Applied Survival Analysis Using R

Visit the link below to download the full version of this book:

https://medipdf.com/product/applied-survival-analysis-using-r/

Click Download Now

Dirk F. Moore

Applied Survival Analysis

Using R

123
Dirk F. Moore
Department of Biostatistics
Rutgers School of Public Health
Piscataway, NJ, USA

ISSN 2197-5736 ISSN 2197-5744 (electronic)

Use R!
ISBN 978-3-319-31243-9 ISBN 978-3-319-31245-3 (eBook)
DOI 10.1007/978-3-319-31245-3

Library of Congress Control Number: 2016940055

© Springer International Publishing Switzerland 2016

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, express or implied, with respect to the material contained herein or for any
errors or omissions that may have been made.

Printed on acid-free paper

This Springer imprint is published by Springer Nature

The registered company is Springer International Publishing AG Switzerland
To Lynne, Molly, and Emily
Preface

This book serves as an introductory guide for students and analysts who need
to work with survival time data. The minimum prerequisites are basic applied
courses in linear regression and categorical data analysis. Students who also have
taken a master’s level course in statistical theory will be well prepared to work
through this book, since frequent reference is made to maximum likelihood theory.
Students lacking this training may still be able to understand most of the material,
provided they have an understanding of the basic concepts of differential and
integral calculus. Specifically, students should understand the concept of the limit,
and they should know what derivatives and integrals are and be able to evaluate them
in some basic cases.
The material for this book has come from two sources. The first source is
an introductory class in survival analysis for graduate students in epidemiology
and biostatistics at the Rutgers School of Public Health. Biostatistics students, as
one would expect, have a much firmer grasp of more mathematical aspects of
statistics than do epidemiology students. Still, I have found that those epidemiology
students with strong quantitative backgrounds have been able to understand some
mathematical statistical procedures such as score and likelihood ratio tests, provided
that they are not expected to symbolically differentiate or integrate complex
formulas. In this book I have, when possible, used the numerical capabilities of the
R system to substitute for symbolic manipulation. The second source of material
is derived from collaborations with physicians and epidemiologists at the Rutgers
Cancer Institute of New Jersey and at the Rutgers Robert Wood Johnson Medical
School. A number of the data sets in this text are derived from these collaborations.
Also, the experience of training statistical analysts to work on these data sets
provided additional inspiration for the book.
The first chapter introduces the concepts of survival times and how right
censoring occurs and describes several of the datasets that will be used throughout
the book. Chapter 2 presents fundamentals of survival theory. This includes hazard,
probability density, survival functions, and how they are related. The hazard
function is illustrated using both life table data and using some common parametric
distributions. The chapter ends with a brief introduction to properties of maximum

vii
viii Preface

likelihood estimates using the exponential distribution as an illustration. Chapter 3

discusses the Kaplan-Meier estimate of the survival function and several related
concepts such as the median survival and its confidence interval. Also discussed
in this chapter are smoothing of the hazard function and how to accommodate left
truncation into the Kaplan-Meier estimate.
Chapter 4 discusses the log-rank test for comparing survival distributions and
also some modified linear rank tests. Stratified tests are also discussed, along with
an example where stratification can reverse the apparent direction of a treatment
effect in a survival example of Simpson’s paradox. In Chapter 5, we present the
Cox proportional hazards model and partial likelihood function in the context of
comparing two groups of survival data. There we illustrate the Wald, score, and
likelihood ratio tests in this basic context. Left-truncated survival data and the partial
likelihood are also discussed.
Chapter 6 presents methods for model selection and extends and illustrates the
proportional hazards model in situations where there are multiple possible predictor
covariates. Chapter 7 presents diagnostic residual plots that are useful for assessing
model assumptions. Chapter 8 discusses how to adapt the survival models discussed
earlier to allow for time-dependent covariates.
The next few chapters discuss some important special situations. Chapter 9
discusses multiple outcomes, which can occur as clustered survival times or in
a competing risks framework, where only the first of multiple outcomes can be
observed. Chapter 10 discusses parametric survival models, and Chapter 11 covers
the critically important design question of how to determine the power and sample
size of a proposed study that has a survival outcome. Finally, Chapter 12 presents
some additional topics, including the piecewise exponential distribution, methods
for handling interval censoring, and the lasso method for handling survival data with
large numbers of predictors. Many of the data sets discussed in the text are available
in the accompanying R package “asaur” (for “Applied Survival Analysis Using R”),
while others are in other packages. All are freely available for download from the
Central R Archive Network at cran.r-project.org. The R-code discussed in the book
is available for download at http://www.springer.com/us/book/9783319312439
A key feature of this book is the integration of the R statistical system with
the survival analysis material. Not only do we show the reader how to use R
functions to fit survival models and how to interpret the results, but we also use R
to illustrate how survival quantities are computed. Typically we use small examples
to illustrate in detail how one constructs survival tests, partial likelihood models,
and diagnostics and then proceed to more complicated examples. Most of the
survival functions will require that the “survival” library be attached using the
“library(survival)” statement. The “survival” package is included by default; other
packages referred to in the text must be explicitly downloaded and installed. The
appendix includes both some basics of the R language and special features relevant
to the survival calculations used elsewhere in the book. Users not already familiar
with the R system should refer to one of the many online resources for more detailed
information.
Preface ix

