Free Analysis Research Tool Version 1.
7
for Sample Size Iterative Estimation
at your fingertips
License: This spreadsheet is freeware. The user is granted conditional use to copy, duplicate, and distribute
provided that the user doesn't remove my name or the identifying information. I would also like to be acknowledged
in the event that your clinical trial is published and you used this spreadsheet to calculate the sample size (fame is
so much more fun than money). The user of this workbook acknowledges that it is not responsible if study fails, is
underpowered, or is overpowered. The user of this workbook acknowledges that sample size estimation is at the
very best just that - an estimate. No guarantees are provided on the success of a study, even if the sample size is
carefully researched and justified. This workbook is not validated, but has been checked against the examples
contained in each spreadsheet in this workbook and against other programs. I am also not responsible in the event
that someone modifies this program in such a way to make it non-functional, and then passes it along. To obtain an
original copy, please contact me at: [email protected], or download from http://individual.utoronto.ca/ddubins
I would like to very warmly thank the the following people for making this workbook possible:
Russell V. Lenth, Professor
Department of Statistics & Actuarial Science
The University of Iowa, Iowa City, IA 52242 USA
http://www.stat.uiowa.edu/~rlenth/
I dedicate this software to Dr. Richard Moisan B.Phm, Ph.D., my former supervisor, mentor, and friend.
You are greatly missed.
References:
Julious, SA. Statist. Med. 2004; 23:1921–1986
Shein-Chung Chow, Jun Shao, Hansheng Wang. Sample Size Calculations in Clinical Research. Marcel Dekker,
Inc. New York • Basel 2003.
http://www.springerlink.com/content/u503p62056413677/fulltext.pdf
http://www.fda.gov/ohrms/dockets/ac/06/slides/2006-4241s2_5.ppt
http://www.fda.gov/ohrms/dockets/ac/06/slides/2006-4241s2_3.ppt
http://www.fda.gov/ohrms/dockets/ac/06/slides/2006-4241s2_4.ppt
http://bebac.at/news/2006-10-06.htm
http://www.fda.gov/cder/Guidance/3616fnl.pdf
http://www.fda.gov/ohrms/dockets/ac/04/briefing/4034B1_07_Bioequivalence%20Requirments-Draft.pdf
David Dubins, B.A.Sc., Ph.D.
Global Bioequivalence Consulting
36 Reiner Road
Toronto, Ontario, Canada
M3H 2L2
[email protected] Superiority, Parallel (Scroll down for an example)
Desired Power 80% Probability of not getting a Type II error (not getting a false negative)
r (note: nA*r=nB) 1 This is the 'allocation ratio' of nB over nA (for equal sample sizes, r=1)
a 5% Probability of a Type I error (false positive) in the event that there is no significant
s 21.4 Common standard deviation
d 22 Difference in means (m2-m1) --> this is the expected clinically relevant difference.
d = d/s 1.028037
s 21.4
nA 16
nA(r+1)-2 30
t(1-alpha/2,nA(r+1)-2) 2.042272
tau1 2.907729
Probt(1) #VALUE!
1-b #VALUE!
nA nB Total
Required Sample Size 16 16 32
NOTE: This is a 2-tailed test. For a 1-tailed test, multiply alpha by 2 (and pre
EXAMPLE:
Source: Julious, SA. Statist. Med. 2004; 23:1921–1986
lse negative)
ple sizes, r=1)
that there is no significant difference between means
cally relevant difference.
tiply alpha by 2 (and pretend it's the same alpha).
Repfactor (for Boss Button)
2
Superiority, Crossover (Scroll down for an example)
Desired Power 90% Probability of not getting a Type II error (not getting a false negative)
a 5% Probability of a Type I error (false positive)
sw 20 Intrasubject standard deviation, taken from the residual line of an ANOVA model
d 10 Difference in means (m2-m1) --> this is the expected clinically relevant difference.
d = d/sw 0.5
sw 20
n 87
n-2 85
t(1-alpha/2,n-2) 1.988268
tau1 3.297726
Probt(1) #VALUE!
1-b #VALUE!
Required Sample Size 87 (total sample size)
NOTE: This is a 2-tailed test. For a 1-tailed test, multiply alpha by 2 (and pre
EXAMPLE:
Source: Julious, SA. Statist. Med. 2004; 23:1921–1986
(Note: this was a typo, Table II gave a total sample size of 87 in the paper).
lse negative)
line of an ANOVA model
cally relevant difference.
iply alpha by 2 (and pretend it's the same alpha).
