Appraisal tool for Cross-Sectional Studies (AXIS)

Critical appraisal (CA) is used to systematically assess research papers and to judge the reliability of the study being presented in
the paper. CA also helps in assessing the worth and relevance of the study [1]. There are many key areas to CA including assessing
suitability of the study to answer the hypothesised question and the possibility of introducing bias into the study. Identifying these
key areas in CA requires good reporting of the study, if the study is poorly reported the appraisal of suitability and bias becomes
difficult.

The following appraisal tool was developed for use in appraising observational cross-sectional studies. It is designed to address
issues that are often apparent in cross-sectional studies and to aid the reader when assessing the quality of the study that they are
appraising. The questions on the following pages are presented in the order that they should generally appear in a paper. The aim of
the tool is to aid systematic interpretation of a cross-sectional study and to inform decisions about the quality of the study being
appraised.

The appraisal tool comes with an explanatory help text which gives some background knowledge and explanation as to what the
questions are asking. The explanations are designed to inform why the questions are important. Clicking on a question will
automatically take you to the relevant section in the help text. The appraisal tool has areas to record a “yes”, “no” or “don’t know”
answer for each question and there is room for short comments as well.

If using this tool please cite the peer review publication: Development of a critical appraisal tool to assess the quality of cross-sectional studies (AXIS);
Martin J Downes, Marnie L Brennan, Hywel C Williams, Rachel S Dean; BMJ Open 2016;6:12 e011458 doi:10.1136/bmjopen-2016-011458

This work is licensed under the Creative Commons Attribution-ShareAlike 4.0 International License.
To view a copy of this license, visit http://creativecommons.org/licenses/by-sa/4.0/.
Contents
Appraisal of Cross-sectional Studies .................................................................................................................................................. 3
Introduction ......................................................................................................................................................................................... 4
Aims ................................................................................................................................................................................................ 4
Methods .............................................................................................................................................................................................. 4
Study Design ................................................................................................................................................................................... 4
Sample Size Justification ................................................................................................................................................................ 4
Target (Reference) Population ........................................................................................................................................................ 4
Sampling Frame .............................................................................................................................................................................. 4
Census ......................................................................................................................................................................................... 5
Sample Selection............................................................................................................................................................................. 5
Non-responders ............................................................................................................................................................................... 5
Measurement Validity & Reliability ............................................................................................................................................... 5
Statistics .......................................................................................................................................................................................... 6
Overall Methods.............................................................................................................................................................................. 6
Results................................................................................................................................................................................................. 6
Basic Data ....................................................................................................................................................................................... 6
Response Rate ................................................................................................................................................................................. 6
Internally Consistent Results .......................................................................................................................................................... 6
Justified Discussions and Conclusions............................................................................................................................................ 7
Aim ............................................................................................................................................................................................. 7
Selection Bias ............................................................................................................................................................................. 7
Non-response .............................................................................................................................................................................. 7
Confounding ............................................................................................................................................................................... 7
Non-significant Results ............................................................................................................................................................... 7
Limitations ...................................................................................................................................................................................... 8
Other ................................................................................................................................................................................................... 8
Conflicts of Interest......................................................................................................................................................................... 8
Ethical Approval ............................................................................................................................................................................. 8
References:.......................................................................................................................................................................................... 8

This work is licensed under the Creative Commons Attribution-ShareAlike 4.0 International License.
To view a copy of this license, visit http://creativecommons.org/licenses/by-sa/4.0/.
Appraisal of Cross-sectional Studies

Don’t know/
Question Yes No
Comment
Introduction
1 Were the aims/objectives of the study clear?

Methods
2 Was the study design appropriate for the stated aim(s)?

3 Was the sample size justified?

Was the target/reference population clearly defined? (Is it clear who the
4
research was about?)

Was the sample frame taken from an appropriate population base so that it
5
closely represented the target/reference population under investigation?

Was the selection process likely to select subjects/participants that were

6
representative of the target/reference population under investigation?

7 Were measures undertaken to address and categorise non-responders?

Were the risk factor and outcome variables measured appropriate to the aims
8
of the study?

Were the risk factor and outcome variables measured correctly using
9 instruments/measurements that had been trialled, piloted or published
previously?

Is it clear what was used to determined statistical significance and/or

10
precision estimates? (e.g. p-values, confidence intervals)

Were the methods (including statistical methods) sufficiently described to

11
enable them to be repeated?

