Results

13. Participant flow It can be useful to describe these features in a flow diagram of the type illustrated in Fig. 10.1. It should though be pointed out that, for many multi-centre trials, the first two boxes may be difficult to complete. For such data to be useful or appropriate, it would be necessary to log, prospectively, every patient with the relevant diagnosis and to record in detail the reasons for not entering the trial. To do this comprehensively is no small task for an individual centre, and extremely difficult for a multi-centre trial coordinator to audit. While the intention is to give an indication of the 'representativeness' of the population actually randomized, it fails to do this because there will always be patients who, through lack of time, organizational problems or even investigator bias are never approached about the trial. Often therefore it will be appropriate for the diagram to begin with the 'randomization' box, and certainly all the remaining box items are important to document, if not in a diagram, then in the text. Deviations from the allocated treatment should be described by group, as should any other deviations from the protocol, for example in the way patients were assessed or followed-up. It is important to quantify the degree of compliance with protocol interventions and the reasons for non-compliance; if good results are achieved despite non-compliance it may indicate potential areas for simplification. Conversely, if'negative' results are associated with poor compliance, could anything be done in the future to improve compliance? If not, then inability to comply is an important conclusion of the trial, if yes, further trials may be needed to determine whether improved compliance is associated with better results.

14. Recruitment It is always important to know the time period over which patients were accrued, partly to place it in its historical context, partly to see how long it took to accrue patients. It may be that some aspects of patient care or assessment have changed over the course of a trial.

15. Baseline data A table of patient characteristics by allocated treatment is one of the first things that should be reported. It serves to demonstrate the characteristics of the actual population entered into the trial (which maybe a small subset of those potentially eligible, which the eligibility criteria themselves describe). It also demonstrates the success, or otherwise, of randomization in achieving approximate balance between the treatment groups with respect to important prognostic variables. As discussed more fully in Chapter 9, it is not necessary to demonstrate balance by carrying out a formal test for the comparability of the groups with respect to each factor, as is sometimes done. If treatment allocation was truly random, differences can only be due to chance.

In addition, prognostic variables are often correlated, and so chance imbalance in one may well be reflected in a similar imbalance in others, without this providing increased evidence of flawed or inadequate randomization. As described in Section 9.4.3, tests for baseline balance are also unhelpful in determining whether, and how, analysis of the main outcome measure should be adjusted for prognostic factors.

16. Numbers analysed The number of patients contributing to various statistics or outcome measures in a trial report may well vary; stating results in absolute numbers makes it clear when patients have been excluded or data are missing.

17. Outcomes and estimation As described in Chapter 9, it is not sufficient to give trial results simply in terms of p-values from hypothesis tests, which are dependent not just on the size of difference observed between groups, but also the sample size. Providing a relevant estimate, and appropriate confidence interval, both for the individual groups and for the measure of difference between them will indicate if statistically nonsignificant results actually provide some indication of what may potentially be a clinically worthwhile effect. Equally, in a large trial, provision of these details will help ensure that the clinical relevance of statistically significant results can be evaluated. Always give exact p-values for all analyses (and argue the case with journals that impose their own constraints) and never use NS (non-significant), and always report the results of all relevant analyses on the primary and secondary outcome measures, not just those which produced 'significant' results.

18. Ancillary analyses Indicate which analyses were pre-specified and justify any that were not. If adjusted analyses of the main endpoint were used, both unadjusted and adjusted results should be presented. The choice of variables for adjustment should be justified (these should have been prespecified see Section 9.4.3).

19. Adverse events Provide estimates of the frequency of the main serious adverse events in each treatment group, and specify any grading system used (e.g. common toxicity criteria).

0 0

Post a comment