The Design and Statistical Analysis of Animal Experiments -

1st Edition

Visit the link below to download the full version of this book:

https://medipdf.com/product/the-design-and-statistical-analysis-of-animal-experi
ments-1st-edition/

Click Download Now

University Printing House, Cambridge CB2 8BS, United Kingdom

Published in the United States of America by Cambridge University Press, New York

Cambridge University Press is part of the University of Cambridge.

It furthers the University’s mission by disseminating knowledge in the pursuit of

education, learning and research at the highest international levels of excellence.

www.cambridge.org
Information on this title: www.cambridge.org/9781107690943

© Simon Bate and Robin Clark 2014

This publication is in copyright. Subject to statutory exception

and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.

First published 2014

Printed in the United Kingdom by <tbc>

A catalogue record for this publication is available from the British Library

Library of Congress Cataloguing in Publication data

Bate, Simon T., 1975–
The design and statistical analysis of animal experiments / Simon Bate, Robin Clark, Huntingdon Life Sciences, UK.
pages cm.
Includes bibliographical references and index.
ISBN 978-1-107-03078-7 (hardback) – ISBN 978-1-107-69094-3 (paperback)
1. Animal experimentation–Statistical methods. 2. Experimental design. I. Clark, Robin A., software developer. II. Title.
HV4930.D47 2002
590.72′4–dc23
2013033851

ISBN 978-1-107-03078-7 Hardback

ISBN 978-1-107-69094-3 Paperback

Additional resources for this publication at www.cambridge.org/9781107030787

Cambridge University Press has no responsibility for the persistence or accuracy of

URLs for external or third-party internet websites referred to in this publication,
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
Contents

Preface page ix
Acknowledgments xi

1 Introduction 1
1.1 Structure of this book 3
1.1.1 Introductory sections 4
1.1.2 Approaches to consider when
setting up a new animal model 4
1.1.3 Approaches to consider when
generating hypotheses 5
1.1.4 Approaches to consider when
testing hypotheses 5
1.2 Statistical problems faced by animal
researchers 5
1.3 Pitfalls encountered when applying
statistics in practice 6
1.3.1 Pitfalls with experimental design 6
1.3.2 Pitfalls with randomisation 9
1.3.3 Pitfalls with statistical analysis 10
1.3.4 Pitfalls when reporting animal
experiments 13
1.4 So where does statistics fit in? 14
1.5 The ARRIVE guidelines 15

2 Statistical concepts 18
2.1 Decision-making – the signal-to-noise ratio 18
2.2 Probability distributions 19
2.2.1 The frequency distribution 20
2.2.2 The density distribution 20
2.2.3 The probability distribution 21
2.2.4 The normal distribution 21
2.2.5 The chi-squared distribution 22
2.2.6 The t-distribution 22
2.2.7 The F-distribution 23