Results
12 Were the basic data adequately described?

13 Does the response rate raise concerns about non-response bias?

14 If appropriate, was information about non-responders described?

15 Were the results internally consistent?

16 Were the results presented for all the analyses described in the methods?

Discussion
17 Were the authors' discussions and conclusions justified by the results?

18 Were the limitations of the study discussed?

Other
Were there any funding sources or conflicts of interest that may affect the
19
authors’ interpretation of the results?

20 Was ethical approval or consent of participants attained?

Taken from Table 2 of: Development of a critical appraisal tool to assess the quality of cross-sectional studies (AXIS); Martin J Downes, Marnie L
Brennan, Hywel C Williams, Rachel S Dean; BMJ Open 2016;6:12 e011458 doi:10.1136/bmjopen-2016-011458
Introduction Question 3 asks if sample size justification was reported, but
it should also be clear what methods were used to determine
The introduction serves to establish the context of the work the sample size. In some cases clustering of observations
that is about to be presented in the text of the paper. Relevant within groups can occur (e.g. patients within hospitals or
primary literature should be discussed and referenced livestock within herds) and this should be taken into account
throughout the introduction. The history and current if sample size has been determined. It should be clear whether
understanding of the problem being researched should be the inferences drawn actually relate to the attributes for which
presented. This should be concluded giving a rational as to the sample size was calculated [7]. If sample size justification
why the current study is being presented and what the aims isn’t given or restrictions make it difficult to reach the desired
and/or hypothesis under investigated are [2,3]. sample size then this should be declared in the text.

Aims Target (Reference) Population

The aim(s) of the study tells us if the study addresses an The target or reference population is the overall population
appropriate and clearly focused question. If the aim is not that the research is directed towards. When doing a cross-
clearly stated or not stated at all, it will be difficult and in sectional study, a target population is the overall population
some cases impossible to assess the extent to which the study you are undertaking the study to make conclusions about or
objectives were achieved. Ideally, an aim should be stated the population at risk of acquiring the condition being
both at the beginning of the abstract and at the end of the investigated [8–10] e.g. the total female population in the UK,
introduction [3]. If the answer to question 1 is no, then it will or all dogs in the USA with cardiovascular disease. (See
make it difficult to assess some of the other questions in the Figure 1) Question 4 asks if this is clearly defined in the study.
critical appraisal process. It is important that this is understood both by the researcher
and the reader; if it is not clearly defined then inferences made
Methods by the researcher may be inappropriate.

The methods section is used to present the experimental study Sampling Frame
design of the paper. The methods should be described clearly
As a reader you need to determine if the sample frame being
in easy to understand language and clearly identify measures,
used is representative of the target population. The study
exposures and outcomes being used in the study [4]. More
population should be taken from the target population; units
specific issues are addressed below.
from this study population have information that is accessible
Study Design and available which allows them to be placed in the study.
The sampling frame is the list or source of the study
Question 2 is used to assess the appropriateness of using a population that the researcher has used when trying to recruit
cross-sectional study to achieve the aim(s) of the study. participants into the study (Figure 1). Ideally it should be
Cross-sectional studies are observational studies that provide exactly the same composition or structure as the target
a description of a population at a given time, and are useful in population. In practice it is generally much smaller, but
assessing prevalence and for testing for associations and should still be representative of the target population.
differences between groups [5]. Examples of cross-sectional Generally, for convenience, the sampling frame is a list of
designs include point-in-time surveys, analysis of records and units that are within the target population e.g. list of telephone
audits of practice [6]. The reader should try and decipher if a owning households, computerised patient records etc. A
cross-sectional study design is appropriate for the questions
being asked by the researcher. Figure 1