I would like to thank Rebecca Moss for permission to use the “pancreatic” data
and Michael Steinberg for permission to use the “pharmacoSmoking” data. Both of
these data sets are used repeatedly throughout the text. I would also like to thank
Grace Lu-Yao, Weichung Joe Shih, and Yong Lin for years-long collaborations
on using the SEER-Medicare data for studying the survival trajectories of prostate
cancer patients. These collaborations led to the development of the “prostateSur-
vival” data set discussed in this text in Chapter 9. I thank the Division of Cancer
Epidemiology and Genetics of the US National Cancer Institute for providing the
“asheknazi” data. I also thank Wan Yee Lau for making the “hepatoCellular” data
publically available in the online Dryad data repository and for allowing me to
include it in the “asaur” R package.

Piscataway, NJ, USA Dirk F. Moore

October 2015
Contents

1 Introduction .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 1
1.1 What Is Survival Analysis? . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 1
1.2 What You Need to Know to Use This Book . . . .. . . . . . . . . . . . . . . . . . . . 2
1.3 Survival Data and Censoring . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 2
1.4 Some Examples of Survival Data Sets . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 6
1.5 Additional Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 9
2 Basic Principles of Survival Analysis . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 11
2.1 The Hazard and Survival Functions . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 11
2.2 Other Representations of a Survival Distribution .. . . . . . . . . . . . . . . . . . 13
2.3 Mean and Median Survival Time . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 14
2.4 Parametric Survival Distributions . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 15
2.5 Computing the Survival Function from the Hazard Function .. . . . . 19
2.6 A Brief Introduction to Maximum Likelihood Estimation . . . . . . . . . 20
2.7 Additional Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 23
3 Nonparametric Survival Curve Estimation. . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 25
3.1 Nonparametric Estimation of the Survival Function . . . . . . . . . . . . . . . 25
3.2 Finding the Median Survival and a Confidence Interval
for the Median .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 30
3.3 Median Follow-Up Time . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 32
3.4 Obtaining a Smoothed Hazard and Survival Function Estimate . . . 32
3.5 Left Truncation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 36
3.6 Additional Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 41
4 Nonparametric Comparison of Survival Distributions . . . . . . . . . . . . . . . . . 43
4.1 Comparing Two Groups of Survival Times . . . . .. . . . . . . . . . . . . . . . . . . . 43
4.2 Stratified Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 49
4.3 Additional Note . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 52

