Comparison of effect estimates from a meta-analysis of summary data from published studies and from a meta-analysis using individual patient data for ovarian cancer studies

Am J Epidemiol. 1997 May 15;145(10):917-25. doi: 10.1093/oxfordjournals.aje.a009051.


To determine the relative merits of two quantitative methods used to estimate the summary effects of observational studies, the authors compared two methods of meta-analysis. Each quantified the relation between oral contraceptive use and the risk for ovarian cancer. One analysis consisted of a meta-analysis using summary data from 11 published studies from the literature (MAL) in which the study was the unit of analysis, and the second consisted of a meta-analysis using individual patient data (MAP) in which the patient was the unit of analysis. The authors found excellent quantitative agreement between the summary effect estimates from the MAL and the MAP. The MAP permits analysis 1) among outcomes, exposures, and confounders not investigated in the original studies, 2) when the original effect measures differ among studies and cannot be converted to a common measure (e.g., slopes vs. correlation coefficients), and 3) when there is a paucity of studies. The MAL permits analysis 1) when resources are limited, 2) when time is limited, and 3) when original study data are not available or are available only from a biased sample of studies. In public health epidemiology, data from original studies are often accessible only to limited numbers of research groups and for only a few types of studies that have high public health priority. Consequently, few opportunities for pooled analysis exist. However, from a policy view, MAL will provide answers to many questions and will help in identifying questions for future investigation.

Publication types

  • Comparative Study
  • Meta-Analysis

MeSH terms

  • Bias
  • Confounding Factors, Epidemiologic
  • Contraceptives, Oral / adverse effects*
  • Data Interpretation, Statistical*
  • Effect Modifier, Epidemiologic*
  • Female
  • Humans
  • Logistic Models
  • Meta-Analysis as Topic*
  • Odds Ratio
  • Ovarian Neoplasms / chemically induced*
  • Reproducibility of Results
  • Risk Factors
  • Time Factors


  • Contraceptives, Oral