Professional Documents
Culture Documents
Full Ebook of Handbook For Clinical Research Design Statistics and Implementation 1St Edition Flora Hammond Online PDF All Chapter
Full Ebook of Handbook For Clinical Research Design Statistics and Implementation 1St Edition Flora Hammond Online PDF All Chapter
https://ebookmeta.com/product/handbook-of-hormones-comparative-
endocrinology-for-basic-and-clinical-research-2nd-edition-
hironori-ando/
https://ebookmeta.com/product/research-methods-the-key-concepts-
michael-hammond/
https://ebookmeta.com/product/research-handbook-on-
implementation-of-human-rights-in-practice-1st-edition-rachel-
murray/
https://ebookmeta.com/product/virtual-and-remote-control-tower-
research-design-development-validation-and-implementation-2nd-
edition-norbert-furstenau/
Activities for Teaching Statistics and Research Methods
1st Edition Jeffrey R. Stowell
https://ebookmeta.com/product/activities-for-teaching-statistics-
and-research-methods-1st-edition-jeffrey-r-stowell/
https://ebookmeta.com/product/using-ibm-spss-statistics-for-
research-methods-and-social-science-statistics-7th-edition-
william-e-wagner/
https://ebookmeta.com/product/bathrooms-and-sanitation-
principles-design-implementation-1st-edition-sibylle-kramer/
https://ebookmeta.com/product/research-methods-the-key-
concepts-2nd-edition-michael-hammond-jerry-wellington/
https://ebookmeta.com/product/fundamentals-of-statistics-for-
aviation-research-1st-edition-michael-a-gallo/
Handbook for
Recommended
Shelving Category:
Medical
Editors
Flora M. Hammond, MD
Chair, Department of Physical Medicine and Rehabilitation
Covalt Professor of Physical Medicine and Rehabilitation
Indiana University School of Medicine
Chief of Medical Affairs
Rehabilitation Hospital of Indiana
Indianapolis, Indiana
Ralph M. Buschbacher, MD
Clinical Professor, Department of Physical Medicine and Rehabilitation
Indiana University School of Medicine
Indianapolis, Indiana
New York
ISBN: 978-1-9362-8754-3
e-book ISBN: 978-1-6170-5099-2
© 2015 Demos Medical Publishing, LLC. All rights reserved. This book is protected by copyright. No part of it may be
reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying,
recording, or otherwise, without the prior written permission of the publisher.
Medicine is an ever-changing science. Research and clinical experience are continually expanding our knowledge, in par-
ticular our understanding of proper treatment and drug therapy. The authors, editors, and publisher have made every effort
to ensure that all information in this book is in accordance with the state of knowledge at the time of production of the book.
Nevertheless, the authors, editors, and publisher are not responsible for errors or omissions or for any consequences from
application of the information in this book and make no warranty, expressed or implied, with respect to the contents of the
publication. Every reader should examine carefully the package inserts accompanying each drug and should carefully check
whether the dosage schedules mentioned therein or the contraindications stated by the manufacturer differ from the state-
ments made in this book. Such examination is particularly important with drugs that are either rarely used or have been newly
released on the market.
Special discounts on bulk quantities of Demos Medical Publishing books are available to corporations, professional asso-
ciations, pharmaceutical companies, health care organizations, and other qualifying groups. For details, please contact:
Special Sales Department
Demos Medical Publishing, LLC
11 West 42nd Street, 15th Floor
New York, NY 10036
Phone: 800-532-8663 or 212-683-0072
Fax: 212-941-7842
E-mail: specialsales@demosmedical.com
14 15 16 17 / 5 4 3 2 1
Contributors ix
Preface xiii
Share Handbook for Clinical Research: Design, 10. Choice of Control Groups in Treatment
Statistics, and Implementation Studies 37
Sureyya S. Dikmen, PhD
PART I: DESIGN
11. Randomization 40
Emilia Bagiella, PhD
1. Development and Testing of Treatments 3
Eric Lenze, MD and John Whyte, MD, PhD 12. Special Issues in Randomized Controlled
Trials 43
2. Qualitative Research 9
Grace Handler and Michael L. Boninger, MD
Elena Gillespie, PhD
13. Secondary Data Analysis 46
3. Single-Case Experimental Designs 13
Kenneth J. Ottenbacher, PhD, OTR,
Robyn L. Tate, MPsychol, PhD and Michael
James E. Graham, PhD, DC, and
Perdices, MA (ClinNeuropsychology), PhD
Amol Karmarkar, PhD, MPH, OTR
4. Studies of Associations 18
14. Scoping Study 50
Michael Schönberger, PhD, Dipl.-Psych.
