Statistical Questions in Evidence-based Medicine is a book of questions and answers designed to help readers improve their skills in critical reading of the medical literature and in evaluation of research evidence. It is intended for medical students, doctors, medical researchers, and all who want to need to read research literature in a medical or health context. It is a companion volume to Martin Bland's An Introduction to Medical Statistics, either second edition or third edition.
The book consists of a set of questions and answers. The questions are printed on the left-hand page with the answers on the right-hand page. The reader can conceal the answers while reading the question, revealing them to check the answer. Some questions are relatively straightforward, for example asking about the meaning of statistical terms, others are more difficult, for example asking why a particular technique was used or was inappropriate. Difficult questions are indicated with a ! symbol in the margin. Questions which cover material which would not usually be covered in undergraduate courses for medical and other healthcare students are also indicated in the margin, by a + symbol. The book has 256 pages.
To illustrate the layout of the book, we have made an Adobe Acrobat PDF version of a couple of pages.
If you do not have Adobe Acrobat, you can download it free from this
link:
You can
download PDF pages here. Use your browser's back arrow to get back to this page.
The questions are almost all based on published medical research, with a few drawn from the general media. The book follows the structure of An Introduction to Medical Statistics, 2nd and 3rd Editions. Answers are complete and self-contained, but have in addition cross-references to An Introduction to Medical Statistics. For example, Section 10.7, Serial data, is referenced in the Adobe Acrobat pages above. You do not need a copy of An Introduction to Medical Statistics (great though it may be!) to use Statistical Questions in Evidence-based Medicine, however. Each chapter starts with a one-page summary of the concepts and methods which it covers, and any good textbook can be used to supplement this.
A Japanese edition was published in November 2002 by Shinohara, Tokyo.
This looks very nice and we are very grateful for all the hard work of the chief translator, Mr. Ken-ichi Adachi.
Four reviews of Statistical Questions in Evidence-based Medicine have been received so far. We have taken the liberty of reproducing them in their entirety. If you know of any others, please let us know.
This book is intended to supplement an introductory course in medical statistics. It presents short summaries or original material from medical research papers, or occasionally from the popular press, and asks questions that test whether the information has been understood, and how it should be interpreted. Sometimes the original presentation is flawed, and the reader is asked to suggest improvements, though this is not the thrust of the majority of examples. The source material and the questions appear on the left-hand pages, with answers on the corresponding right-hand pages.
The organisation of the book follows Martin Bland's An Introduction to Medical Statistics, the third edition of which was reviewed briefly in Biometrics 57, pp. 329-330. There are seventeen chapters covering topics including design of experiments and observational studies, data summary and presentation, probability, estimation and testing, regression and correlation, rank-based methods, cross-tabulated data, clinical measurement and mortality statistics. A final chapter gives five more general and extensive examples, which are not linked to a single topic. Each chapter begins with a single-page summary of the main themes that are explored. The examples themselves are well chosen, and are taken predominantly from the medical literature of the last ten years. Questions that go beyond what is usually covered in an undergraduate degree in health sciences are marked with a special symbol. A different symbol is used to indicate questions that the authors consider particularly difficult..
The authors have clearly gone to considerable trouble to ensure that their answers are clear and thorough. This makes the book ideal for self-study, and most students will benefit from reading all the answers, not just those for the questions that they are unable to answer. Many of the answers end with a reference to a particular section of An Introduction to Medical Statistics that can be consulted for additional background. The third edition of that book retains essentially the same numbering system as the second edition, so that either edition may be used.
One example, chosen by opening the book haphazardly, will give a flavour of the material. It begins with a single-sentence summary of a study of the incidence of pre-term births in the UK (reported as 7.5%) with a quoted standard error and confidence interval. The initial questions ask what is meant by the SE and the confidence interval, what distribution is used to calculate them, and how the confidence interval would change if the level of confidence were increased or decreased. The final question introduces further data from a larger Danish study, and asks why this gives a narrower interval, and whether the reader thinks there is a real difference between the two populations.
A dry humour pervades a few of the examples. We can, for example, eavesdrop on three British Members of Parliament as they struggle (unsuccessfully) to understand what is meant by the median. Another example is a study, published in the British Medical Journal which investigated the vexed question of whether people's ears become larger as they get older. In a cross-sectional study, the ears of patients over the age of 30 were measured 'using a transparent plastic ruler', and the correlation with age was calculated. Bland and Peacock rightly point out that the positive correlation obtained could be due to birth cohort differences rather than a genuine growth effect and that a definitive answer would require a longitudinal study; we can only hope that such studies are under way.
