5. Sample size for precision and power, multiple hypothesis testing

Report
Section V
Sample Size and Power
Multiple hypothesis testing
Sample size (n) based on
estimation precision-CI width
Can plan sample size so that standard
errors (SEs) and the corresponding
confidence intervals are sufficiently
small.
(How small? How small do you need it
to be?)
Sample size for precision/CIs
π = proportion with TB in the population (prevalence)
p = proportion with TB in a sample of size n,
SE(p) = √[π(1- π)/n]
95% confidence interval for π: p ± 1.96(SE)
Precision: want to estimate true prevalence (π) within ± 6%
Solve for n: 1.96(SE) = 1.96 √[π(1- π)/n] = 0.06
n = 1.962 π(1- π)/(0.06)2 = 3.84 π(1- π)/0.0036
Can estimate π using the observed p or use maximum at π=0.5
If π=0.15, n = 3.84 (0.15)(0.85)/0.0036 = 136
At π=0.50, n = 3.84 (0.50)(0.50)/0.0036 = 267 (worst case)
Rule of thumb: For 95% CI for π, conservative n for precision w
is n = 1/w2
Sample size based on
hypothesis testing
Hypothesis test decision table
Test:
Do not reject null
Test:
Reject null
hypothesis
No difference
in population
(Null is true)
Actual
difference
(Null is false)
1-α
(correct)
β
(Type II error)
α
(Type I error)
1-β = power
(correct)
Determinants of power
Power (1-β) depends on
δ = delta = true difference
σ = sigma = true SD or true variation
α = alpha = significance criteiron
n = sample size
(Or, n depends on δ, σ, α, 1-β )
Alpha versus Power
The top distribution shows
the sampling distribution of
a test statistic under the
assumption that delta is zero
(the null hypothesis is true).
•
α
-3.0
-2.0
-1.0
0.0
1.0
2.0
3.0
1-β
-1.0
0.0
1.0
2.0
3.0
The bottom distribution shows
the true population distribution
(unknown at the time of testing),
with a true population delta=2.5.
1-β
4.0
5.0
6.0
Power calculation
Zpower = Zobs – Zα
Treatment
n
Mean HBA1c
chg
SD
SE
Liraglutide
5
-1.24
0.99
0.44
Sitaglipin
4
-0.90
0.98
0.49
Difference
5.5 – 4 = 1.5
√[0.442 + 0.492] = 0.66
Zobs=0.34/0.66 = 0.516 (<1.96, so not statistically
significant, p=0.622)
Zpower = Zobs - Z α = 0.516 – 1.96 = -1.44
From the Gaussian table, Zpower=-1.44 yields power
about 7%.
Interpretation of power
If test is “statistically” significant (p < α), we have a
“positive” or “significant” outcome (& accept the
false positive probability of α).
If test is not statistically significant (p > α) either
there is no relationship (“negative” outcome) or
sample size is inadequate (inconclusive
outcome).
If power is low for a given δ, results are
inconclusive, not negative.
If power is high, results are affirmatively negative.
(But better to quote Confidence interval after the
study is published)
Power when testing difference between 2 means
Two independent groups, each with sample size
n = 2 (Zpower + Zα)2 (σ/δ)2
Zα = 1.96 and Zpower = 0.842 (for power of 80%), so
n = 2(0.842 + 1.96 )2 (σ/δ)2 = 15.7 (σ/δ)2
or
n/group approximately ≈ (range/δ)2
(since 15.7 ≈16, 16(σ/δ)2 = (4σ/δ)2 and the range ≈4σ )
Power for increasing delta
δ =2.5
δ=0
-3
-2
δ = 3.5
-1
0
1
2
3
4
5
6
7
Areas under the curves and right of the vertical line are
α for the black curve and power for the other curves.
The power is larger for the red curve than for the blue.
Power Summary
Power increases as
•
•
•
•
True difference (δ) increases
Sample size (n) increases
α increases (less strict significance criterion)
Patient heterogeneity (σ) decreases
Generally, we set α = 0.05 & power = 1 – β = 0.80
To determine n, we need to estimate δ and σ.
Often we use values of δ/σ for the calculation
For time to event outcome (survival), n also depends on f/u time.
