I'm working with mixed effects cox models and I can't find the way to produce a "predicted" survival curve.
All I've read is about coxph objects but nothing about how to interpret the random effect on coxme objects.
It is not clear to me what exactly you mean with "predicted survival curve".
I see two possible interpretations here. First: you want to display an overall survival. Since no groups are involved you can just use a simple Kaplan-Meier or similar estimator for this.
I think however that it is more likely that you want to calculate survival curves for multiple groups adjusted for some confounders. You already fitted a mixed-effects cox model to adjust for the confounders and now you want to graphically display this analysis, right?
If that is indeed what you are looking for, the G-Formula method (also called G-Computation, corrected group prognosis method or direct standardization in epidemiology) is a good way to do this. Assuming your model is correctly specified you will get unbiased estimates of the survival curves that would have been observed if your whole sample was exposed to each level of your grouping variable. This method is described in many papers, one of which can be found here: Article Corrected Group Prognostic Curves and Summary Statistics
Basically what you need to do is follow these steps:
1. Fit the model to your whole dataset
2. Create a copy of the dataset for each level of your grouping variable
3. Set the group variable to one level in each data set
4. Use your model to predict the survival probability of each individual at some points in time for each dataset (the predict method of the coxme object might help you here)
5. Take the average of these predicted values at each time point in each dataset
Finally you can display those average predicted survival estimates in a nice plot (possibly using the ggplot2 package). Usually simple cox-models are used for this method, but you can use any model you want as long as it allows you to predit survival probabilities given covariates and a point in time.
Thank you so much Robin Denz
I'm new in the use of these models so I'm not very familiar with the technical language. Maybe I used the phrase "predicted survival curve" in a wrong way.
I'm gonna read the article you suggested to me and try to find any useful information.
Thanks again
- Badges
- Science topic
- Similar topics
- Philosophy
- Esthetics
- Interpretation