Xinsheng Cai, PhD
5. Observational Studies: Retrospective Versus
15. Systematic Reviews 53
Prospective 23
Naomi Lynn Gerber, MD and
Brian Keogh, Jr, MD, and
Xinsheng Cai, PhD
Katherine W. Stenson, MD
16. Meta-Analysis 57
6. Historical Controls 27
Xinsheng Cai, PhD
Whitney Pratt, MD, PhD and
Katherine W. Stenson, MD 17. Recommendations for Reporting Research
Studies 62
7. Subject as Own Control 30
Ronald T. Seel, PhD
Lisa A. Lombard, MD
18. Developing and Evaluating Systematic Reviews
8. Longitudinal Cohort Versus
and Practice Guidelines 68
Cross-Sectional Cohort Studies 32
Ronald T. Seel, PhD
Zachary Timothy Glynn Siegel, MD and
Katherine W. Stenson, MD
9. Survey Research 34
Amanda Leigh Harrington, MD
52. Planning Grants 208 69. Participant Recruitment and Enrollment 266
Eric Lenze, MD Flora M. Hammond, MD
53. Developing the Idea With Stakeholder 70. Participant Retention 271
Input 211 Flora M. Hammond, MD
John D. Corrigan, PhD
71. Data Collection 276
54. Research Questions, Hypotheses, Aims, and Elena Gillespie, PhD and Flora M. Hammond, MD
Abstract 214
72. Case Report Forms 280
Flora M. Hammond, MD
Marybeth Whitney, BSN, CRA and Flora M.
55. Reviewing the Literature 218 Hammond, MD
Dawn Neumann, PhD and
73. Database Development 283
Edward Garay, MD, PhD
Emilia Bagiella, PhD
56 Background and Significance 222
74. Data Dictionary 286
Flora M. Hammond, MD
Emilia Bagiella, PhD
57. Preliminary Studies and Experience 224
75. Data Management 288
Flora M. Hammond, MD
Emilia Bagiella, PhD
58. Methods and Design 226
76. Plan of Operation 291
Kathleen R. Bell, MD
Angelle M. Sander, PhD
59. Types of Measures 231
77. Evaluation 295
James F. Malec, PhD
Angelle M. Sander, PhD
60. Letters of Support 234
78. Regulatory Binder and Essential
James F. Malec, PhD
Documents 298
61. Budget and Budget Justification 236 Marybeth Whitney, BSN, CRA and Flora M.
LaMoria Patterson, BS and Rob Dimmitt, MBA, BS Hammond, MD
62. Preaward Management 241 79. Adverse Events 302
LaMoria Patterson, BS and Rob Dimmitt, MBA, BS Marybeth Whitney, BSN, CRA and Flora M.
Hammond, MD
Conducting the Research
80. Protocol Deviations and Violations 307
63. Post-Award Management 245 Marybeth Whitney, BSN, CRA and Flora M.
Rob Dimmitt, MBA, BS and LaMoria Patterson, BS Hammond, MD
64. Good Clinical Practices 249 81. Data and Safety Monitoring 310
Shawn Axe, CIP and John R. Baumann, PhD Emilia Bagiella, PhD
65. Research Misconduct 252 82. Multicenter Trials 315
John R. Baumann, PhD, Shelley Bizila, MS, CIP, Flora M. Hammond, MD
CCEP, and Rebecca Hopson, BA
83. Site Monitoring and Oversight 318
66. Study Protocol 256 Marybeth Whitney, BSN, CRA and Flora M.
Flora M. Hammond, MD Hammond, MD
67. Manual of Procedures 260
Index 323
Flora M. Hammond, MD
68. Treatment Manuals 263
Kathleen R. Bell, MD
We started working on this text at the encouragement And so, the Handbook for Clinical Research:
of Dr. Ralph Buschbacher to develop a quick refer- Design, Statistics, and Implementation took shape.