v
vi Contents

2.3 The hypothesis testing procedure 23 3.4.6 Balanced incomplete

2.3.1 The null and alternative block design 55
hypotheses 23 3.4.7 More than one block – the
2.3.2 The p-value 25 row-column block design 56
2.3.3 The significance level 25 3.4.8 Row-column block
2.3.4 Significant stars 26 designs based on Latin squares 57
2.3.5 Type I and Type II errors 26 3.4.9 Crossover designs 59
2.4 Exploratory vs. confirmatory 3.5 Factorial design 63
experiments 27 3.5.1 Randomisation 64
2.5 The estimation process 28 3.5.2 Categorical factors and
interactions 64
3 Experimental design 30
3.5.3 Small factorial designs 66
3.1 Why design experiments? 30
3.5.4 Large factorial designs 68
3.1.1 Practical reasons 30
3.5.5 Factorial designs with
3.1.2 Statistical reasons:
continuous factors 79
Variability, the signal
3.5.6 Final thoughts on factorial
and bias 31
designs 84
3.2 What does an experimental
3.6 Dose-response designs 85
design involve? 32
3.6.1 The four- and five-
3.2.1 Variables to be recorded 32
­­parameter logistic curve 85
3.2.2 Set of treatments 35
3.6.2 Experimental design
3.2.3 The experimental unit and
considerations 86
the observational unit 37
3.6.3 Including the control group 89
3.2.4 Effects and factors 37
3.7 Nested designs 91
3.2.5 Fixed and random factors 39
3.7.1 Sample size and power 92
3.2.6 Categorical factors and
3.7.2 Higher-order nested
continuous factors 42
designs 92
3.2.7 Crossed factors and
3.8 Repeated measures and dose-
nested factors 42
escalation designs 112
3.2.8 Repeatedly measuring the
3.8.1 Repeated measures
animal 45
designs 112
3.3 Summary of design types 46
3.8.2 Dose-escalation designs 117
3.3.1 Block designs 46
3.9 Split-plot designs 119
3.3.2 Factorial designs 47
3.9.1 Animals as whole-plots 119
3.3.3 Dose-response designs 47
3.9.2 Animals as subplots 120
3.3.4 Nested designs 47
3.10 Experimental designs in practice 120
3.3.5 Split-plot designs 48
3.11 A good design should result in… 121
3.3.6 Repeated measures and
dose-escalation designs 48 4 Randomisation 123
3.3.7 Designs applied in practice 48 4.1 Practical reasons to randomise 123
3.4 Block designs 49 4.1.1 Bias reduction 123
3.4.1 Practical reasons to block 49 4.1.2 Blinding 125
3.4.2 Statistical reasons to block 49 4.2 Statistical reasons to randomise 125
3.4.3 How to block 51 4.2.1 Estimating the variability 126
3.4.4 Complete block designs 53 4.2.2 Deciding upon the
3.4.5 Incomplete block designs 54 statistical analysis strategy 126
 Contents vii

4.2.3 Repeatedly measured 5.5.3 Survival analysis 235

responses 128
6 Analysis using InVivoStat 239
4.3 What to randomise 130
6.1 Getting started 239
4.4 How to randomise 131
6.1.1 Data import 239
4.4.1 Mechanical methods 131
6.1.2 Importing a dataset into
4.4.2 Computer software 131
InVivoStat – Excel import 241
5 Statistical analysis 133 6.1.3 Importing a dataset into
5.1 Introduction 133 InVivoStat – text file import 241
5.1.1 InVivoStat 134 6.1.4 Data management 241
5.1.2 A recommended five-stage 6.1.5 Running an analysis 241
parametric analysis procedure 134 6.1.6 Warning and error messages 242
5.2 Summary statistics 136 6.1.7 Log file 242
5.2.1 Parametric measures of 6.1.8 Exporting results 242
location 136 6.2 Summary Statistics module 242
5.2.2 Parametric measures of 6.2.1 Analysis procedure 243
spread 139 6.2.2 Worked example 244
5.2.3 Non-parametric measures 6.3 Single Measure Parametric
of location 140 Analysis module 244
5.2.4 Non-parametric measures 6.3.1 Analysis procedure 244
of spread 141 6.3.2 Worked example 246
5.3 Graphical tools 141 6.3.3 Technical details 249
5.3.1 Scatterplots 141 6.4 Repeated Measures Parametric
5.3.2 Box-plots 142 Analysis module 253
5.3.3 Histograms 143 6.4.1 Analysis procedure 253
5.3.4 Categorised case profiles 6.4.2 Worked example 256
plot 145 6.4.3 Technical details 256
5.3.5 Means with SEMs plot 146 6.5 P-Value Adjustment module 259
5.4 Parametric analysis 152 6.5.1 Analysis procedure 260
5.4.1 Parametric assumptions 153 6.5.2 Worked example 260
5.4.2 The t-test 164 6.6 Non-Parametric Analysis module 261
5.4.3 Analysis of variance 6.6.1 Analysis procedure 261
(ANOVA) 169 6.6.2 Worked example 263
5.4.4 Repeated measures 6.7 Graphics module 263
analysis 180 6.7.1 Analysis procedure 263
5.4.5 Predicted means from the 6.7.2 Example plots 264
parametric analysis 196 6.8 Power Analysis module 264
5.4.6 Analysis of covariance 6.8.1 Analysis procedure 264
(ANCOVA) 200 6.8.2 Worked example 267
5.4.7 Regression analysis 212 6.9 Unpaired t-test Analysis module 268
5.4.8 Multiple comparison 6.9.1 Analysis procedure 268
procedures 213 6.9.2 Worked example 271
5.5 Other useful analyses 229 6.10 Paired t-test/within-subject
5.5.1 Non-parametric analyses 230 Analysis module 271
5.5.2 Testing the difference 6.10.1 Analysis procedure 272
between proportions 233 6.10.2 Worked example 275
viii Contents

