Two-stage methods for the analysis of pooled data

Epidemiologic studies of disease often produce inconclusive or contradictor y results due to small sample sizes or regional variations in the disease i ncidence or the exposures. To clarify these issues, researchers occasionall y pool and reanalyse original data from several large studies. In this pape r we explore the use of a two-stage random-effects model for analysing peal ed case-control studies and undertake a thorough examination of bias in the pooled estimator under various conditions. The two-stage model analyses ea ch study using the model appropriate to the design with study-specific conf ounders, and combines the individual study-specific adjusted log-odds ratio s using a linear mixed-effects model; it is computationally simple and can incorporate study-level covariates and random effects. Simulations indicate that when the individual studies are large, two-stage methods produce near ly unbiased exposure estimates and standard errors of the exposure estimate s from a generalized linear mixed model. By contrast, joint fixed-effects l ogistic regression produces attenuated exposure estimates and underestimate s the standard error when heterogeneity is present. While bias in the poole d regression coefficient increases with interstudy heterogeneity for both m odels, it is much smaller using the two-stage model. In pooled analyses, wh ere covariates may not be uniformly defined and coded across studies, and o ccasionally not measured in all studies, a joint model is often not feasibl e. The two-stage method is shown to be a simple, valid and practical method for the analysis of pooled binary data. The results are applied to a study of reproductive history and cutaneous melanoma risk in women using data fr om ten large case-control studies. Copyright (C) 2001 John Wiley & Sons, Lt d.