Computer Methods and Programs in Biomedicine 64 (2001) 121 – 124

www.elsevier.com/locate/cmpb

Sample size and power calculations in repeated

measurement analysis
Chul Ahn *, John E. Overall, Scott Tonidandel
Uni6ersity of Texas Health Science Center at Houston, 6431 Fannin St., MSB 1.112, Houston, TX 77030, USA

Received 8 November 1999; received in revised form 3 March 2000; accepted 20 April 2000

Abstract

Controlled clinical trials in neuropsychopharmacology, as in numerous other clinical research domains, tend to
employ a conventional parallel-groups design with repeated measurements. The hypothesis of primary interest in the
relatively short-term, double-blind trials, concerns the difference between patterns or magnitudes of change from
baseline. A simple two-stage approach to the analysis of such data involves calculation of an index or coefficient of
change in stage 1 and testing the significance of difference between group means on the derived measure of change
in stage 2. This article has the aim of introducing formulas and a computer program for sample size and/or power
calculations for such two-stage analyses involving each of three definitions of change, with or without baseline scores
entered as a covariate, in the presence of homogeneous or heterogeneous (autoregressive) patterns of correlation
among the repeated measurements. Empirical adjustments of sample size for the projected dropout rates are also
provided in the computer program. © 2001 Elsevier Science Ireland Ltd. All rights reserved.

Keywords: Repeated measures; Sample size estimate; Power calculation; Dropouts

1. Introduction line. The analysis of data from the randomized,

Controlled clinical trials tend to employ a par-
allel-groups repeated measurements design in
which individuals are randomly assigned between
treatment groups, evaluated at baseline, and then
evaluated at intervals across a treatment period of
fixed total duration. The repeated measurements
are usually equally spaced, although not necessar-
ily so. The hypothesis of primary interest in short-
term efficacy trials of perhaps 6 – 8 weeks duration
concerns the difference between treatment groups
in patterns and magnitudes of change from base-
* Corresponding author.
parallel-groups design often focuses on the differ-
ence between experimental and control groups in
the average rate of change as represented by
slopes of regression lines fitted to the mean re-
sponse patterns.
In several recent articles, we have examined the
actual type I error and power provided by alter-
native general linear mixed model formulations,
including procedures that utilize maximum likeli-
hood and related solutions to model random ef-
fects and error covariance structures of the
repeated measurements [1,2]. The purpose of this
article is to make available a computer program
for calculation of sample sizes and power for this
most common clinical trials design based on

