Week 9(1).pdf
Week 9(1).pdf
Week 9(1).pdf
INTRODUCTION not be able to detect the difference between test groups, making
the study unethical.
One of the pivotal aspects of planning a clinical study is the
calculation of the sample size. It is naturally neither practical On the other hand, if we study more subjects than required,
nor feasible to study the whole population in any study. we put more individuals to the risk of the intervention, also
Hence, a set of participants is selected from the population, making the study unethical, and waste precious resources,
which is less in number (size) but adequately represents the including the researchers’ time.
population from which it is drawn so that true inferences about
the population can be made from the results obtained. This set The calculation of an adequate sample size thus becomes
of individuals is known as the “sample.” crucial in any clinical study and is the process by which we
calculate the optimum number of participants required to be
In a statistical context, the “population” is defined as the able to arrive at ethically and scientifically valid results. This
complete set of people (e.g., Indians), the “target population” is article describes the principles and methods used to calculate
a subset of individuals with specific clinical and demographic the sample size.
characteristics in whom you want to study your intervention
(e.g., males, between ages 45 and 60, with blood pressure Generally, the sample size for any study depends on the:[1]
between 140 mmHg systolic and 90 mmHg diastolic), and • Acceptable level of significance
“sample” is a further subset of the target population which • Power of the study
we would like to include in the study. Thus a “sample” is a • Expected effect size
portion, piece, or segment that is representative of a whole. • Underlying event rate in the population
• Standard deviation in the population.
ATTRIBUTES OF A SAMPLE Some more factors that can be considered while calculating the
• Every individual in the chosen population should have an final sample size include the expected drop-out rate, an unequal
equal chance to be included in the sample. allocation ratio, and the objective and design of the study.[2]
• Ideally, choice of one participant should not affect the
chance of another’s selection (hence we try to select the LEVEL OF SIGNIFICANCE
sample randomly – thus, it is important to note that random
sampling does not describe the sample or its size as much Everyone is familiar with the “p” value. This is the “level of
as it describes how the sample is chosen). significance” and prior to starting a study we set an acceptable
value for this “p.” When we say, for example, we will accept
The sample size, the topic of this article, is, simply put, the a p<0.05 as significant, we mean that we are ready to accept
number of participants in a sample. It is a basic statistical that the probability that the result is observed due to chance
principle with which we define the sample size before we start (and NOT due to our intervention) is 5%. To put it in different
a clinical study so as to avoid bias in interpreting results. If words, we are willing to accept the detection of a difference 5
we include very few subjects in a study, the results cannot be out of 100 times when actually no difference exists (i.e., get a
generalized to the population as this sample will not represent “false positive” result). Conventionally, the p value of 5% (p
the size of the target population. Further, the study then may = 0.05) or 1% (p = 0.01), which means 5% (or 1%) chance of
erroneously reporting a significant effect is accepted.
Address for correspondence:
Dr. Supriya S. Bhalerao, Department of Clinical Pharmacology,
TNMC and BYL Nair Hospital, Mumbai Central, Mumbai 400 001, POWER
India. E-mail: [email protected]
Sometimes, and exactly conversely, we may commit another
DOI: 10.4103/0974-7788.59946 type of error where we fail to detect a difference when actually
there is a difference. This is called the Type II error that detects UNDERLYING EVENT RATE IN THE POPULATION
a false negative difference, as against the one mentioned above
where we detect a false positive difference when no difference The underlying event rate of the condition under study
actually exists or the Type I error. We must decide what is the (prevalence rate) in the population is extremely important
false negative rate we are willing to accept to make our study while calculating the sample size. This unlike the level of
adequately powered to accept or reject our null hypothesis significance and power is not selected by convention. Rather,
accurately. it is estimated from previously reported studies. Sometimes it
so happens that after a trial is initiated, the overall event rate
This false negative rate is the proportion of positive instances proves to be unexpectedly low and the sample size may have
that were erroneously reported as negative and is referred to to be adjusted, with all statistical precautions.
