It's tempting to use a methodology filter to eliminate undesirable study designs because it seems like an easy way to reduce the size of large retrieval sets, but it's alway preferable not to limit by the type of study. In most cases, such limits will result in the exclusion of relevent studies because of the large variety of terms used to describe study types. Moreover, indexing in the major databases for study types has only recently been a priority, limiting the effectiveness of controlled vocabulary in culling retrieval sets. For non-randomized designs the problem is especially severe since many authors never mention methodologies in either the title or abstract, and indexing is inconsistent. Fortunately, in many cases, an effective search involving a patient population, and intervention and an outcome will produce a managable set of results that requires no methodology filter. Define the scope of your study with this in mind.
Sometimes retrieval sets are truly unmanagable and you'll need or want to employ a methodology filter. Attempts have been made to validate many different filters. The most successful are for randomized controlled trials. These are effective because authors frequently identify their randomization methods in the title or abstract, and since the mid-1990s, controlled vocabulary has been applied in regular ways in both MEDLINE and Embase. Validated filters, which have been tested against gold-standard sets of bibliographic records, are available for other types of designs. If at all possible use one of these validated filters. You'll get decent results and will have a way of justifying your choice to the reviewers and readers of your paper. See the subtabs to this page for information on specific design classes.