Statistics Collaborative - Design and analysis for biomedical research



Meta-analysis involves combining data from different sources in a statistically sound way in order to arrive at an overall estimate of an outcome of interest. Statistics Collaborative, Inc. (SCI) is particularly helpful to clients in first gaining a thorough understanding of the clinical and statistical issues that might arise when data are combined from various sources.

Some of the statistical matters involving meta-analysis that SCI can address with substantial experience include:

  • whether a fixed or random-effects model is more appropriate in the particular circumstances at hand for combining data from various studies;
  • whether the odds ratio or the relative risk is the more appropriate statistic to generate from the meta-analysis;
  • how to estimate the effect robustly when only a small number of studies are available for inclusion in the meta-analysis; and
  • how to assess the size and impact of the variability of the data. Combining study populations that are quite different will affect the variability of the combined estimate. SCI performs sensitivity analyses to assess the impact of any particular study on the overall estimate of the outcome of interest.

Examples of SCI's work in meta-analyses:

  • Antimicrobial agent for preventing pneumonia: SCI conducted a systematic literature review identifying 90 articles reporting the results of randomized clinical trials. We abstracted data from 25 relevant articles and performed a meta-analysis of them, addressing the statistical issues of heterogeneity and the appropriate way to combine the data statistically. We found a statistically significant odds ratio indicating a treatment effect.
  • Chromium supplementation for treating patients with diabetes: SCI conducted a systematic literature review identifying approximately 40 articles reporting the results of randomized clinical trials. We abstracted data from 12 relevant studies and combined them using a fixed-effect model. Although the combined estimate did not show a study treatment effect, SCI thoroughly explored the impact of combining the selected studies. We summarized the results both in tabular form and in forest plots.
  • Vaccine efficacy for influenza: SCI combined data from nine trials, some previously unpublished, that evaluated the use of the FluMist®, an inhaled influenza vaccine, in pediatric populations. Because only a small number of studies were included in the meta-analyses (as few as three), SCI determined that it would be best to use a fixed-effects model rather than adopting the typical random-effects approach for combining studies in a meta-analysis.