It depends on your study objectives. If your goal is to get an unbiased (this is a very strong statement; alternatively, you can say a less biased estimate) and a more precise estimate, then you need to adjust for multiple observations per subject. You can use the GEE or mixed-effect model in this case. For a prediction-only problem (the goal is not getting a precise estimate), you can use the logistic regression.
Hi Robel Hussen Kabthymer . Yes, you should provide more info about Your study. When you say "simple", the answer would probably be "NO", you have to use Regression analyses specifically meant for longitudinal data, e.g. repeated measure ANOVA and GEE. If you have only two timepoints, though, you have more options.
If it is a count (which cannot go negative) and you have three time points , I would think of using a Negative Binomial Distribution random effects model (that is a potentially over-dispersed Poisson) http://www.bristol.ac.uk/cmm/software/mlwin/mlwin-resources.html#discrete