in statistics by the letter β. The “power” of the study then is
equal to (1 − β) and is the probability of failing to detect a
STANDARD DEVIATION (SD OR Σ)
difference when actually there is a difference. The power of a
study increases as the chances of committing a Type II error Standard deviation is the measure of dispersion or variability
decrease. in the data. While calculating the sample size an investigator
needs to anticipate the variation in the measures that are being
Usually most studies accept a power of 80%. This means that
studied. It is easy to understand why we would require a smaller
we are accepting that one in five times (that is 20%) we will
sample if the population is more homogenous and therefore
miss a real difference. Sometimes for pivotal or large studies,
has a smaller variance or standard deviation. Suppose we
the power is occasionally set at 90% to reduce to 10% the are studying the effect of an intervention on the weight and
possibility of a “false negative” result. consider a population with weights ranging from 45 to 100
kg. Naturally the standard deviation in this group will be great
EXPECTED EFFECT SIZE and we would need a larger sample size to detect a difference
between interventions, else the difference between the two
We can understand the concept of “effect size” from day-to- groups would be masked by the inherent difference between
day examples. If the average weight loss following one diet them because of the variance. If on the other hand, we were to
program is 20 kg and following another is 10 kg, the absolute take a sample from a population with weights between 80 and
effect size would be 10 kg. Similarly, one can claim that a 100 kg we would naturally get a tighter and more homogenous
specific teaching activity brings about a 10% improvement in group, thus reducing the standard deviation and therefore the
examination scores. Here 10 kg and 10% are indicators of the sample size.
claimed effect size.
In the above-mentioned formula σ is the standard deviation Calculating for a 10% drop-out rate one would need to
(estimated) and Δ the difference in effect of two interventions complete approximately 400 patients per arm to be able to
which is required (estimated effect size). say with any degree of confidence whether a difference exists
between the two treatments.
This gives the number of sample per arm in a controlled
clinical trial.
LIMITATIONS OF THE CALCULATED SAMPLE
EXAMPLE SIZE
This issue of the Journal has an article describing the benefits The sample size calculated using the above formula is based on
of ayurvedic treatment AyTP in patients of migraine in an some conventions (Type I and II errors) and few assumptions
open uncontrolled trial design.[3] If anyone wishes to confirm (effect size and standard variation).
these results using a randomized controlled trial design where
the effect of the ayurvedic intervention will be compared to The sample size ALWAYS has to be calculated before initiating
standard of care in headache as measured by VAS how would a study and as far as possible should not be changed during
we plan the sample size? the study course.
As seen above, we need the following values: Zα, Z1−β, σ, The sample size calculation is also then influenced by a few
practical issues, e.g., administrative issues and costs.
standard deviation (estimated), and Δ, the difference in effect
of two interventions. Let us assume we will accept a p<0.05
as acceptable and a study with 80% power; using the above REFERENCES
tables, we get the following values: Zα, is 1.96 (in this case we
will be using a two-tailed test because the results could be 1. Kirby A, Gebski V, Keech AC. Determining the sample size in
bidirectional). Z1−β is 0.8416. The standard deviation (based on a clinical trial. Med J Aust 2002;177:256-7.
2. Larsen S, Osnes M, Eidsaunet W, Sandvik L. Factors
the data in the published paper) would be approximately 0.7. influencing the sample size, exemplified by studies on
For Δ, the paper describes that the ayurvedic therapy has given gastroduodenal tolerability of drugs. Scand J Gastroenterol
a 35% effect. Previously it has been reported that sumatriptan 1985;20:395-400.
3. Prakesh B, Babu SR, Sureshkumar K. Response of Ayurvedic
at 50 mg improves headache by 50%.[4] Thus, the effect size therapy in the treatment of migraine without aura. Int J
would be 15% (i.e., 0.15). Ayurved Res 2010;1:29-35.
4. Cady RK, Sheftell F, Lipton RB, O'Quinn S, Jones M, Putnam
G, et al. Effect of early intervention with sumatriptan on
The sample size for the new study will be migraine pain: Retrospective analyses of data from three
clinical trials. Clin Ther 2000;22:1035-48.
n = 2(1.96 + 0.8416)2(0.72)2
(0.15)2
Source of Support: Nil, Conflict of Interest: None declared
= 362 per arm.