Statistical thinking for non-statisticians in drug regulation Richard Kay (2007) ISBN 0470319712; 296 pages; £39.95, €52.90, $80.00 Wiley; http://www.wiley.com/
Article first published online: 19 MAR 2009
Copyright © 2009 John Wiley & Sons, Ltd.
Volume 10, Issue 3, page 282, May/June 2011
How to Cite
Mitchell, L. (2011), Statistical thinking for non-statisticians in drug regulation Richard Kay (2007) ISBN 0470319712; 296 pages; £39.95, €52.90, $80.00 Wiley; http://www.wiley.com/. Pharmaceut. Statist., 10: 282. doi: 10.1002/pst.372
- Issue published online: 19 MAR 2009
- Article first published online: 19 MAR 2009
This excellent book presents a more effective use of statistical arguments within the regulatory process and helps improving critical reviews of reports and publications. There are 16 chapters in the book in which statistical methods in the pharmaceutical industry and their relationship with current regulatory guidelines are very well presented.
Chapter 1 introduces basic ideas in clinical trial design such as control groups, placebos, blinding, randomisation, bias and precision, between and within patient designs, signal and noise, confirmatory and exploratory trials, superiority, equivalence, non-inferiority, data types, choice of endpoints together with important notes for ICH E9 and ICHE10 guidelines and examples. Sampling and inferential statistics based on sample and population, median and mean, standard deviation, the normal distribution sampling and the standard errors are introduced in Chapter 2.
Topics such as calculation and interpretation of confidence intervals, testing hypotheses and p-values can be found in Chapter 3.
Chapter 4 takes us into the area of statistical tests for continuous and binary data: paired and unpaired t-tests and their interpretations, chi-square test for binary data, categorical and ordinal, between and within patient designs and continuous data, Fisher's exact test, odds ratio, relative risk, relative risk reduction, number needed to treat.
Chapter 5 continues to discuss methods for continuous, binary, categorical and ordinary data including the treatment-interaction investigation. Also, this chapter brings rationale for multi-centre trial based on ICH E9: to recruit sufficient numbers of patients within an appropriate timeframe to allow the evaluation of the homogeneity of the treatment effect and provide a basis for generalizability.
Chapter 6 introduces concepts of adjusted analyses and it discusses statistical methods such as analysis of covariance (ANCOVA), linear regression, multiple regressions and logistic regression. There are also recommendations from the ICH E9 and CPMP (2003).
Chapter 7 has an introduction into intention-to-treat analysis sets, per-protocol set, missing data, methods for missing data and key points on the avoidance of missing data based on CPMP (2001).
Type I and Type II errors are clearly explained in Chapter 8.
Chapter 9 returns us to the topics of Chapter 3, confidence interval and p-values and misinterpretation of p-values. This chapter also briefly discusses the distinction between statistical significance and clinical importance.
Chapter 10 deals with multiple testing, inflation of the Type I error and corresponding quotes from ICH E9 (1998).
Chapter 11 discusses non-parametric methods such as: Mann–Whitney U-test, Wilcoxon signed rank test, advantages and disadvantages of non-parametric methods.
Chapter 12 brings us into area of equivalence and non-inferiority. Two-sided confidence intervals should be used for equivalence, while a one-sided confidence interval for the non-inferiority; p-value has no role to play in equivalence or non-inferiority. Statistical and clinical justification should be provided for the equivalent or non-inferiority margin. Switching between non-inferiority and superiority, analyses sets and principle of bioequivalence are also included in this chapter.
The analysis of survival data is discussed in Chapter 13. Time-to-event data, their censoring, Kaplan–Meier, event rates and relative risk, median event times, hazard ratio (constant and non-constant HR), stratified methods, proportional hazard regression, accelerated failure time model are presented with corresponding examples in this chapter.
Interim analyses and data monitoring committees are described in Chapter 14. Stopping rules for interim analysis, stopping for efficacy and futility, futility and conditional power, monitoring safety and data monitoring committees and corresponding notes from CHMP (2005) Guideline on Data Monitoring Committees and FDA (2006) are also discussed. As a part of this chapter, adaptive design is introduced and main points from CHMP 2006 Reflection Paper on Methodological Issues in Confirmatory Clinical Trials with Flexible Design and Analysis Plan. I wish that this chapter had a longer discussion and more details regarding adaptive designs.
Chapter 15 introduces meta-analyses in a regulatory setting with recommendation that meta-analyses should be pre-planned analyses. This chapter also discusses the difference between pooling data and meta-analyses.
The importance of statistical thinking during clinical development is discussed in the last chapter, Chapter 16. This chapter describes main roles of statisticians in pharmaceutical companies, such as: discussing and writing the statistical methods section of the protocol, the data validation plan, the blind review, the statistical analysis plan, reporting the analyses, sensitivity and robustness and the regulatory submissions.
The book is clearly written and explains statistical methods in an easy and understandable way. The author assumed that readers are familiar with general aspects of the drug development process and have previous knowledge of the phase I to phase IV framework, placebos, control groups and other basic elements of clinical trials. Although there is little of the material that is really new for statisticians and non-statisticians in the pharmaceutical industry, the book is an excellent guideline for everyday work in the pharmaceutical environment and it can improve communication between statisticians and non-statisticians. The book is low on mathematical equations, but strong on explanation of statistical methods used in the pharmaceutical environment and their relationship with regulatory requirements. My only suggestion for updated editions of this book is to include more discussion on Bayesian methods, analyses of safety data and more details for adaptive designs. In summary, I will definitely use this book a lot during my everyday work and I highly recommend it to both statisticians and non-statisticians.