In the preface, the authors of this superb text (now in its 4th edition) state that they envision the book to be useful for both “consumers” of statistics (medical professionals critically reading the medical literature), and “doers” of statistics (those who design their own studies and analyze data for these projects). The authors have done an excellent job in carrying out these goals.
The authors cover the topics that one would expect in an introductory biostatistics text. These include chapters covering the following: describing and displaying categorical and continuous data, probability basics, distributions (binomial, poisson, and normal), confidence intervals, statistical inference, and common statistical tests (t-tests, correlation, regression, and non-parametric tests which are used for data not normally distributed). They also devote two chapters dealing with types of medical studies (e.g., observational studies and randomized controlled trials); this material is often seen in epidemiology books, and is increasingly appearing in medical statistics books. Finally, there is material which should be covered in biostatistics books, but often is not (i.e., about 15 pages each on survival analysis, sample size determinations and reliability/method comparison studies). A unique feature of the book is a final chapter on common pitfalls in analyzing data, which unfortunately lead to common statistical errors in the published medical literature.
There are many nice features of the book. These include numerous examples and exercises of analyses taken directly from the medical literature, a relatively inexpensive soft-cover book price ($37.50), illustrations of actual “generic” output from statistical computer packages, and solutions (not just the answers!) to all of the exercises. I wonder whether, as a convenience to the instructor, the solutions should only have been provided for the even or odd numbered exercises.
Past users of this text would be interested to know that the amount of material has increased by about 50% (in terms of number of pages) compared to the previous edition (see the reference below). There are now exercises after each chapter, with solutions (as previously mentioned). The authors also provide increased coverage on sample size calculations, method comparison studies, and survival analysis.
There are nevertheless a few shortcomings. With this edition, there is still virtually no mention of analysis of variance (a commonly used method when one is comparing means of three or more groups). This criticism was also mentioned in a review of the 3rd edition by Silke Lange which appeared in a medical statistics journal (Statistical Methods in Medical Research 2000; 9: 181-182). Further, for the analysis of 2 X 2 tables, both Fisher’s exact test and Yates’ correction for chi-square tests are discussed; the authors could have mentioned that only the former is generally used in statistical practice, since this computer intensive method has largely replaced the latter, which can be too conservative in detecting statistical significance.
It would have been also desirable to see more material on meta-analyses, which were barely mentioned; a few pages on how to interpret a typical meta-analysis would have been helpful, since these types of studies very frequently appear now in the medical literature. One technical error occurs on page 156, where the authors state that one of the assumptions underlying the test of statistical significance of the Pearson correlation is that at least one of the two variables has a normal distribution; actually, both variables need to be normal.
Finally, on page 208, the authors state that some computer programs give a p-value associated with kappa, which is a measure of agreement of two categorical variables (e.g., agreement of two raters in classifying 100 patients as positive or negative for a disease). The authors’ advice is to ignore this p-value (which tests no agreement vs. statistically significant agreement); however, there are instances (especially when agreement is modest) where one would wish to examine the p-value. But overall, the reviewer feels that the above deficiencies are relatively minor in nature, in comparison to the fine comprehensive coverage of introductory medical statistics that is presented here.
Curiously, little mention was made of statistical computer packages; consequently, an instructor using this book would need to provide supplementary material on this topic, if it was desired to work out the examples and exercises on a computer. Possibly some basic computer syntax could be provided which would precede the corresponding “generic” output that is sprinkled in the text. Fortunately, most of the examples and exercises can be done with a hand calculator (which can be a blessing, as it requires students to acquire the skills of doing basic statistical calculations and obtain a “sense of feel” in working with numerical data). The previous edition had a “final exam” of multiple choice questions, which was dropped from this edition. This probably should have been retained, since health profession students often take exams of this type.
In conclusion, this reviewer recommends without any reservation this book for an introductory course in medical statistics. The length of 290 pages (excluding solutions, statistical tables, references and index) would appear to be just about the right amount of material for a standard three-credit semester course. The authors state in the preface that the aim of training courses in medical statistics is not to turn students into medical statisticians but rather to help them interpret the published literature (“consumer”) and appreciate how to design studies and analyze data arising from their own projects (“doers”). The authors have admirably achieved this aim.
Campbell MJ, Machin D (1999); Medical statistics: A Commonsense Approach, 3rd edition. Chichester: John Wiley.
Martin Feuerman is a biostatistician in the Department of Academic Affairs at Winthrop-University Hospital in Mineola, Long Island, New York. He has also taught elementary statistics at Mercy College. Martin holds a B.S. degree in Mathematics from Brooklyn College and an M.S. degree in Applied Mathematics from New York University. He is delighted to have had this opportunity to prepare this review for MAA Reviews.
Chapter 1 Uses and abuses of medical statistics.
1.2 Why use statistics?
1.3 Statistics is about common sense and good design.
1.4 Types of data.
1.5 How a statistician can help.
1.6 Further reading.
Chapter 2 Describing and displaying categorical data.
2.1 Summarising categorical data.
2.2 Displaying categorical data.
2.3 Points when reading the literature.
Chapter 3 Describing and displaying quantitative data.
3.1 Summarising continuous data.
3.2 Displaying continuous data.
3.3 Within-subject variability.
3.5 Points when reading the literature.
Chapter 4 Probability and decision making.
4.1 Types of probability.
4.2 Diagnostic tests.
4.3 Bayes' Theorem.
4.4 Relative (receiver)-operating characteristic (ROC) curve.
4.5 Points when reading the literature.
Chapter 5 Distributions.
5.2 The Binomial distribution.
5.3 The Poisson distribution.
5.4 Probability for continuous outcomes.
5.5 The Normal distribution.
5.6 Reference ranges.
5.7 Points when reading the literature.
5.8 Technical details.
Chapter 6 Populations, samples, standard errors and confidence intervals.
6.3 The standard error.
6.4 The Central Limit Theorem.
6.5 Standard errors for proportions and rates.
6.6 Standard errors of differences.
6.7 Confidence intervals for an estimate.
6.8 Confidence intervals for differences.
6.9 Points when reading the literature.
6.10 Technical details.
Chapter 7 p-Values and statistical inference.
7.2 The null hypothesis.
7.3 The p-value.
7.4 Statistical inference.
7.5 Statistical power.
7.6 Confidence intervals rather than p-values.
7.7 One-sided and two-sided tests.
7.8 Points when reading the literature.
7.9 Technical details.
Chapter 8 Tests for comparing two groups of categorical or continuous data.
8.2 Comparison of two groups of paired observations - continuous outcomes.
8.3 Comparison of two independent groups - continuous outcomes.
8.4 Comparison of two independent groups - categorical outcomes.
8.5 Comparison of two groups of paired observations - categorical outcomes.
8.6 Non-Normal distributions.
8.7 Degrees of freedom.
8.8 Points when reading the literature.
8.9 Technical details.
Chapter 9 Correlation and linear regression.
9.3 Linear regression.
9.4 Comparison of assumptions between correlation and regression.
9.5 Multiple regression.
9.6 Logistic regression.
9.7 Correlation is not causation.
9.8 Points when reading the literature.
9.9 Technical details.
Chapter 10 Survival analysis.
10.1 Time to event data.
10.2 Kaplan-Meier survival curve.
10.3 The logrank test.
10.4 The hazard ratio.
10.5 Modelling time to event data.
10.6 Points when reading the literature.
Chapter 11 Reliability and method comparison studies.
11.5 Method comparison studies.
11.6 Points when reading the literature.
11.7 Technical details.
Chapter 12 Observational studies.
12.2 Risk and rates.
12.3 Taking a random sample.
12.4 Questionnaire and form design.
12.5 Cross-sectional surveys.
12.6 Non-randomised studies.
12.7 Cohort studies.
12.8 Case-control studies.
12.9 Association and causality.
12.10 Points when reading the literature.
12.11 Technical details.
Chapter 13 The randomised controlled trial.
13.2 Why randomise?
13.3 Methods of randomisation.
13.4 Design features.
13.5 Design options.
13.7 The protocol.
13.8 Checklists for design, analysis and reporting.
13.9 Number needed to treat (NNT).
13.10 Points when reading the literature.
Chapter 14 Sample size issues.
14.2 Study size.
14.3 Continuous data.
14.4 Binary data.
14.6 Subject withdrawals.
14.7 Internal pilot studies.
14.8 Points when reading the literature.
14.9 Technical details.
Chapter 15 Common pitfalls.
15.2 Using the t-test.
15.3 Plotting change against initial value.
15.4 Repeated measures.
15.5 Clinical and statistical significance.
15.6 Exploratory data analysis.
15.7 Points when reading the literature.
Solutions to exercises.