6.11 Dose-Response Analysis 6.12.1 Analysis procedure 285

module 275 6.12.2 Worked example 287
6.11.1 Technical details on 6.13 R-runner module 288
curve fitting 275 6.14 Nested Design Analysis module 288
6.11.2 Fitting logistic curves to 6.14.1 Analysis procedure 289
data 277 6.14.2 Worked example 293
6.11.3 Analysis of quantitative 6.15 Survival Analysis module 294
assays 277 6.15.1 Analysis procedure 294
6.11.4 Analysis procedure 280 6.15.2 Worked example 295
6.11.5 Worked example – a
7 Conclusion 296
biological assay 283
6.11.6 User-defined equation
Glossary 299
option 284
References 301
6.12 Chi-squared Test and Fisher’s
Index 307
Exact Test module 285
Preface

This book is aimed at practitioners who do not have a

statistics degree and yet wish to apply statistics to help
them arrive at valid and reliable conclusions while
minimising the animal numbers required. Descriptions
of the mathematical methods underpinning the topics
covered in the book are purposefully kept to a min-
imum. If readers wish to gain a better understanding of
the mathematics behind experimental design and stat-
istical analysis then reading a more advanced textbook
would help further their understanding.
The solutions to practical problems encountered
when conducting animal experiments are explained
using non-technical approaches. We believe that in
many situations advanced statistical ideas can be
employed successfully by researchers with no statistical
qualification, using a combination of common sense
and modern statistical analysis software packages. In
our experience statistical ideas are often introduced
to scientists using mathematical terminology. This can
be off-putting to non-mathematicians and can leave
researchers with, at best, only rudimentary statistical
tools and at worst a fear of statistics.
To keep the descriptions of the statistical tools covered
in this book as simple as possible, we shall occasionally
give pragmatic explanations. While such explanations
may not apply in all cases and in all scientific disciplines,
this approach does allow us to introduce methods in a
clear and concise way. By allowing ourselves the freedom
to simplify the problems pragmatically, we aim to make
statistical tools more accessible. The reader is invited,
once they have familiarised themselves with (and hope-
fully found the benefit of using) the tools described in
this book, to read more advanced texts on the subject.

ix
x Preface

This book is divided into seven chapters which InVivoStat. The final chapter draws some conclusions
loosely correspond to the procedure a researcher about the ideas contained within the text.
should take when planning the experimental design, A scientist can apply all of the methodology
running the experiment and evaluating the data gener- described in this book. Certain topics covered are
ated. Following an introductory chapter and a second more advanced than others and while we aim to make
describing certain statistical concepts, the third chapter all subjects accessible, the reader should be aware that
covers different types of designs. Designs are outlined, the help of a professional statistician may be advisable
where possible, in simple non-technical language. This when first implementing some of the more advanced
is followed by a chapter describing the randomisation tools. However, once the readers have familiarised
of the experimental material. The fifth chapter discusses themselves with the ideas contained within this book,
the statistical analysis of animal experiments and this is we hope they will have a fuller appreciation of the help
followed by a chapter describing how these methods statistics can offer to improve the conclusions that can
can be applied within the statistical software package be made when running animal experiments.
Acknowledgments

We would like to thank all our colleagues, both past

and present, for providing so much inspiration for this
book. We hope this text reflects the discussions, consul-
tations and the range of examples that we have worked
on together. There are simply too many scientists to
mention, but you know who you are! We hope the wide
variety of interesting practical problems we have faced
together is reflected throughout this book.
Special thanks also to the statisticians who have
advised us over the years and shaped our views on this
subject matter. Simon would especially like to thank
Philip Overend and Andrew Lloyd for their input and
encouragement and Janet and Ed Godolphin for their
help and support over the years.
We would like to give an extra special thanks to
Gillian Amphlett, Gill Fleetwood, David Wille, Clare
Stanford and Nathalie Du Sert for giving up their time to
review sections of the book, and providing ideas, advice
and feedback. Their input has been most helpful and
improved the text considerably.
Finally we would like to thank our friends and family
for putting up with the long unsociable hours we spent
writing this book.

