I need to plot cumulative incidence curves, which correspond to 1 - Kaplan-Meier-values. Doing so for a factor x with four groups gives me the following curves: unadjusted curves

But in fact, I have many variables beside factor x and I need to adjust the plot for these variables. Plotting the incidence curves after adjusting for these variables gives me the following result: adjusted curves

As you can see adjusting for control variables changes the plot. This is not my problem, but actually even what I expected (theoretically meaningful). My problem is, however, that ggsurvplot does not show the number at risk once one adjusts for other variables. This was discussed here. In the link provided, the suggstion is to simply use the number of risk values of the unjusted analysis for the plot of the adjusted analysis. Is this a valid approach?

The reason why I am sceptical: Thinking of number at risk as the number of people beeing alive in a certain group to a certain time suggests that the number of risk values do not change if one adjusts for other variables. On the other hand, I am confused because the unadjusted plot shows that the cumulative events are high in group 1, meaning that people in group 1 die more often during the study. The adjusted plot shows the exact opposite: people in group 1 have lowest cumulative incident rates, meaning less people die here. But if the number of people dying vary for a certain group between the unadjusted and the adjusted analysis, how can be the number of risk the same? I think I have a misconception either about number at risk or the cumulative incidence values. Thanks in advance.

  • $\begingroup$ I've talked to the professor of biostatistics of my university and he says that after adjusting for other variables the number at risk is calculated from the adjusted model and thus the values do not correspond to those of an unadjusted analysis, e.g. to the actual observed number at risk in a certain group to a certain time. However, he said it would be alrigtht to use the number at risk of the unadjusted analysis for the plot of the adjusted analysis if one explicitly states it in the paper. But it would be inappropriate to simply use the unadjusted values without mentioning it. $\endgroup$ – user213325 Feb 11 '19 at 12:39

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy