data obtained from a number of different studies. A
new drug application usually consists of many
different studies. The approval of the application
is not based on any single study rather on the
synthesis of the information obtained from all the
studies. Data from some studies will have to be
combined and analyzed. Such an analysis, called
meta-analysis, must be planned for in advance, too.
Two examples of meta-analysis are the integrated
summary of safety (ISS) and the integrated sum-
mary of efficacy (ISE). These analyses must be
planned for just as if the combined database repre-
sented data from a new study. Ideally, plans for
meta-analyses should be made at the time the
individual studies are planned. This is not always
possible, but this is the best way to assure that the
meta-analysis database used for the analysis is
coherent. For example, if the adverse events infor-
mation is colleted in two studies using different
data collection forms, the combination of the indi-
vidual databases may be difficult and some infor-
mation may be lost.
Epilogue
An anonymous cynic once said that there are three
types of liars: liars, damned liars and statisticians.
This statement reflects the discomfort many
researches feel when working with statisticians.
The image of the statistician taking the data to his
or her dark room, performing incomprehensible
manipulations behind closed doors and coming
back with results, charts and magic numbers,
throwing around vaguely understood terms, is
unfortunate. It is my hope that this chapter helps
to disperse the haze and clarify the statistician’s role
and mode of thinking. The statistician is neither a
liar nor a magician; rather, the statistician is a
professional trained in scientific methods devised
to establish causal relationships under conditions of
uncertainty. My goal in writing this chapter was not
to turn the reader into a statistician. Instead, it was
to bring the statistician out of the dark room into the
open, and by reviewing the issues he or she is
concerned about and clarifying basic terminology,
to facilitate communication between the statistician
and the rest of the study team.References and
additional readingArmitage P. 1971.Statistical Methods in Medical
Research. John Wiley & Sons: New York.
Bloomberg Business News. 2004. ‘U.S. doctors may
seek rules for drug makers to disclose studies’.
Bloomberg Business News, June 15.
Bok S. 1974. ‘The ethics of giving placebos’.Sci. Am.
231: 17.
Fisher RA. 1956.Statistical Methods and Scientific
Inference. Hafner: New York.
Freedman D, Pisani R, Purves R. 1978.Statistics. W.W.
Norton & Co.: New York.
Friedman LM, Furberg CD, DeMets DL. 1981.Funda-
mentals of Clinical Trials. John Wright: Boston.
Gribbin M. 1981. ‘Placebos: cheapest medicine in the
world’.New Sci. 89: 64–65.
International Conference on Harmonization (ICH).- ‘E9: Statistical Principles for Clinical Trials’.
No. 1,Federal Register, 63; 179, 49583–49598 (16
September 1998).
PMA Biostatistics and Medical Ad Hoc Committee on
Interim Analysis. 1993. ‘Interim analysis in the
pharmaceutical industry’.Control. Clin. Trials14:
160–173.
Pocock SJ. 1983.Clinical Trials: A Practical Approach.
John Wiley & Sons, Ltd: Chichester.
Popper K. 1959.The Logic of Scientific Discovery.
Basic Books: New York.
The Washington Post. 2004a. ‘Drugmakers prefer
silence on test data. Firms violate U.S. law by not
registering trials’.The Washington Post, 6 July.
The Washington Post. 2004b. ‘Drug firms flout law by
failing to report test data, FDA says’.The Washington
Po st, 7 July.
REFERENCES AND ADDITIONAL READING 343