The book has no index, and as is pointed out on the first author's website, the last two pages of references have accidentally been omitted. I didn't notice very many typographical errors, though one that appears twice in the answer to question 12.3.5 will probably render that answer unintelligible to many readers. Typographical errors will be listed on the website, though at the time of writing only one is given.
The book is targeted primarily at people who need to understand presentations of medical data in reports and papers. I think it will be a tremendous help in developing the critical abilities that this requires and I recommend it strongly.
M. S. Ridout
Institute of Mathematics and Statistics
University of Kent at Canterbury,
Canterbury, U.K.
This book is a companion to the senior author's popular textbook "An Introduction to Medical Statistics" currently in its 3rd edition. The book has its beginnings in a course in medical statistics that the authors give to their students in the form of seminars where the students read material provided to them in advance and prepare answers to a series of questions. The material comprises articles from medical journals, and the book carries more than 150 of them in the references section. These articles provide examples of how different types of methods have been applied by researchers to investigate defined problems, and how the information has been interpreted. Seventeen different topics in statistics are discussed in the form of questions and answers, the questions being on the left hand page and the answers on the right. Many of the answers carry references to sections in "Introduction to Medical Statistics" for further reading. The questions and answers relate to one or more published works referenced in the References section. This approach has the advantage of providing critical information relating to statistical aspects of the study being discussed. Because of this unique approach, the information comes in small doses, and is easy to assimilate. The great advantage of the book is that the information is provided in the form of 'normal' language, and statistical jargon or complex formulae are avoided. The book alone will not teach statistics, but when used with the companion text it promises to make understanding the role of statistics in medical research easy. The final chapter "General Questions" brings all the information together and discusses issues of study design, analysis and interpretation.
Author unknown.
This book presents the fundamental concepts of medical statistics, focusing on real case studies taken from the published medical literature, particularly the Lancet and the BMJ. The material covers a wide variety of topics related to design of clinical trials and epidemiological studies, sampling and data collection, summarizing and presenting data, basic probability concepts and data analysis. The analytical methods presented range from estimation, significance tests (parametric and non parametric), determination of sample size and regression models to the analysis of categorical data, mortality statistics, comparison of two methods of measurement and multifactorial methods.
While avoiding mathematical technicalities when possible, topics are presented in a wide range of depth. For example, the chapter on sample size determination illustrates common basic faults (e.g. omitting information on variability, using unrealistically large differences in the computations or using methods that assume normality when the distribution of the data is highly skewed) as well as problems that appear less frequently (e.g. the computation of sample size to estimate sensitivity) and more advanced topics such as the computation of sample size in cluster randomisation.
Each chapter starts with a succinct one-page introduction to the most important concepts. The main body of the chapters consists of well chosen case-studies. On the left hand page one finds one or two case studies with a collection of questions about them. On the facing, right hand, page the answers to the questions are presented, together with further discussion. The reader can thus cover up the answers while studying the questions before proceeding to the right hand page. Some of the questions are marked according to the degree of difficulty thus helpfully highlighting particularly challenging and interesting points.
The emphasis on the case studies and questions, as opposed to theory, is a highly distinctive didactic feature of the book, presenting material that is not readily available elsewhere. The book is designed to be read as a companion to Martin Bland's "An introduction to Medical Statistics" (Oxford UP) but it will also be a valuable supplement to any other standard statistics text. This book can be recommended very highly to anyone who is interested in medical statistics; in addition to explaining the fundamental concepts it is thought provoking and fun to read.
Dr Nora Donaldson
The third edition of Martin Bland's Introduction to Medical Statistics was reviewed by Huson (2001). He praised it as 'an introductory statistical text aimed at medical students, medical researchers, and others concerned with medical data . . . written in a style which particularly addresses the needs of students, containing . . . a wealth of examples based on real data'. In his view 'it should be the first choice for any student wanting a serious introduction to the practice of medical statistics'.
Bland and Peacock's Statistical Questions in Evidence-based Medicine is the companion volume. In its introductory first chapter the authors remark that there has been a shift from authority-based to evidence-based practice in medicine and that the fundamental level at which evidence can be accessed by practitioners is the research paper, followed by review articles. The aim of these two books is to enable medical students and doctors to understand research literature.
The authors have concentrated on the design of medical studies, the presentation of data and data analysis and interpretation. The early chapters in the book are focused on basic statistical methodology, including the design of experiments, estimation and significance tests. Later chapters deal with topics such as the choice of the statistical method, clinical measurement, mortality statistics, multifactorial methods and the determination of sample size.
