Size of Trials Flashcards
What is the null hypothesis?
The null hypothesis states that, in truth, there is no overall difference between treatments in their impact on the measured outcome i.e. that the treatments have the same effect
What is the alternative hypothesis?
The alternative hypothesis states that there is a true difference between the two treatments in respect of their impact on some outcome of interest. In calculating sample sizes we are required to state what this difference might be.
What five key factors are important for determining sample size?
Outcomes - suitable outcome to reflect purpose of the trial
Analysis - binary or continuous data
Results expected in the control group
Treatment difference - what is clinically meaningful
Degree of certainty
What properties define the standard normal deviation curve?
Mean of 0
SD of 1
Area under the curve =1
What variables increase required sample size?
Lower than expected event rate in control arm
Effect difference decreasing
Decreasing significance
Increasing power
Other than statistical power, what other factors may have an effect on overall sample size
Practical factors
Ethical factors
Economic factors
What adjustments may be required to make to sample size after mathematical calculation?
Loss to follow-up = 1/(1-Q)
Participants who receive other treatment = 1 - (1 Q1 - Q2 … Qn)^2
Participants who stop treatment 1-(1-Q)^2
Unequal allocation ratios of ration r:1 = (r+1)^2/4r
What effects may a small study have on outcomes measured?
The observed difference could be far from true difference
The risk of a false negative is increased
The study will be underpowered to detect realistic clinically important differences
The treatment effect will be imprecisely measured
May predispose to publication bias
What are small trials useful for?
Use in meta-analyses
What strategies may be employed to avoid some of the issues associated with a small/underpowered study?
Multi-centre collaboration
Avoid restrictive elibilbility criteria
Avoid trials of more than two treatment groups (they require larger samples) therefore consider parallel two arm study or factorial design if possible
What is the definition of power?
The power of the trial is the probability of being able to detect a specified difference between the treatment groups, given such a difference exists.
What values are most commonly chosen for Type I and Type II error rates in sample size calculation?
0.05 and 0.9 respectively