literature review, december 2003-march 2004
TRANSCRIPT
PHARMACEUTICAL STATISTICS
Pharmaceut. Statist. 2004; 3: 143–146 (DOI:10.1002/pst.117)
Literature review, December 2003–
March 2004
Simon Day1,*,y and Meinhard Kieser2
1Medicines and Healthcare Products Regulatory Agency, Room 13-205, Market Towers,
1 Nine Elms Lane, London SW8 5NQ, UK2Department of Biometry, Dr Willmar Schwabe Pharmaceuticals, Karlsruhe, Germany
INTRODUCTION
We see a slight change to the line-up of journals being published
this year. From the end of 2003, The Statistician (Series D of the
Journal of the Royal Statistical Society) will no longer be
published. The coverage of the other three series has been
revised, and whilst Series B remains about the same, most of the
content of Series D will now go to either Series A or Series C
(Applied Statistics). In addition, a new journal appears, and a
story behind it: Controlled Clinical Trials, whilst owned and
published by Elsevier, was the official journal of the Society for
Clinical Trials. The Society has now decided to publish its own
journal (in collaboration with Hodder Arnold) called Clinical
Trials. The first issue (of six each year) has appeared. We will be
including both Controlled Clinical Trials and Clinical Trials in
these reviews.
This review, with its slightly different coverage, includes the
following journals received during the period from the middle
of December 2003 to the middle of March 2004:
* Applied Statistics, volume 53, part 1.* Biometrical Journal, volume 45, part 8 and volume 46,
part 1.* Biometrics, volume 59, part 4.* Biostatistics, volume 5, part 1.* Clinical Trials, volume 1, number 1.* Drug Information Journal, volume 38, part 1.* Journal of the American Statistical Association, volume 98,
part 4.
* Journal of the Royal Statistical Society, Series A, volume
167, part 1.* Statistics in Medicine, volume 23, parts 1–6.* Statistical Methods in Medical Research, volume 13,
parts 1, 2.* The American Statistician, volume 57, part 4.
SELECTED HIGHLIGHTS FROM THE
LITERATURE
We welcome the new journal, Clinical Trials, and give
references here to all of the major papers published so that
you can see the initial breadth of coverage:
* Goodman SN. Editorial: A birthday, an anniversary, and
an agenda for Clinical Trials. Clinical Trials 2004; 1:1–2.* Burke DS. Lessons learnt from the 1954 field trial of
poliomyelitis vaccine. Clinical Trials 2004; 1:3–5.* Ellenberg SS. Editorial: Monitoring data on data monitor-
ing. Clinical Trials 2004; 1:6–8.* Lavori PW, Dawson R. Dynamic treatment regimes:
practical design considerations. Clinical Trials 2004; 1:9–20.* Xie H, Heijtman DF. Sensitivity analysis of causal inference
in a clinical trial subject to crossover. Clinical Trials 2004;
1:21–30.* Scott RB, Farmer E, Smiton A, Tovey C, Clarke M,
Carpenter K. Methodology of neuropsychological research
in multicentre randomized clinical trials: a model derived
from the International Subarachnoid Aneurysm Trial.
Clinical Trials 2004; 1:31–39.* Kiri A, Tonascia S, Meinert CL. Treatment effects
monitoring committees and early stopping in large clinical
trials. Clinical Trials 2004; 1:40–47.