Repfactor (for Boss Button)
1
Equivalence, Parallel (Scroll down for an example)
Desired Power 80% Probability of not getting a Type II error (not getting a false negative)
r (note: nA*r=nB) 1 This is the 'allocation ratio' of nB over nA (for equal sample sizes, r=1)
a 2.5% Probability of a Type I error (false positive).
s 100 Common standard deviation
d 10 Maximum allowable difference in means (m2-m1) --> this is the expected clinica
D 0 Expected difference in means (m2-m1 expected).
d = d/s 0.1
s 100
nA 2103
nA(r+1)-2 4204
t(1-alpha,nA(r+1)-2) 1.9605284333
tau2 -3.242684073
tau1 3.2426840734
Probt(2) #VALUE!
Probt(1) #VALUE!
1-b #VALUE!
nA nB Total
Required Sample Size 2103 2103 4206
EXAMPLE:
Source: Julious, SA. Statist. Med. 2004; 23:1921–1986
getting a false negative)
equal sample sizes, r=1)
-m1) --> this is the expected clinically relevant difference.
Repfactor (for Boss Button)
2
Equivalence, Crossover (Scroll down for an example)
Desired Power 90% Probability of not getting a Type II error (not getting a false negative)
a 2.5% Probability of a Type I error (false positive).
sw 20 Expected within-subject population standard deviation
d 10 Maximum allowable difference in means (m2-m1) --> this is the expected clinica
D 0 Expected difference in means (m2-m1 expected).
d = d/sw 0.5
sw 20
n 106
n-2 104
t(1-alpha,n-2) 1.9830375265
tau2 -3.640054945
tau1 3.6400549446
Probt(2) #VALUE!
Probt(1) #VALUE!
1-b #VALUE!
Required Sample Size 106
EXAMPLE:
Source: Julious, SA. Statist. Med. 2004; 23:1921–1986
getting a false negative)
-m1) --> this is the expected clinically relevant difference.
Repfactor (for Boss Button)
1
Non-inferiority, Parallel (Scroll down for an example)
Desired Power 80% Probability of not getting a Type II error (not getting a false negative)
r (note: nA*r=nB) 1 This is the 'allocation ratio' of nB over nA (for equal sample sizes, r=1)
a 5.0% Probability of a Type I error (false positive)
s 40 Common standard deviation
d 10 Maximum allowable difference in means (mA-mB) --> this is the expected clinically r
D 0 Expected difference in means (mA-mB expected)
d = d/s 0.25
s 40
nA 338
nA(r+1)-2 674
t(1-alpha,nA(r+1)-2) 1.647118
tau1 3.25
Probt(1) #VALUE!
1-b #VALUE!
nA nB Total
Required Sample Size 338 338 676
EXAMPLE:
Source: Julious, SA. Statist. Med. 2004; 23:1921–1986
lse negative)
ple sizes, r=1)
is the expected clinically relevant difference.
Repfactor (for Boss Button)
2
Non-inferiority, Crossover (Scroll down for an example)
Desired Power 90% Probability of not getting a Type II error (not getting a false negative)
a 2.5% Probability of a Type I error (false positive)
sw 20 Expected within-subject population standard deviation
d -10 Maximum allowable difference in means (mA-mB) --> this is the expected clinically r
D 0 Expected difference in means (mA-mB expected)
d = d/sw -0.5
sw 20
nA 87
nA(r+1)-2 85
t(1-alpha,nA(r+1)-2) 1.988268
tau1 3.297726
Probt(1) #VALUE!
1-b #VALUE!
Required Sample Size 87
EXAMPLE:
Source: Julious, SA. Statist. Med. 2004; 23:1921–1986
lse negative)
is the expected clinically relevant difference.
Repfactor (for Boss Button)
2
Bioequivalence, Crossover (Scroll down for an example)
Desired Power 80%
a 5%
intra-subject CV (ISV) 25.0% ISV = sqrt(Mean Square Error) / ((mA + mB)/2). You can also get a quick and dirty e
Expected mT/mR Ratio 95% (to 105.3%)
Lower Equivalence Limit 80.0% 75
Upper Equivalence Limit 125.0% 133
sw 0.246221
n 37
n-2 35
t(1-alpha,n-2) 1.689572
tau2 -4.794063
tau1 3.002006
Probt(2) #VALUE!
