Nephrol Dial Transplant (2015) 0: 1–6 doi: 10.1093/ndt/gfv068 NDT Perspectives Lag-censoring analysis: lights and shades Giovanni Tripepi 1 , Georg Heinze 2 , Kitty J. Jager 3 , Vianda S. Stel 3 , Friedo W. Dekker 3,4 and Carmine Zoccali 1 1 CNR-IFC/IBIM, Clinical Epidemiology and Physiopathology of Renal Diseases and Hypertension of Reggio Calabria, Reggio Calabria, Italy, 2 Section for Clinical Biometrics, Center for Medical Statistics, Informatics and Intelligent Systems, Medical University of Vienna, Vienna, Austria, 3 ERA–EDTA Registry, Department of Medical Informatics, Academic Medical Center, University of Amsterdam, Amsterdam, The Netherlands and 4 Department of Clinical Epidemiology, Leiden University Medical Centre, Leiden, The Netherlands Correspondence and offprint requests to: Giovanni Tripepi; E-mail: gtripepi@ifc.cnr.it ABSTRACT ‘Intention-to-treat’ (ITT) analysis is the recommended ap- proach for the data analysis of randomized clinical trials (RCT). ITT analysis considers patients in the active or in the control arm as originally allocated by randomization, inde- pendently of their actual adherence to the assigned treatment. Lag-censoring analysis is a statistical method which takes into account the compliance of patients to the study protocol because the investigator censors a patient when or shortly after he/she stops the treatment being tested. Herein we describe the methodology underlying lag-censoring analysis in general terms and by considering the application of this technique in the analysis of a large RCT in haemodialysis patients, the Evaluation of Cinacalcet Hydrochloride Therapy to Lower Cardiovascular Events (EVOLVE) trial. Use and misuse of this technique are discussed. Keywords: intention-to-treat analysis, lag-censoring analysis INTRODUCTION The randomized controlled trial (RCT) is the gold standard study design for testing scientific hypotheses in a clinical scenario. RCTs are conducted to test the efficacy of medical in- terventions and to collect information about adverse effects of the same interventions [1]. The key feature of standard RCTs is that participants are randomly assigned to undergo the treatment being tested or other alternative therapies. After ran- domization, the two (or more) study groups are followed up by an identical protocol, the only difference between the care patients receive (clinical tests, outpatient visits, etc.) is intrinsic to the interventions being compared. The fundamental advan- tage of randomization is that it prevents bias by prognosis and that any differences in known and unknown prognostic factors in the groups being compared are due to chance [1]. Unfortu- nately, randomized controlled trials often suffer from major problems for measuring efficacy, such as noncompliance, protocol deviations, and patient withdrawals, whether these problems are related to side effects or not, and these drawbacks pose important methodological concerns during data analysis and interpretation. ‘Intention-to-treat’ (ITT) is the fundamental approach to the analysis of RCTs. By this approach, all RCTs ‘as rando- mized’, regardless of the compliance to the treatment they actually receive [1]. In ITT analysis of superiority trials (i.e. RCTs aimed at demonstrating that the experimental treatment is superior to placebo/previous therapy for reducing the risk of a given disease/event), the estimate of treatment effect is usually conservative because in patients who discontinue the experimental drug or start taking a non-study drug or who are submitted to an unplanned co-intervention, all events occur- ring after treatment discontinuation or after a co-intervention continue to be attributed according to randomization, i.e. to the active or the control arm (Figure 1, upper panel). As a consequence, in ITT analysis, the difference in efficacy is typically smaller than that which we would expect assuming as effective the experimental drug, thus making it harder to reject the null hypothesis of no difference. In contrast, the smaller differences commonly observed in ITT analysis have an obvious opposite effect when testing equivalence or non- inferiority between drugs. © The Author 2015. Published by Oxford University Press on behalf of ERA-EDTA. All rights reserved. 1 NDT Advance Access published April 16, 2015 by guest on April 22, 2015 http://ndt.oxfordjournals.org/ Downloaded from