Individual Participant Data Meta-Analysis. Группа авторов

Figure 4.9 Date (shown by year‐month) participants were allocated to treatment and control in a trial excluded from an IPD meta‐analysis, because participants in one group (‘arm 1’) were generally recruited earlier than those in the other group (‘arm 0’).

      Source: Lesley Stewart.

As statistical software and database packages allow dates to be converted to days of the week, another simple check is to look at the number of participants allocated to the research and control groups on each day of the week.^7,106 This method would highlight, for example, if participants were being allocated to treatment on particular clinic days (pseudo‐random allocation) or if participants were being allocated on the weekends, which would be unusual for trials in chronic conditions in many countries. For example, in a trial included in an IPD meta‐analysis examining pre‐operative chemotherapy for lung cancer,⁸⁸ there were no weekend randomisations, and the numbers randomised to each treatment group were well balanced on the weekdays (Figure 4.10(a)). In contrast, for a trial included in an IPD meta‐analysis examining post‐operative radiotherapy for lung cancer,¹⁰⁷ there appeared to be an unusually high number of weekend randomisations, and large imbalances in the number of participants allocated to each group on each day (Figure 4.10(b)). When this was brought to the attention of the trial investigator, they discovered problems with the management of the trial data, and went back to individual participant records, to ensure that the appropriate information was supplied, and the issues were resolved.

      As mentioned in Section 4.4.1, if the dates of randomisation have been redacted from trial IPD for de‐identification purposes, it will not be feasible for the central research team to employ these checking procedures, but the trial statistician may be able to run these on their behalf. Moreover, it is still possible to visually check whether baseline characteristics appear reasonably balanced by group, as we would expect with a robust randomisation process. However, balance will never be perfect, and imbalances may be more pronounced in small trials or those with simple (i.e. non‐stratified) randomisation methods; indeed we might be concerned if everything appeared too perfectly balanced. Note that we do not advocate statistical tests of baseline balance.¹⁰⁸

      4.6.2 Deviations from the Intended Interventions

      While robust randomisation procedures should ensure the unbiased assignment to, and comparison of, participants between treatment groups, this can only be guaranteed if all participants are analysed according to the treatments initially assigned: an intention‐to‐treat approach.^109–111 Even if a trial has not been analysed appropriately, as long as the IPD have been provided with the original treatment allocation recorded, then participants can be grouped according to this treatment allocation, rather than the treatment they received, enabling an intention‐to‐treat analysis of effectiveness.^109–111 However, there may be value in conducting certain analyses based on a subset of participants randomised, such as an analysis of toxicity in just those who received most of their allocated treatment, or sensitivity analyses according to treatment received, to explain differences between published trial results and those used in the meta‐analysis.

      There may be sufficient detail in the trial dataset to check whether participants who deviated from intended interventions did so for pre‐specified or otherwise rational reasons. For example, if the data indicate that a participant had experienced an adverse event, this might explain why treatment was stopped early, and would also need to be considered in any analysis of adverse outcomes. It may also be possible to assess whether deviations from planned treatment are similar, and for comparable reasons across treatment groups (more so than with aggregate data).

If a treatment is a major procedure, such as surgery, or particularly toxic, then those delivering treatments and participants will usually be aware of the assigned treatment. Provided that outcomes are objectively measured or ‘hard’, such as mortality, this is unlikely to introduce bias. However, a carer might inadvertently or otherwise deliver a treatment or measure a more subjective outcome, such as an adverse effect, differently if they are aware of which treatment a participant received. Similarly, if a participant is aware of their assigned treatment, it might influence a patient‐reported outcome, such as pain or quality of life. Therefore, in this scenario, there is a potential for bias in this domain, which cannot be alleviated by the collection of IPD. However, the contact with trial teams, that is intrinsic to collaborative IPD meta‐analyses, can provide useful clarification of the methods used to blind participants, carers or outcome assessors, to help determine whether these are appropriate, and therefore allow the risk of bias to be judged with more accuracy.