Queensland University of Technology
School of Nursing and Midwifery
Faculty of Health
Institute of Health and Biomedical Innovation

Alternative Analytical Methods for the Identification of
Cancer-Related Symptom Clusters

Helen Mary Skerman
DipTeach, BSc, GradDipCompEd, MSocSc (App)
This thesis is submitted
to fulfil the requirements for the
Award of Doctor of Philosophy

MAY 2010



KEY WORDS

Symptom clusters; symptoms; cancer; symptom experience; symptom management
strategies; literature review; empirical methods; multivariate methods; exploratory
factor analysis; common factor analysis; cluster analysis; principal axis factoring;
stability; longitudinal analysis; chemotherapy; outpatients; nursing research;
oncology; Theory of Unpleasant Symptoms

i


ABSTRACT

Advances in symptom management strategies through a better understanding
of cancer symptom clusters depend on the identification of symptom clusters that are
valid and reliable. The purpose of this exploratory research was to investigate
alternative analytical approaches to identify symptom clusters for patients with
cancer, using readily accessible statistical methods, and to justify which methods of
identification may be appropriate for this context. Three studies were undertaken: (1)
a systematic review of the literature, to identify analytical methods commonly used
for symptom cluster identification for cancer patients; (2) a secondary data analysis
to identify symptom clusters and compare alternative methods, as a guide to best
practice approaches in cross-sectional studies; and (3) a secondary data analysis to
investigate the stability of symptom clusters over time.

The systematic literature review identified, in 10 years prior to March 2007,
13 cross-sectional studies implementing multivariate methods to identify cancer
related symptom clusters. The methods commonly used to group symptoms were
exploratory factor analysis, hierarchical cluster analysis and principal components
analysis. Common factor analysis methods were recommended as the best practice
cross-sectional methods for cancer symptom cluster identification.

A comparison of alternative common factor analysis methods was conducted,
in a secondary analysis of a sample of 219 ambulatory cancer patients with mixed
diagnoses, assessed within one month of commencing chemotherapy treatment.
Principal axis factoring, unweighted least squares and image factor analysis

ii


identified five consistent symptom clusters, based on patient self-reported distress
ratings of 42 physical symptoms. Extraction of an additional cluster was necessary
when using alpha factor analysis to determine clinically relevant symptom clusters.

The recommended approaches for symptom cluster identification using nonmultivariate normal data were: principal axis factoring or unweighted least squares
for factor extraction, followed by oblique rotation; and use of the scree plot and
Minimum Average Partial procedure to determine the number of factors.

In contrast to other studies which typically interpret pattern coefficients
alone, in these studies symptom clusters were determined on the basis of structure
coefficients. This approach was adopted for the stability of the results as structure
coefficients are correlations between factors and symptoms unaffected by the
correlations between factors. Symptoms could be associated with multiple clusters as
a foundation for investigating potential interventions.

The stability of these five symptom clusters was investigated in separate
common factor analyses, 6 and 12 months after chemotherapy commenced. Five
qualitatively

consistent

symptom

(Musculoskeletal-discomforts/lethargy,

clusters

were

identified

Oral-discomforts,

over


time

Gastrointestinal-

discomforts, Vasomotor-symptoms, Gastrointestinal-toxicities), but at 12 months two
additional clusters were determined (Lethargy and Gastrointestinal/digestive
symptoms). Future studies should include physical, psychological, and cognitive
symptoms. Further investigation of the identified symptom clusters is required for
validation, to examine causality, and potentially to suggest interventions for
symptom management. Future studies should use longitudinal analyses to investigate

iii


change in symptom clusters, the influence of patient related factors, and the impact
on outcomes (e.g., daily functioning) over time.

