risk and Died columns from Figure

22-2 for that group. Columns D and E pertain to the Control group and reprint the At risk and Died

columns from Figure 22-2 for that group.