Introduction to Systematic Reviews

cal experience and training with high quality scientific evidence. Because the volume of new scientific knowledge is increasing continuously, practising clinicians simply do not have sufficient time to read all the important primary research reports. Indeed, to keep in touch with the top clinical psychiatry journals, a psychiatrist would need to read over 5000 articles a year (Geddes et al., 1999). Hence, we all need a reliable system of knowledge management and an essential component will be provided by reviews. Reviews are usually narrative or systematic in approach. The Journal of Psychopharmacology will henceforth be hosting summaries of systematic reviews completed under the guidelines of the Cochrane collaboration (Cochrane Collaboration, 1995) to help readers keep up to date with the best available evidence. By way of introduction, we will reprise here the advantages, as we see them, of systematic over narrative methodology. Narrative reviews, primarily based on the experience and subjective judgement of the author(s) – often expert in the area – are the traditional approach to reviews of any body of knowledge. The usual absence of a clear and objective methodology can be an importance weakness (Murlow, 1987), because a personal interpretation of the literature may lead to bias. Alternatively, in avoiding bias, some reviews fail to synthesize the data and answer any question at all. By contrast, systematic reviews (or overviews) use explicit and therefore reproducible methodological strategies to synthesize the findings of primary research studies (Carney and Geddes, 2002). They provide excellent training for inexperienced investigators and should provide the prelude to any serious research enterprise, especially a clinical trial. A standard format also allows systematic reviews (e.g. in the Cochrane Library) to be updated periodically to incorporate new evidence. Systematic reviews can focus on specific clinical questions with improved reliability and accuracy. Nevertheless, the results are rarely unequivocal and still require integration with clinical judgement and the patient’s preferences. A systematic review encourages the formulation of a clear question. For example, to decide which of competing treatments is better, the most reliable study design is a randomized controlled trial. The key element is randomization because it avoids any systematic tendency to produce an unequal distribution of prognostic factors between the experimental and control treatments, influencing the outcome (Altman and Bland, 1999). Systematic reviews frequently combine randomized trials of similar design to perform a so-called meta-analysis. This increases the power to answer a therapeutic question. There is simply no substitute for large numbers in any experiment. Questions of diagnosis or aetiology can also be addressed by systematic review: guidelines exist for undertaking reviews (and meta-analyses) of diagnostic tests (Irwig et al., 1994) and the observational epidemiological designs used in aetiological research (Stroup et al., 2000). Systematic reviews and meta-analysis present the results of individual studies and the ‘pooled’ weighted average result. The confidence interval shows the extent to which the ‘true’ effect may differ from that estimated by the studies by the play of chance alone, but it does not provide any indication of the likelihood of bias (Altman, 1992). Systematic bias comes in several forms. One of the most important is the selection of primary research studies to be included in the review. Most obviously, a reviewer may only select those studies that support his prior beliefs. More subtly, publication bias – the tendency of investigators, reviewers and editors differentially to submit or accept manuscripts for publication based on the direction or strength of the study findings – and biases introduced by an over reliance on electronic databases or single research cultures may also be important. Results from different trials are rarely identical: meta-analysis can investigate quantitative variations between the results of individual studies. If more heterogeneity exists than can be reasonably explained by the play of chance alone, attempts should be made to identify the reasons for such heterogeneity and whether or not it is reasonable to combine the studies in the first place (Thompson, 1994). By also relating heterogeneity to the methods of the identified primary studies, systematic reviews and meta-analyses can identify common methodological weaknesses and errors that materially affect the results – these may be missed using a less systematic approach. For example, in a meta-regression analysis, the heterogeneity observed in the results of trials comparing atypicals with haloperidol was related to the dose of haloperidol used (Geddes et al., 2000). Journal of Psychopharmacology 18(2) (2004) 249–250 © 2004 British Association for Psychopharmacology ISSN 0269-8811 SAGE Publications Ltd, London, Thousand Oaks, CA and New Delhi 10.1177/0269881104042629 Introduction to systematic reviews