Copyright # 2004 John Wiley & Sons, Ltd.Received \60\re /teci
*Correspondence to: Simon Day, Medicines and HealthcareProducts Regulatory Agency, Room 13-205, Market Towers,1 Nine Elms Lane, London SW8 5NQ, UK
yE-mail: [email protected]
* Sydes MR, Altman DG, Babiker AB, Parmer MKB,
Spiegelhalter DJ, the DAMOCLES group. Reported use
of data monitoring committees in the main published
reports of randomized controlled trials: a cross-sectional
study. Clinical Trials 2004; 1:48–59.* Sydes MR, Spiegelhalter DJ, Altman DG, Babiker AB,
Parmer MKB, the DAMOCLES group. Systematic quali-
tative review of the literature on data monitoring commit-
tees for randomized controlled trials. Clinical Trials 2004;
1:60–79.* Eldridge SM, Ashby D, Feder GS, Rudnicka AR,
Ukoumunne OC. Lessons for cluster randomized trials in
the twenty-first century: a systematic review of trials in
primary care. Clinical Trials 2004; l:80–90.* The Complications of Age-Related Macular Degeneration
Prevention Trial study group. The Complications of Age-
Related Macular Degeneration Prevention Trial (CAPT):
rationale, design and methodology. Clinical Trials 2004;
1:91–107.* Carino T, Sheingold S, Tunis S. Using clinical trials as a
condition of coverage: lessons from the National Emphy-
sema Treatment Trial. Clinical Trials 2004; 1:108–121.* Dawson L. The Salk Polio Vaccine Trial of 1954: risks,
randomization and public involvement in research. Clinical
Trials 2004; 1:122–130.* Marks HM. A conversation with Paul Meier. Clinical Trials
2004; 1:131–138.
The themes for Statistical Methods in Medical Research were:
* Part 1: Nonparametric longitudinal data analysis (pp 1–82).* Part 2: Outcome measures in clinical trials (pp 87–176).
Of particular interest from the latter issue are the following:
* Fairclough DL. Patient reported outcomes as endpoints in
medical research. Statistical Methods in Medical Research
2004; 13:115–138.* Cook J, Drummond M, Heyse JF. Economic endpoints in
clinical trials. Statistical Methods in Medical Research 2004;
13:157–176.
One special issue of Statistics in Medicine has appeared:
* Kryscio RJ, Schmitt FA (eds). Statistical methodology in
Alzheimer’s disease research II. Statistics in Medicine 2004;
23:163–367.
Phase I/II
We regularly review ‘continual reassessment’ methods in this
column – and here is another, but on a slightly different theme.
Here, finding the optimum dose is not the objective, but finding
the optimal dose duration is (TITE-CRM stands for ‘time-to-
event’ continuous reassessment method). In principle, the
problems are the same as those for dose finding, but different
design considerations are necessary to ensure the duration of
the study does not become excessive.
* Braun TM, Levine JE, Ferrara JLM. Determining a
maximum tolerated cumulative dose: dose reassignment
within the TITE-CRM. Controlled Clinical Trials 2003;
24:669–681.
At a slightly later stage in development, two-stage designs are
common in cancer: recruit a small number of patients and then
either stop (essentially for futility), or stop (and go straight to
phase III), or continue recruiting to obtain a little more
information. Jung et al. look at a variety of types of designs
with this underlying theme and a develop a strategy to search
for the best design in any particular situation:
* Jung S, Lee T, Kim K, George SL. Admissible two-stage
designs for phase II cancer clinical trials. Statistics in
Medicine 2004; 23:561–569.
General study design
This paper just has to be admired for courage: a 26 factorial
design of single or combined treatments of antiemetics. No
doubt statisticians designing agricultural field trials would not
consider this anything special – but in medicine it might be
unique.
* Apfel CC, Korttila K, Abdalla M, Biedler A, Kranke P,
Pocock SJ, Roewer N. An international multicenter
protocol to assess the single and combined benefits of
antiemetic interventions in a controlled clinical trial of a
2� 2� 2� 2� 2� 2 factorial design (IMPACT). Controlled
Clinical Trials 2003; 24:736–751.
Non-inferiority
The decision that the significance level for a one-sided
significance test should be half that of the two-sided level for
showing departure from the null hypothesis in either direction is
reasonably well agreed. Certainly there can be arguments
against the simple splitting of alpha in half, but if we do not
take this approach then we are at risk of having inconsistent
inferences – particularly if we want to switch from non-
inferiority to superiority. However, the following does look at
cases where unequal splits may be made for one-sided tests and
tries to defend such an approach:
* Neuh.aauser M. The choice of a for one-sided tests. Drug
Information Journal 2004; 38:57–60.