ence guide to research primarily for rehabilitation Part I covers the basics of research design: the variety
clinicians. Initially, we imagined a relatively thin of ways in which studies can be organized to address
volume that focused on the key elements of rehabili- questions of association or causation; the appropri-
tation research. However, as we began to formalize ate sequencing of studies in a particular area to move
the outline, the scope of the book rapidly grew to most efficiently from demonstration of concept to a
the 80-plus chapters in the present volume. As we definitive and rigorous trial; methods and appropriate
recruited authors for the various chapters and began rigor of control conditions; retrospective and prospec-
initial editing, we quickly realized that the content tive trials; qualitative and quantitative analyses; and
of these chapters was applicable not just to rehabil- methods for summarizing, evaluating, and reporting
itation, but to other disciplines involved in clinical clinical research in a particular area. Part II, statistics,
research, eg, other medical specialties, psychology, begins with a discussion of selecting the correct statis-
nursing; and the scope of the book expanded further. tical approach and when to consult a statistician. Part
The potential impact of the book also became more II then reviews the varieties of data types; descriptive
apparent to us. This book gave us a chance to out- and inferential statistics; methods for demonstrat-
line not only the key elements of clinical research ing associations, hypothesis testing, and prediction;
but also to put down in writing the “trade secrets,” specialized methods, such as survival modeling; and
shortcuts, practical insights, and helpful hints to con- specialized methods for epidemiological studies and
ducting clinical research that we and the other authors measure construction. Part III, implementation, begins
have learned over the years by word of mouth from with a number of chapters on developing successful
mentor to mentee and colleague to colleague. Our grant applications from planning and seeking consumer
excitement grew about developing a text that might input to developing specific sections of research grant
be a quick reference guide for clinical researchers in proposals, the project budget, and ancillary materials.
any discipline—a book that could assist them in their The final chapters of this section cover the nuts and
daily work planning and conducting research studies, bolts of the timely and successful completion of the
journal and grant reviews, and mentoring. We began research project; developing procedural manuals and
to imagine a book that might serve as a supplemen- case report forms; collecting, managing, and securing
tary text to courses on research design and analysis data; operational structure and ongoing monitoring
as well as grant writing courses and workshops, or and evaluation of the project; and ethical and regula-
that might be a good primary text on research for tory concerns in research with human subjects.
any clinical training program with a clinical research Many of the authors (including ourselves) at
component. first found the succinct, focused format of the book to
be challenging. Most of us would have found it easier Our hope in assembling this cogent overview of
to write the traditional 30-odd-page chapter replete the broad realm of clinical research is to provide a text
with extensive referencing and interesting (at least to that both established and apprentice clinical research-
us) intellectual alleyways. However, all of the authors ers will find to be a useful guide and reference.
rose to the challenge and boiled down their insights in
each topic area to the key elements, practical consid- Flora M. Hammond, MD
erations, potential pitfalls, and helpful hints that this James F. Malec, PhD
text was designed to communicate. Chapters include Todd G. Nick, PhD
a few references for those seeking a more in-depth Ralph M. Buschbacher, MD
treatment of the topic area.
ries about how changes in a proximal clinical target The minimal effect that has clinical relevance in the
(referred to as the target of treatment) will influence management of patients.
distal clinical targets (referred to as treatment aims).
■ Target of treatment: The functional variable, that is,
ments in all patients, as when a patient with leg ● Step 1: Manualization or protocolization of the
weakness and ataxia fails to improve in ambula- intervention;
tion (an aim) despite strengthening exercises that ● Step 2: Conducting iterative case series, often
improve the weakness (the target). In such cases, called an open-label trial, to test and modify both
one may perform a confirmatory trial demonstrat- the intervention and all aspects (such as inclusion
ing that the treatment reliably improves weak- criteria, outcome measures) of the research;
ness (the target), but one must assess its practical ● Step 3: Conducting a pilot RCT to demonstrate
impact on ambulation in a separate research with feasibility of doing an RCT.
a very different design (see below). ■ Treatment development is required prior to treat-
● Later tests may explore whether those that the underlying treatment theory is false.
treatment-induced changes in the target are However, a negative result may also occur when
reliable across a heterogeneous population the treatment theory is correct but one of the
of patients, and when delivered by hetero- following failures occurred:
geneous clinicians (effectiveness in treating ● Treatment is given in an inadequate dose.
the target). ● Treatment dose/duration is adequate to be
■ Testing of the practical clinical impact of potent, but is poorly implemented by study
treatment clinicians (ie, poor treatment fidelity).
● When testing for practical impact on a ● Treatment implementation by clinicians is
distal clinical aim (eg, participation in social adequate, but patient adherence too low.
or role activities), one must have an enable- ● Patient adherence to treatment is ade-
ment/disablement model in mind of how the quate, but missing data rate is too high (due
treatment target relates to the distal aim, and to patient dropout/refusal or other reasons).
what other strengths and limitations may ● Recruitment is poor, leading to inadequate
also affect the aim. sample size.
regarding what is being done well and what ● Good recruitment and retention
needs to change in order to achieve good treat- ● Even though recruitment and retention
ment fidelity. for treatment may have been adequate in
● Measure fidelity. the case series, introduction of blinding or
■ It is often advisable to verify that the control randomization may lead to a drop in recruit-
treatment is lacking in the treatment’s active ment or retention that must be anticipated
ingredients, even prior to a pilot RCT stage. and corrected.
