My interpretation of the GRADE review is to ignore the pooled estimates for the observational studies for sure because of the poor quality. The issue is, as Stefan highlights, confounding by indication. The available RCT data do not support transfusion as increasing risk of NEC. I think the most elegant study to try and better understand what is going on and by-pass the confounding by indication study is the JAMA https://www.ncbi.nlm.nih.gov/pubmed/?term=26934258 where they are able to parse out anemia from transfusion. I am still only a fellow, but I base my transfusion practice vis a vis NEC risk on the RCT data from the GRADE review and this paper highlighting the anemia as the key feature. This conclusion is consistent with the finding in the RCTs of less NEC in the liberal transfusion group because the liberal transfusion group would presumably have less (and likely less severe) anemia.