Can censored data act as the dependent variable for a logistic regression? I need to do the research about the risk factors that contribute to heart failure.  The data I have just the censored data with some risk factors of heart failure.  And I need to use logistic regression and exponential regression to find out the risk factors that contribute to heart failure. The censored data define as 0=censored, 1=died, can I use this as my dependent variable for logistic regression? If not, if someone got data that related the risk factors that contribute to heart failure which can use exponential and logistic regression to find out the risk factors of heart failure, is possible to share with me? 
 A: Time-to-event analysis (the field of survival analysis) exists primarily because of the need to handle censoring.  When subjects are followed different lengths of time, the binary logistic model has no way to handle this.  For example suppose that one subject was followed only one week and was event-free as of that time.  Her time-to-event is recorded and used in survival analysis as 7+ days.  Another subject followed one year without an event would be used as 365+ days.  The 7+ days subject contributes very little to the analysis - it's almost as if she isn't there.  The 365+ days subject contributes in a significant way.  Treating these values as right-censored is the way to account for different follow-up periods, and to account for what we don't know.
Even were there no censoring (loss to follow-up), using a binary logistic model would be a mistake because it treats a failure at one day as no worse as a failure at 5 years.  This is a huge information loss that leads to loss of statistical power and precision.
Finally there is the "how useful is the output" question.  A single survival model can be used to estimate the probability of surviving at least t days for many different t.
A: You just need to connect your data to your inference properly. If the inference is not the one you want to make, then you need to change the model.
it might help to think of two scenarios. the first you observed everyone for say 10 years. Then your model is really a prediction of "person lives for at least 10 years".
for the second suppose that you observed the 0s for just 1 year, but the 1s were watched for 10 years. Then you can perfectly predict the outcome in this sample by simply having a variable "person was watched for 2 years or more".
your real data will be somewhere in between, but you need to include "time exposed to risk" somehow in your model.
when including an "exposed to risk" in your model, your inferences may change, and your predictions now become dependant on this.
you may also want to include or test for interactions between the "exposed to risk" variable and your other predictors.