■ Fidelity assessment should focus on the ● Comparator treatment is satisfactory (eg,
essential ingredients of the theory underlying no excessive dropout, includes no active
the new intervention. treatment elements)
■ In research treatment development and test- ● Outcome and process measures are practi-
ing, the following treatment fidelity steps should cal, reliable, and responsive
pilot data demonstrating feasibility and a possible important part of treatment development but also
treatment effect may enhance the chances of funding. the hardest to get buy-in from grant or journal
astrology would not have changed over that same of data without assessing for meaning to examine other
time period. data within the set for the significance of those data.
■ Phenomenology: Observing and analyzing data, the ● For example, correlation of two variables may
research participant and their environs for the subjec- not equal causation, although it might. Each vari-
tive experience of the participant. able is isolated and analyzed before such a conclu-
■ Hermeneutics: Reflexively analyzing data situ- sion is reached (if it can be—there may be other
ated within investigator’s biases and epistemological variables present influencing the outcome).
values. ■ Ethnography: Field studies or case reports exploring
that the family members are not intellectually determined methods and rituals for producing and
challenged in their ability to understand what is maintaining social order.
being explained to them, but are not familiar with ● Example: The examination of the decision-
the medical terminology used and the course of making process utilized by American families
recovery. The practitioner also understands that precipitant to installing an elderly family member
the family is undergoing a period of emotional into a nursing home. These would include factors,
stress, so that extra time and care must be taken such as the family dynamic, financial outlook, and
to explain possible outcomes, perhaps multiple health needs of the family member in question.
times, in a manner, that is sensitive to their social ■ Grounded theory: Post hoc exploration of the data to
and cultural milieu. The investigator has followed identify observations and trends not identified a priori
a hermeneutic process based on their experience to the study. The analysis of data is led or grounded by
and observations. where the data leads; the investigators return repeatedly
to assess the data for the next step. The hypothesis is ■ Mixed methodology: A combination of both quan-
then formulated based on where the information led titative and qualitative methods in the conduct of a
the investigators. clinical research study.
■ Field study: A set of data—usually interviews— ● Example: A study of the frequency, locations,
gathered from participants in the area of interest. and size of injection doses of Botulinum toxin A
● Involves narratology, that is, storytelling for the amelioration of spasticity in patients with
● Can be structured, semi-structured, or nonstruc- stroke, and interviewing a subset of these subjects
tured interviews in a structured interview about the quality of their
■ Participatory action research (PAR): Research in mobility over a 3-month period.
which the participants not only provide data but are ■ Content analysis: Themes in the transcribed inter-
also asked about their experience(s) with the topic and views are identified and coded, then further broken
are actively enlisted to create the study’s hypothesis. down into more minute similarities and differences,
● The investigators concern themselves with and organized into categories.
the methodology and implementation. PAR may
involve specifically designing a study to increase
the understanding of the participants regarding a INTRODUCTION
particular social or medical issue and involve them
in social change related to that issue. ■ Qualitative methodology is the study of the individ-
● Example: Several groups of the relatives of ual subject’s perceived experience.
■ Assessing the “mean” or average experience of
patients with traumatic brain injury (TBI) are
enlisted to discuss how best to obtain access to many is not its goal, but rather the purpose of qualita-
federal monies to aid the patients to return to work tive research is the identification of a few variables
more effectively, with guidance from the investiga- that impact a small sample, that would probably not
tors. The study then continues to monitor the group come to light in a quantitative study.
● Rigor—clarity and the economical application
and determines how many of them become activ-
ists as a result of their new level of understanding of logic—are fervently applied in its utilization.
● Central criteria: Can the answer to the question add
of federal processes and TBI.
■ Case study: A write-up of an individual case to
knowledge to the field of inquiry about this topic?
exemplify either its singular qualities different from
the mean of the illness or syndrome, or to emphasize
typifications across the population. Can be descrip- IMPLICATIONS
tive, analytic, or collective (or more than one).
■ Critical social research: Analyzing the means of ■ Can deliver deep insights into treatment effects and
communication within interviews to determine how clinical impact not measurable by quantitative surveys
subjects derive meaning. or other measures.
● Example: A field study of a group of patients to ■ Not appropriate for use for quantitative measure-
assess how their sense of identity has changed after ment, that is, provides only nominal level data.
a stroke.
