Cox proportional hazard regression with time varying covariate in spss

This video shows how to assess the effect of heart transplantation using data from Stanfort Heart Transplant study using SPSS.

[@ZeroZoom-tm2kz]

Your video was extremely helpful and clear. many thanks.

[@melodys2467]

Hi Ayumi, Your video was extremely helpful. I have a quick question. What if instead of time to transplant we were looking to time to a diabetes diagnosis as the time varying covariate. However, there is missing data for this covariate after a certain interval. For example if there are four intervals the data is 0, 0, ., . However, even though information is missing the patient has not been lost to follow up and their total time in the study continues. Can this covariate data that is available contribute to the study - similar to censoring in terms of the event? Or should this data not be included in the analysis? Best, M

[@zlw3877]

I learned a lot from this video. But I had a question. In this video (Time 7:15-7:40), when you show how to create a T_COV, what you had typed into the box was not complete. What you input is "T_

[@Xstr1ke]

Hi, thank you for your clear video, really appreciate it! I was wondering whether this model also works if someone was transplanted on the day he/she was put on the transplant list (thus t=0). Or do I need to use a value that approximates 0, like 0.00001? Thanks!

[@mfberty]

Dear Prof. Shintani, thank you for this comprehensive and valuable video. Just one question to consider, if the patients need to survive longer to receive "Event", does the regression model take age differences into account? I.e. maybe the groups differ significantly over time in age, which in turn may influence survival?

[@ayumishintani7044]

In that case, you can adjust for age at baseline in a regression. I would adjust for baseline age in the analysis, and I think you do not need to adjust for time-varying age effect. Because Cox regression can model baseline hazard which can take time effect into consideration already in Cox regression. So you do not need to model increasing age effect in the Cox regression I think.

[@weijingfeng3820]

Hi Ayumi Shintani, thanks for your excellent video. I have a question. If we have more than two covariates, which are time-varying covariates, how to deal with them by time-varying cox regression. Thank you!

[@ayumishintani7044]

You can have two or more time-varying covariates. You can simply create a new row in a dataset when either of the two variables changes a state. For example ID Stattime Stoptime exp1 exp2 event 1 0 1 1 0 0 1 1 3 0 0 0 1 3 4 0 1 1

[@t_ito7774]

Dear Prof Shintani, your video was extremely helpful. I would like to ask how is the best way to graphically represent that. The Kaplan-Meier method can be used?

[@ayumishintani7044]

+Rafael Cavalcante We can do time-varying K-M curve. Just use the same dataset you use for the time-varying Cox. You need to use a software which can do K-M with 2 time variables (start time and stop time).

[@t_ito7774]

Dear Prof Shintani, many thanks for your response. I am assessing the effect of infection on mortality. I did in SPSS by plotting the curve of the whole population censoring the patients with infection at the day they got infected and considering them alive against the curve of the infected patients considering the day of infection as day zero. What do you think of that analysis?

[@mitch4349]

What do you do when your time to transplant is longer than time with transplant, as SPSS will see this as "no transplant"

[@JpKayyy]

Hallo, your video was very helpful, yet in the analysis I´m conducting the covariate can change more than once. To argue with Your example: The transplant can be inserted and removed various times. Is there a way to consider this? I would appreciate an answer. Yours Jan

[@ayumishintani7044]

Jp Klein Yes, to do this you just need to create dataset to reflect in and out. For example, ID stat.time stop.time event exp1 0 5 0 11 5 7 0 01 7 9 1 1Although, SPSS cannot handle this as it does not have robust or cluster option to compute p-value correctly reflecting repeatedly entered data.

[@JoepS85]

+Jp Klein Hi there! I'm facing a similar problem like yours. Did you manage to work it out? Please let me know! :)