Stopping a randomized trial early: from protocol to publication. Commentary to Thome at al.: outcome of extremely preterm infants randomized at birth to different PaCO2 targets during the first seven days of life (Biol Neonate 2006;90:218-225).

even reverse itself, and so fi xed stopping rules dictated by predetermined signifi cance levels for differences in outcome are increasingly discouraged [3] . Additional information such as the medical plausibility of the difference, ethical considerations both for the trial participants and for the wider target population, evidence from other relevant trials and the nature of the disease under consideration may also be important [2, 4–6] . The decision to stop a trial for harm can also be based on adverse events, although attribution of the cause of an adverse event may be diffi cult [3] . There is less guidance available on when and how trials should be stopped for futility. The 2005 DAMOCLES study [3] highlights a recognition in the literature that termination of a trial may be recommended where it is considered that there is no longer a reasonable chance of accumulating suffi cient evidence against the null hypothesis (usually due to poor recruitment or to external information becoming available [7] ). However, it is unclear whether trials should be stopped on these grounds [3, 4] . Another reason for stopping for futility is that the trial is unlikely to come to a sound conclusion [3] . This is illustrated in the trial by Thome et al. The diffi culty with this trial was that, at the time of the interim analysis, the majority of the patients randomized to the active arm (minimal ventilation to achieve higher PaCO 2 ) had not An important, often neglected element to planning a clinical trial is to consider the possibility of stopping the trial early, before either recruitment or follow-up is completed. A trial may be stopped for futility, harm or benefi t. A trial that is stopped on the grounds of benefi t is one that early on identifi es important treatment effects unlikely to be due to chance. A trial stopped for harm provides evidence that the treatment under investigation is unsafe for trial subjects. A trial stopped for futility requires a decision that the trial would not provide suffi ciently useful information to warrant continuation. Stopping a trial early is always a diffi cult decision to make and consideration must be given to who will be making these decisions and on what grounds they will be made. This issue of Biology in the Neonate publishes a trial by Thome et al. investigating the use of increased PaCO 2 targets on outcomes in extremely preterm neonates. It appears that this trial was stopped for futility on the grounds that the increased target levels of PaCO 2 were not being reached in the active arm of the trial. Guidance for stopping a trial for benefi t or harm is well documented [1] although approaches remain controversial. Statistical methods exist and may contribute to deciding whether the observed data imply that benefi t is very likely or, in the case of stopping for harm, very unlikely, or that the new treatment is harmful [2] . An early trend in the data under consideration may fl uctuate and Published online: April 19, 2006