iv


TABLE OF CONTENTS
Keywords

i

Abstract

ii


Table of Contents

v

List of Tables

x

List of Figures

xi

Declaration of Authorship

xii

Glossary of Acronyms and Terms

xiii

Publications from the Research Program

xv

Statement of Contribution of Co-authors

xvi

Funding for the Research Program


xvii

Acknowledgements

xviii

Chapter 1: Introduction
1.1 Introduction

1

1.2 The Burden of Cancer

2

1.3 Rationale and Significance of the Research

3

1.4 Research Purpose and Objectives

7

1.5 Research Questions

8

1.6 Thesis Outline

9


v


Chapter 2: Background
2.1 Introduction

13

2.2 The Cancer Symptom Experience

13

2.2.1 Change in the Symptom Experience

14

2.2.2 Change in the Symptom Experience Over Time

15

2.3 Concept of a Symptom Cluster

16

2.4 The Clinical Relevance of Symptom Clusters

19

2.5 The Symptom Experience and Symptom Management Models


21

2.5.1 The Theory of Unpleasant Symptoms (TOUS)

22

2.5.2 The Symptoms Experience Model (SEM)

23

2.5.3 The Model of Symptom Management

24

2.5.4 Symptom Interaction Framework

25

2.6 Summary

25

Chapter 3: Symptom Cluster Identification
3.1 Introduction

27

3.2 Measuring Symptoms


27

3.3 Approaches to Symptom Cluster Identification

31

3.3.1 The Clinical Approach

31

3.3.2 The Empirical Identification of Symptom Clusters

32

3.4 Towards Best Practice Methods

36

3.5 Summary

38

vi


Chapter 4: A Systematic Literature Review
4.1 Introduction

41


4.2 Method

42

4.3 Multivariate Methods to Identify Cancer-Related Symptom Clusters

43

4.4 Summary

83

Chapter 5: Methods
5.1 Introduction

85

5.2 The Conceptual Framework

86

5.3 Study Design

90

5.3.1 The Parent Study
5.4 Measures

5.5


90
94

5.4.1 Measures in Parent Study - Ambulatory Care Project

94

5.4.2 Selected Measures in Current Study

96

Statistical Analysis

99

5.5.1 Data Quality

99

5.5.2 Missing Data

100

5.5.3 Complete Data for Exploratory Factor Analysis

102

5.5.4 Study 2: Identification of Symptom Clusters

102


5.5.5 Number of Factors to Retain

104

5.5.6 Simple Structure and Symptom Cluster Identification

106

5.5.7 Alternative Methods of Common Factor Extraction

107

5.5.8 Study 3: Identification of Symptom Clusters over Time

110

5.6 Sample size

112

5.7 Limitations

112

vii


Chapter 6: The Empirical Identification of Symptom Clusters
6.1 Introduction


115

6.2 Identification of Cancer-Related Symptom Clusters: An Empirical

116

Comparison of Exploratory Factor Analysis Methods
6.3 Summary

145

Chapter 7: The Empirical Identification of Symptom Clusters over Time
7.1 Introduction

147

7.2 Cancer-Related Symptom Clusters, 6 and 12 Months after Commencing
Chemotherapy: An Empirical Investigation

148

Chapter 8: Final Discussion and Conclusions

177

8.1 Key Findings

178


8.1.1 Multivariate Methods for Symptom Cluster Identification

180

8.1.2 EFA Decisions for Symptom Cluster Identification

181

8.1.3 Stability of Symptom Clusters Identified at Different Times

185

8.2 Strengths and Limitations

187

8.3 Implications of the Findings

195

8.3.1 Conceptual Implications

195

8.3.2 Analytical Implications

199

8.3.3 Implications for Clinical Practice


203

8.3.4 Implications for Future Research

204

8.4 Conclusion

viii

206


Appendices
Appendix 1: Systematic Literature Review Summary Sheets

209

Appendix 2: Modified Rotterdam Symptom Checklist

213

Appendix 3: Pattern Matrix, Principal Axis Factoring at T1

217

Appendix 4: Structure Matrix, Principal Axis Factoring, at T2

219


Appendix 5: Structure Matrix, Principal Axis Factoring, at T3

221

References

225

ix


LIST OF TABLES
Table 4.1 Symptom clusters identified by common factor analysis methods

70

Table 4.2 Symptom clusters identified by principal components analysis

73

Table 4.3 Symptom clusters identified by hierarchical cluster analysis

75

Table 5.1 Distribution of missing data at each time point

101

Table 6.1 Patients’ clinical characteristics


134

Table 6.2 Symptom Clusters identified by Principal Axis Factoring

135

Table 6.3 Symptom Clusters by alternative extraction methods

136

Table 7.1 Patients’ clinical characteristics

156

Table 7.2 Prevalence of patients’ symptom distress over time

158

Table 7.3 Vasomotor-symptoms and Oral-discomforts symptom clusters

160

identified within 1, at 6, and 12 months after commencing
chemotherapy
Table 7.4 Gastrointestinal-related symptom clusters identified within 1,

161

at 6, and 12 months after commencing chemotherapy
Table 7.5 Musculoskeletal discomforts/lethargy symptom clusters identified


