RCR class #9 Flashcards
Def power
The abilty to correctly declare a treatment benefit (find an effect if it exists)
power equation
power = 1-B where B is the prob of a type II error (failure to reject the null when the alt is true
What power do we want in an RCT
80% to 90% these are important converntions
T/f power is the prob of finding a stat sig difference
of size delta
F, it is the probability of finding a difference between the two interventions
If a trial is underpowered it should…
not be done
Too high a power does what
too much he wasted resources and declared a tiny irrelevant effect.
Just right power
80-90%
trial 80% 90% is for what effect size
the minimum effect size that would be clinically signifcatn
Is it better to over or under estimate power
OVer is better , if we overestimate can always stop early for benefit.
Why are RCTs underpowered f? (2)
clinical benefit of a new intervention is oft lower than estimated.
- benefit of a new intervention found in an RCT is generally lower than in a population or clinic samples.
2. unaccounted for LTFup and cross over
effect size?
the minimal effect detected within a given sample size.
Key question about effect size?
What is the minimal effect size that would be clinically relevant.
Dichotomous/categorical outcome delta?
Difference in proportions between two arms
Cont outcome delta
difference in means in two arms/ SD
Time to event delta
HR reduction for control relative to experimental
where get estimates for power analysis
phase II studies. if not availbe look at other studies or ask clinicians to estimate
4 elements to sample size calc of a cateogorical variable
pc,
p1
power (1-B)
type I error =0.05 two sided
confusing abbrev for std
delta
when computing sample size and LTFup how do adjust
can adjust the effect size to estimate the sample size that reflects the adjusted effect size.