Missing data
Selecting endpoints that are easy to measure is a good strategy
for helping to reduce missing values – provided, of course, that
the endpoint chosen is still of clinical relevance. The following
Copyright # 2004 John Wiley & Sons, Ltd. Pharmaceut. Statist. 2004; 3: 143–146
LITERATURE REVIEW144
paper has the ‘originally chosen endpoint’ for some patients and
the newly selected, simpler, measurement for most. Good
overlap allowed the authors to explore the potential for
differences between the patients who had missing data and
those who did not. In their example (acupuncture to treat
migraine and chronic tension headaches) they found good
agreement, so were able to use the simpler measurement to
impute missing values. Each therapeutic area – perhaps each
trial – may need to be considered on its own merits.
* Vickers AJ, McCarney R. Use of a single global assessment
to reduce missing data in a clinical trial with follow-up at
one year. Controlled Clinical Trials 2003; 24:731–735.
Multiple imputation is an increasingly popular method to
deal with missing data, not least due to its implementation in
software packages. Some packages that support multiple
imputation assume that the joint distribution of the variables
in the data set is multivariate normal. Applying this approach
to possibly missing data which are not normal but, for example,
binary will lead to implausible values. Rounding the imputed
normal is one of the proposed solutions. The following article
cautions against this method because it can introduce bias to
the parameter estimate:
* Horton NJ, Lipsitz SR, Parzen M. A potential for bias when
rounding in multiple imputation. American Statistician
2003; 57:229–232.
Surrogate endpoints
Validating surrogate endpoints seems a controversial topic. If
done well, it is probably valuable and reliable; if done badly – as
is often the case – it can be useless and potentially misleading.
The following paper is very thorough, looking at the problem of
using an ordinal or binary endpoint (tumour progression) as a
surrogate for survival in studies for advanced colorectal cancer:
* Burzykowski T, Molenberghs G, Buyse M. The validation
of surrogate end points by using data from randomized
clinical trials: a case-study in advanced colorectal cancer.
Journal of the Royal Statistical Society, Series A 2004;
167:103–124.
The literature on validation of surrogate endpoints tends to
focus on univariate responses. The paper by Alonso et al.
introduces a method for the situation where both the surrogate
and the true endpoint are measured repeatedly over time. The
new methodology is illustrated with meta-analysis data from
schizophrenia trials.
* Alonso A, Geys H, Molenberghs G, Kenward MG,
Vangeneugden T. Validation of surrogate markers in
multiple randomized clinical trials with repeated measure-
ments. Biometrical Journal 2003; 45:931–945.
Sample size calculation and recalculation
The Pearl Index is a widely used measure to describe the
reliability of contraceptive methods. Guidelines on the clinical
investigation of steroid contraceptives in women require that,
for key clinical studies in this field, the width of the 95%
confidence interval for the Pearl Index should not exceed a
given margin. The following paper develops the related sample
size calculation methodology:
* Benda N, Gerlinger C, van der Meulen EA, Endrikat J.
Sample size calculation for clinical studies on the efficacy of
a new contraceptive method. Biometrical Journal 2004;
46:141–150.
Papers on the internal pilot study design have almost a
regular place in this review. The currently available procedures
use only the data of those patients for sample size recalculation
who have already completed the study. However, as recruit-
ment is usually ongoing when the re-estimation of sample size is
carried out, a number of patients will not have completed the
whole treatment phase but will only have reached some
intermediate point of the follow-up time. The following paper
shows how inclusion of these data leads to an improvement of
the precision of the sample size estimate:
* W .uust K, Kieser M. Blinded sample size recalculation for
normally distributed outcomes using long- and short-term
data. Biometrical Journal 2003; 45:915–930.
Interim analyses
The following paper looks at flexible approaches to stopping
rules in group sequential trials. This is not a new topic, but the
authors show how their approaches compare (in some cases
exactly and in others not) with other existing methods:
* Burington BE, Emerson SS. Flexible implementations of
group sequential stopping rules using constrained bound-
aries. Biometrics 2003; 59:770–777.
