A systematic review of comparisons of effect sizes derived from randomised and non-randomised studies.

摘要: BACKGROUND There is controversy about the value of evidence effectiveness healthcare interventions from non-randomised study designs. Advocates for quasi-experimental and observational (QEO) studies argue that randomised controlled trials (RCTs) often difficult or impossible to obtain, inadequate answer question interest. RCTs point out QEO are more susceptible bias refer published comparisons suggest estimates tend find a greater benefit than RCT estimates. However, literature cited selectively, may be unsystematic have failed distinguish between different explanations any discrepancies observed. OBJECTIVES The aim was investigate association methodological quality magnitude by comparing systematically derived studies. Quantifying such should help decision-makers judge strength Two strategies were used minimise influence differences in external validity studies: comparison intervention, where both reported single paper specified interventions, papers. authors also sought identify designs been proposed address one problems found with conventional RCTs. METHODS DATA SOURCES Relevant identified from: Cochrane Library, MEDLINE, EMBASE, DARE, Science Citation Index. References relevant papers already experts. Electronic searches very design yielded few first strategy when identifying CHOICE OF INTERVENTIONS TO REVIEW FOR STRATEGIES 1 AND 2: For 1, intervention eligible. 2, which population, outcome investigated anticipated homogeneous across selected review: Mammographic screening (MSBC) women reduce mortality breast cancer. Folic acid supplementation (FAS) prevent neural tube defects trying conceive. EXTRACTION QUALITY ASSESSMENT: Data extracted author checked second author. Disagreements negotiated reference concerned. scored using checklist assess whether same populations, assessment outcomes 'blinded', extent estimate took account possible confounding. detailed instrument on four dimensions: reporting, generalisability results, subject All assessments carried three people. SYNTHESIS ANALYSIS: pairs classified as high low quality. Seven indices size effect frequency calculated, possible, each comparison. Distributions direction compared high- low-quality comparisons. STRATEGY THREE ANALYSES WERE CARRIED OUT: Attributes described k statistics, percentage agreement, Cronbach's values. Regression analyses -variations (ABSTRACT TRUNCATED)