Sample Size Justification

Sample size justification is crucial as sample size
profoundly affects the significance of the
outcomes of the study. If the sample size is too
small then the conclusions drawn from the study
will be under powered and may be inaccurate.
This can occur by failing to detect an effect which sample of units is selected from the study population to take
truly exists (type II error) sometimes referred to as a “false part in the study and is generally only a small proportion of
negative”. The probability of a type I error is also taken into the study population (see Sample Selection below) - this
account when determining sample size. A type I error is proportion ratio is known as the sampling fraction. It is very
drawing significant conclusions when no real difference important that the sampling frame is representative of the
exists and is a function of the p-value (see Statistics section target population as results from the study are going to be
below) sometimes referred to as a “false positive”. used to make assumptions about the target population [8–10].
Convenience sampling can be carried out in some situations made about the health of the worker population but is only
and are used because the participants are easy to recruit. relevant to healthy workers and not ill workers. A veterinary
Convenience samples generally lead to non-representative or example of this is a researcher trying to do a cross-sectional
biased samples and therefore cannot be used to make study to determine health factors in the general dog
assumptions about the characteristics of the target population population and decides to sample from a local park.
[11]. Convenience samples are often used for pilot or Unfortunately there is a tendency to over select healthy
analytical studies where the need for a representative sample animals as sick animals will tend to be left at home and not
is not required [12], however the authors should make this taken for a walk. This will in turn lead to inference been made
clear in the text. about the health of the dog population but is only relevant to
healthy dogs and not sick dogs.
Census
Self-selection is another example of selection bias that can be
A census is where the target population and the study
introduced and should be assessed [13]. For example, when
participants are the same at the time the census is taken. In
using a postal questionnaire to examine eating habits and
theory questions 5, 6 and 7 don’t apply to census studies.
weight control, people who are overweight might read the
However even if a study is described as a census it should be
survey and be less inclined to complete and return the survey
very clearly stated where the study participants have been
than those with normal weight leading to over representation
recruited from, and the reader should make the decision if the
of people with normal weight. Similarly, if using a postal
study truly is a census. A census may include all the
questionnaire to examine mastitis levels on cattle farms,
population from the sample frame, but not all the target
farmers that have a high somatic cell counts (SCC) might be
population; in this scenario questions 5 to 7 need to be
less inclined to complete the survey than those with normal
addressed.
or low SCC leading to over representation of farms with good
SCC (see Non-responders below).
Sample Selection
Question 6 is used to establish how the researchers got from Non-responders
the sample frame to the participants in the study. It examines
Non-response in cross-sectional studies is a difficult area to
the potential for selection bias and how the researcher
address. A non-responder is someone who does not respond
developed methods to deal with this. The sample selection
either because they refuse to, cannot be contacted, or because
process is important in determining to what extent the results
their details cannot be documented. As a rule, if participants
of the study are generalizable to the target population. For
don’t respond it is often difficult and sometimes impossible
question 6 we are looking in depth at how the sample (study
to gain any information about them. However other baseline
participants) was selected from the sampling frame. It is
statistics may exist that can be used as a comparator to assess
important to know if there were any inclusion or exclusion
how representative the sample is [14] e.g. age, sex, socio-
criteria used, as inappropriate criteria can dramatically shift
economic classification. Methods used, if any, should be well
how representative the sample is of the target population
described so that the results from the analyses can be
[8,10,13].
interpreted. This is important as non-responders may be from
Selection bias can occur if every unit in the sample frame a specific group, which can lead to a shift in the baseline data
doesn’t have an equal chance of been included in the final away from that group. This shift can lead to results that don’t
study [11,14]. Randomisation is used to ensure that each represent the target population. In some situations the
participant in the sampling frame has an equal chance of sampling frame doesn’t have a finite list or a fully defined
being included in the sample. If methods of randomisation are baseline population. This also makes it difficult, and in some
not used, not described or are not truly random, this may lead cases impossible, to quantify non-response and it may be
to a non-representative sample being selected and hence inappropriate to do so in these situations. If the researchers
affect the results of the study [10,11]. are using non-defined populations this should also be
declared clearly in the materials and methods section [15,16].
There are many other situational issues to take into account
when determining if the population in the sample is likely to Measurement Validity & Reliability
represent the target population. Often these issues are outside Measurement validity is a gauge of how accurately the study
the control of the researcher, but sometimes are overlooked. measurements used assess the concepts that the researcher is
