Journal of Clinical Epidemiology 57 (2004) 341–348 The Gini coefficient as a measure for understanding accrual inequalities in multicenter clinical studies Anna-Bettina Haidich a , John P.A. Ioannidis a,b, * a Clinical Trials and Evidence-Based Medicine Unit, Department of Hygiene and Epidemiology, University of Ioannina School of Medicine, Ioannina 45110, Greece b Division of Clinical Care Research, Department of Medicine, Tufts-New England Medical Center, Tufts University School of Medicine, Boston, MA 02111, USA Accepted 15 September 2003 Abstract Objective: Clinical sites participating in multicenter trials may have unequal performance in recruiting subjects. We propose using the Gini coefficient as a quantitative measure of site accrual inequalities. Study Design and Setting: We evaluated the relationship of this metric to other study characteristics across 166 clinical studies (27,865 subjects) conducted by the AIDS Clinical Trials Group between 1986 and 1999. Results: Overall there was a modest recruitment inequality among clinical centers (mean Gini = 0.33). In multivariate modeling, site accrual inequalities were higher when there was more protracted enrollment, and a larger number of sites and were lower in antiretroviral studies than other studies. In long-term studies, the site accrual inequality increased significantly over time (P = 0.004). In efficacy trials, a higher Gini coefficient was associated with higher likelihood of the study results being statistically significant (P = 0.010). Conclusion: The Gini coefficient may be easily and routinely incorporated in the description of the characteristics of a clinical study and may provide insights about its enrollment pattern.  2004 Elsevier Inc. All rights reserved. Keywords: Clinical centers; Enrollment; Gini coefficient; Multicenter clinical study; Site accrual inequality; Statistical significance 1. Introduction Multicenter clinical research depends on recruitment of patients across several clinical sites. Recruitment is consid- ered one of the parameters that characterize the performance of a clinical site, and various participating centers may often differ considerably in their capacity to enroll subjects into a clinical study. Some sites may join the protocol at a later date, while even in studies with strictly predefined sites, certain sites may open earlier while others later. Site recruit- ment may sometimes be very uneven with some sites car- rying the lion’s share of the burden of total recruitment; however, there probably is substantial variability in this regard. Some clinical trials may have a more-or-less equal accrual across the participating sites, while others may have very unequal site accrual rates. * Corresponding author. Tel.: +3026510-97807; fax: +3026510-97867. E-mail address: jioannid@cc.uoi.gr (J.P.A. Ioannidis). 0895-4356/04/$ – see front matter  2004 Elsevier Inc. All rights reserved. doi: 10.1016/j.jclinepi.2003.09.011 It would be useful to be able to present the inequality of accrual across participating sites using a standardized ap- proach that would be applicable to all clinical studies. It also would be useful to evaluate empirically the extent of this inequality in a large number of clinical studies. Finally, it is unknown whether a large inequality across sites in their recruitment performance is likely to be related with other study attributes. Here, we used the Gini coefficient, a standardized measure to describe inequalities across groups to describe site inequalities in accrual, and we evaluated whether any study characteristics were related with site ac- crual inequality. Moreover, the change in the Gini coefficient over time was assessed within clinical trials during their conduct. Prior work has shown that early accrual, especially the first month’s accrual, was positively associated with finding statistically significant results in the final analysis of a randomized efficacy trial [1]. It would be interesting to investigate whether the site accrual inequality can offer any predictive information on the significance of the results. These issues were addressed using a dataset on clinical stud- ies conducted by a large multicenter clinical trials group in