I have just started to read about survival analysis,

I have read in this interesting article about survival analysis, and why use it rather than the famous multiple linear regression to estimate the time of the event occurance. Mainly because linear regression cannot handle right censoring (i.e. if the event has not yet occured to some of objects of study after the study duration).

My question is, hypothetically, what if I have the data of heart disease patients, say 1M, 0.5M patients have had a heart attack, and 0.5M patients haven't had a heart attack, in a 10 year period. What if I use only the 0.5M patients that had a heart attack to build a regression model to estimate the time period it will take for a patient to have a heart attack?. Or, to make it more thorough, use both data to predict whether a patient will have a heart attack or not, and if so, estimate when the event will happen. This way, the problem with the right censoring issue will be eliminated.

Would there be any faulty/wrong thought process to this methodology? If there is, could you also show me how survival analysis (or perhaps another alternative method) tackles it?

I am trying to get a deeper understanding of the subject. Thank you, in advance.

  • $\begingroup$ If you use a specific criteria to separate the patients selected for the modeling, such as smoker / non-smoker for example, that could be applied to future patients to determine if the model you have created can be applied to patients outside the study group. If you select for modeling only the patients who have already had a heart attack with no other criteria, how could that model be applied to patients from outside the study group? And so this type of data selection does not provide a predictive model, even though it can be an informative analysis. $\endgroup$ – James Phillips Mar 14 at 11:10

Your Answer

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

Browse other questions tagged or ask your own question.