xi
 1

Introduction

Many researchers, either directly or indirectly, rely on Every time any particle of statistical method is properly used,
statistical ideas when carrying out animal experiments. fewer animals are employed than would otherwise have been
While some statistical tools are well known and are necessary.
applied routinely, other tools are less well understood Many authors have since highlighted how important the
and so are less well used. The overall aim of this book is use of good experimental design is when conducting
to discuss statistical methodologies that can be applied animal experiments; see Festing (1994, 2003a, 2003b)
throughout the many stages of the experimental pro- and the references contained within. Some of the more
cess. Researchers should be able to carry out most of practical, as well as statistical, aspects of experimen-
the techniques described, although the advice of a pro- tal design and statistics when applying the 3Rs are
fessional statistician is advisable for some of the more described in the book by Festing et al. (2002). There
advanced topics. Making use of these techniques will have also been surveys into the use of statistics in ref-
ensure that experiments are conducted in a logical and ereed journals; see McCance (1995) and more recently
efficient way, which should result in reliable and repro- Kilkenny et al. (2009). The latter draws attention to
ducible decisions. some of the mistakes that can be made by researchers
The particular types of study addressed in this book, when designing and analysing animal experiments.
as the title suggests, are studies involving animals. The reliability of the reporting of animal experiments
We attempt to cover all of the statistical tools that the has been considered in, for example, Macleod et al.
animal researcher should use to run successful stud- (2009) and Rooke et al. (2011). These articles highlight
ies. Of course many of the problems faced by the ani- that papers describing experiments that do not use
mal researcher are common to other disciplines, and suitable randomisation and/or blinding may contain
hence the ideas contained within this book can be biased results.
applied to other areas. It should be noted that certain The main goal of this text is to demonstrate how sta-
topics described in the text have been simplified to tistics can aid the reduction and refinement of animal
allow non-statisticians to apply the ideas without pro- studies. The efficient use of statistics, both in terms of
fessional statistical support. Such pragmatic descrip- complex experimental design and powerful statistical
tions, while simplifying the technical details, are not analysis, can reduce the number of animals required.
universal and will not be applicable in all scientific Statistics can also help the researcher understand the
disciplines. processes that underpin the animal model and help
There has been much interest in the use of statistics identify factors that are influencing the experimental
in animal research, in particular in the application of results. Such an understanding will inevitably lead to
the 3Rs, replacement, reduction and refinement, as a refinement in the experimental process and a reduc-
described by Russell and Burch (1959): tion in the total number of animals used.

1
2 Introduction

Statistics, as a discipline, provides researchers with 100

tools to help them arrive at valid conclusions. However,
statistics, along with the application of some common

Cholesterol level (% of baseline)

80
sense, can also increase the understanding of the ani-
mal model through the application of graphical and
mathematical techniques. For example, graphical tools 60
play an important role in helping the researcher under-
stand the effect of the features of the experimental
40
design and also uncover overall patterns present in the
data. The application of a formal statistical test, with-
out first investigating the data graphically, can lead to 20
the researcher drawing incorrect conclusions from the
data. Consider the following real-life case study, which 0
used graphical, as well as statistical, tools. If a conven- control drug
tional statistical analysis had been carried out, without Treatment
first investigating all of the information gathered within
Figure 1.1. Plot of treatment means with standard errors
the experiment, then the conclusions would have been
for the percentage of baseline cholesterol response for
misleading.
Example 1.1.
Example 1.1: Reducing blood cholesterol levels in mice

