The answer to your question depends on your research question. Are you interested in endline and/or follow-up data? You are the one who must choose but you must be able to justify what you choose. There is no problem dealing with multiple effect sizes per study coming from multiple timepoints. Most meta-analysis software will average together the effect sizes and variances for the multiple timepoints to yield one effect size and one variance per study. There is a fancier formula for combining the variances from multiple timepoints in a given study but it is not commonly used, mainly because a needed number (i.e., correlation between timepoints) is usually not provided in the study report.

Dear Gordon L Warren, thank you so much for your answer.

I agree with the answer above. You can legitimately pool these together, but these differences must be recognized in case your pooled effect has high heterogeneity. You may also do subgrouping for both follow-up and endline data in revman to avoid confusion and isolate the effects. This may not be applicable in your case though since you only have 4 studies. You may do sensitivity analysis if you want to check the robustness of your result (i.e. remove the effect of the follow-up data only and check if the result will be similar). If the outcome variable is continuous in nature, your must convert this to a mean with SD (for article 4). Cochrane specifies some techniques on how to convert to mean and SD if other variables are available (e.g. median, range, SE etc).

Thank you Louella Patricia D. Carpio for your valuable answer

You can pool, but should be cautious on interpreting the results as studies have varying time points of effect measure and this may introduce high heterogeneity.

I suggest you have a look into: https://onlinelibrary.wiley.com/doi/book/10.1002/9780470743386