Alternative visual displays of metaanalysis of malaria treatment trials to facilitate translation of research into policy Piero Olliaro a,b, ⁎ ,1,2 , Michel T. Vaillant c,d,1,2 a UNICEF/UNDP/WB/WHO Special Programme for Research and Training in Tropical Diseases (TDR), 20, avenue Appia CH-1211 Geneva 27, Switzerland b Centre for Tropical Medicine and Vaccinology, Nuffield Department of Medicine, University of Oxford, Churchill Hospital, Oxford OX37LJ, UK c Clinical Epidemiology and Public Health Unit, Center for Health Studies, CRP-Santé, L-1445 Strassen, Luxembourg d Unité 3677, Bases thérapeutiques des inflammations et infections, Université Victor Segalen Bordeaux 2, 33076 Bordeaux, France Received 5 November 2009; accepted 6 August 2010 Abstract Typically, metaanalyses show relative effects and heterogeneity, but not absolute effects—an essential element in policy decision. Data obtained through a systematic review of antimalarial treatment trials and virtual trials were used to generate a display that shows and quantifies absolute and relative effects as well as heterogeneity for comparative trials results. A plot of failure rates (with 95% confidence intervals) of the test drug on the y axis against the risk difference (RD) versus the comparator drug on the x axis is proposed; the area is divided into 4 quadrants by a vertical line (no RD) and a horizontal line (maximum tolerated failures, e.g., 10% for antimalarials). This allows identifying where a drug can be used (meeting efficacy requirements) and quantifying differences (versus another treatment option). The area of the polygon connecting the study points expresses heterogeneity. This graphic display is simple to prepare and interpret and combines in 1 graph both measures of absolute treatment effect and difference, as well as heterogeneity. It may complement current methods and provide useful information in policy decision making. © 2010 Elsevier Inc. All rights reserved. Keywords: Metaanalysis; Malaria; Olliaro-Vaillant plot; Forest plot; Policy decision making 1. Introduction 1.1. The question Systematic reviews of available evidence on benefits and risks of medical interventions are increasingly used to inform decision making in clinical practice and public health (Buschman and Wang, 1998; Egger et al., 1997). Such reviews are, whenever possible, based on metaanalysis— defined as “a statistical analysis which combines or integrates the results of several independent clinical trials considered by the analyst to be combinable” (Yusuf et al., 1985). Although the underlying statistical methods are generally well developed (Whitehead, 2002), the results and inter- pretation of such analyses remain difficult to understand for those who are less acquainted with statistics. The prevailing statistical approach to metaanalysis is to estimate the magnitude of combined results across studies (referred to as effect size estimate) with a confidence interval (CI) around it (Hedges et al., 1992; Oakes, 1986). This provides information not only on whether the null hypothesis (e.g., that the test treatment has no effect over the reference treatment) should be rejected at a given significance level or not, but also on whether the observed treatment effect is large enough to be considered of practical import. Between- study variation in effects can be treated as fixed or random (Hedges and Olkin, 1985), depending on whether the model assumes that the population effect size is a single fixed Available online at www.sciencedirect.com Diagnostic Microbiology and Infectious Disease 68 (2010) 422 – 431 www.elsevier.com/locate/diagmicrobio ⁎ Corresponding author. Tel.: +41-22-791-37-34; fax: +41-79-815-46-55. E-mail address: Olliarop@who.int (P. Olliaro). 1 Both authors contributed equally to this paper. 2 Both authors contributed equally to the concept, conduct and write-up of this research. 0732-8893/$ – see front matter © 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.diagmicrobio.2010.08.004