xi
xii Contents

5 Regression Analysis Using the Proportional Hazards Model .. . . . . . . . . 55

5.1 Covariates and Nonparametric Survival Models .. . . . . . . . . . . . . . . . . . . 55
5.2 Comparing Two Survival Distributions Using
a Partial Likelihood Function .. . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 56
5.3 Partial Likelihood Hypothesis Tests . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 59
5.3.1 The Wald Test. . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 60
5.3.2 The Score Test . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 60
5.3.3 The Likelihood Ratio Test. . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 60
5.4 The Partial Likelihood with Multiple Covariates . . . . . . . . . . . . . . . . . . . 63
5.5 Estimating the Baseline Survival Function.. . . . .. . . . . . . . . . . . . . . . . . . . 64
5.6 Handling of Tied Survival Times . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 65
5.7 Left Truncation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 69
5.8 Additional Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 71
6 Model Selection and Interpretation . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 73
6.1 Covariate Adjustment . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 73
6.2 Categorical and Continuous Covariates . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 74
6.3 Hypothesis Testing for Nested Models . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 78
6.4 The Akaike Information Criterion for Comparing
Non-nested Models.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 81
6.5 Including Smooth Estimates of Continuous Covariates
in a Survival Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 84
6.6 Additional Note . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 86
7 Model Diagnostics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 87
7.1 Assessing Goodness of Fit Using Residuals . . . .. . . . . . . . . . . . . . . . . . . . 87
7.1.1 Martingale and Deviance Residuals . . .. . . . . . . . . . . . . . . . . . . . 87
7.1.2 Case Deletion Residuals. . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 92
7.2 Checking the Proportion Hazards Assumption .. . . . . . . . . . . . . . . . . . . . 94
7.2.1 Log Cumulative Hazard Plots . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 94
7.2.2 Schoenfeld Residuals . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 96
7.3 Additional Note . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 100
8 Time Dependent Covariates . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 101
8.1 Introduction .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 101
8.2 Predictable Time Dependent Variables .. . . . . . . . .. . . . . . . . . . . . . . . . . . . . 106
8.2.1 Using the Time Transfer Function . . . . .. . . . . . . . . . . . . . . . . . . . 107
8.2.2 Time Dependent Variables That Increase
Linearly with Time . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 109
8.3 Additional Note . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 110
9 Multiple Survival Outcomes and Competing Risks. . . . . . . . . . . . . . . . . . . . . 113
9.1 Clustered Survival Times and Frailty Models. . .. . . . . . . . . . . . . . . . . . . . 113
9.1.1 Marginal Survival Models. . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 115
9.1.2 Frailty Survival Models . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 116
9.1.3 Accounting for Family-Based Clusters
in the “ashkenazi” Data . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 117
Contents xiii

9.1.4 Accounting for Within-Person Pairing of Eye

Observations in the Diabetes Data . . . . .. . . . . . . . . . . . . . . . . . . . 120
9.2 Cause-Specific Hazards . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 121
9.2.1 Kaplan-Meier Estimation with Competing Risks . . . . . . . . . 121
9.2.2 Cause-Specific Hazards and Cumulative
Incidence Functions . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 123
9.2.3 Cumulative Incidence Functions for Prostate
Cancer Data .. . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 126
9.2.4 Regression Methods for Cause-Specific Hazards . . . . . . . . . 128
9.2.5 Comparing the Effects of Covariates on
Different Causes of Death .. . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 131
9.3 Additional Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 134
10 Parametric Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 137
10.1 Introduction .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 137
10.2 The Exponential Distribution .. . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 137
10.3 The Weibull Model .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 138
10.3.1 Assessing the Weibull Distribution as a Model
for Survival Data in a Single Sample . .. . . . . . . . . . . . . . . . . . . . 138
10.3.2 Maximum Likelihood Estimation of Weibull
Parameters for a Single Group of Survival Data . . . . . . . . . . 141
10.3.3 Profile Weibull Likelihood . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 142
10.3.4 Selecting a Weibull Distribution to Model
Survival Data . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 143
10.3.5 Comparing Two Weibull Distributions Using
the Accelerated Failure Time and Proportional
Hazards Models. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 146
10.3.6 A Regression Approach to the Weibull Model .. . . . . . . . . . . 148
10.3.7 Using the Weibull Distribution to Model
Survival Data with Multiple Covariates . . . . . . . . . . . . . . . . . . . 149
10.3.8 Model Selection and Residual Analysis with
Weibull Survival Data . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 151
10.4 Other Parametric Survival Distributions . . . . . . . .. . . . . . . . . . . . . . . . . . . . 153
10.5 Additional Note . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 154
11 Sample Size Determination for Survival Studies . . . .. . . . . . . . . . . . . . . . . . . . 157
11.1 Power and Sample Size for a Single Arm Study .. . . . . . . . . . . . . . . . . . . 157
11.2 Determining the Probability of Death in a Clinical Trial .. . . . . . . . . . 161
11.3 Sample Size for Comparing Two Exponential Survival
Distributions .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 163
11.4 Sample Size for Comparing Two Survival Distributions
Using the Log-Rank Test . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 165
11.5 Determining the Probability of Death
from a Non-parametric Survival Curve Estimate . . . . . . . . . . . . . . . . . . . 166
11.6 Example: Calculating the Required Number of Patients
for a Randomized Study of Advanced Gastric Cancer Patients .. . . 169
xiv Contents

