Biology Reference
In-Depth Information
number of genes. In contrast to the estimation scenario, in this case it
may be advantageous rather than harmful to draw upon multiple, het-
erogeneous sources. Heterogeneity should tend to increase the robust-
ness of inferences, thereby enhancing the generalizability of study
conclusions. The effects of within-study bias might also be reduced, as
we would expect different biases in different studies.
3. Individual Patient Data (IPD)
Data for a combined analysis can be obtained in various ways, including
extraction of aggregate data (e.g. test statistics) from published reports
or collection of data at the individual patient level (e.g. from publicly
available databases). Systematic reviews of summaries based on individual
patient data (IPD) are sometimes considered as the gold standard as
compared to meta-analysis based on published summaries.
5
Basing a meta-analysis on published summaries can be problematic
for a number of reasons. Information on patient characteristics or out-
comes may be lacking. In addition, the study may be ongoing after pub-
lication, so the data on which the publication is based may be out of date.
When raw data are available for each study, more flexible and detailed analy-
ses (e.g. subgroup analyses) not carried out by the original investigators can
be performed. Finally, there is scope for further improvement of data
quality when the original data are available for independent scrutiny.
4. SwissBrod: Swiss Breast
Oncology Database
4.1. Difficulties with Public Data Sources
Information such as original data files from public repositories or supple-
mentary files provided as support for a publication are largely article-ori-
ented and tuned to support the paper's claims. Without further refinement,
these are unsuitable for meta-analysis for the following reasons:
(1)
Lack of independent patient cohorts
. Multiple articles may have
used some of the same arrays or the same patients using other