■ Ethical research: Analysis of the foundations of an
■ Edmund Husserl, Martin Heidegger, and Hans- ● Example: A qualitative investigator does not
George Gadamer added to the area of inquiry by believe that socioeconomic class impacts treat-
assisting to define ontology—the integration of human ment compliance and therefore unconsciously only
consciousness and its perceptions within the real recruits subjects that own their own car for trans-
world. Gadamer in particular described the “fusion of port to the study site.
horizons” of understanding through language (herme- ■ The investigator is a part of the study. Meticulous
neutic approach). field notes and observations are kept on the investiga-
■ Foucault observed that epistemology—what we tor’s experience in relation to participant enrollment
value as knowledge and define as “data”—is driven and data collection.
by the predominant group in power, not by “truth.” ● Example: In a group round table setting, the
■ Max van Manen, Clark Moustakas elaborate on investigator, after listening to the audio recording
writing and storytelling as a means of exploration. multiple times, finds that he or she subconsciously
interrupted subjects who were reflecting an oppos-
ing perspective to the investigator’s hypothesis,
STRENGTHS thus effectively discounting that subject’s data and
impacting the flow of the discussion.
■ Requires Institutional Review Board (IRB) approval
■ Can consider questions or clinical observations that
are not captured by quantitative measures and that and informed consent document. Using patient names,
impact clinical outcomes. demographics, and other significant identifiers is not
■ Qualitative measurement is a highly useful addition permissible in the final manuscript.
■ Recruitment similar to RCTs; advertisements, word-
to a clinical study in which the primary end point is a
quantitative measure. of-mouth, and letter campaigns.
■ Sample is generally no greater than nine
● Example: A placebo-controlled, double-blind
randomized controlled trial (RCT) for efficacy of participants.
■ Inclusion and exclusion criteria are designated
an intervention also conducts a blinded field study
of a subset of subjects from each group and discov- before study initiation. These may change—document
ers that the practitioners’ clinical manner and style the rationale for such change.
in each arm of the study were widely different, thus
impacting how the intervention’s effectiveness was PITFALLS
perceived, and consequently, the strength of the
placebo effect. ■ Results represent subjects’ personal subjective expe-
rience; care must be taken not to extrapolate data to an
unsampled group.
WEAKNESSES ● Example: A field study of the attitudes of
Hispanic men toward paraplegia after a disabling
■ Transcription is time-consuming and laborious. accident could not be directly correlated to Hispanic
■ Methodology is not widely accepted by funding women, due to cultural and gender differences.
agencies or journals. ■ Qualitative studies are not designed to find the “aver-
that is, his or her biases and background, and docu- HELPFUL HINTS
ment them before beginning. The investigator is the
“metacode,” that is, the medium by which the data are ■ Do not overenroll. Six to nine subjects will provide
filtered and interpreted. a rich continuum of information to examine.
or intervention) is systematically manipulated priate for different types of interventions and suitable
(applied and withdrawn) across the phases. for answering different types of research questions.
● The dependent variable (ie, symptom being ■ This chapter describes the three most common types
treated or the target behavior) is measured repeat- of SCEDs and their strengths, weaknesses, and chal-
edly and frequently throughout each phase. lenges compared to the traditional group-based design.
■ Other terms used for SCEDs:
● Medical sciences use the term “n-of-1 trial,”
specifically referring to the randomized, multiple IMPLICATIONS
crossover design in a single patient.
● Behavioral sciences also use terms such as “sin- ■ In 2011, the Oxford Centre for Evidence-Based
gle-case design,” “single-participant design,” and Medicine (http://www.cebm.net/index.aspx?o=5653)
“single system design” (the older literature also introduced the Levels of Evidence 2 table in which
uses the term “single-subject design”). the n-of-1 trial is classified as Level 1 evidence for
treatment decision purposes in the individual patient, ■ Are used when there are carry-over effects
comparable to the systematic review of multiple and treatment cannot be meaningfully with-
RCTs. drawn (eg, acquisition of skills, therapeutic
■ SCEDs can serve as a good prelude to a Phase I instructions).
clinical trial. ● Alternating treatment designs (see Figure 3.3):
■ SCEDs may be the only means of empirically estab- ■ Two or more treatments are compared con-
lishing the effectiveness of a treatment in an individ- currently in an alternating manner, rather than
ual patient. sequentially as in the withdrawal design.
■ SCEDs provide an opportunity to use evidence- ■ A baseline condition is not necessary.
4
Target Behavior
1
A Baseline B Treatment A Baseline B Treatment
0
1 3 5 7 1 3 5 7 9 11 13 15 17 19 21 23
Sessions
Target Behavior 1 20
15
10
5
A Baseline B Treatment
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
Sessions
20
Target Behavior 2
15
10
5
A Baseline B Treatment
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
Sessions
20
Target Behavior 3
15
10
5
A Baseline B Treatment
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
Sessions
FIGURE 3.2 Hypothetical data for a multiple baseline design across behaviors.
