CHEO Research Institute Logo
Decrease Text SizeIncrease Text SizeFacebookTwitterYoutube

Guidance with Respect to Sample Size Calculations 

What is a clinical trial?
A clinical trial is a study that measures the outcomes of human participants who receive an intervention. There are numerous elements involved when a REB is deciding whether a clinical trial is ethical. One of those considerations is whether the clinical trial is likely to result in scientifically valid data. An appropriate sample size is essential to making that determination.

What should a research protocol contain regarding sample size?
The research protocol for a clinical trial must identify the specific outcome(s) that will be used to assess the efficacy of an intervention. If several outcomes will be used, one must be identified as the primary outcome. In order to determine whether the results of a clinical trial are statistically significant and therefore have the potential to improve health care, the study must plan to enrol a sufficient number of participants. The research protocol must indicate the desired study sample size and provide a suitable justification for it.

Why are REB’s interested in sample size calculations?
Sample size calculations are inherently ethical. A clinical trial that plans to recruit too few participants may be unable to accomplish its statistical objectives, thereby jeopardizing its scientific validity. This reduces the study’s potential to benefit society, threatening the risk/benefit ratio presented to participants during the informed consent process. Conversely, a clinical trial should not expose more participants than absolutely necessary to an unproven and potentially harmful intervention.

Should I consult a statistician?
Yes. These guidelines are intended to outline requirements for the simplest of sample size calculations and will not be applicable to all studies. Investigators contemplating clinical trials are strongly encouraged to consult with a statistician regarding sample size calculations prior to submitting a proposal to the REB. When applicants do not provide a suitable justification for the proposed sample size, they will be asked to do so prior to final REB approval.

What about pilot studies?
The REB recognizes that many clinical trials being proposed are innovative and may not be able to provide all parameters needed to justify the sample size. When faced with this situation, it is generally recommended to first conduct a small pilot study. Pilot studies provide investigators with an opportunity to assess the feasibility of their study methods with a small number of participants prior to undertaking a large clinical trial. In addition, one of the objectives of a pilot study is to gather data on the outcome measure that will be used in sample size calculations for future clinical trials. If an investigator is unable to provide the REB with any credible data on which to base a sample size calculation, and they are in fact proposing to conduct a pilot study, a statement to that effect may be provided in lieu of a sample size calculation.

Steps for calculating a sample size for a clinical trial with a continuous outcome measure:
  1. Specify the primary outcome and outcome measure
    It is best to choose a primary outcome that is clinically relevant and for which a validated and responsive outcome measure is available. Although multiple outcomes may be collected, one must be identified as the primary outcome.
  2. Identify the minimal clinical important difference (MCID) in the primary outcome measure.  This is the threshold difference below which clinicians do not feel the improvement noted is important enough to change their practice. This can be established from previous studies or may have been reported when the outcome measure was validated.
  3. Specify the expected variance in the outcome measure
    The clinical presentation of participants will often vary over time and can be established from a pilot study, previous studies for the same intervention/indication/population, or may have been reported when the outcome measure was validated.
  4. Select the desired power level
    The power level varies from 0 to 1 (or 0-100%) and represents the likelihood that a study will report statistically significant results with the parameters entered into the sample size calculation.
  5. Select the type-I error rate
    The type-I error rate also varies from 0 to 1 (or 0-100%) and represents the likelihood that a study will report statistically significant results when no true difference exists between the groups being compared.
  6. Select 1-sided or 2-sided hypothesis testing
    A 1-sided sample size calculation can only report whether the intervention is statistically significantly superior to the control. A 2-sided sample size calculation can report whether the intervention is statistically significantly superior to the control, as well as whether the control is in fact superior to the intervention. Investigators should always select a 2-sided sample size calculation unless there are compelling reasons not to do so.
  7. Estimate study loss to follow-up: The sample size calculation reports the final number of participants required for data analysis, not the starting number of participants enrolled. It must therefore be increased to adjust for projected study dropouts.
  8. Input parameters into sample size statistical software.


