Epidemiology: Study Design and Data Analysis

Citation

Woodward, M. (2013). Epidemiology: Study Design and Data Analysis (3rd ed.). Chapman & Hall/CRC.

Chapter Summary

The book has fourteen chapters and three appendices (materials on the book's website, statistical tables, and additional data sets for exercises). Every chapter ends with exercises. The publisher describes it as self-contained: it introduces study design, reviews the standard methods, and then takes the reader through analysis for cohort, case-control and intervention studies, sample size, statistical modelling, meta-analysis, risk forecasting and computer-intensive methods.

Chapter 1: Fundamental Issues

The opening chapter asks what epidemiology is, using the work of Doll and Hill as a case study. It covers populations and samples, the measurement of disease and of risk factors, causality, studies using routine data, an overview of study design, and an overview of data analysis.

Chapter 2: Basic Analytical Procedures

A review of elementary statistics for readers who need it: types of variables, tables and charts, inference for categorical variables, descriptive techniques for quantitative variables, inference about means and for non-normal data, and the measurement of agreement and the assessment of diagnostic tests.

Chapter 3: Assessing Risk Factors

This chapter develops the standard measures of association: risk and relative risk, odds and odds ratio (and how to choose between them), prevalence studies, tests of association, risk factors measured at several levels, attributable risk, rates and relative rates, and measures of difference. It includes Stata EPITAB commands.

Chapter 4: Confounding and Interaction

The concepts of confounding and interaction are introduced, with ways to identify, assess and handle each, standardization and Mantel-Haenszel methods, and tests for interaction.

Chapter 5: Cohort Studies

Design and analytical considerations for cohort studies are followed by cohort life tables, Kaplan-Meier estimation, comparison of survival probabilities, competing risk, the person-years method and period-cohort analysis.

Chapter 6: Case-Control Studies

The chapter covers basic design and analysis, selection of cases and controls, matching and the analysis of matched studies, and variants including nested case-control, case-cohort and case-crossover designs.

Chapter 7: Intervention Studies

Ethics and the avoidance of bias are discussed, followed by parallel group, cross-over and sequential designs, allocation to treatment groups, and the analysis of trials as cohorts.

Chapter 8: Sample Size Determination

Power is introduced, then sample size for testing a mean, a difference between means, a proportion and a relative risk, for case-control studies, and for complex sampling designs.

Chapter 9: Modeling Quantitative Outcome Variables

Statistical models for continuous outcomes, from one explanatory variable to several, model building and checking, general linear models, confounding, splines, panel data and non-normal alternatives.

Chapter 10: Modeling Binary Outcome Data

This chapter explains why standard regression is a problem for binary outcomes and develops logistic regression: interpretation of coefficients, multiple models, tests, confounding and interaction, quantitative explanatory variables, model checking and measurement error. Case-control studies, outcomes with several levels, longitudinal data, binomial regression and propensity scoring complete the chapter.

Chapter 11: Modeling Follow-up Data

Survival-time functions and the estimation of the hazard function lead to proportional hazards models, including the Cox and Weibull models, model checking, competing risk, Poisson regression and pooled logistic regression.

Chapter 12: Meta-analysis

The chapter treats reviewing evidence and systematic review, a general approach to pooling, heterogeneity, pooling tabular data, individual participant data, study quality, publication bias, and the strengths and limitations of meta-analysis.

Chapter 13: Risk Scores and Clinical Decision Rules

New in this edition. It covers risk scores built from statistical models, discrimination, calibration and recalibration, the accuracy of predictions, reclassification, validation, presentation of risk scores and impact studies.

Chapter 14: Computer-Intensive Methods

Also new. It covers the bootstrap, bootstrap confidence intervals and hypothesis testing, permutation tests and missing values, including naive imputation and univariate and multivariate multiple imputation.

Key Concepts

Chapters 1–4: Foundations

• Relative risk and odds ratio: the two main measures of association, with guidance on choosing between them (Chapter 3).
• Confounding and interaction: ways a third factor distorts or modifies an association (Chapter 4).
• Standardization and Mantel-Haenszel methods: ways to adjust for confounding (Chapter 4).

Chapters 5–8: Designs

• Kaplan-Meier estimation and the person-years method: analysis of cohort follow-up (Chapter 5).
• Matching and nested case-control designs: variants of the case-control approach (Chapter 6).
• Power and sample size: planning how large a study must be (Chapter 8).

Chapters 9–14: Models and advanced methods

• Logistic regression and propensity scoring (Chapter 10); proportional hazards models (Chapter 11).
• Meta-analysis, heterogeneity and publication bias (Chapter 12).
• Discrimination and calibration of risk scores (Chapter 13).
• Bootstrap, permutation tests and multiple imputation (Chapter 14).

Critical Analysis

Strengths

• Statistics in an epidemiological setting. The book is written by a biostatistician and treats each design with its own analysis: cohort, case-control and intervention studies each have a dedicated chapter, and the modelling chapters are organised by outcome type (quantitative, binary and follow-up).
• Coverage kept current. The third edition adds chapters on risk scores and computer-intensive methods, and sections on binomial regression, competing risk, information criteria, propensity scoring and splines.
• Practical support. The publisher provides real data sets, SAS and Stata programs and a sample size spreadsheet on the book's web page, and the exercises use Stata and SAS.
• Reviewer response. In The American Statistician (November 2014), Gregory E. Gilbert wrote that the author writes extremely well and that the text is full of exercises. Reviews of earlier editions praised the approachable level of technical difficulty and the use of nothing more complicated than summation signs for sophisticated models.

Limitations

• A methods book, not a first epidemiology course. The book's emphasis is quantitative; the conceptual material on causation and study logic is concentrated in the first chapter, so readers new to epidemiology will want a conceptual text alongside it.
• Software choice. Worked programs are in SAS and Stata. Readers who use other software will need to translate the examples.
• Survey of many models. A reviewer of an earlier edition (Short Book Reviews of the International Statistical Institute) called the tour of commonly used models decent but very quick, so readers needing depth in one model family should consult a specialised text.

Where It Fits in the Course

This is the quantitative, analysis-focused book in the epidemiology course. Gordis's Epidemiology, Epidemiology: An Introduction and Essential Epidemiology are the conceptual and introductory texts: they explain how disease is measured, how studies are designed and how causation is judged, with little or no regression. They are the right place to start; Woodward then adds the statistical machinery.

Modern Epidemiology is the graduate-level comprehensive reference and treats causal inference and the quantitative assessment of bias in depth. Woodward is shorter and more directly tied to analysis and software, so it is a practical companion when the task is to analyse a data set, and Modern Epidemiology is the next step for theory and bias.

How to Use This Book

The book suits postgraduate students in epidemiology, public health and medical statistics, and researchers who need to analyse cohort, case-control or trial data. It helps to have taken an introductory epidemiology course and a first statistics course; Chapter 2 reviews elementary statistics, and the later chapters assume that regression is not entirely new.

A practical plan:

• Core: Chapters 1 and 3–7 for design and association measures; Chapter 4 on confounding underlies the later modelling chapters.
• Analysis: Chapters 9–11 on regression models, in order; each builds on the preceding one.
• As needed: Chapter 8 when planning a study, Chapter 12 for reviews, and Chapters 13–14 for prediction and resampling methods.
• Practice: download the data sets and programs, reproduce the worked examples in Stata or SAS, and attempt the exercises at the end of each chapter.