A scientist wanted to test the hypothesis that a novel compound had

2000
a beneficial effect on reducing high-density lipoprotein (HDL) chol-
esterol levels in a transgenic C57Bl/6J strain of mice. A blood sam-
ple was taken pretreatment and the baseline cholesterol level for
HDL terminal cholesterol (mg g/dL)

each animal measured. The mice were then randomised to either the 1500
drug treatment group or the control group and dosed with either the
drug treatment or vehicle twice daily for 2 weeks. At the end of this
period, a terminal blood sample was taken and the HDL cholesterol
level measured. 1000
As the scientist wanted to make use of the baseline informa-
tion in the statistical analysis, it was decided that the percentage
change from baseline would be a suitable response to investigate.
This would, the scientist hoped, effectively remove the animal-to- 500
animal differences by normalising to the baseline level. While there
was evidence of a decrease in HDL cholesterol level in the group
of animals administered the drug treatment (a 20% decrease from
baseline in the drug treatment group compared to a 10% decrease 0
control drug
in the control group) this was not deemed statistically significant
Treatment
using an unpaired t-test (p = 0.191). A means with standard errors
of the mean (SEMs) plot of the data (see Section 5.3.5) is presented Figure 1.2. Plot of treatment means with standard errors for
in Figure 1.1. the terminal HDL cholesterol for Example 1.1.
As a follow-up the scientist also analysed the terminal HDL
cholesterol level. From this analysis it appeared that there was a
statistically significant increase in cholesterol level in the drug- by a simple scatterplot of the measured HDL cholesterol levels. If
treated group compared to the control. A plot of the means we plot terminal vs. baseline HDL cholesterol levels, an underlying
with SEMs of the terminal HDL cholesterol level is presented in problem with the experiment becomes clear. The scatterplot is pre-
Figure 1.2. sented in Figure 1.3.
Based on the results of this experiment, should we conclude the From the plot it can be seen that there are two distinct groupings
drug increases cholesterol levels? And why did the two analyses along the X-axis. The plot reveals that, in terms of the HDL baseline
give such different conclusions? These questions can be answered cholesterol level, the animals belong to one of two ­sub-populations.
 Structure of this book 3

control of p-values. Graphs can be the best and simplest way to achieve
drug this and should always be considered first, ideally by plotting the
2500 individual data points.
HDL terminal cholesterol (mg g/dL)

In reality the treatment allocation observed in this example is

quite extreme and perhaps indicates a biased selection process. It
2000 should be noted, however, that there is always a chance, however
small, that the randomisation will generate a significant treatment
effect due to the baseline. This will occasionally happen, even
1500
when the allocation process is valid. As long as the randomisa-
tion is performed correctly, then we should not be too concerned
that effects present at the baseline will influence the treatment
1000
comparisons.

500

500 1000 1500 2000 2500 1.1 Structure of this book

HDL baseline cholesterol (mg g/dL)

Figure 1.3. Scatterplot of terminal HDL cholesterol vs. The majority of the remainder of this text is split up into
baseline HDL cholesterol, categorised by treatment for three main chapters. In Chapter 3 we describe families
Example 1.1. of experimental designs that can be employed when
conducting animal research. The chapter consists of
a description of each design and practical examples
Unless we are careful how these baseline differences are accounted of their use. Also given is an explanation of when and
for, we could draw incorrect conclusions from the analysis. Given where to apply each design. The section attempts to
that there appears to be a correlation between baseline and ter-
introduce each experimental design, without overuse
minal cholesterol levels, this baseline difference is probably the
most important feature of the experiment that influences the con- of mathematical terminology.
clusion – perhaps more so than the treatment effect itself. The treat- In Chapter 4 some general issues involving random-
ment effect observed in the experiment will be influenced by the isation are discussed. We consider why the experimen-
allocation of the mice (within each sub-population) to the treatment tal material should be randomised and describe the
groups. In this case most of the mice that were allocated to the
influence this has on the statistical analysis. Techniques
novel drug group were from the sub-population with the high base-
line level. So it is not surprising that, when analysing the terminal that can be employed to perform the randomisation are
HDL cholesterol level, it appears that the terminal cholesterol level is also given.
higher in the treatment group. Obviously the researcher was unlucky When conducting a statistical analysis, one of the first
that the randomisation of the mice to the treatment groups pro- steps in the process is to define the statistical model
duced such an allocation.
that will be used to explain the observed data. There are
The solution is twofold. Firstly, and most importantly, the
researcher should try to identify what is causing the baseline dif- several ways to justify the choice of statistical model.
ferences. We can then account for this effect in the experimental Given that the animal researcher has control over the
design. However, if we fail to identify what is causing the baseline experimental design, it seems sensible to make use of
differences, then the randomisation should be carried out so that the design when deciding which statistical model to
the treatment replication is equal in both sub-populations. This will,
apply. One way of linking the experimental design to
of course, depend on whether the baseline information is available
when allocating the mice to the treatment groups. the statistical analysis is by considering the random-
If the researcher produces the scatterplot of Figure 1.3, then it isation applied to the experimental material. Most
will become apparent that there are possible effects due to baseline analyses, and certainly those considered in this book,
cholesterol levels. However, without such a graphical investigation make assumptions about the allocation of animals to
of the data the problem may not have been identified. It is import-
treatment groups. For a valid statistical analysis of a
ant at this stage of the book to note that a valid statistical investi-
gation of a dataset is more about understanding the information designed experiment, a suitable randomisation should
contained within the data. It does not just involve the calculation have been carried out.
4 Introduction