20
18
16
Target Behavior
14 Treatment 1
12
10 Treatment 2
8
6
4
2
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Session
WEAKNESSES ■ Identify the design that is best suited for the inter-
vention and the clinical research situation (see
■ Poorly designed SCEDs are subject to risk of bias Background).
and threats to internal validity (see Strategies that, if ■ Consider the use of randomization. In SCEDs, ran-
● Generalization is not an issue in the clinical set- pist, as well as blind to the phase of treatment.
ting when the question is whether the treatment ■ When possible, blind the participant and the person
was effective for the specific patient being treated. conducting the intervention to the phase of treatment
■ It is important to establish baseline stability, but (eg, baseline vs treatment).
there are no specific or agreed-upon criteria for its ■ In pharmacological treatments, plan for run-in and
researcher and participant because generally, assess- being used as a research tool.
ment and therapy sessions are frequent and can be ■ Involve the patient and family in selecting the symp-
CONTROLS
After the Study Starts
■ In an SCED, the participant serves as his or her own
control, by having the investigator manipulate the ■ Establish interrater reliability of the measure for
phase (Condition 1, Condition 2). each phase (recommended for ≥20% of observations
per phase, with recommended result of ≥80% agree-
ment/kappa ≥0.6; see Kratochwill et al, 2010).
STRATEGIES ■ Ensure sufficient sampling of behavior occurs in
Barlow DH, Nock MK, Hersen M. Single Case Experimental PsycBITE (http://www.psycbite.com) has archived more
Designs. Strategies for Studying Behaviour Change. 3rd than one thousand designs using single-case methodol-
ed. Boston, MA: Pearson; 2009. ogy, which are also rated for method quality
Kratochwill TR, Hitchcock J, Horner RH, et al. Single-case Reporting guidelines, in the CONSORT tradition, for both
designs technical documentation. 2010. Available at http:// single-case experimental designs (see Tate et al., 2012) and
ies.ed.gov/ncee/wwc/pdf/wwc_scd.pdf. n-of-1 trials (see Vohra et al.) are currently being prepared.
Studies of Associations
the relationship between two variables; for example, eral clusters on the basis of their characteristics as
amount of medication administered plotted against measured by a set of variables
subjective well-being
■ Repeated measurement studies: Studies in which
power of several variables is compared in a repeated ■ Studies of association are concerned with relation-
measurement study. Provides evidence regarding the ships between measured quantities.
direction of causality between two or more variables ● Such associations can have different forms (linear
■ Path analyses: A statistical technique in which a or nonlinear) and they can be causal or noncausal.
model of interrelationships between variables that ● An example of an association that involves a
have been defined by the investigator is tested using causal relationship is the association between
empirical data patients’ engagement in rehabilitation (cause) and
■ Structural equation modeling: A form of path analy- their outcome (effect), but even this simple exam-
sis that contains latent variables (ie, factors underlying ple demonstrates the complexity of cause–effect
observed variables) relationships in clinical research: Patients’ experi-
■ Multilevel modeling: A flexible form of multiple ence of improvement might well have a positive
regression that is appropriate when the data structure effect on their engagement in rehabilitation. Often,
is hierarchical (such as patients measured in several complex causal models are required to describe
units in several hospitals; repeated measurement of clinical reality appropriately.
participants) ■ Associations can exist between variables with all
■ Multicollinearity: Strong correlations between pre- kinds of scale properties—nominal, ordinal, interval,
dictors in a regression analysis; makes it difficult to and ratio scales.
determine each individual predictor’s contribution to ● The appropriate statistical procedure to deter-
the prediction of the dependent variable mine associations between two or more variables
■ Explorative factor analysis (EFA): Statistical tech- depends on
nique with which dimensions underlying a set of vari- ■ The number of dependent and independent
ables/test items are determined variables in the analysis
■ Confirmatory factor analysis (CFA): Statistical tech- ■ The variables’ scale properties
nique with which a model of dimensions underlying ■ The distribution of values on these scales
■ The assumptions regarding interrelation- ● Path analyses and structural equation modeling
ships between predictor variables and between are often used to test assumptions about complex
dependent variables relationships between variables.
■ Most importantly, the research question that ● Multilevel modeling is appropriate for the analy-
one is aiming to answer sis of multicenter study data.
■ The purpose of this chapter is to give an overview
therapist characteristics and the quality of the client– surement designs are used, studies of associations
practitioner relationship). allow detailed examination of the complex interplay
■ In clinical practice, service providers are required between factors related to outcome after clinical inter-
to secure and improve the quality of their services. ventions (eg, by using path analysis and structural
Studies of association can help to determine which equation modeling).
services contribute to a positive outcome.