One such issue is the healthy worker effect which is a well- attempting to explore. Measurement reliability is a gauge of
known phenomenon in human cross-sectional studies [13]. the accuracy of the measurements taken or the procedures
An example of this is, a researcher trying to do a cross- used during the study. Question 8 is used to address the
sectional study to determine health factors in a factory concepts of measurement validity, and is specifically aimed
population and decides to sample from workers at work on a to address the appropriateness of the measurements being
particular day. Unfortunately there is a tendency to over select used.
healthy workers as ill workers may tend to be at home on the
day of selection. This will in turn lead to inferences been
The importance of measurement validity is that it gives Results
weight to applying the statistical inferences from the study to
members of the target population. If inappropriate measures The results section of a paper is solely for the purpose of
are used in the study it could lead to misclassification bias and declaring the results of the data analysis and no opinion
it will be difficult to determine to what extent the study results should be stated in this section. This gives the reader the
are relevant to the target population [12,17]. opportunity to examine the results unhindered by the opinion
of the researcher. It is important for the reader to form their
Question 9 is an attempt to gauge the measurement reliability own ideas or opinions about the results before progressing to
of the study measures. Measurements must be able to be the discussion stages.
reproduced and produce identical results if measured
repeatedly, so that the measurements would be exactly the Basic Data
same if performed by another researcher. With this in mind,
Question 12 asks for a description of the basic data. Basic
the measurements must be of international or globally
descriptive analysis aims to summarise the data, giving
accepted standards (e.g. IU standards) where possible and
detailed information about the sample and the measurements
appropriate. If they are being used for the first time they must
taken in the study. The basic data gives an overview of the
be trialled, or in the case of questionnaires, they should be
process of recruitment and if the sampling methods used to
piloted before being used.
recruit individuals were successful in selecting a
representative sample of the target population. If the sampling
Statistics
methods are unsuccessful in selecting a representative sample
While interpretation of statistics can be quite difficult, a basic of the target population, those participants included in the
understanding of statistics can help you to assess the quality study can often be different to the target population; this leads
of the paper. Often many
different methods can be used Figure 2
correctly to test the same data,
but as there is such a wide range
available, knowing what tests
are most appropriate in
particular situations can be hard
to decipher. There is an
expectation that the researcher
has this understanding or has at
least sought statistical
assistance to ensure that the
to inaccurate estimates of prevalence, incidence or risk
correct methods are used. Therefore for question 10 the
factors for disease. Descriptive data of the measurements
emphasis for the reader is that the statistical methods,
taken in the study give an overview of any differences
software packages used and the statistical significance levels
between the groups, and may give insight into some of the
are clearly stated even if the paper is just presenting
reasons for statistical inferences that are made later in the
descriptive statistics. The statistical significance level is
paper.
usually described as a p-value. In most cases the p-value, at
which the null hypothesis is rejected, is set at 0.05. The higher Response Rate
the p-value is set the greater the possibility of introducing a
type I error. Confidence intervals should also be declared with As stated previously it can often be difficult to deal with non-
p-values or instead of p-values as an indication of the responders. Question 13 requires that there is some attempt
precision of the estimates. It is usual to present a confidence made to quantify the level of non-response by the researchers
interval of 95% which means that the researchers were 95 per and asks the reader to interpret if the response rate is likely to
cent confident that the true population value of the outcome lead to non-response bias. Question 14 is examining if any
lies between these intervals. This can be used to compare information on non-responders was available and if so were
groups where an overlap would suggest no difference and a they comparable to those that did respond as this could help
gap between confidence intervals would suggest a difference in answering question 13. Non-response bias occurs if the
(Figure 2). non-responders are substantially different to the rest of the
population in the sample [15].
Overall Methods
Internally Consistent Results
Question 11 asks if the methods are sufficiently described to
enable them to be repeated. If there are sections or even small Question 15 is an exploration of the basic data and asks that
pieces of information missing it could make a great difference the reader spends some time exploring the numbers given in
for the reader when interpreting the results and the discussion the results; in the text, figures and tables. Information about
as they may be unsure if the correct methods are being used. the level of missing data should also be declared in the results.
It is important to check that the numbers add up in the tables There is an expectation that the researcher discusses selection
and the text. If the study has recruited 100 participants, the biases and takes these into account when interpreting the
tables and the text should include data about 100 participants. results of the study. This also gives a clear view of whether
If not, the missing data should be clearly declared and the the researcher has an overall understanding of the study
reason for its non-appearance explained. design. (See notes on selection bias in the methods section).

