Lecture 7: Sample size calculation

Question 1

What is done in an analysis of RCT data for one follow-up measurement?

Answer 1

• Comparison of changes between baseline and follow-up between the intervention and control group

Question 2

What are two different ways to compare changes between baseline and follow-up from an RCT with one follow-up measurement?

Answer 2

1. Absolute change
2. Relative change

Question 3

What problem can arise when analysing change scores?

Answer 3

Regression to the mean

Question 4

What is meant by regression to the mean?

Answer 4

Regression to the mean refers to the probability that extreme outcomes are followed by more moderate one.

Question 5

What is the consequence of regression to the mean?

Answer 5

Due to randomization, groups in an RCT are assumend to be the same because they were created from the same source population, however…

… when two groups differ in their mean values (at baseline) and the intervention does not have an effect, the group with the highest average value will go down, while the group with the lowest average value will go up

Question 6

1. When would you use ‘relative change’ method?
2. When wouldn’t you use the ‘relative change’ method and what alternative method can be used?

Answer 6

1. The relative change method can be used when the effect of the intervention is negative.
2. If the effect is positive, you could do an analysis of covariance

Question 7

What are two methods to analyse data of an RCT with more than one follow-up measurement?

Answer 7

1. Analyse the two follow-up measurements seperately
2. Analyse the two follow-up measurements in one analysis/ model

Question 8

What is the best way to analyse data of an RCT with more than one follow-up measurement and why?

Answer 8

Analyse the two follow-up measurements in one analysis/model
= mixed model analysis :)

Because this takes into account the dependency of the observations within the same individual

Question 9

What kind of data structure is used for a longitudinal analysis of covariance?

Answer 9

A long data structure (instead of wide)

Question 10

What is reported after performing a longitudinal analysis of covariance?

Answer 10

1. The overall effect on average over time
2. Effect at the follow-up measurements

Question 11

Why would you perform a sample size calculation?

Answer 11

It is necessary for:
- Grant providers
- Ethical committees
- The ‘consort statement’

Question 12

What components are needed for the sample size calculation?

Answer 12

• The expected effect (clinical relevant effect)
• Standarddeviation of the outcome variable
• Power and significance

Question 13

Name two arguments for not performing a sample size calculation

Answer 13

1. Sample size calculations are highly dependent on arbitrary assumptions
“With a few small (but realistic) changes in the parameters, the calculated sample size can differ enormously”

2. Small studies (e.g., in children) would not be possible if sample size calculations were strictly adhered to

Question 14

Sample size calculations are based on…

Answer 14

…testing theory (significance and power)

Question 15

Based on what arguments should the sample size of a study be determined?

Answer 15

• Logistical capabilities
• Financial capabilities
• Ethical considerations