Individual Participant Data Meta-Analysis. Группа авторов
over the conventional aggregate data approach (Table 2.1).7,9,43,44 A key benefit is the potential to improve the quantity and quality of data, because there is no need to be limited by what has been published. For example, IPD from unpublished trials can be included (Section 4.2.3), as can any outcomes that were not reported for published trials, or even participants who were inappropriately excluded from the original trial analyses.7,9,43 As well as helping to circumvent potential reporting biases,46 this can increase the quantity of information available for analysis and, therefore, boost the statistical power to detect genuine effects.47 In addition, there is greater ability to standardise outcome and covariate definitions across trials (Section 4.5), which not only facilitates the conduct of meta‐analysis, but also aids the interpretation of findings. Detailed data checking helps to ensure the completeness, validity and internal consistency of data items for each trial, further enhancing data quality (Section 4.5.4),7,9,43,44 as well as providing independent scrutiny of the trial data.
Table 2.1 Key potential advantages of an IPD meta‐analysis project compared with a conventional systematic review and meta‐analysis of aggregate data focusing on the synthesis of randomised trials to evaluate treatment effects, adapting those shown by Tierney et al.9
Source: Adapted from Tierney et al.,9 with permission, © 2015 Tierney et al. (CC BY 4.0).
| Aspect of systematic review or meta‐analysis | Advantages of an IPD meta‐analysis project |
|---|---|
| Trial identification and inclusion | Ask collaborative group (trial investigators and other experts in the clinical field) to help identify eligible trials (particularly those that are unpublished or ongoing)*Clarify a trial’s eligibility with the trial’s investigators* |
| Data completeness and uniformity | Include data from trials that are unpublished or not reported in full*Include unreported data (e.g. unpublished subgroups, outcomes and time‐points), more complete information on outcomes, and data on participants excluded from original trial analyses*Check each trial’s IPD for completeness, validity and consistency, and resolve any queries with trial investigatorsDerive new or standardised outcome definitions across trials or translate different definitions to a common scaleDerive new or standardised classifications of participant‐level characteristics, or translate different definitions to a common scaleUpdate follow‐up of time‐to‐event or other time‐related outcomes beyond those reported* |
| Risk of bias assessment | Clarify trial design, conduct and analysis methods with trial investigators*Resolve unclear risk of bias assessments (i.e. based on trial reports) through direct contact with investigators*Examine trial IPD directly for evidence of potential bias in trial design and conduct, and resolve any queries with trial investigatorsObtain extra data where necessary to alleviate or mitigate against potential biases* |
| Analyses | Apply a consistent method of analysis for each trial (independent of original trial analyses)Analyse all important outcomes irrespective of whether published*Explore validity of analytical assumptions e.g. normality of residuals in a linear regression analysisDerive outcomes and measures of effect directly from IPD (independent of trial reporting), potentially at multiple time‐points of interestUse a consistent unit of analysis for each trial (e.g. consistently analyse preterm birth events per mother rather a mix of per mother and per baby in trials that include twin pregnancies)Account for complexities in each trial in the analysis, such as cluster randomised trials or multi‐centre trialsAnalyse continuous outcomes on their continuous scale and adjust for baseline valueAdjust for a pre‐defined set of prognostic factorsApply consistent definitions for categorised data (e.g. stage of cancer)Conduct more detailed and appropriate analysis of time‐to‐event outcomes (e.g. handling of censored observations, generating Kaplan Meier curves, examination of non‐proportional hazards)Achieve greater power for assessing interactions between effects of interventions and participant‐level characteristicsModel associations at the participant level, including potential non‐linear relationshipsUse appropriate but non‐standard models (e.g. that account for repeated measurements or correlation between multiple outcomes) or measures of effectExplain potential heterogeneity and inconsistency in network meta‐analysisAddress additional important questions over and above efficacy, or not considered by original trials e.g. to explore the natural history of disease, prognostic factors or surrogate outcomes |
| Interpretation | Discuss implications for clinical practice and research with a multi‐disciplinary group of collaborators including trial investigators who supplied data, and patient research partners* |
| Dissemination | Achieve more widespread dissemination though collaborative group networks and patient groups |
*These advantages accrue from direct contact with trial investigators (rather than the IPD per se), so potentially could be achieved for conventional systematic reviews if more active communication with trial investigators were adopted. This is seldom done in practice.
In general, having access to IPD also supports more flexible and sophisticated analyses than are possible with only existing aggregate data. IPD are vital for a thorough investigation of participant‐level associations, for example to identify treatment effect modifiers (Chapter 7).7,9,43 For instance, an IPD meta‐analysis project by the Early Breast Cancer Trialists Group, which combined IPD from 37,000 women in 55 randomised trials, established that the drug tamoxifen works better in the subgroup of breast cancer patients who are classed as oestrogen receptor positive.15
With IPD, there is no need to rely on, or be restricted by, the original trial methods of analysis. For example, the IPD meta‐analysis research team could opt for alternative effect measures (e.g. hazard ratios rather than odds ratios) or assumptions (e.g. non‐proportional rather than proportional hazards), as appropriate, and consider a broader set of outcomes than originally reported. Collecting IPD also allows continuous variables to be analysed on their continuous scale; potential non‐linear relationships to be examined; and the analysis of outcomes, covariates (e.g. prognostic factors) and time‐points that were recorded, but not originally analysed by trial investigators.
As most IPD meta‐analysis projects are collaborative endeavours, direct contact with trial investigators can help to identify trials that may not be easily identifiable via other forms of searching,7,9,43 and to clarify the eligibility of potentially relevant trials. Trial investigators can also provide extra information leading to more reliable risk of bias assessments