Using filters for non-randomized or observational designs is fraught because authors use terminology inconsistantly and because of the diversity in controlled vocabularies implemented in the various databases. A filter will often exclude desirable studies while failing to eliminate much of the unwanted material. Nevertheless, validated filters for most study designs have been developed and can be used succesfully in your review, if necessary.
The PubMed Clinical Queries filters have been validated and provide good results for observational studies (etiology). Filters are available for therapies, diagnosis, prognosis, and prediction guides, too. They are based on a body of work by Haynes et al.
Here's a decent general purpose filter for observational designs. It's modified from the "fixed method B" of Furlan, et al.. The designers of the original found that it can retrieve between 95 to 100% of relevant articles, though in actual use may not be as efficient. Here it's presented as an expanded search that will retrieve case-control studies, too. Expect more retrieval but also more noise. PubMed syntax in red; Embase in blue. The Embase search returns about 30% more citations than the PubMed search because of the structure of Emtree.
"cohort studies"[mesh] OR "case-control studies"[mesh] OR "comparative study"[pt] OR "risk factors"[mesh] OR "cohort"[tw] OR "compared"[tw] OR "groups"[tw] OR "case control"[tw] OR "multivariate"[tw]
'clinical article'/exp OR 'controlled study'/exp OR 'major clinical study'/exp OR 'prospective study'/exp OR 'cohort analysis'/exp OR 'cohort':ti,ab OR 'compared':ti,ab OR 'groups':ti,ab OR 'case control':ti,ab OR 'multivariate':ti,ab
Two other sources for filters:
1. Furlan AD, Irvin E, Bombardier C. Limited search strategies were effective in finding relevant nonrandomized studies. J Clin Epidemiol. 2006 Dec;59(12):1303-11. Epub 2006 Jul 11. PubMed PMID: 17098573.