Biology Reference
In-Depth Information
2. Combining Information
For combination of separate study information to be sensible, relevant
quantities emanating from the studies should be linked in some way, even
if only conceptually. How to combine data appropriately depends on the
(joint) alternative of interest. Below, we describe several ways in which
information may be combined.
There is a recent and increasing literature on combining microarray
studies via meta-analysis. The most commonly used meta-analytical
methods, described, for example, by Sutton et al ., 4
typically combine
either parameter estimates or p -values.
A major difficulty with synthesizing information is the occurrence of
study heterogeneity. In general, studies carried out by different research
groups may vary in a number of ways: scientific research goals, popula-
tion of interest, design, quality of implementation, subject inclusion and
exclusion criteria, baseline status of subjects (even with the same selection
criteria), treatment dosage and timing, management of study subjects,
outcome definition or measures, scoring of outcomes (e.g. pathologist
scoring of tumors), data integrity, and statistical methods of analysis.
Additional issues more specific to the microarray context include the
following: differences in the technology used for the study, heterogeneity
of measured expression from the same probe occurring multiple times on
the array, multiple (different) probes for the same gene, variability in the
probes used by different platforms, differences in quantification of gene
expression even when the same technology is used, and large “batch”
effects, due, for example, to a particular laboratory and/or technician.
Genomic studies differ from traditional epidemiological or clinical
trials in several respects. One obvious difference is that the number of
variables measured in genomic studies is usually in the thousands per
sample, rather than the perhaps tens per sample for a clinical trial.
Microarray study sample sizes are also typically much smaller, putting
additional impetus on effective combination. Study goals ordinarily dif-
fer as well.
In a clinical trial, the overall goal is primarily to obtain a combined
estimate treatment effect. Genomic studies more often focus on combin-
ing evidence supporting the role of a gene or to rank evidence for a large
Search WWH ::




Custom Search