Comprehensive Description of Results Non-response

Was there an interpretation of the results that included non-
It is important to check that all the methods described response? This is particularly important if the response rate
previously lead to data in the results section (question 16). was low, as non-responders may be a specific group, and lead
Sometimes the results from all analyses are not described. If to a shift in the baseline data (See notes on non-response in
this is noted it will be unclear whether the researcher found the methods section).
non-significant results or just didn’t describe what was found.
If there are results missing that you would expect to find, Confounding
there is a concern that these missing results may not have been
Confounding is a major threat to the validity of practical
what the researcher wanted to see and hence the authors have
inferences made from statistical analyses about cause and
omitted them. It is also important that the significance level
effect. Confounding occurs when the outcome of interest is
declared in the methods is adhered to. As the reader, it is
associated with two different independent variables and one
important to watch out for phrases such as “tended towards
of those variables is closely associated with the outcome only
significance” in the text, and if these are used to pay close
because it is closely associated with the other variable
attention to the results.
(confounder). This can sometimes be accounted for using
Discussion statistical methods however sometimes these associations are
missed because the confounder isn’t measured or isn’t
The discussion of a paper should summarise key results of considered to be a confounder in the analyses. What then
the study objectives. It should give an overall interpretation happens is an erroneous conclusion is made; that the variable
of the results of the study keeping in mind the limitations might have a causal relationship with the outcome. The
and the external validity of the document. The discussion researcher should consider confounding both in the analyses
section should also address both significant and non- and in the interpretation of the results [18]. An example
significant findings of the study and make comparisons with would be where in a study on cancer a researcher concludes
other research, citing their sources [2,4]. that increased alcohol intake causes lung cancer; however
there was confounding in the sample that the researcher didn’t
Justified Discussions and Conclusions discover. People in the study that were inclined to drink more
In question 17 there is an expectation that the researcher gives alcohol were also inclined to smoke more (the confounder)
an overall summary of the main findings of the study and and smoking was the cause of lung cancer not increased
discusses these in detail. It is important that the reader alcohol intake. Similarly, a study was undertaken to examine
considers the study as a whole when reading the researcher’s surgical deaths in cats. The researcher concluded that cats that
conclusion. If the researcher’s conclusion is different or is had gaseous anaesthesia were more likely to die during
more definitive than the study suggests it should be, it can be surgery than those that had just injectable anaesthesia. There
an indication that the researcher has misunderstood their own was confounding in the sample: cats that underwent surgery
study or has other motives or interests for coming to that using gaseous anaesthesia were more likely to be ill or
conclusion. undergoing major surgical procedures (the confounders) and
this was the cause for cats being more likely to die during
It is up to the reader to explore the discussion fully in order to surgery and not the use of gaseous anaesthetics.
answer question 17. The following points should be taken into
account: Non-significant Results
Discussing non-significant results is as important as
Aim discussing significant results and should also be included in
In the discussion section the researcher should discuss all the discussion, especially if they have a direct association
results that pertain to the overall aim of the study, even if they with the aim being investigated. Non-significant results can
are not significant. If some results are overlooked in the be influenced by factors associated with study design and
discussion it could suggest that the researcher either doesn’t sample size. If there are biases introduced during the study
believe the results, or doesn’t want to draw attention to design this can lead to non-significant results that in reality
controversial discoveries from the study and may therefore be may be significant (this can work the other way around as
giving a biased overview of the research conducted. well). If there are only small differences between groups,
non-significant results may be apparent because the sample
