When and how should I combine patient-level data and literature data in a meta-analysis?
Jonathan L. French
Meta-analysis is an integral part of the model-based drug development paradigm . While meta-analysis of individual patient data (IPD) is the gold-standard against which other types of meta-analyses are compared, IPD is not always available for all studies included in a meta-analysis. In particular, a sponsor will typically have access to IPD from their internal compounds, but only have access to aggregate level data (AD) from literature sources for studies which they did not conduct. When both IPD and AD are available, it seems intuitively attractive to combine both types of data into a single model. In this talk we will discuss three approaches for doing this: a two-stage approach in which the IPD are reduced to AD, a hierarchical model approach [2,3] in which a model for the AD is derived from an IPD model, and a Bayesian approach in which the AD is used to form prior distributions for parameters in a model for the IPD. We will demonstrate some of the difficulties with all three of these approaches, including the potential for ecological bias when constructing non-linear models under the hierarchical or Bayesian approach [4,5]. We conclude with some recommendations about when and how best to combine IPD and AD in a meta-analysis.
 Lalonde RL, Kowalski KG, Hutmacher MM et al. Model-based drug development. Clin Pharmacol Ther. 2007; 82: 21-32.
 Goldstein H, Yang M, Omar R et al. Meta-analysis using multilevel models with an application to the study of class size effects. Appl Statist. 2000; 49: 399-412.
 Sutton AJ, Kendrick D and Coupland CAC. Meta-analysis of individual- and aggregate-level data. Statist. Med. 2008; 27: 651-69.
 Berlin JA, Santanna J, Schmid CH et al. Individual patient- versus group-level data meta-regressions for the investigation of treatment effect modifiers: ecological bias rears its ugly head. Statist. Med. 2002; 21: 589-624.
 Wakefield J. Ecological studies revisited. Annu. Rev. Public Health. 2008; 29: 75-90.