WEAKNESSES
BACKGROUND
■ In the case of convenience samples, representative-
■ From a philosophy of science point of view, causal
ness may be questionable.
relationships cannot be proven. Hill defined criteria
■ No experimental variation of conditions.
that should be considered when deciding whether a
■ Evidence for causality is weaker than in randomized
causal relationship between two variables should be
controlled trials.
assumed:
● Strength of association
● Consistency across persons, places, circum-
stances, and times STRATEGIES
● Specificity of association (eg, specific clinical
services are associated with specific outcomes) Sample
● Temporality of association (what came first?)
● Biological gradient (dose–response curve) ■ Often a single group of subjects
● Plausibility ■ As is always the case in clinical research, ideally the
● Coherence of empirical findings sample should be randomly selected from the target
● Change in the dependent variable as an experi- population in order to secure representativeness of the
mental condition is changed sample.
● Analogy of empirical findings to established ■ However, studies of association often use conve-
■ Retrospective (measuring/making use of data from functional limitations following traumatic brain
the past; eg, comparative effectiveness studies) versus injury predict depression, but depression does not
prospective (real-time, single cross-sectional mea- predict functional status.
surement or repeated measurements over time) ● Test of mediator relationships. In the example
■ Single-site studies versus multicenter studies (Figure 4.2), the difference between brain-injured
individuals and healthy controls on a complex
selective attention task is mediated by group differ-
Methods (See Specific Chapters in Part II ences on a simple selective attention task, but not
for More Information About Statistical by group differences on a working memory task.
Methods) ● Test of moderator relationships (multigroup
path analysis or addition of interaction terms to the
■ If an association between two variables is to be model): Using multigroup path analysis, whether
determined: Compute a correlation, t test, analysis the same explanatory model of interrelationships
of covariance, or their nonparametric equivalents, between variables holds true in different popula-
depending on the quality of the measurement scales tions can be tested.
and distributions. ■ In order to determine whether the relationship
■ If the relationship between two variables is to be between two variables is moderated by a third
determined and the effect of a third variable is to be variable, an interaction term can be added to a
partialled out: Partial correlation. predictive model.
■ If a continuous dependent variable is to be predicted ■ If a path analysis contains factors that are not
from one or several predictors: Standard multiple observed directly but measured by a set of indica-
regression. tors (eg, depression and anxiety, measured with the
● Careful with predictors that correlate strongly Hospital Anxiety and Depression Scale): Structural
with each other (multicollinearity problem). Equation Modeling.
■ If a dichotomous/ordinal variable is to be predicted from ● The first step in Structural Equation Modeling
one or several predictors: Logistic/ordinal regression. is to compute one or several confirmatory factor
■ If one or several dependent variables are to be pre- analyses in order to test whether sets of items are
dicted from one or several observed variables, and the indicators of underlying factors.
interrelationships among the predictors and among the ● Once this is confirmed, then assumptions regard-
dependent variables are to be modeled: Path analysis, ing the interrelationships between these underlying
for example, factors (and possibly also additional, observed
● Cross-lagged panel analysis testing temporal variables) can be tested.
relationships. This technique provides evidence ● In the example in Figure 4.3, mood changes
regarding the direction of causality between two as well as cognitive and behavioral changes were
or more variables. In the example (Figure 4.1), measured with several questionnaire items.
.09
.27 –.32
–.31
SCID Depression SCID Depression
Z2
6 months .34 12 months
FIGURE 4.1 An example for a path analysis testing the temporal association between functional status (GOSE) and
depression after traumatic brain injury in the form of a cross-lagged panel design analysis.
Note: GOSE = Glasgow Outcome Scale-Extended (measure of functional status); SCID = Structured Clinical Interview for DSM-
IV-TR. Curved lines represent correlations; straight lines represent regression coefficients. Correlations and standardized regression
coefficients are displayed. Z1 and Z2 are the residuals of the dependent variables. Coefficients in the model with P < .05 are printed
boldly. The analysis was statistically controlled for covariates age, gender, level of education, and PTA duration. These covariates are
omitted from the figure.
Source: Originally published in Schönberger et al (2011a). Reprinted with the publisher’s permission.
Z1 Group Z2
1 4 – 1
LNS SSAT
– 8
CSAT
Z3
FIGURE 4.2 Example of a path analysis testing a mediator relationship: Can the difference between head-injured
and nonhead-injured individuals on a complex selective attention task be explained by (is it mediated by) their per-
formance on tests of working memory and simple selective attention?
Note: Group = Group membership (head injury vs healthy control); LNS = Letter Number Sequencing Task (working memory); SSAT
= Simple Selective Attention; CSAT = Complex Selective Attention. Standardized coefficients are shown with significant coefficients
printed in bold. Z1 through Z3 are the residual variances of the dependent variables in the model. Reprinted with the publisher’s
permission.