Sample Size Checklist
Effect size (δ) = smallest clinically important difference
(n increases as δ decreases)
Variability = patient heterogeneity = group SD
(n increases as variability increases)
Power = prob of detecting the effect, set at 80% or higher
(n increases with power)
α level = prob of rejecting when δ=0, usually set at 0.05, two-sided
(n decreases with larger α)
** for time to event (survival) outcomes ***
Time of comparison = how long it takes to achieve the effect (n decreases
with time)
Follow up time = time each patient is followed (n reduced if patients followed
longer). In survival “n” is the number who have the outcome.
Also consider
percentage who will agree to participate
patient accrual rate, dropout / loss rate
Sample size for selected δ/σ - means
(2 means difference= δ, SD=σ, two-sided α=0.05)
δ/σ
70% power
80% power
90% power
0.10
1,234
1,570
2,102
0.15
549
698
934
0.20
309
392
525
0.25
198
251
336
0.50
49
63
84
0.75
22
28
37
1.00
12
16
21
1.25
8
10
13
1.50
5
7
9
Sample size for two proportions
Smaller of P1 &
P2
Difference between P1 and P2
|P1- P2|=δ
0.05
0.10
0.15
0.20
0.05
434
140
79
45
0.10
685
199
99
62
0.15
904
250
120
72
Pseudo replication
Most variation is between persons, not within person.
Two blood samples on n=10 is not a sample size of 20.
Observed value = true population mean
+ between person variation (σp)
+ within person variation (σe)
Example: To estimate the mean
1. Compute a mean for each person using her “m”
observations per person.
2. Compute the group mean from the “n” person means.
SEM = √[σp2/n + σe2/nm], usually σe < σp
Statistical vs Medical “significance”
(“A difference, in order to be a difference, must
make a difference”–Gertrude Stein?).
Average drop in weight (kg) after 3 months
Diet Mean Drop
I
0.50
II
10.0
p
95% CI
<0.001 (0.45,0.55)
0.16
(-5.0, 25.0)
Multiple Hypothesis testing
Multiple efficacy endpoints /outcomes
Multiple safety endpoints/outcomes
Multiple treatment arms and/or doses
Multiple interim analyses
Multiple patient subgroups
Multiple analyses
Exploratory vs confirmatory
Who killed Tweety Bird?
Did Sylvester do it?
Motivation (class discussion)
Tweety Bird is murdered by a cat who left a
DNA sample. The particular DNA profile
found in the sample is known to occur in one
of every one million cats. There is also about
a 0.1% false positive rate for this test.
Is the level of evidence (guilt) equal in these
two scenarios?
1. Sylvester only is tested and is a match.
2. A DNA database on 1 million cats, including
Sylvester, is searched and Sylvester is a
match.
Motivation (class discussion)
The “disease score” ranges from 2 (good) to
12 (worst).
Scenario A: Due to prior suspicion (prior
information), only patients 19 and 47 are
measured and both have scores of 12. We
report that they are “significantly” ill.
Scenario B: The score is measured on 72
patients. Only patients 19 and 47 have
scores of 12. We report that they are
“significantly” ill.
Is the amount of “evidence” or “belief” that
patients 19 and 47 “really” are very ill (have
“true” score of 12) the same in both
scenarios? The data for patients 19 and 47
are the same in both scenarios.
Most would agree that, if both patients were
retested (confirmation step), and came out
with lower scores, this would decrease the
belief that there “true” score is 12. If they
came out with 12 again, this would increase
the belief that the true score is 12.
Multiple testing
“If you torture the data long enough, it will eventually
confess”
Two different situations for new arthritis treatment compared
to aspirin.
A. Only pain (0-10) and swelling (0-10) are measured. Both
are significantly better at p < 0.05 on the new treatment
compared to aspirin.
B. Ten different outcomes measured: pain, swelling, activities
of daily living, quality of life, sleep, walking, bending, lifting,
grinding, climbing. Only the two that are significant are
reported after all 10 are evaluated. (fraud?)
Confirmatory studies specify outcomes in advance. Misleading
to report only statistically significant results.
How to really lie with stats for fun and profit
1. Bet on the horse after you know who
won (Movie -> The “Sting”)
2. Send financial advice after you know how
the market did (example in class)
Multiple Testing
Assume all results are reported, not just the significant ones.
Bonferroni: out of m independent tests at 0.05, num significant by
chance alone, even when all null hypotheses are true (assumes
independence).