0169-2607/01/$ - see front matter © 2001 Elsevier Science Ireland Ltd. All rights reserved.
PII: S 0 1 6 9 - 2 6 0 7 ( 0 0 ) 0 0 0 9 5 - X
122 C. Ahn et al. / Computer Methods and Programs in Biomedicine 64 (2001) 121–124
adaption of more general equations previously
provided by [3] and elaborated computationally for
a two-stage generalized least squares solution by
[2].
For sample size and power calculations, the
computer program to be described considers a
two-stage analysis of the repeated measurements in
which an index or coefficient of change is calculated
for each individual in stage 1, and the significance
of the difference between group means on the
derived measure of change is evaluated against the
within-groups variability of that measure in stage
2 using analysis of variance (ANOVA) or analysis
of covariance (ANCOVA) methods. Stage 1 of the
analysis can specify three different definitions of
subject-specific rate of change, which represents
endpoint change, ordinary least squares (OLS), or
generalized least squares (GLS) regression analysis.
Each provides an estimate of the slope of a regres-
sion line relating the repeated measurements to
corresponding assessment times. For the intent-to-
treat analysis, the number of available repeated
measurements to which the regression model is
fitted is permitted to differ for provided dropouts
and completers.
Dropouts tend to attenuate the power of tests for
response measure in two treatment groups can be
obtained as the area that lies beyond a critical Zb
value in the unit normal curve. The required sample
size can be calculated directly from the power
calculation formula. Type I error can be calculated
by making the treatment effect size D equal to zero
from the power calculations formula. The formulas
presented here for sample size and power calcula-
tions are simplified version in [3], which present
equations for calculating Zb for tests of the ‘equal
change hypothesis’ using simple endpoint change
or ordinary least squares (OLS) regression slopes
as dependent variables. An equation for calculating
power for tests on generalized least squares (GLS)
regression slopes results from substituting x%=
z%C − 1 into calculation of effect size D. The general
'
power equation is of the form
n
Zb = D − Za
2
where Zb is the critical Z-score delineating an area
under the unit normal curve equal to power, Za is
the critical value corresponding to the desired
one-sided or two-sided alpha level, n is the sample
size per groups, and D is the effect size which
depends on the particular definition of change.
evaluating differences in patterns of change across
time in a repeated measurements design. Simula-
x%Cx
tion methods are used to examine the attenuation
of power due to dropouts, and they concluded that
the common practice of increasing the ‘dropout
free’ sample size by the anticipated number of
dropouts is a useful rule-of-thumb [4]. Such
dropout adjusted sample sizes are thus also pro-
vided by the computer program. It first calculates
sample sizes and/or power, with or without the
baseline value entered as a covariate, using all three
definitions of change mentioned above, and then it
also provides adjusted sample size estimates aimed
at maintaining the same desired power for an
intent-to-treat analysis that includes data for an
anticipated 20 or 30% dropouts.
x%d
D=
s
where x% is the contrast vector for endpoint, OLS,
or GLS analysis, d is the vector of differences
between group means, C is the within-groups
correlation matrix for the repeated measurements,
and s is the within-groups standard deviation
which is assumed to be constant across the repeated
measurements. It is common to covary baseline
scores for the simple endpoint analysis, which
involves adjustment of the error variance by the r 2
correlation between baseline and endpoint in calcu-
lating D. The following simplified equations for
calculating Zb are obtained by considering the
different x% linear contrasts for endpoint, OLS, or
GLS analysis.
'
Endpoint analysis without baseline covaried:
2. Sample size and power calculations m − m2t n
Zb = 1t − Za
sw 4(1− r)
The power of a test of significance for the where m1t − m2t is the difference between endpoint
difference between means of a normally-distributed means only, assuming that the expected value of
 C. Ahn et al. / Computer Methods and Programs in Biomedicine 64 (2001) 121–124 123

the baseline difference in the two treatment groups 3. Program description

is zero due to randomization, and r is the within-
groups correlation between baseline and endpoint Power depends on the pattern of treatment
scores. effects across time, the within-groups variance
Endpoint analysis with baseline covaried: which is assumed to be constant across the different

Zb =
m1t −m2t ' n
−Za
time points, the number of equally-spaced repeated
measurements, and the level and pattern of their
sw 2(1 −r 2) intercorrelations, as well as on parameters common
to estimation of sample sizes for simple randomized
OLS slope difference without baseline covaried:

'
designs. Herein we describe the requirements for
x%d n use of a computer program called POWER.EXE,
Zb = − Za which has been written in Microsoft Fortran, to
sw 2(x%Cx)
perform sample size and power calculations for
where x% is the mean-corrected vector of linearly comparing differences in rates of change from
increasing time coefficients, x% =[ −4 − 3 − 2 − 1 baseline produced by two treatments in a controlled
0 1 2 3 4] for a total of nine measurements, d is the clinical trial. The interactive program is easily
mean-corrected vector of postulated linearly in- implemented. It calculates: (1) the required sample
creasing differences between group means, and C size per group for desired level of power at a
is the within-groups correlation matrix among the specified alpha level given a specified minimum (or
repeated measurements. meaningful) endpoint treatment difference; or (2)
OLS with baseline covaried: the statistical power against a specified minimum

Zb =
x%d' n
−Za
(or meaningful) endpoint treatment difference
given specified alpha level and sample size per
sw 2
2(1−r )(x%Cx)
b group.
where rb = x%Cx(1)/(x%Cx)1/2 is the correlation be- To implement the program POWER.EXE, sim-
tween baseline scores and the time-weighted combi- ply type POWER at the DOS prompt. The program
nation of repeated measurements defined by asks whether you want to compute the required
x%= [− 4 −3 − 2 − 1 0 1 2 3 4] with x %(1) = [1 0 sample size per treatment group (two groups) or the
0 0 0 0 0 0 0] for a total of nine measurements. statistical power provided by a specified sample
GLS slope differences without baseline covaried: size. The program next asks whether you want to

