Multivariable Model-Building A Pragmatic Approach To Regression Anaylsis Based On Fractional Polynomials For Modelling Continuous Variables

Multivariable regression models are of fundamental importance in all areas of science in which empirical data must be analyzed. This book proposes a systematic approach to building such models based on standard principles of statistical modeling.

Multivariable regression models are widely used in all areas of science in which empirical data are analysed. Using the multivariable fractional polynomials (MFP) approach this book focuses on the selection of important variables and the determination of functional form for continuous predictors. Despite being relatively simple, the selected models often extract most of the important information from the data. The authors have chosen to concentrate on examples drawn from medical statistics, although the MFP method has applications in many other subject-matter areas as well.

Multivariable Model-Building:

• Focuses on normal-error models for continuous outcomes, logistic regression for binary outcomes and Cox regression for censored time-to-event data.
• Concentrates on fractional polynomial models and illustrates new approaches to model critisism and stability.
• Provides comparisons with and discussion of other techniques such as spline models.
• Features new strategies on modelling interactions with continuous covariates which are important in the context of randomized trials and observational studies
• Does not consider high-dimensional data, such as gene expression data.
• Is illustrated throughout with working examples from more than 20 substantial real datasets, most data sets and programs in Stata are available on a website enabling the reader to apply techniques directly
• Is written in an accessible and informal style making it suitable for researchers from a range of disciplines with minimal mathematical background

This book provides a readable text giving the rationale of, and practical advice on, a unified approach to multivariable modelling. It aims to make multivariable model building simpler, transparent and more effective. This book is aimed at graduate students studying regression modelling and professionals in statistics as well as researchers from medical, physical, social and many other sciences where regression models play a central role.

Patrick Royston DSc, is a senior statistician and cancer clinical trialist at the MRC Clinical Trials Unit, London, an honorary professor of statistics at University College London, and a fellow of the Royal Statistical Society. He has authored many research papers in biostatistics, and has published over 150 articles in leading statistical journals. Patrick is an experienced statistical consultant, Stata programmer and software author.

Willi Sauerbrei PhD, is a senior statistician and professor in medical biometry at the IMBI, University Medical Center Freiburg. He has authored many research papers in biostatistics, and has published over 100 articles in leading statistical and clinical journals. He worked for more than two decades as an academic biostatistician and has extensive experience of cancer research, with a particular concern for breast cancer.

Multivariable regression models are of fundamental importance in all areas of science in which empirical data must be analyzed. This book proposes a systematic approach to building such models based on standard principles of statistical modeling. The main emphasis is on the fractional polynomial method for modeling the influence of continuous variables in a multivariable context, a topic for which there is no standard approach. Existing options range from very simple step functions to highly complex adaptive methods such as multivariate splines with many knots and penalisation. This new approach, developed in part by the authors over the last decade, is a compromise which promotes interpretable, comprehensible and transportable models.