Probt(1) #VALUE!
1-b #VALUE!
Required Sample Size 37
44.4
EXAMPLE:
(Note: Table VII provided sample size estimates at 90% power.)
Source: Julious, SA. Statist. Med. 2004; 23:1921–1986
Dave's note: in industry for BE studies, the tendency is to use a Test/Reference ratio of 1.05 and a power of 80%.
also get a quick and dirty estimate taking 0.5 * total CV (which is just published standard deviation / mean)
If you can find a BE study in the literature that reports the
Test/Reference Ratios and 90%CI for Cmax, AUCt, or AUCinf, you
can use this macro to back-calculate the ISV. If you're lucky
sometimes authors will publish the ISV directly.
and a power of 80%.
Repfactor (for Boss Button)
1
Bioequivalence, Replicate (Scroll down for an example)
Desired Power 80% 3 period replicate (ABB/BAA)
a 5% Method C1 - EMEA Approach
intra-subject CV (ISV) 90% ISV = sqrt(Mean Square Error) / ((mA + mB)/2) - estimate with 1/2 of total CV
Expected mT/mR Ratio (%) 95% (to 105.3%)
Lower Equivalence Limit 55.69% Reference-Scaled Parameters
Upper Equivalence Limit 179.57% k (Method C1)
k (= sb2 / sw2) 1 (used for 2-Period Repl. only) sW0 (Method C2)
sWR 0.770277 θ: (Method C2)
n 257 # Subjects (Method D)
n-2 255
t(1-alpha,n-2) 1.650851 EMEA: Method C1, 3- or 4-period replica
tau2 -9.370054 FDA: Method C2, 3-period replicate (TR
tau1 7.860338
Probt(2) #VALUE!
Probt(1) #VALUE!
1-b #VALUE!
Required Sample Size 193
EXAMPLE:
Source: Julious, SA. Statist. Med. 2004; 23:1921–1986
wn for an example)
estimate with 1/2 of total CV
Scaled Parameters
0.760
0.294
0.57606836
24
Method C1, 3- or 4-period replicate, k=0.760, maximum of 69.84 – 143.19%.
Method C2, 3-period replicate (TRR/RRT), sW0=0.294
Multiply n by:
2 period re 1.5
3 period re 0.75
4 period r 0.5
2
0.75
Reference-Scaled BE Lower CI: Upper CI:
No Reference Scaled BE Limits 0.8 1.25
Method A - Direct Expansion of BE 0.75 1.33
Method C1 - EMEA Approach 0.556877 1.795728
Method C2 - FDA Approach 0.55731 1.794334 alpha
Method D 0.508844 1.965239 1.717144
3
beta
1.321237
Bioequivalence, Parallel (Scroll down for an example)
Desired Power 90%
r (note: nT*r=nR) 1 This is the 'allocation ratio' of nR over nT (for equal sample sizes, r=1)
a 5%
inter-subject CV 80%
Expected mT/mR Ratio 100%
Lower Equivalence Limit 80%
Upper Equivalence Limit 125%
s 0.7033464593
nT 216
nT*(r+1)-2 430
t(1-alpha,nT(r+1)-2) 1.6484049694
tau2 -3.2970604726
tau1 3.2970604726
Probt(2) #VALUE!
Probt(1) #VALUE!
1-b #VALUE!
nA nB Total
Required Sample Size 216 216 432
EXAMPLE:
Source: Julious, SA. Statist. Med. 2004; 23:1921–1986
equal sample sizes, r=1)
Repfactor (for Boss Button)
1
Scaled Bioequivalence Limits - Highly Variable Drugs
IntraCV of Reference Product 30.0% Note: This CV will be used to calculate BE limits for each method on this
Proposed FDA Approach for Scaled Bioequivalence Limits
06 October 2006 meeting of the Advisory Committee for Pharmaceutical Sciences (ACPS)
Scaled average bioequivalence, based on within subject variability of reference
Method C2 (Haidar et al, 2008)
Permitted SD (σw0): 0.2936 (Note: Presentation by Barbara Davit suggests 30% cutoff fo
θ: 0.577639 If CVWR > 30%, FDA will use the reference-scaled average B
Corresponding Permitted IntraCV: 30.0% If CVWR < 30%, FDA will use the unscaled average BE appro
σwr^2: 0.086178
Limits of μT - μR (±sqrt(θ*σwr^2)): 0.223144
Bioequivalence Limits: 80.00 - 125.00%
Bioequivalence Limits for Point Estimate: 80.00 - 125.00%
Sources:
http://www.springerlink.com/content/u503p62056413677/fulltext.pdf
http://www.fda.gov/ohrms/dockets/ac/06/slides/2006-4241s2_5.ppt
* Calculation of BE limits assumes that the 30% cutoff suggested by Barbara Davit be used as the Permitted SD.