162

within 1, at 6, and 12 months after commencing chemotherapy
Table 7.6 Symptom clusters identified ONLY at 12 months after commencing 163
chemotherapy
Table A3.1 Factors identified within one month of commencing chemotherapy 218
Table A4.1 Factors identified 6 months after commencing chemotherapy

220

Table A5.1 Factors identified 12 months after commencing chemotherapy

222

x


LIST OF FIGURES

Figure 5.1 Conceptual Framework of the Study Adapted from the Theory of
Unpleasant Symptoms

89

Figure 5.2 Flow of Participants in Parent Study

93

xi



QUEENSLAND UNIVERSITY OF TECHNOLOGY
DECLARATION OF AUTHORSHIP

The work contained in this thesis has not been previously submitted to meet
requirements for an award at this or any other higher education institution. To the
best of my knowledge and belief, the thesis contains no material previously
published or written by another person except where due reference is made.

Signature: …………………………………….
Date ………………………………………….
Helen Skerman DipTeach BSc GradDipCompEd MSocSc (App)

Copyright Statement
© 2010 Helen Skerman
This thesis is copyright. No part of this thesis may be reproduced, stored in a
retrieval system, or transmitted, in any form, or by electronic, mechanical,
photocopying, recording or otherwise, without prior permission of the author. All
rights reserved.

xii


GLOSSARY OF ACRONYMS AND TERMS

AFA

Alpha Factor Analysis


CARES-SF

Cancer Rehabilitation Evaluation System, Short Form

CFA/FA

Common Factor Analysis

ECOG

European Cooperative Oncology Group

EFA

Exploratory Factor Analysis

GEE

Generalized Estimating Equations

GLS

Generalized Least Squares

HCA

Hierarchical Cluster Analysis

IFA


Image Factor Analysis

KMO

Kaiser-Meyer-Olkin

LSS

Life Satisfaction Scale

MAP

Minimum Average Partial

MAR

Missing At Random

MCAR

Missing Completely At Random

MDASI

M.D. Anderson Symptom Inventory

MNAR

Missing Not At Random


ML, MLFA

Maximum Likelihood Factor Analysis

MSAS

Memorial Symptom Assessment Scale

NIH

National Institutes of Health

PA

Parallel Analysis

PAF

Principal Axis Factoring

PCA

Principal Components Analysis

QUT

Queensland University of Technology
xiii



RMSEA

Root Mean Square Error of Approximation

RSCL

Rotterdam Symptom Checklist

SDS

Symptom Distress Scale

SEM

Structural Equation Modeling

SMC

Squared Multiple Correlation

SSQT

Social Support Questionnaire for Transactions

TOUS

Theory of Unpleasant Symptoms

UCSF


University of California, San Francisco

ULS

Unweighted Least Squares

Clinically relevant

Important to the patient’s experience and has some practical
consequence for symptom management and patient outcomes

Communality

Common variance of a variable, shared with other variables

Multivariable

Relationships between independent and dependent variables

Multivariate

Relationships among multiple dependent variables

Stability

Consistent/replicated for a group at a point in time or over
time, or in individuals over time, and for different patient
populations (subgroups)

xiv



PUBLICATIONS FROM THE RESEARCH PROGRAM

Skerman, H. M., Yates, P. M., & Battistutta, D. (2009). Multivariate methods to
identify cancer-related symptom clusters. Research in Nursing & Health,
32(3), 345-360.
(This manuscript is presented in Chapter 4).

Skerman, H., Yates, P., & Battistutta, D. (2007). A path analysis modeling the
symptom experience of cancer patients commencing adjuvant treatment in
ambulatory clinics. Oncol Nurs Forum, 34(1), 214.

Skerman, H. M., Yates, P. M., & Battistutta, D. (2009). Identification of CancerRelated Symptom Clusters: An Empirical Comparison of Common Factor
Analysis Methods. J Pain Symptom Manage (Under review).
(The submitted manuscript is presented in Chapter 6).