# tests=m
Probability reject at least one
1
0.0500
2
0.0975
3
0.1426
4
0.1855
5
0.2262
10
0.4013
20
0.6415
25
0.7226
50
0.9231
Multiple testing-What to do?
Option 1: Use nominal alpha level for
significance. Creates too many false
positives.
Option 2: Use Bonferroni criterion –Declare
significance if p < α/m if “m” tests are
made. Has too many false negatives.
Option 3: Use Holm/Hochberg criterion – a
compromise
Holm/Hochberg criterion
Rule for m (not necessarily independent) significance
tests. Keeps overall false positive rate at α.
1)Sort the “m” p values from lowest to highest.
2)Declare the ith ordered p significant if it is less
than α/(m+1-i). If p > α/(m+1-i), this & all
larger p values are declared non significant.
Overall type I error rate (FWER) is ≤ α.
FWER = family wise error rate
Holm/Hochberg Example for m=5, α=0.05
i
1
2
3
4
5
p value
p1-smallest
p2
p3
p4
p5-largest
α/(6-i)
α/5
α/4
α/3
α/2
α
0.05/(6-i)
0.0100
0.0125
0.0167
0.0250
0.0500
No adjustment vs Hochberg vs Bonferroni
0.06
m=5, α=0.05
significance criterion
0.05
no adjustment
0.04
Bonferroni
Hochberg
0.03
p value
0.02
0.01
0
1
2
3
i
4
5
m=5, alpha=0.05
i
no adjustment Bonferroni Hochberg p value
1
0.05
0.01
0.0100
0.007
2
0.05
0.01
0.0125
0.011
3
0.05
0.01
0.0167
0.014
4
0.05
0.01
0.0250
0.044
5
0.05
0.01
0.0500
0.049
FWER vs FDR
If a “family” of “m” hypothesis tests are carried
out, the family wise error rate (FWER) is the
chance of any “false positive” type I error
assuming that the null is true for all m tests.
Rather than control the FWER, it may be
preferable to control the number of “positive”
tests (not all tests) that are false positives. This
is called controlling the false discovery rate
(FDR), a less stringent criterion.
For FDR, the ith ordered p value must be less
than (i/m)α which is larger than α/(m+i-i) for
FWER.
FDR vs FWER
errors committed when testing “m” null hypotheses
Declared non sig
Declared sig
Total
Truth-Null true
U
V
m0
Truth-Null false
T
S
m-m0
total
m-R
R
m
FWER= V/m (prob V ≥ 1)
FDR = V/R (average V/R)
FDR is more liberal
FWER vs FDR significance criteria
m=5 hypothesis, 5 p values
α=0.05
p value
FDR criteria
FEWR criteria
p1-smallest
(1/5) α=0.01
α/5=0.01
p2
(2/5) α=0.02
α/4=0.0125
p3
(3/5) α=0.03
α/3=0.0167
p4
(4/5) α=0.04
α/2=0.025
p5-largest
α=0.05
α=0.05
Multiple testing & primary outcomes
As “m’, the number of outcomes, increases, individual αi
for each outcome must be smaller so n must be larger if
overall α is to stay constant (ie at α=0.05).
But not all outcomes are equally important. Designate
important outcomes “primary” & the rest secondary so m
is only the number of primary outcomes. Assumes less
concern if there is a false positive finding among
secondary outcomes.
Must designate primary vs secondary outcomes in
advance, before study results are known. It is not fair to
declare which outcomes are primary and which are
secondary based on their p values.
Statistical Analysis Plan
Statistical models and methods to answer study questions
Conclusions = data + models (assumptions)
Each specific aim needs a stat analysis section.
Sample size and power follows the analysis plan.
Outline:
•Outcomes: denote primary & secondary
•Primary predictors or comparison groups
•Covariates/confounders/effect modifiers
•Methods for missing data, dropouts
•Interim analyses (for efficacy, for safety)
Common Methods
Univariate analysis
Continuous outcome: Means, SDs, medians
Time to event:
Survival curves
Discrete:
Proportions
Multivariate analysis
Continuous outcome: Linear regression,correlation
Positive integers:
Poisson regression
Binary (yes/no):
Logistic regression
Time to event:
Proportional hazard regression
ANOVA and t-test are special cases of linear regression

similar documents