If a Permitted SD of 0.25 is used in this method, it will yield the same results numerically as Dr. Midah.
Other Discussed Approaches
Scaled average bioequivalence, based on within subject variability of reference
( 0 . 223
)
sw0 is the SD at which the BE limits ar
BE limits, upper, lower = EXP ± ¿ σ wr swr is either the residual SD (ABE) or t
σ w0
Permitted SD (σw0): 0.25 (Note: Dr. Midha and Dr. Sam Haidar propose a value of 0.25 as a perm
Corresponding Permitted IntraCV: 25.4%
Intra-subject Standard Deviation: 0.29356 (reference product)
BE limits, ln-scale (±) 0.262024
Relationship between σw0 and k of Me
Bioequivalence Limits: 76.95 - 129.96%
Sources:
http://www.fda.gov/ohrms/dockets/ac/06/slides/2006-4241s2_3.ppt
http://www.fda.gov/ohrms/dockets/ac/06/slides/2006-4241s2_4.ppt
http://bebac.at/news/2006-10-06.htm
Expansion of Bioequivalence Limits Based on Fixed Sample Size
Method B (Haidar et al 2008)
Reference/Reference Ratio: 1
Number of Subjects (Completed): 24
σwr^2 0.086178
Standard Error of Difference 0.084744
t(0.05,n-2) 1.717144
Bioequivalence Limits: 86.46 - 115.66%
*Note: for this method, the ISV would have to be >47%, with n=24 and Ref/Ref = 1.
If Ref/Ref <>1, BE limits are not symmetrical about unity.
Perhaps take the reciprocal of the largest confidence limit to rectify? Issue not discussed in the Haidar paper.
Source:
http://www.springerlink.com/content/u503p62056413677/fulltext.pdf
Widening of Bioequivalence Limits Based on Reference Variability
Method C1 (Haidar et al, 2008)
k: 1
σwr 0.29356
Bioequivalence Limits: 74.56 - 134.12%
Sources:
http://www.springerlink.com/content/u503p62056413677/fulltext.pdf
http://www.fda.gov/ohrms/dockets/ac/04/briefing/4034B1_07_Bioequivalence%20Requirments-Draft.pdf
Widening of Bioequivalence Limits Based on Reference Variability
Method C3 (Haidar et al, 2008)
Permitted SD (σw0): 0.25
σwt^2 0.086178 (IntraCV of Test)
σwr^2 0.086178 (IntraCV of Reference)
σd^2 0.018387 (Subject by formulation interaction)
θi 1.596689
Limits for (uT - uR): 0.34527
Bioequivalence Limits: 70.80 - 141.24% (Individual BE)
Sources:
http://www.fda.gov/cder/Guidance/3616fnl.pdf
http://www.springerlink.com/content/u503p62056413677/fulltext.pdf
Expansion of Bioequivalence Limits Based on Sample Size and Scaling
Method D (Haidar et al, 2008)
# Subjects Desired in Study (2n): 24
Type I Error (alpha) 5% (consumer risk)
Type II Error (beta) 10% (producer risk, corresponding to a power of 90% to determine bioequival
σwr 0.29356
t(alpha,2n-2) 1.717144
t(beta/2,2n-2) 1.717144
Bioequivalence Limits: 74.75 - 133.78%
Sources:
http://www.springerlink.com/content/u503p62056413677/fulltext.pdf
http://www.fda.gov/ohrms/dockets/ac/04/briefing/4034B1_07_Bioequivalence%20Requirments-Draft.pdf
iable Drugs
imits for each method on this worksheet.
