What if the independent variable changes value during the course of the study?

Let's say that over the course of a longitudinal study a subject's value changes on an independent variable. This may happen because the patient quits (or starts) a habit such as smoking, begins a new medicine, or develops a new symptom or illness. How can you deal with this in your analysis? The answer is that within proportional hazards analysis you can create time-dependent variables. These variables change value at a particular point in time. So, instead of having a variable such as smoking at baseline (yes/no), you create a time-dependent variable, where each subject is 0 (nonsmoker) or 1 (smoker) at a particular point of survival time.

In the simplest case, time-dependent variables change their value only once (e.g., a nonsmoker starts to smoke – the variable is 0, 0, 0, at several points in time before the subject begins smoking and then the variable changes value to 1 at the time the subject begins smoking and remains 1 for the remainder of the observational time). It is also possible to construct time-dependent variables that change their value back and forth multiple times (reflecting what sometimes happens when smokers try to quit).