Pharmaceut. Statist. 2006; 5: 6769
Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/pst.204
Simon Day1,*,y and Scott D. Patterson2
1Medicines and Healthcare Products Regulatory Agency, Room 13-205, Market Towers,
1 Nine Elms Lane, London SW8 5NQ, UK2GlaxoSmithKline Pharmaceuticals, 1250 South Collegeville Road, Collegeville, PA
We bring two changes to the Literature Reviews section of the
journal, beginning with this issue. Firstly, we are introducing a
separate non-clinical review, written by non-clinical experts,
headed by Professor Ludwig Hothorn. As there is rather less
non-clinical statistical material published, that review will
appear biannually, in issues 1 and 3 of the journal. Secondly,
the clinical review (as we might now call it) has a slight change in
authorship and, with that, a very slight change in potential
coverage. But we hope its usefulness will remain at a steady level.
This review covers the following journals received during the
period from the middle of September 2005 to end of December
* Applied Statistics, volume 54, part 5.* Biometrical Journal, volume 47, part 5.* Biometrics, volume 61, parts 2, 3.* Biometrika, volume 92, part 4.* Biostatistics, volume 6, part 4.* Clinical Trials, volume 2, part 5.* Computational Statistics and Data Analysis, volume 50,
parts 13.* Drug Information Journal, volume 39, part 4.* Journal of Biopharmaceutical Statistics, volume 15, part 6.* Journal of the Royal Statistical Society, Series A, volume
168, part 4.* Statistics and Probability Letters, volume 74, parts 14.* Statistics in Medicine, volume 24, parts 2023.* Statistical Methods in Medical Research, volume 14, parts 5, 6.
SELECTED HIGHLIGHTS FROM THELITERATURE
The theme of Statistical Methods in Medical Research was:
* Part 6: Statistics in oral health research (pp. 537602).
One tutorial has appeared in Statistics in Medicine:
* Gurrin LC, Scurrah KJ, Hazelton ML. Tutorial in
biostatistics: spline smoothing with linear mixed models.
Statistics in Medicine 2005; 24:33613381.
There is also a tutorial in Clinical Trials: this one on an
important topic of missing data. It compares various ap-
proaches for imputing missing values. . . from likelihood basedmethods to simple (their term) methods:
* Beunckens C, Molenberghs G, Kenward MG. Direct
likelihood analysis versus simple forms of imputation for
missing data in randomized clinical trials. Clinical Trials
Determining the maximum tolerated dose is the primary goal of
phase I research, and increasingly adaptive methods are being
used in clinical trials where the dose given next is dependent on
previous responses. In oncology trials for example, if a toxicity
is observed, the next patient may receive a lower dose. If no
such event is observed, the next patient receives a higher dose.
This paper reviews coherence principles relating primarily to
modied continual-reassessment methods:
* Cheung Y. Coherence principles in dose-nding studies.
Biometrika 2005; 92:863873.
Copyright # 2006 John Wiley & Sons, Ltd.Received \60\re /teci
*Correspondence to: Simon Day, Medicines and HealthcareProducts Regulatory Agency, Room 13-205, Market Towers, 1Nine Elms Lane, London SW8 5NQ, UK.
Nonlinear modelling of pharmacokinetic and, increasingly,
pharmacodynamic and safety repeated-measures data is becom-
ing more and more common in phase I and clinical
pharmacology research. This paper describes alternatives to
the multivariate normal distribution, most often used for this
* Lindsey J, Lindsey P. Multivariate distributions with
correlation matrices for nonlinear repeated measurements.
Computational Statistics and Data Analysis 2005; 50:
In a (regulatory) trials framework, Bayesian methods certainly
do not come to prominence reasons may be quite varied.
Perhaps phase II might be where some deviation from the usual
might nd a place and this is where Wang et al. review using
Bayesian methods to consider the posterior probability that a
treatments effect is of a given magnitude. For the true, die-hard
Bayesians, however, justifying such approaches on the grounds
that they control frequentist error rates must be anathema!
* Wang Y-G, Leung DH-Y, Li M, Tan S-B. Bayesian designs
with frequentist and Bayesian error rate considerations.
Statistical Methods in Medical Research 2005; 14:445456.
A completely different phase II problem is addressed by Lu
and colleagues who combine total response rate and rate of
complete response as the endpoint. They develop a two-stage
design and give guidance on stopping rules, etc.
* Lu Y, Jin H, Lamborn KR. A design of phase II cancer
trials using total and complete response endpoints. Statistics
in Medicine 2005; 24:31553170.
The following paper perhaps has little to take away and use
and its mathematical content is a bit heavier than the general
reading we usually highlight but all the same is worth a quick
look for those interested in the ongoing discussion of how to
dene surrogate endpoints:
* Baker SG, Izmirlian G, Kipnis V. Resolving paradoxes
involving surrogate end points. Journal of the Royal
Statistical Society, Series A 2005; 168:753762.
Multiplicity ought (perhaps) to be a problem that is designed
out of a trial but if that is not done, then clear, up-front, rules
on how to handle it are essential. Whether statistics should be
an art, or more rule-based might be a debatable point . . .although pre-specication may need some art, once methods/
approaches are pre-specied, then inevitably some element of
following set rules necessarily follows. This paper gives an
example of setting out rules for families of related endpoints:
* Chen X, Capizzi T, Binkowitz B, Quan H, Wei L, Luo X.
Decision rule based multiplicity adjustment strategy.
Clinical Trials 2005; 2:394399.
Procedures to adjust for multiple hypotheses are of similar
importance. This paper describes a procedure for evaluating a
family of hypotheses:
* Wiens B, Dmitrienko A. The fallback procedure for
evaluating a single family of hypotheses. Journal of
Biopharmaceutical Statistics 2005; 15:929942.