Skerman, H. M., Yates, P. M., & Battistutta, D. (2009). Stability of Cancer-Related
Symptom Clusters, 6 and 12 Months after Commencing Chemotherapy.
Support Care Cancer (Under review).
(The submitted manuscript is presented in Chapter 7).

xv



FUNDING ATTRACTED BY RESEARCH PROGRAM
Queensland University of Technology Postgraduate Research Award
(APRA) Funding received from 2005 to 2008.
School of Nursing

Funding received for a presentation at the Palliative Care Australia conference in
2007.
Grant in Aid
Funding received for an international presentation at the Oncology Nursing Society
for Cancer Nursing Research, in 2007.
Institute of Health and Biomedical Innovation
Funding received for a presentation at the ANZ Society of Palliative Medicine 2008
conference.

xvii


ACKNOWLEDGMENTS
I wish to express my sincere thanks to my Chief Supervisor, Professor Patsy Yates,
for her belief in this topic, and her support, encouragement and patience during the
course of this thesis. I wish to acknowledge my Associate Supervisor, Associate
Professor Diana Battistutta, for her willingness to discuss all aspects of this project,
and in particular, her mentoring on the statistical perspectives developed.

In the early years of this project, Dr Cameron Hurst provided informative statistical
discussions on factor and cluster analysis. I would like to express my thanks to Dr
Anne Walsh for her friendship and support in the early years. I would also like to
acknowledge the continued support and encouragement of my research colleagues in
the School of Nursing and IHBI, particularly the Palliative Care Group.

I would like to acknowledge the financial support for this research, received from the
Queensland University of Technology, the School of Nursing, and the Institute of
Health and Biomedical Innovation.

Finally, I wish to acknowledge the endless support of my husband, Rob, in allowing

me to achieve this goal. My family has been a tower of strength and distraction,
helping to maintain some reality. I thank my friends for their support and patience
during this time.

xviii


CHAPTER 1
INTRODUCTION

1.1

The Research Problem
An individual’s symptom experience and ability to function in everyday life

are increasingly recognised as important health outcomes for individuals with a
chronic disease such as cancer (Lipscomb, Gotay, & Snyder, 2005; Sullivan, 2003).
Patients with cancer experience multiple symptoms, so the investigation of
individual symptoms to understand their symptom experience provides a limited
perspective. Dodd, Janson et al. (2001) challenged oncology researchers to consider
symptom clusters (i.e., a grouping of related, concurrent symptoms), in order to
broaden the current perspectives on possible mechanisms underlying cancer related
symptoms, and to suggest strategies that may advance cancer symptom management.
Conceptually, Dodd et al. proposed that if a key symptom in a group of commonly
occurring symptoms could be treated, the associated symptoms may also be relieved.
This assumes that either the key symptom is etiologically related or subject to
common treatment approaches with other symptoms in the cluster. The potential
benefits of an approach which considers multiple symptoms in symptom
management may therefore be: reduced or optimized polypharmacy, better outcomes
for patients, increased clinician satisfaction, and reduced health care costs.


At the commencement of this project in 2005, symptom cluster research was
in its early stages. There were many gaps in our knowledge and understanding of
symptom clusters. Few studies identified symptom clusters empirically, and with no

1


specific guidance about analytical methods for determining valid and reliable
symptom clusters, there was a risk that research in this field would not realise its
potential. Hence, the overall aim of this study was to investigate analytical
approaches that are conceptually and contextually relevant for symptom cluster
identification. It was intended that the outcomes of this investigation would provide
guidance for the most effective analytic methods for identifying valid and reliable
symptom clusters, to support the advancement of symptom management in
oncology.

1.2

The Burden of Cancer
In the developed world, despite a reduction in the incidence of some cancers,

the overall incidence of cancer cases continues to rise each year. This trend is
expected to continue, due in part to the ageing population (WHO, 2003).

In

Australia, cancer is a leading cause of death, and based on 2005 data, an estimated
111,000 new cases will be diagnosed in 2009. The current expectation is that 1 in 2
men, and 1 in 3 women, will be diagnosed with cancer before the age of 85 years

(Cancer Council Australia, 2009). In 2005, the most common cancers, excluding
non-melanoma skin cancer, were prostate (16, 349 cases), colorectal (13, 076), breast
(12,265), skin melanomas (10,684), and lung cancer (9,182). Apart from sex-specific
cancers, almost all cancers occur at higher rates in men than women. The cost of
health care services for cancer in Australia exceeds $3.8 billion.

Nevertheless, the outlook for cancer patients has improved, due to earlier
detection through screening, improved prevention strategies, technologies, and
methods of treatment (Aziz & Rowland, 2003). In Australia, more than half the
newly diagnosed cases will be successfully treated and over 60% of cancer patients
2


will survive more than 5 years after diagnosis (Cancer Council Australia, 2009).
Hence, with the increasing number of cancer survivors, many of whom are expected
to live for longer periods, further research is necessary, to understand the experience
of cancer patients in the short and long term, and to adequately address their needs
(Haylock, 2006).