Regulatory issues
Although available on the EMEA website, Statistics in
Medicine has published the recently adopted (May 2003)
guidance on adjustment for baseline covariates:
* Committee for Proprietary Medicinal Products (CPMP).
Points to consider on adjustment for baseline covariates.
Statistics in Medicine 2004; 23:701–709.
The following is a paper that matches perfectly with this topic:
conventional adjustment techniques treat the covariate as
binary, nominal or continuous, and this may be problematic
when applied to ordinal covariates. Berger et al. propose new
methods to adjust for ordinal covariates that do not require a
LITERATURE REVIEW 145
Copyright # 2004 John Wiley & Sons, Ltd. Pharmaceut. Statist. 2004; 3: 143–146
dichotomy of an ordinal variable or to impose a continuous
variable where none exists.
* Berger VW, Zhou YY, Ivanova A, Tremmel L. Adjusting
for ordinal covariates by inducing a partial ordering.
Biometrical Journal 2004; 46:48–55.
Miscellaneous
‘Large, simple trials’ seems to be a phrase commonly used to
describe trials. No doubt behind such a description is a
multitude of complexities and simplifications. Although tradi-
tionally the domain of academic clinical trialists, companies do
sometimes get involved and regulatory authorities sometimes
consider such trials and results. Collins et al. describe one
‘large, simple trial’, the DIG trial. A special issue of Controlled
Clinical Trials was devoted to various aspects of it, this being an
overview:
* Collins JF, Egan D, Yusuf S, Garg R, Williford WO, Geller
N on behalf of the DIG Investigators. Overview of the DIG
trial. Controlled Clinical Trials 2003; 24:269S–276S.
When a patient is treated with an active drug in clinical
practice or in a clinical trial, an improvement may be attributed
to the active chemical component of the drug, to non-specific
aspects of the treatment (‘placebo effect’), or a combination of
both. Tarpey et al. claim to provide a method to identify and
separate placebo responders and true drug responders among
drug-treated patients. The idea is to estimate outcome profiles
for each subject, to determine representative profiles for each
category of response, and to classify the patients accordingly.
* Tarpey T, Petkova E, Ogden RT. Profiling placebo
responders by self-consistent partitioning of functional
data. Journal of the American Statistical Association 2003;
98:850–858.
The sign test is a suitable and frequently used non-parametric
procedure to assess the median difference of paired samples
when the distribution is not symmetrical. In case of ‘zero’
differences, the sign test may not perform well. Proposed
solutions result in an increased alpha level (when the ties are
excluded or equally assigned as positive or negative) or are too
conservative. The following paper proposes a simple modifica-
tion that shows better performance than conventional methods:
* Fong DYT, Kwan CW, Lam KF, Lam KSL. Use of the sign
test for the median in the presence of ties. American
Statistician 2003; 57:237–240.
Pharmacoepidemiology ought to begin way back in the
development phase of a drug, but, perhaps inevitably, most
interest arises late in the day. Gutterman gives an overview of
safety assessment of newly approved products, acknowledging
the gaps that may have arisen during the clinical trials for
registration. She uses healthcare claims databases to look at the
impact of drugs on a broad population and discusses the pros
and cons of such databases:
* Gutterman EM. Pharmacoepidemiology in safety evalua-
tions of newly approved medications. Drug Information
Journal 2004; 38:61–67.
We rarely (if ever, yet) highlight papers concerned with
production, so we are pleased to close this review with such an
example. This paper uses quality control of a pill production
process to investigate the strategy of 100% sampling, rejecting
defective items, and then sampling again. The resampling is
because it is recognized that the inspection process is not perfect
and defective items may be inadvertently passed. Of course,
such a procedure is not confined to the production end of drug
development, so there may be lessons for many of us to learn.
* Gasparini M, Nusser H, Eisele J. Repeated screening with
inspection error and no false positive results with applica-
tion to pharmaceutical pill production. Applied Statistics
2004; 53:51–62.
Copyright # 2004 John Wiley & Sons, Ltd. Pharmaceut. Statist. 2004; 3: 143–146
LITERATURE REVIEW146