Zb =
z%C − 1d ' n
− Za
model the correlational structure as autoregressive
(order 1) or compound symmetry. The program
sw 2(z%C − 1z) then asks user to specify the desired minimal
detectable endpoint mean difference and the
where z% is the vector of linearly increasing time within-groups standard deviation.
coefficients as used for OLS calculations, z%= [− 4 Limitations of the program include the fact that
− 3 −2 − 1 0 1 2 3 4] for a total of nine it calculates sample size and power for a two-group
measurements, and C − 1 is the inverse of the design only, considers fitting linear equations to the
within-groups correlation matrix or model thereof. patterns of change across equally-spaced repeated

'
GLS slope differences with baseline covaried: measurements, and assumes that any dropouts will
z%C − 1d n tend to be relatively uniformly distributed across
Zb = −Za time in the repeated measurements design.
sw 2(1− r 2c )(z’C − 1z)
where rc = z%z(1)/(z%C − 1z)1/2 is the correlation be-
tween baseline scores and the transformed GLS 4. Example
definition of change, where z%= [− 4 − 3 − 2 − 1
0 1 2 3 4] and z%(1) =[1 0 0 0 0 0 0 0 0] for a total Schizophrenia symptomatology is assessed using
of nine measurements. the total score from the Brief Psychiatric Rating
124 C. Ahn et al. / Computer Methods and Programs in Biomedicine 64 (2001) 121–124
Scale (BPRS). Scores have a potential range from
0 to 108, with higher score indicating more severe
pathology. Assessments are carried out at baseline
(week zero) and at weeks 1, 2, 3, 4, 5, and 6. The
primary objective of the study is to compare
gradients of change in BPRS total scores from
baseline to week 6 for placebo and a new therapy
for schizophrenia. The investigator wants to esti-
mate the sample size needed to detect a treatment
effect of medium magnitude with 80% power.
Cohen [5] (p. 26) has characterized a difference of
one-half standard deviation as a treatment effect
of ‘medium magnitude.’ The within-groups corre-
lation structure of the repeated measurements is
assumed to conform to autoregressive (AR(1))
with the (population) baseline-to-endpoint corre-
lation equal to 0.5. Table 1 shows the results from
using the computer program POWER.EXE to
estimate the required sample size.
Table 1
It can be appreciated that power and sample
size for the GLS solution approach those for the
simple endpoint difference-score analysis as the
correlation structure of the repeated measure-
ments approaches a true AR(1) pattern and g
approaches 1. The GLS solution approaches OLS
definition of change when the correlation struc-
ture approaches compound symmetry. The end-
point analysis, and consequently GLS regression,
have superior power and thus require smaller
sample sizes than OLS regression in the presence
of an autoregressive correlation structure. The
OLS and GLS regression provide superior power
in the presence of uniform correlation. An unpub-
lished parametric study examined the sample size
and power requirements for various intermediate
levels of serial dependence, and the conclusion
followed that it is better to use the autoregressive
(AR(1)) option for all cases where any meaningful
degree of serial dependence among the repeated
measurements is present. This is also a conserva-
tive approach as confirmed by the larger sample
sizes that are required for desired power under
autoregressive conditions.
Acknowledgements
This work was supported in part by NIH grants
MH32457 and MO1RR02588.
References
[1] C. Ahn, S. Tonidandel, J.E. Overall, issues in use of SAS
PROC.MIXED to test the significance of treatment effects
in controlled clinical trials. J. Biopharm. Stat., 10 (2000)
265 – 286.
[2] J.E. Overall, C. Ahn, C. Shivakumar, Y. Kalburgi, Prob-
lematic formulations of SAS PROC.MIXED models for
repeated measurements, J. Biopharm. Stat. 9 (1999) 189 –
216.
[3] J.E. Overall, S.R. Doyle, Estimating sample sizes for re-
peated measurements designs, Control. Clin. Trials 15
(1994) 100 – 123.
[4] J.E. Overall, G. Shobaki, C. Shivakumar, J. Steele, Adjust-
ing samples size for anticipated dropouts in clinical trials,
Psychopharmacol. Bull. 34 (1998) 25 – 33.
[5] J. Cohen, Statistical Power Analysis for the Behavioral
Sciences: 2nd ed. Hillsdale, NJ, Lawrence Erlbaum, 1988.