1.3

Rationale and Significance of the Research
A significant aspect of the cancer experience is the experience of disease- and

treatment-related symptoms. Research has typically conceptualized symptom
experiences in terms of two attributes: (a) symptom occurrence (frequency and
duration), and (b) symptom distress (Armstrong, 2003; Goodell & Nail, 2005; Lenz,
Pugh, Milligan, Gift, & Suppe, 1997), where frequency is the number of times the
symptom is experienced in a given time interval, and duration indicates how long
that experience lasts. The occurrence of symptoms often causes distress, which may

be physical or mental upset, anguish, or suffering (Rhodes & McDaniel, 1999).
Importantly, an individual’s perceived degree of symptom distress has been
identified as the main stimulus for individuals to act to relieve symptoms (Fu,
Anderson, McDaniel, & Armer, 2002; Sweed, Schiech, Barsevick, Babb, &
Goldberg, 2002), not simply the occurrence of the symptom. For example, patients
distressed by severe pain, or fatigue, are likely to be motivated to relieve, decrease,
or prevent distress, whereas patients experiencing minimal or no distress may not
bother (Fu, LeMone, & McDaniel, 2004). A person’s symptom experience has thus
been described as subjective, reflecting changes in an individual’s biological,
psychological, cognitive, and social functioning (Dodd, Janson et al., 2001). For
instance, the sensation of fatigue reflects changes in an individual’s ability to
perform their usual roles and activities (Curt et al., 2000). While such
3


conceptualizations emphasize the impact of individual symptoms for cancer patients,
they also suggest that when multiple symptoms occur together, an individual’s
distress is likely to be magnified. Thus, an investigation of cancer related symptoms
should reflect the patient’s perspective of their experience, and incorporate the
potential of symptom clusters.

Despite advances in early diagnosis and insights into the causes and
treatments for cancer, strategies to manage an individual’s experience of cancerrelated symptoms and the side-effects of cancer have not progressed at the same
pace. In 2002, the National Institutes of Health (NIH) Symptom Management
Conference Panel resolved multi-symptom research provided an opportunity to
improve symptom management. The NIH research directive was to focus initially,
on the most common side-effects of cancer and treatment, namely, pain, depression
and fatigue (National Institutes of Health State-of-the-Science Panel, 2003).
Subsequently, a number of correlation based studies investigated the relationships
between cancer-related symptoms, their predictors and impact on daily living (Dodd,

Miaskowski, & Paul, 2001; Francoeur, 2005). A potential benefit of a symptom
cluster approach was that treating one symptom in a cluster may, directly or
indirectly, resolve related symptoms, based on the assumption of a clinical,
etiological relationship (e.g., treatment based) versus a statistical, least squares
relationship (e.g., regression).

The investigation of symptom clusters is complex, given the variety of cancer
diagnoses, treatments, patient characteristics and methods to collect and analyse
data. Conceptual frameworks to understand the symptom experience and determine

4


strategies to manage symptoms have been proposed (Armstrong, 2003; Dodd, Janson
et al., 2001; Lenz et al., 1997) and may be modified as new knowledge unfolds.
Furthermore, there is no agreed definition of the composition and fundamental
characteristics of symptom clusters; there is limited understanding of how and why
particular symptoms occur simultaneously; and there is a paucity of research to
investigate multiple symptoms, as the disease progresses and treatments proceed. To
ensure symptom cluster research is of high scientific quality and adds to the
knowledge in this field, a range of issues require further consideration, including the
study design (e.g., heterogeneous/homogeneous samples), symptom measurements
(e.g., scales, domains), the most useful timing of the symptom assessment, clinically,
and appropriate analytical methods for symptom cluster identification.

Currently, there is limited knowledge of the methodological decisions for
symptom cluster research that may have important implications for research
findings. For example, a variety of symptom assessment tools exist, enabling the
assessment of variation in individual symptoms, common to specific diagnoses and
treatments (Beck, 2004; Miaskowski, Dodd, & Lee, 2004). As a consequence,

different symptom clusters are likely, although a core set of common symptoms may
form a cluster, consistently.

There are two approaches to symptom cluster identification: the clinical
approach and the empirical approach. A clinically identified symptom cluster
comprises symptoms that are observed to occur together, and may be associated.
Frequently co-occurring and distressing symptoms, such as pain, depression, fatigue,
and insomnia have been targeted (Barsevick, 2007a, 2007b). However, in this pre-

5