Impact of individual components on sample size
Here is the impact of varying each component of the sample size calculation on the sample size:

Component

Impact

Primary outcome measure

A more responsive outcome measure decreases sample size

MCID

A larger MCID decreases sample size

Variance

A smaller variance decreases sample size

Power

A smaller power decreases sample size

Type-I error rate

A larger type-I error rate decreases sample size

Tails

A 1-tailed analysis decreases sample size

Loss to follow-up

A lower rate of loss to follow-up decreases sample size


Other comments
When a clinical trial will measure multiple outcomes that are of equal importance, investigators can provide a sample size calculation for each outcome in the research protocol. The largest sample size among these equally important outcomes can then be used for enrolment purposes.

Rather than providing the result of a sample size calculation as a single number (i.e. n=48 participants), investigators are encouraged to provide a range of numbers based on different assumptions regarding the parameters. A research protocol could provide a table or graph reporting the required sample size for a range of values for each parameter (i.e. MCID, standard deviation, power, type-I error rate, tails, loss to follow-up) if there is uncertainty requiring their estimated values.

Example
A clinical trial is designed to determine the efficacy of an analgesic compared to a placebo for post-surgical pain.

Here are the parameters reported in the research protocol for the sample size calculation:

  1. The primary outcome is pain and the outcome measure is the visual analogue scale (VAS).
  2. The MCID for post-surgical pain using the VAS is 2.0.
  3. The expected standard deviation in VAS in this population is 1.5.
  4. A 2-sided calculation is chosen since both possibilities are of clinical interest.
  5. A power of 90% is chosen.
  6. A type-I error rate of 0.01 is chosen.
  7. A loss to follow-up rate of 25% is expected.
  8. The above parameters are entered into sample size calculation software, with the result that each study group must enroll 24 participants for a total of 48 participants.

Impact of individual components on sample size

Here is the impact of varying each component of the sample size calculation on the sample size:

Component

Impact

Primary outcome measure

A more responsive outcome measure decreases sample size

MCID

A larger MCID decreases sample size

Variance

A smaller variance decreases sample size

Power level

A smaller power level decreases sample size

Significance Type-I error rate

A larger significance type-I error rate decreases sample size

Tails

A 1-tailed analysis decreases sample size

Loss to follow-up

A lower dropout rate of loss to follow-up decreases sample siz


Other comments

When a clinical trial will measure multiple outcomes that are of equal importance, investigators can provide a sample size calculation for each outcome in the research protocol. The largest sample size among these equally important outcomes can then be used for enrolment purposes.

Rather than providing the result of a sample size calculation as a single number (i.e. n=48 participants), investigators are encouraged to provide a range of numbers based on different assumptions regarding the parameters. A research protocol could provide a table or graph reporting the required sample size for a range of values for each parameter (i.e. MCID, standard deviation, power level, significance Type-I error rate, tails, loss to follow-up) if there is uncertainty requiring their estimated values.

Example

A clinical trial wants is designed to determine the efficacy of an analgesic compared to a placebo for post-surgical pain.

Here are the parameters reported in the research protocol for the sample size calculation:

  1. The primary outcome is pain and the primary outcome measure is the visual analogue scale (VAS).
  2. It is established from previous studies that the MCID for post-surgical pain using the VAS is 2.0.
  3. It is established from previous studies that the expected standard deviation in VAS in this population pain is 1.5.
  4. A 2-tailed sided calculation is chosen since both possibilities are of clinical interest.
  5. A power level of 0.9 (i.e. 90%) is chosen.
  6. A, with a significance Type-I error rate of 0.01 is chosen.
  7. A dropout loss to follow-up rate of 25% is expected.
  8. The above parameters are entered into sample size calculation software, with the result that each study group must enroll 24 participants for a total of 48 participants.

sample size

powerlevel
 

Take Action
Quick Links

Our Researchmagnifying glass

abcefg hijklmnopqrst uvwxyz