Print ISSN: 2319-2003 | Online ISSN: 2279-0780

IJBCP International Journal of Basic & Clinical Pharmacology

Educational Forum

Small size sampling?

ABSTRACT
Department of Pharmacology,
C.U. Shah Medical College, Though small size samples can be planned and justified based on scarcity of
Surendranagar 363001, India time, money and manpower, there are situations making more accuracy a must
and needing larger samples sizes. That’s why rare adverse drug reactions are
Received: 21 September 2012 identified only after a drug comes into the market and a large population is
Accepted: 25 September 2012 exposed to it. There are many more reasons for increasing the sample size and
requirement of the study decides which criteria of accuracy should be
*Correspondence to: tightened the most (e.g. avoiding type I error is more important or type II
Dr. Rakesh R. Pathak, error).
E-mail: [email protected]
Keywords: Type I error, Type II error, Cohort, Statistical predictors

Yes, the same heading as in the previous issue of the study of pharmacovigilance called post marketing
journal1 but with a “question mark” this time! Would the surveillance.
small size sample work all the while? Are all the large
sample studies just wastage of time, money and effort? Many more reasons are there for larger size sampling as
we can verify from the sample size calculating program.5
A quote from the bible of pharmacology would be In a population of 20,000, error level of 5% and 95%
exemplary - “a cohort study must follow at least 10,000 confidence level leads to a maximal samples size of 377.4
patients who are receiving the drug to detect with 95% It’s to be noted that at 50% response distribution i.e. when
confidence one event that occurs at a rate of 1 in 3300 and there is equal chance of the outcome to occur or not, the
that event can be attributed to the drug only if doesn’t sample size is maximum.
occur spontaneously in the control population”!2
But error level of 1% with the same 95% confidence
Let’s solve this sum ourselves. If we permit 1% margin of level, leads to a maximal samples size of 6489 and if error
error and consider the situation worldwide (population level of 1% with the confidence interval of 99% is
>300,000,000) we can find the number by standard chart considered, the required sample size becomes 9068.4 In
of the research advisors (2006) or online calculator.3,5 the last case, relieving only the error level to 5% (and
confidence level remaining up to 99%) decreases the
But here the case is not that simple. Here the  (mean sample size down to 643.4
event rate) is 1 in 3,300 or 0.0003030303….for this , and
the power (the probability that the mean event rate is Thus the required level of accuracy of conclusion also
greater than that specified) of 0.95, and critical tolerance influences the sample size. Now “how do these different
= 1 (i.e. the prevalence is greater than the mean event rate statistical predictors of accuracy of conclusion (viz.
we have already set) - the sample size of 10,000 (for an confidence level, permitted level of error etc.) manifest in
incidence rate of 0.0003, it is 9986 - to be exact) can be the real life problems” is hinted ahead.
found from the table4 - that too if the incidence of the
adverse event (supposed to be due to the drug) is “not Just to start with, by reducing the risk of type I error (or 
spontaneously occurring in the population” as already error - i.e. false positive, erroneously rejecting the null
stated.2 hypothesis in favor of alternate hypothesis) we tend to
increase type II errors (or  error i.e. false negative,
Up to phase III, total number of people exposed to a new erroneously accepting null hypothesis).6
drug is not as high as this and that’s why many rare
adverse drug reactions are identified only after a drug Seeing the result of a clinical trial, type I error means that
comes into the market and a large population is exposed the test treatment is taken to be effective when actually it
to it and this exposed population is reassessed in phase IV is not. And FDA or any other supervising body would try

www.ijbcp.com International Journal of Basic & Clinical Pharmacology | September-October 2012 | Vol 1 | Issue 2 Page 118
 Pathak RR. Int J Basic Clin Pharmacol. 2012 Oct;1(2):118-119

to minimize this error and prohibit unwarranted 2. Oates JA. The science of drug therapy. In: Brunton
permissions.7 LB, Lazo JS, Parker KL, editors. Goodman &
Gilman's The Pharmacological Basis of
But type II error implies that test treatment is taken as Therapeutics. 11th ed. New York, NY: McGraw-Hill;
ineffective when actually it is. And the drug companies 2005:135.
investing a lot of money in discovering a drug are more 3. The Research Advisors. Sample Size Table.
interested to minimize this type II error. Accordingly, the Available at http://research-
two approaches vary in fixing the criteria of accuracy.7 advisors.com/tools/SampleSize.htm. Accessed 9
September 2012.
Suppose a camera system installed in a car tries to 4. Explanations and Tables for
determine whether certain blobs it sees on the road are sample size for rare events. Available at
pedestrians or not. And if it misses anyone to http://www.stattools.net/SSizRareInc_ExpTab.php.
acknowledge as pedestrian, there are increased chances of Accessed 10 September 2012.
hitting someone - obviously type II error is more 5. Raosoft. Sample size calculator. Available at
important here.8 http://www.raosoft.com/samplesize.html. Accessed
10 September 2012.
Alike is the case where medical imaging software tries to 6. Bland M. An introduction to medical statistics, Third
find and mark suspicious regions, which could be a Edition. Oxford University Press, Great Britain,
serious tumor. Likewise, if a medication with some fatal 2000.
side effect is being tested, type II error would be 7. Chow SC. Controversial Statistical Issues in Clinical
obviously more important.8 Trials. Chapman and Hall/CRC Press, Taylor and
Francis, New York; 2011:22.
But suppose a prosecutor is trying to prove that the 8. Introduction to probability and statistics for
defendant is guilty using a blood test. Here type I error is computer engineers. Available at
more important as innocent person should not be http://www.cmpe.boun.edu.tr/courses/cmpe343/fall2
punished.8 The same is true when a factory wishes to 010/PS8%20Key.pdf. Accessed 9 September 2012.
make sure whether a costly innovation is worth it or not.9 9. Levine DM, Krehbiel TC, Berenson ML. (2010).
Business Statistics: A First Course, 5th ed., Upper
For screening test of HIV in blood donors, false positive Saddle River, NJ: Prentice Hall; 2010: 295.
is not that dangerous and small type I error is allowed and 10. Type-1-error. Available at
cheaper tests are used but when it comes to the diagnosis http://www.nationmaster.com/encyclopedia/Type-1-
in patients, much costly test are used to avoid this false error. Accessed 5 September 2012.
positive labeling.10

REFERENCES

1. Pathak RR. Small size sampling. Int J Basic Clin

International Journal of Basic & Clinical Pharmacology | September-October 2012 | Vol 1 | Issue 2 Page 119