Table 1 Summary of opportunities for evidence synthesis to inform design, conduct and analysis of a clinical trial

Prior to design

To justify the need for a new trial in light of the existing evidence base.

A systematic literature review and, where appropriate, quantitative synthesis, could be used to assess the need for the new trial [31, 32]. If there are no relevant previous trials, a search strategy might be requested by funders, such as the National Institute for Health Research (NIHR), to support this. Relevant systematic reviews might include existing clinical trials, early-phase trials, nonrandomised comparisons, animal studies or qualitative research studies [7].

Design

Choice of population.

A systematic review may highlight the population and particular subgroups that warrant further investigation [17].

Choice of interventions and comparators.

Results from evidence syntheses, including network meta-analyses, decision models and value of information analyses, can be used to choose which interventions and comparators to trial [33] and characteristics of these, e.g. dose or duration of treatment [4, 17].

Choice of outcomes and length of follow-up.

A systematic review may help inform the choice of outcomes [4, 7] in a new trial and how they should be defined and, if relevant, the duration of follow-up [17]. For example, a systematic review may highlight adverse events that should be monitored, in particular events that are expected and related.

Sample size calculations

A systematic review and/or meta-analysis may provide information on the parameters needed for sample size calculations [4, 17] such as the standard deviation, control group outcome rates, plausible effect sizes, loss to follow-up and correlation coefficients [7]. Alternatively, expected value of sample information calculations can be used to assess the ability of a new trial to inform cost-effectiveness assessment of the intervention and reduce decision uncertainty [29].

Recruitment and consent.

For example, good or poor recruitment rates in previous relevant trials can inform site selection in a new multicentre trial [4].

Monitoring (conduct)

To deal with adverse events

Observed adverse event rates can be compared with predictions from a synthesis of historic data to see if they are higher than expected by chance [34].

To decide whether to stop an ongoing trial

Emerging trial results considered in the context of results from previous studies might be used to make the decision to stop a trial early [17].

Analysis

To inform the statistical analysis plan

Factors, such as measures of effects (event rates, mean difference, etc.) from previous trials, might influence the choice of statistical model. Prognostic or predictive factors identified though evidence synthesis may be used to stratify or adjust trial analyses [17]. Choice of the most important covariates to be recorded for imputation modelling might be informed by patterns of missing data in previous trials.

To assess the trial treatment effect in the context of existing evidence

An existing meta-analysis might be used to form a prior distribution for the treatment effect in a new study which can then be updated using the trial data in a Bayesian statistical analysis.

To adjust for potential biases.

External evidence about typical biases associated with undesirable study characteristics, e.g. inadequate blinding, might come from ‘meta-epidemiological’ studies [35], allowing the analyst to assess the sensitivity of the findings to alternative model assumptions.

To inform secondary parameters.

External evidence might be used to improve the estimation of ‘nuisance’ parameters involved in trial analysis which are often poorly estimated, such as the intra-class correlation coefficient (ICC) in cluster randomised trials [10] and between-centre variability in multicentre trials.

Reporting

To report the new trial results in the context of the wider evidence base.

An updated systematic review [12] or meta-analysis including the new trial results [36] should be reported to put the results in the context of the wider evidence base [17].