11.7 Example: Calculating the Required Number of Patients

for a Randomized Study of Patients with Metastatic
Colorectal Cancer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 170
11.8 Using Simulations to Estimate Power . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 171
11.9 Additional Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 174
12 Additional Topics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 177
12.1 Using Piecewise Constant Hazards to Model Survival Data . . . . . . . 177
12.2 Interval Censoring .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 187
12.3 The Lasso Method for Selecting Predictive Biomarkers . . . . . . . . . . . 192

A A Basic Guide to Using R for Survival Analysis . . . .. . . . . . . . . . . . . . . . . . . . 201

A.1 The R System .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 201
A.1.1 A First R Session . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 202
A.1.2 Scatterplots and Fitting Linear Regression Models .. . . . . . 204
A.1.3 Accommodating Non-linear Relationships .. . . . . . . . . . . . . . . 207
A.1.4 Data Frames and the Search Path for Variable Names .. . . 209
A.1.5 Defining Variables Within a Data Frame . . . . . . . . . . . . . . . . . . 211
A.1.6 Importing and Exporting Data Frames . . . . . . . . . . . . . . . . . . . . 211
A.2 Working with Dates in R . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 212
A.2.1 Dates and Leap Years.. . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 213
A.2.2 Using the “as.date” Function .. . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 213
A.3 Presenting Coefficient Estimates Using Forest Plots . . . . . . . . . . . . . . . 215
A.4 Extracting the Log Partial Likelihood and Coefficient
Estimates from a coxph Object . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 217
References .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 218

Index . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 223

R Package Index.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 225

Chapter 1
Introduction

1.1 What Is Survival Analysis?

Survival analysis is the study of survival times and of the factors that influence
them. Types of studies with survival outcomes include clinical trials, prospective and
retrospective observational studies, and animal experiments. Examples of survival
times include time from birth until death, time from entry into a clinical trial until
death or disease progression, or time from birth to development of breast cancer
(that is, age of onset). The survival endpoint can also refer a positive event. For
example, one might be interested in the time from entry into a clinical trial until
tumor response. Survival studies can involve estimation of the survival distribution,
comparisons of the survival distributions of various treatments or interventions, or
elucidation of the factors that influence survival times. As we shall see, many of the
techniques we study have analogues in generalized linear models such as linear or
logistic regression.
Survival analysis is a difficult subject, and a full exposition of its principles would
require readers to have a background not only in basic statistical theory but also in
advanced topics in the theory of probability. Fortunately, many of the most important
concepts in survival analysis can be presented at a more elementary level. The aim
of this book is to provide the reader with an understanding of these principles and
also to serve as a guide to implementing these ideas in a practical setting. We shall
use the R statistical system extensively throughout the book because (1) it is a
high-quality system for doing statistics, (2) it includes a wealth of enhancements
and packages for doing survival analysis, (3) its interactive design will allow us
to illustrate survival concepts, and (4) it is an open source package available for
download to anyone at no cost from the main R website, www.R-project.org. This
book is meant to be used as well as read, and the reader is encouraged to use R
to try out the examples discussed in the text and to do the exercises at the end of
each chapter. It is expected that readers are already familiar with the R language; for

© Springer International Publishing Switzerland 2016 1

D. Moore, Applied Survival Analysis Using R, Use R!,
DOI 10.1007/978-3-319-31245-3_1