This preview shows pages 1–2. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.View Full Document
Unformatted text preview: Meta-Analysis in Clinical Trials Meta-analyses allow researchers to combine results from a number of published studies. Seminal book: Hedges and Olkin. A shorter guide is available free online from BMJ (Egger). Used perhaps more in observational studies, but also quite a bit of use in clinical trials. See QUORUM statement for guidelines for clinical trials. Reasons to carry out a meta-analysis in clinical trial setting: 1. To summarize a large complex body of literature. 2. To resolve conflicting reports. 3. To increase power or improve precision over many small studies. 4. To clarify, quantify strengths and weaknesses of study. 5. To document need for Phase III RCT, or trial in subgroup. 6. To avoid time and expense of major RCT. 7. To investigate sources of variation in published reports. 8. To improve generalizability of previous reports. Conduct of a meta-analysis It is absolutely critical to have a written protocol for doing a meta-analysis. The protocol should pay particular attention to: Searching the literature for previously reported studies. Inclusion and exclusion criteria for reported studies. Data extraction from written reports, including definitions of treatment and outcome. Statistical analysis plan. Quality control and reproducibility of the decisions made and data reported for each trial. The literature search is critical. Here are some helpful guidelines and questions. Define what period of time will be covered by the search (by when published? by when trial conducted? Both?) What journals will be used (only peer reviewed? only in English?) Dont just trust computer searches! Example: A MEDLINE search by a trained librarian identified only 29% of trials in neonatal hyperbilirubinemia and only 56% of trials for intraventricular hemorrhage that were listed in the Oxford Database on Perinatal Trials. Many trials produce multiple publications; decide in advance how to handle these. Abstracts may not contain enough information, though they can confirm existence of studies. Decide how you will reduce and identify publication bias. Will you include unpublished results? This has both strengths and weaknesses....
View Full Document
- Winter '06