When I use the survival package (R) for a one-thought study, the covariates do not satisfy the proportional risk assumption but the exposures satisfy the proportional risk assumption. Then I add time dependent coefficients with the tt parameter for the covariates that don't satisfy the conditions so that all variables satisfy the proportional risk assumption. Can I just read the exposure coefficients returned by the coxph function as ln(HR) values? Do I need to consider the time-dependent coefficients of the covariates again?
Thank you so much for this helpful video! Do you have to center your time variable before putting it in interaction with the time-dependent covariate? Thanks!
Thanks for your comment Ally! I am not entirely sure what you mean. Centering is typically something you do for independent variables to provide a meaningful interpretation for the constant in regression analyses. The time variable is not added as an independent variable to the model (contrary to how it typically works with interaction terms).
Firstly, it depends on the nature of overlap. Is it a clear pattern, or are they small overlaps here and there? The latter is often to be expected, the prior often needs to be taken into account (as is the case for the video example). If there are more than two time-fluctuations, there might be more elaborate models that can deal with this but I am (not yet) aware of those at this moment.
Hi Alyson, in the most practical sense, yes. However, typically in research you would have to have a solid theoretical underpinning as to why you would expect the curves to show a different pattern from a certain timepoint. Then you follow this up by a-priori adding this time-depedent covariate. Adding the covariate based on the results of the Kaplan-Meier curve is possible, but is data driven, thus it depends heavily on the field of study and available theory.
hello, would you like to do multivariate analysis with time dependent cox regression? do you interract each variables with time or not? and how do you know it happened after time of 16?
Hi, that is a more complex question I am afraid I cannot assist you with at this time. Regarding when the hazards become disproportional: this typically comes from theory. Often, there is some biological reasoning why this occurs and after how many days/weeks/months it occurs. For instance, in medicine we may be aware that some substances are only synthesized by the body after a certain amount of time. In the case of this video, I used the graph to pinpoint the time it occured. This is a data-driven approach however, and often not the preferred method.
Thank you so much for your class. You have no idea how valuable it is for a doctor studying statistics!
Leuk om dit tegen te komen van de VU van een oude werkgroepdocent terwijl ik een brede zoekterm in TH-cam deed op mijn werk, wat toevallig! Bedankt!
Leuk dat je op die manier hier terecht bent gekomen :). Dank!
When I use the survival package (R) for a one-thought study, the covariates do not satisfy the proportional risk assumption but the exposures satisfy the proportional risk assumption. Then I add time dependent coefficients with the tt parameter for the covariates that don't satisfy the conditions so that all variables satisfy the proportional risk assumption. Can I just read the exposure coefficients returned by the coxph function as ln(HR) values? Do I need to consider the time-dependent coefficients of the covariates again?
Thank you so much for this helpful video! Do you have to center your time variable before putting it in interaction with the time-dependent covariate? Thanks!
Thanks for your comment Ally! I am not entirely sure what you mean. Centering is typically something you do for independent variables to provide a meaningful interpretation for the constant in regression analyses. The time variable is not added as an independent variable to the model (contrary to how it typically works with interaction terms).
Thank you for informative session.if the kalpan meier curves meet in morethan 3 points, can we still do cox regression with time dependent covariate
Firstly, it depends on the nature of overlap. Is it a clear pattern, or are they small overlaps here and there? The latter is often to be expected, the prior often needs to be taken into account (as is the case for the video example). If there are more than two time-fluctuations, there might be more elaborate models that can deal with this but I am (not yet) aware of those at this moment.
How do we obtain the value for ‘expression for T_COV’? By eyeballing where the 2 survival curves cross each other?
Hi Alyson, in the most practical sense, yes. However, typically in research you would have to have a solid theoretical underpinning as to why you would expect the curves to show a different pattern from a certain timepoint. Then you follow this up by a-priori adding this time-depedent covariate. Adding the covariate based on the results of the Kaplan-Meier curve is possible, but is data driven, thus it depends heavily on the field of study and available theory.
@@niels_bal thank you very much!
hello, would you like to do multivariate analysis with time dependent cox regression? do you interract each variables with time or not? and how do you know it happened after time of 16?
Hi, that is a more complex question I am afraid I cannot assist you with at this time. Regarding when the hazards become disproportional: this typically comes from theory. Often, there is some biological reasoning why this occurs and after how many days/weeks/months it occurs. For instance, in medicine we may be aware that some substances are only synthesized by the body after a certain amount of time. In the case of this video, I used the graph to pinpoint the time it occured. This is a data-driven approach however, and often not the preferred method.
Kan ik dit ook uitvoeren met een continue variabele die time dependent is?
En hoe zou je dit noteren in de resultaten? Dat je dus gebruikt hebt gemaakt van soort afkapwaarde (16 in uw geval)
Do you also know how to perforn this in R?
Unfortunately I have not previously carried out this analysis in R.