Clinical Trials

Clinical Trials: A Practical Guide ■❚❙❘Table 1. Summary of the most common types of bias in clinical trials and methods of bias control.Type of biasSelection biasBias in study managementObserver ascertainment biasBias introduced by exclusions after randomizationPublication biasMethod of bias controlRandomized treatment assignmentConcealment of treatment assignmentStandardized study proceduresStandard equipmentTraining and certification of research personnelBlinding or maskingIntention-to-treat analysisWorst-case scenario analysisProspective registration of clinical trialsPublication of ‘negative’ trialsby odd/even date of birth or hospital number. However, it should be noted that,even when carried out properly, simple randomization does not guarantee theelimination of selection bias – it only reduces the possibility of this unwanted effect.To further help minimize selection bias, stratification of randomization is used.This involves patients first being classified into subgroups, or ‘strata’, by one ormore characteristics that may influence treatment response, such as age orseverity of disease. Patients within each stratum are then randomized separatelyto ensure that, based on the stratification characteristics, the patients are wellbalancedacross the treatment groups.With respect to randomization, there are two processes of equal importance:• creation of a random treatment assignment code• concealment of that code until treatment allocation occursSome investigators working in clinical research appreciate the code-generatingprocess of randomization but then disregard concealment. Without satisfactoryconcealment, even the best, most unpredictable randomization codes may beundermined. Chalmers et al. reported manifold overestimation of treatmenteffect in trials without adequate concealment of treatment allocation [4]. By usingproper concealment procedures, such as keeping individual treatment codes insealed opaque envelopes and making them accessible only to authorizedpersonnel, those who are admitting volunteers into a study are protected fromknowing the treatment allocation that will be used.Selection bias can also be introduced if a highly selected group is enrolled into a trial(eg, in order to ease the demands of patient-recruitment or minimize inter-subjectvariability). The treatment effect in such a group might well be different from that57