Selection Bias size is too small (see sample size justification). Again it is
important that the researcher has a clear understanding of this 8 Groves RM, Fowler FJJ, Couper MP, et al. Chapter 1: An
and conveys that in the discussion. Introduction to Survey Methodology. In: Survey methodology.
Hoboken, NJ: 2004. 1–
37.http://www.loc.gov/catdir/toc/wiley041/2004044064.html
Limitations http://www.loc.gov/catdir/bios/wiley047/2004044064.html
http://www.loc.gov/catdir/description/wiley041/2004044064.
In question 18 we explore whether limitations are discussed. html
Unfortunately all forms of research have some limitations.
9 Lohr SL. Chapter 6: Coverage and Sampling. In: Leeuw ED,
The question here is whether the researcher has an
Hox JJ, Dillman DA, eds. International handbook of survey
understanding of the limitations involved in their study design. methodology. New York, NY [etc.]: Erlbaum 2008. 97–
If this issue is not explored, this is cause for concern that the 112.http://library.wur.nl/WebQuery/clc/1877609 (accessed 26
limitations don’t stop at the design and that the researcher has Nov2012).
a poor understanding of the study as a whole. 10 Woodward M. Chapter 1: Fundamental Issues. In:
Epidemiology: Study Design and Data Analysis, Second
Edition. Chapman and Hall/CRC 2004. 1–32.
Other
11 Groves RM, Fowler FJJ, Couper MP, et al. Chapter 3: Target
Conflicts of Interest Populations, Sampling Frames and Coverage Error. In: Survey
methodology. Hoboken, NJ: 2004. 67–
It is very important that conflicts of interest or bodies 92.http://www.loc.gov/catdir/toc/wiley041/2004044064.html
involved in funding the study are declared in the text http://www.loc.gov/catdir/bios/wiley047/2004044064.html
http://www.loc.gov/catdir/description/wiley041/2004044064.
(question 19). This can give an impression as to background html
reasons for carrying out the study. Where studies are funded
12 Dohoo IR, Martin SW, Stryhn H. Capther 2: Sampling. In:
by a specific agency the researcher may unconsciously Veterinary Epidemiologic Research. Charlottetown, Canada: :
interpret in favour of the agencies’ ideals; if the researcher Ver Books 2009. 27–52.
has worked in a specific area their own ideas and beliefs may 13 Rothman KJ. Chapter 5: Bias in Study Design. In:
affect the interpretation of the results. It is up to the reader to Epidemiology: An Introduction. New York: : Oxford
identify these and come to the conclusion as to whether these University Press 2002. 94–112.
conflicts of interest are relevant or not. This can be declared 14 Bruce N, Pope D, Stanistreet D. Quantitative Methods for
in different areas of the text and should be stated. Health Research: A Practical Interactive Guide to
Epidemiology and Statistics. 1st ed. Wiley-Blackwell 2008.
Ethical Approval 15 Groves RM, Fowler FJJ, Couper MP, et al. Chapter 2:
Inference and Error in Surveys. In: Survey methodology.
Question 20 deals with ethical approval and participant Hoboken, NJ: 2004. 39–
consent. It is important that these are sought before carrying 65.http://www.loc.gov/catdir/toc/wiley041/2004044064.html
out research on any animal or person. http://www.loc.gov/catdir/bios/wiley047/2004044064.html
http://www.loc.gov/catdir/description/wiley041/2004044064.
html
References:
16 Lynn P. Chapter 3: The Problem of Nonresponse. In: Leeuw
1 Young JM, Solomon MJ. How to critically appraise an article. ED, Hox JJ, Dillman DA, eds. International handbook of
Nature Clinical Practice Gastroenterology & Hepatology survey methodology. New York, NY [etc.]: Erlbaum 2008.
2009;6:82–91. doi:10.1038/ncpgasthep1331 35–96.http://library.wur.nl/WebQuery/clc/1877609 (accessed
26 Nov2012).
2 Lebrun J-L. Scientific Writing 2.0: A Reader and Writers
Guide. World Scientific 2011. 17 Hox JJ. Chapter 20: Accommodating Measurement Errors. In:
Leeuw ED, Hox JJ, Dillman DA, eds. International handbook
3 Masic I. How to Search, Write, Prepare and Publish the of survey methodology. New York, NY [etc.]: Erlbaum 2008.
Scientific Papers in the Biomedical Journals. Acta Inform http://library.wur.nl/WebQuery/clc/1877609 (accessed 26
Med 2011;19:68–79. doi:10.5455/aim.2011.19.68-79 Nov2012).
4 Von Elm E, Altman DG, Egger M, et al. The Strengthening 18 Rothman KJ. Epidemiology: An Introduction. New York: :
the Reporting of Observational Studies in Epidemiology Oxford University Press 2002.
(STROBE) statement: guidelines for reporting observational
studies. The Lancet 2007;370:1453–7. doi:10.1016/S0140-
6736(07)61602-X
5 Leeuw ED, Hox JJ, Dillman DA. Data Analysis. In:
International handbook of survey methodology. New York,
NY [etc.]: Erlbaum 2008. 317–
86.http://library.wur.nl/WebQuery/clc/1877609 (accessed 26
Nov2012).
6 Rothman KJ. Chapter 4: Types of Epidemiologic Study. In:
Epidemiology: An Introduction. New York: : Oxford
University Press 2002. 57–93.
7 Woodward M. Chapter 8: Sample Size Determination. In:
Epidemiology: Study Design and Data Analysis, Second
Edition. Chapman and Hall/CRC 2004. 381–426.