al Sciences (ACPS)
ability of reference (Dr. Haider notation)
Davit suggests 30% cutoff for IntraCV)*
e reference-scaled average BE approach
e unscaled average BE approach
d as the Permitted SD.
ability of reference (Dr. Midah notation)
e SD at which the BE limits are permitted to be widened (set by an agency)
her the residual SD (ABE) or the SD of the reference product (replicate design)
se a value of 0.25 as a permitted SD)
hip between σw0 and k of Method C1 (below)
the Haidar paper.
ents-Draft.pdf
and Scaling
90% to determine bioequivalence)
ents-Draft.pdf
Given Precision, Parallel (Scroll down for an example)
r (note: nA*r=nB) 1 This is the 'allocation ratio' of nB over nA (for equal sample sizes, r=1)
a 5% Probability of a Type I error (false positive)
s 25 Common standard deviation
w 5 Half-width of the confidence interval desired.
d 0.2
s 25
nA 194
nA(r+1)-2 386
t(1-alpha/2,nA(r+1)-2) 1.966129
(r+1)t^2*s^2/(rw^2) 193.2831
nA nB Total
Required Sample Size 194 194 388
EXAMPLE:
Source: Julious, SA. Statist. Med. 2004; 23:1921–1986
ple sizes, r=1)
Given Precision, Crossover (Scroll down for an example)
a 5% Probability of a Type I error (false positive)
sw 10 Common standard deviation
w 5 Half-width of the confidence interval desired.
d 0.5
sw 10
n 34
n-2 32
t(1-alpha/2,n-2) 2.036933
2t^2*s^2/(w^2) 33.19278
Required Sample Size 34
EXAMPLE:
Source: Julious, SA. Statist. Med. 2004; 23:1921–1986
Difference Between Proportions (Binomial Test)
(Scroll down for an example)
Desired Power 80%
a 5% Probability of a Type I error (false positive)
p0 10% Predefined reference level (non-responder)
p1 30% Minimum cure rate level we would like to detect
d 20% (This is like d in the other spreadsheets)
rcrit 5 This is the critical number of responders above which the null hypothesis is reject
n 25
1-b 0.806512
Required Sample Size 25 (including responders + non-responders)
Required Responders 5 (# responders out of the total sample size needed to reject the null hypothesis)
(good for smaller sample sizes, <= 170)
H0: a difference of d does not exist between treatment and control.
H1: a difference exists between treatment and control.
If the results of the study see more than rcrit responders, you can reject H0 and conclude there is a more than d % differe
EXAMPLE:
New York • Basel 2003.
mial Test)
ve which the null hypothesis is rejected.
eded to reject the null hypothesis)
lude there is a more than d % difference.
Variance Correction (Baseline and Post-dose Measures)
Is the data you're looking at to plan your trials baseline corrected?
If so the variance may be too high to plan your sample size around.
The variance in baseline-corrected trials can be corrected for:
The variance can simply be divided by the factor on the right.
s2 reported 20
p 3 Number of baseline measures taken per individual
r 0.5 Pearson Correlation Coefficient between observations
Variance Correction Factor:
0.625
s2 adjusted: 12.5
2. Post-dose Measures Correction.
s2 reported 20
r 5 number of post-dose measures
r 0.5 correlation between post-dose measures
Variance Correction Factor:
0.6
s2 adjusted: 12
3. Accounting for Baseline(s) and Post-dose Measures as Covariates
s2 reported 20
r 5 number of post-dose measures
r 0.5 correlation between post-dose measures
p 4 Number of baseline measures taken per individual
Variance Correction Factor:
0.2
s2 adjusted: 4
Source: Julious, SA. Statist. Med. 2004; 23:1921–1986
se Measures)
Allied Research International Inc.
Study Design
Mississauga, Canada
Intra-Subject Variability Calculator
POINT ESTIMATE AND CONFIDENCE INTERVALS to ISV
Ratio 1.0072 Ratio of Test to Reference
Lower 0.9022 Lower Confidence Interval
Upper 1.1122 Upper Confidence Interval
N 35 # Subjects
CI range 90%
DF 33
T90 1.6923603
SEEL 0.0650531
SEEH 0.0585962
MSEL 0.0740583
MSEH 0.0600864
ISVL 27.725368
ISVH 24.885405
ISV 26.3 %
858624173.xls 03/04/2025 - 15:55:18