Chapter 5 describes the statistical analysis techniques This section introduces the types of experimental
that the researcher should employ when analysing data design that are available to the researcher. Information
generated using the designs described in Chapter 3. We is given on when and why they should be used.
approach the subject in a practical way, without the use Section 3.7.3 – Sample size calculation
of mathematical formulae. We assume the researcher This section discusses the factors that influence sam-
has access to an advanced statistical package, such as ple size and gives information on how to calculate suit-
InVivoStat, to compute all numerical results. The tools able sample sizes in animal experiments.
described in this book are flexible enough to cope with Sections 4.2 and 4.4 – Randomisation
most experimental situations. Any assumptions made These sections describe why we need to randomise
during the statistical analysis are also discussed. When our experimental material and give practical examples
these assumptions do not hold alternative approaches of how to carry out the randomisation.
are given. Section 5.1.2 – Introduction into statistical analysis
In Chapter 6 we describe how the researcher can This section is a description, including a worked
perform the analyses discussed in Chapter 5 using example, of our preferred analysis procedure using the
the InVivoStat statistical software package. For each InVivoStat software package.
InVivoStat module the analysis procedure is described, Section 5.4 – Types of parametric analysis
including input and output options, and a worked This section describes the types of parametric ana-
example given. Where appropriate, a technical descrip- lysis, some of their properties and gives information on
tion of the implementation of the analysis methodology when to use them.
is also presented.
This text contains many ideas that the researcher
1.1.2 Approaches to consider when setting up
may need to employ during the course of the experi-
a new animal model
mental process. Some of the techniques will be applied
frequently during routine work and hence will be of When setting up a new animal model, or perhaps try-
interest to all readers. Other sections describe tech- ing to replicate a model described in the literature,
niques that are more advanced and would only be used there may be many factors that influence the animals’
occasionally, for example when setting up a new ani- responses, which will need to be quantified. Perhaps
mal model. the researcher needs to decide which sex to use, how
old the animals need to be, how long to dose the ani-
mals prior to testing… the list goes on. A common
1.1.1 Introductory sections
approach taken by researchers is to investigate each of
The following sections should be read by the casual these factors one at a time. However, there are better
reader who wishes to get a simple overview of the ideas and more efficient (not to say more informative) ways
contained within this text. These sections provide a fla- to conduct these investigations. Use:
vour of some of the more advanced sections. • Large factorial designs (Section 3.5.4) to assess fac-
Sections 2.1 and 2.3 – Statistical concepts tors and factor interactions and to maximise the win-
Readers should familiarise themselves with these sec- dow of opportunity
tions as they provide the framework for all following sec- • Nested designs (Section 3.7) to decide on the replica-
tions on experimental design and statistical analysis. tion required within the experiment
Section 3.1 – Why design studies? • Power analysis to select suitable sample sizes (Section
This section is an introduction into why we should be 3.7.3)
designing animal experiments and some information • Parametric analysis tools, such as ANOVA (Section
on the benefits that can be gained from an understand- 5.4.3), repeated measures analysis (Section 5.4.4) or
ing of experimental design. graphical tools (Section 5.3) to investigate how the
Section 3.3 – Summary of design types factors relate to each other