From Chapter 2 onwards each of the remaining 18 chapters has a brief introduction, followed by a series of questions on the left-hand pages with discursive answers on the right-hand pages. There are from five to 17 questions with subdivisions in each chapter. The material is taken from published work (mainly from the Lancet and the British Medical Journal).
There are a number of cross-references to Bland's Introduction to Medical Statistics and the two books are structured similarly. The authors have tried, however, with some success, to make Statistical Questions in Evidence-based Medicine usable with 'any good introductory book on medical statistics'.
In his review Huson (200l) commented that there is much material in these two books for the student to assimilate. I agree. Martin Bland's preferred course method is one based on seminars with appropriate preparation by students beforehand. The time spent in this manner, or in attending a course of lectures based on the books, would repay the learner with an excellent and rewarding understanding of medical statistics.
Reference
Huson, L. (2001) Review of Introduction to Medical Statistics
(by M. Bland). Statistician, 50, 548.
Freda Kemp
St Andrews University
This book consists of questions and answers based on examples from the research literature and newspapers. The authors describe it as a companion volume to Martin Bland's book, An Introduction to Medical Statistics, but they also note that it will work with any good textbook.
I found the examples very useful for teachers. Furthermore, they cover so many types of research and research questions that they are also interesting for the teacher's own education. It is really curious what people sometimes choose to study, for what reason and how, which makes some of the examples outrightly amusing. One of the more odd ones describes a hospital which refuses to tell a couple the sex of their unborn child. A spokesman for the hospital said that they did not do this because of the possibility of getting it wrong: 'We can only really be accurate in half of the cases'. Which means that the hospital's technique for determining sex is no better than tossing a coin.
It is brave that the authors have included some of their own mistakes. For example, using the term quintiles for groups rather than for values which define groups. As anyone who has conducted courses in critical appraisal will know, if the teacher includes his/her own brilliant research papers, there is a great risk that it will be torn into pieces by clever students.
Each chapter starts with a summary page describing what it is about. These summaries are too brief for people without a good background knowledge of statistics to understand, which is in line with the idea that the book is a companion volume. In the introduction the authors quite rightly note that the reader should not take any paper at face value and they explain their worries by noting that most agricultural research is not done by farmers, whereas in health care, research is largely done by practising clinicians. I readily admit that I am not a statistician but more like the farmer with a long-standing interest in agricultural research, but perhaps I can be excused for writing this review by the fact that I do not practice medicine any longer.
The examples are well thought through and I found it hard to find much with which to disagree. On page 9 the authors note that random allocation can be done by tossing a coin. As this could give the naive farmer the idea that it is an adequate randomization method, the example is not very pedagogical. It has been shown that when the allocation process is not concealed, the estimated effect is exaggerated by about 40 per cent, on average. Therefore, the method can only be used for generation of the allocation sequence, not for the allocation itself, which is best done with the help of a computer to avoid biased assignments.
On page 23 the authors mention that a case control study can be used to study the natural history of disease. It would have been more appropriate to call for inception cohort studies as the natural course of the disease means we have to study cohorts from the time of diagnosis until the patients get cured or die.
On page 93, P is defined as the probability of a test statistic as far from what would be expected as that observed, if the null hypothesis were true. The authors are surely in good company here, since many excellent textbook writers have often forgotten to note that the probability of a singular event is usually very small. The proper expression is: as far from or farther from.
Peter C. Gøtzsche
The Nordic Cochrane Centre
Selected questions and answers can be read on the Web.
People sometimes want examples of statistical mistakes in published papers, for teaching purposes. Although it was not our main purpose to collect such mistakes, our book contains quite a few interesting examples, from both the medical literature and the popular media.
Anyone in search of such examples could try the following questions from
Statistical Questions in Evidence-based Medicine:
2.15, 3.4, 3.16, 3.17, 4.2, 4.6, 4.10, 4.12,
5.1-5.4,
5.6, 5.7, 5.8, 6.1-6.5, 6.7, 6.8, 7.1-7.3,
7.5,
8.3, 9.2, 9.4-9.7,
9.8,
10.2-10.3, 10.6, 11.7, 11.8, 11.11,
12.3, 13.4, 13.11,
14.7, 15.2, 15.9,
16.2, 16.3, 17.6,
17.7, 18.4.
Back to Martin Bland's home page.
This page maintained by Martin Bland.
Last updated: 28 July, 2005.