Meta-analysis methods involve combining and analysing quantitative evidence from related studies to produce results based on a whole body of research. As such, metaanalyses are an integral part of evidence based medicine. Traditional methods for meta-analysis synthesise aggregate study level data obtained from study publications or study authors, such as a treatment effect estimate (for example, an odds ratio) and its associated uncertainty (for example, a standard error or confidence interval). An alternative but increasingly popular approach is meta-analysis of individual participant data, or individual patient data, in which the raw individual level data for each study are obtained and used for synthesis.1 In this article we describe the rationale for individual participant data meta-analysis and illustrate through applied examples why this strategy offers numerous advantages, both clinically and statistically, over the aggregate data approach.1 2 We outline when and how to initiate an individual participant data meta-analysis, the statistical issues in conducting one, how the findings should be reported, and what challenges this approach may bring. What are individual participant data?The term "individual participant data" relates to the data recorded for each participant in a study. In a hypertension trial, for example, the individual participant data could be the pre-treatment and post-treatment blood pressure, a treatment group indicator, and important baseline clinical characteristics such as age and sex, for each patient in each study (table). A set of individual participant data from multiple studies often comprises thousands of patients; this is the case in the table, so for brevity we do not show all rows of data here. This concept is in contrast to the term "aggregate data," which relates to information averaged or estimated across all individuals in a study, such as the mean treatment effect on blood pressure, the mean age, or the proportion of participants who are male. Such aggregate data are derived from the individual participant data themselves, so individual participant data can be considered the original source material. What is an individual participant data meta-analysis?As with any meta-analysis, an individual participant data meta-analysis aims to summarise the evidence on a particular clinical question from multiple related studies, such as whether a treatment is effective. The statistical implementation of an individual participant data meta-analysis crucially must preserve the clustering of patients within studies; it is inappropriate to simply analyse individual participant data as if they all came from a single study. Clusters can be retained during analysis by using a two step or a one step approach. 3 In the two step approach, the individual participant data are first analysed in each separate study independently by using a statistical method appropriate for the type of data being analysed; for example, a linear regression model might be fitted for continuous responses such as blo...

BackgroundA fundamental aspect of epidemiological studies concerns the estimation of factor-outcome associations to identify risk factors, prognostic factors and potential causal factors. Because reliable estimates for these associations are important, there is a growing interest in methods for combining the results from multiple studies in individual participant data meta-analyses (IPD-MA). When there is substantial heterogeneity across studies, various random-effects meta-analysis models are possible that employ a one-stage or two-stage method. These are generally thought to produce similar results, but empirical comparisons are few.ObjectiveWe describe and compare several one- and two-stage random-effects IPD-MA methods for estimating factor-outcome associations from multiple risk-factor or predictor finding studies with a binary outcome. One-stage methods use the IPD of each study and meta-analyse using the exact binomial distribution, whereas two-stage methods reduce evidence to the aggregated level (e.g. odds ratios) and then meta-analyse assuming approximate normality. We compare the methods in an empirical dataset for unadjusted and adjusted risk-factor estimates.ResultsThough often similar, on occasion the one-stage and two-stage methods provide different parameter estimates and different conclusions. For example, the effect of erythema and its statistical significance was different for a one-stage (OR = 1.35, ) and univariate two-stage (OR = 1.55, ). Estimation issues can also arise: two-stage models suffer unstable estimates when zero cell counts occur and one-stage models do not always converge.ConclusionWhen planning an IPD-MA, the choice and implementation (e.g. univariate or multivariate) of a one-stage or two-stage method should be prespecified in the protocol as occasionally they lead to different conclusions about which factors are associated with outcome. Though both approaches can suffer from estimation challenges, we recommend employing the one-stage method, as it uses a more exact statistical approach and accounts for parameter correlation.

ObjectivesIndividual participant data (IPD) meta-analyses often analyze their IPD as if coming from a single study. We compare this approach with analyses that rather account for clustering of patients within studies.Study Design and SettingComparison of effect estimates from logistic regression models in real and simulated examples.ResultsThe estimated prognostic effect of age in patients with traumatic brain injury is similar, regardless of whether clustering is accounted for. However, a family history of thrombophilia is found to be a diagnostic marker of deep vein thrombosis [odds ratio, 1.30; 95% confidence interval (CI): 1.00, 1.70; P = 0.05] when clustering is accounted for but not when it is ignored (odds ratio, 1.06; 95% CI: 0.83, 1.37; P = 0.64). Similarly, the treatment effect of nicotine gum on smoking cessation is severely attenuated when clustering is ignored (odds ratio, 1.40; 95% CI: 1.02, 1.92) rather than accounted for (odds ratio, 1.80; 95% CI: 1.29, 2.52). Simulations show models accounting for clustering perform consistently well, but downwardly biased effect estimates and low coverage can occur when ignoring clustering.ConclusionResearchers must routinely account for clustering in IPD meta-analyses; otherwise, misleading effect estimates and conclusions may arise.