Sample size calculation and recalculation
Following from multiplicity (above), designing studies when
effects are necessary to be seen on more than one endpoint is
not simple. Xiong et al. use Alzheimers disease as an example
and show how to calculate power and sample size for an
appropriate intersectionunion test.
* Xiong C, Yu K, Gao F, Yan Y, Zhang Z. Power and sample
size for clinical trials when efcacy is required in multiple
endpoints: application to an Alzheimers treatment trial.
Clinical Trials 2005; 2:387393.
When many correlated outcomes or endpoints are involved,
simulation-based procedures may be useful.
* Bang H, Jung S, George S. Sample size calculation for
simulation-based multiple testing procedures. Journal of
Biopharmaceutical Statistics 2005; 15:957967.
Interim analyses and Data Monitoring Committees
There seem to be two distinct types of interim analysis: those in
which there is (virtually) complete follow-up on all the patients
who have been recruited so far (but not all the pre-planned
patients have been recruited); and those where all patients have
been recruited but many of them do not have complete follow-
up. The rst case is about curtailing recruitment; the second case
is about curtailing follow-up. In most cases, standard (group)
sequential methods (designed for case one) are used even in case
two. Troendle et al. look more closely at this problem. It is
important to understand that, in this second case, different
hypotheses are being tested at different follow-up times.
* Troendle JF, Liu A, Wu C, Yu KF. Sequential testing for
efcacy in clinical trials with non-transient effects. Statistics
in Medicine 2005; 24:32393250.
Designing studies to compare population pharmacokinetic
response is not a setting where statisticians have traditionally
Copyright # 2006 John Wiley & Sons, Ltd. Pharmaceut. Statist. 2006; 5: 6769
been involved. The need for involvement in design to support
such studies is discussed in:
* Narukawa M, Yafune A. A note on power and sampling
schedule in population pharmacokinetic studies. Drug
Information Journal 2005; 39:353359.
Data analysis issues
Handling missing data can be a complex problem depending on
the mechanism (typically, missing at random, or not). These
papers reviewed analysis methods when missing data are
* Ali M, Talukder E. Analysis of longitudinal binary data
with missing data due to dropouts. Journal of Biopharma-
ceutical Statistics 2005; 15:9931007.* Sheng X, Carrie`re K. Strategies for analysing missing item
response data with an application to lung cancer. Biome-
trical Journal 2005; 47:605615.
When data are not necessarily missing but are just sparse, it
may be of interest to test for homogeneity of treatment response
across different strata. This paper considers the topic of
difference in risk:
* Lui K. A simple test of the homogeneity of risk difference in
sparse data: an application to a multi-centre study.
Biometrical Journal 2005; 47:654661.
Analysis of multivariate data such as that encountered in
microarrays and imaging is a similarly complex topic with the
potential for false positives inherent to such large data sets with
correlated responses being of most concern to drug developers.
* DeCook R, Nettleton D, Foster C, Wurtele E. Identifying
differentially expressed genes in unreplicated multiple-
treatment microarray timecourse experiments. Computa-
tional Statistics and Data Analysis 2005; 50:518532.* Bowman F. Spatio-temporal modelling of localised brain
activity. Biostatistics 2005; 6:558575.
Some have suggested methods for combining data across trials
to assess non-inferiority rather than predening a xed margin.
This paper discusses several aspects of such proposals:
* Lawrence J. Some remarks about the analysis of active
control trials. Biometrical Journal 2005; 47:616622.
Modelling of doseresponse is a complex topic, complicated
further when the response is an unintended side-effect. This
paper discusses a Bayesian approach to the topic:
* Johnson T, Taylor J, Ten Haken R, Eisbruch A. A Bayesian
mixture model relating dose to critical organs and functional
complication in 3D conformal radiation therapy. Biostatistics
An issue we do not often think of when taking a pill is
whether the expiry date is in the past, present, or future.
Different conditions of light, temperature, etc. can have a
dramatic effect on pharmaceutical products. Please check those
expiry dates, and to nd out more on how they are derived
* Verbon F, van den Heuvel E, Vermaat C. The cluster design
for the postmarketing surveillance program. Drug Informa-
tion Journal 2005; 39:369371.
A special section of the Drug Information Journal is dedicated
to the topic of ICH E14, the clinical evaluation of QT/QTc
interval prolongation and proarrythmic potential for non-
antiarrythmic drugs. Several of the papers contained in the
special section are statistical in nature:
* Dmitrienko A, Sides G, Winters K, Kovacs R, Rebhun D,
Bloom J, Groh W, Eisenberg P. Electrocardiogram refer-
ence ranges derived from a standardised clinical trial
population. Drug Information Journal 2005; 39:395405.* Patterson S, Jones B, Zariffa N. Modelling and interpreting
QTc prolongation in clinical pharmacology studies. Drug
Information Journal 2005; 39:437445.* Hosmane B, Locke C. A simulation study of power in
thorough QT/QTc studies and a normal approximation for
planning purposes. Drug Information Journal 2005; 39:447
Finally, if you are proud to be a pharmaceutical statistician and
not just a general, medical statistician (apologies to readers
who may quite legitimately be pharmaceutical but denitely not
medical), then Grieves outgoing Presidential address to the
Royal Statistical Society will inspire you. And to readers who
are not proud to be pharmaceutical statisticians, it is still worth
a read for a better understanding of how our profession is
* Grieve AP. The professionalization of the shoe clerk.
Journal of the Royal Statistical Society Series A 2005;
Literature Review 69
Copyright # 2006 John Wiley & Sons, Ltd. Pharmaceut. Statist. 2006; 5: 6769