Nonrandomized treatments

The specific treatments being randomized will rarely be the only treatments which the patient receives, but these 'other' treatments constitute another source of variability which might be controlled or encouraged. They range from ancillary or supportive care such as antibiotics or antiemetics to more fundamental aspects of treatment. For example, in high grade glioma trials comparing the addition of adjuvant chemotherapy to standard treatment, there are possible sources of variability in standard treatment, for example the extent of surgery a patient had, and the radiotherapy schedule they received. Should any of the 'other treatments' be controlled? It is useful to consider the extremes. The early, exploratory, trial aiming to limit variability wherever possible, might choose to do so - specifying a prophylactic anti-emetic schedule or a specific radiotherapy dose and fractionation, or even imposing a strategy for supportive care which is identical in both arms of the trial, even if this would not be done out of the trial setting. The pragmatic trial would need to consider what would happen outside the trial - is there really any evidence that these variations in non-randomized treatments have an impact on outcome? The MRC BA06 trial in locally advanced bladder cancer [9] addressed the question - does CMV (cisplatin, methotrexate, vinblastine) chemotherapy, in addition to local therapy, improve survival? Internationally, opinion was and is divided on the best local treatment, some centres performing cystectomy, others radical radiotherapy and still others combining both treatments. These differences in routine practice have evolved, as such things do in the absence of trials. However, because there was no clear evidence of superiority of one over the other and, as importantly, no clear evidence that the impact of chemotherapy would differ in a patient undergoing cystectomy as local treatment and one undergoing radiotherapy, the trial permitted any of these forms of local treatment. This added a source of variability, but also doubled the number of centres which could take part and was one of the reasons why the trial recruited so successfully. It also helped us to provide evidence on the value of adding chemotherapy to both local treatments. These issues crop up regularly in relation to meta-analysis, in which a common concern is the combination of trials addressing the same basic question, but perhaps in different groups of patients and against a background of different nonrandomized treatments. In meta-analysis this is addressed by ensuring that comparisons are only made within trials, and the within-trial differences are effectively combined. In an individual trial, this is accomplished by stratifying randomization appropriately and, if necessary, stratifying the analysis too.

0 0

Post a comment