Source: Willmott et al (2009). Reprinted with the publisher’s permission.
Spine Limb
–.12
.10
–.39
PTA 1
Employ-
.13 ment 1 yr
.11
–.34
.18
Mood 1
Educa- –.37
tion changes
–.15
.69 –.30
Pre-inj. Behavioral 1
employ. –.29 changes
Age Sex
FIGURE 4.3 Example of a full Structural Equation Model predicting outcome 1 year after traumatic brain injury.
Note: Final, modified model with improved model fit. The observed variables that indicate the mood, cognitive, and behavioral
change factors were entered into the Structural Equation Model but are omitted from this figure. Curved lines represent correlations;
straight lines represent regression coefficients. Correlations and standardized regression coefficients are displayed. PTA, posttrau-
matic amnesia.
Source: Originally published in Schönberger et al (2011b). Reprinted with the publisher’s permission.
■ In a first step, it was confirmed that the items When Not to Use
really were indicators of mood, cognitive, and
behavioral changes, respectively. ■ If the study goal is to determine the effectiveness or
■ Once this was confirmed, the factor model efficacy of a new intervention, then randomized con-
was entered into the full Structural Equation trolled trials should be conducted, rather than studies
Model displayed in Figure 4.3. of association.
■ Whether this model was confirmed by the
■ Retrospective: The population of interest is defined ■ In both retrospective and prospective studies,
after an outcome has occurred. Subjects do not have the disease or outcome state pres-
■ Prospective: The population of interest is defined ent prior to the baseline observation measurement.
before an outcome has occurred. ■ Retrospective observational studies:
■ Cohort: Group of patients, sharing a common char- ● A population is selected with a common starting
acteristic, assembled to analyze or observe over time. point, outcome, or timing of implementation of an
■ Selection bias: When selection results in group dif- intervention.
ferences that may be related to and affect variations in ● The required information to test the hypothesis
outcome. This results in a nonrepresentative sample of interest is already available.
of the population. ● Records are reviewed in order to gather the
■ Measurement bias: When knowledge of the presence data.
or absence of exposure to a risk factor or intervention ● Analyses are performed to determine what risk
influences the assessment of disease or outcome on factors may have contributed to the development of
follow-up. an outcome or disease.
■ Confounding: When a factor, other than the risk fac- ● Further data and follow-up are not obtained.
tor of interest, affects the outcome of a study. May ● For example, a study by Bennet et al (1999) ret-
be accounted for statistically if the confounder is rospectively examined the pooled chart data from a
recognized. cohort of 101 patients with complex regional pain
syndrome type I (CRPS I) who underwent spinal
cord stimulation (SCS) implantation between 1995
INTRODUCTION and 1998. The authors were able to conclude that
those who underwent implantation of a dual-lead
■ The purpose of this chapter is to compare and con- octapolar SCS with high frequency capabilities,
trast prospective and retrospective observational study compared to those who had single-lead quadrapo-
design with a discussion of the components, strengths, lar systems, had a significantly greater reduction
and weaknesses, while providing examples of each in pain scores and surgical revisions to maintain
approach. paresthesia coverage.
■ This was a retrospective study because the data ●Time efficient, since the measurements have
were collected prior to defining the cohort. already occurred.
■ Prospective observational studies: ● Cost effective.
● A sample is identified with a common expo- ● Able to establish predictor variables of the
sure and followed over time to assess for a given outcome.
outcome. ● Reduced chance of bias during measurements as
● Data are collected from the group during the the research question and expected outcome were
follow-up period. not known.
● When the follow-up period is complete, the ● Good method for studying rare diseases.
group is analyzed to determine what risk factors ■ Prospective:
or exposures contributed to the development of the ● Allows for more accurate data collection since
given disease or other outcome. the determination of which variables to measure is
● The analyzed data are obtained in the future. made prior to the outcome, allowing customization
● For example, in Wilson et al (2012), motor recov- of follow-up.
ery following traumatic spinal cord injury (SCI) in ● Considered higher quality evidence as compared
84 patients was compared for those undergoing to retrospective observational studies because study
surgery less than 24 hours after injury and those conditions can be controlled prospectively.
who underwent surgery more than 24 hours after ● A strong inference of association can be made
injury. There was a statistically significant positive due to the time sequencing of events, that is, the
effect associated with early surgery with regard to measurement occurs prior to the outcome.
American Spinal Injury Association (ASIA) motor ● Valuable for studying fatal outcomes, since measure-
score after adjustment for injury level and preop- ment of variables are obtained directly from the research
erative state. participant rather than from records or a proxy.
■ This is an example of a prospective study ■ For example, the quantity of cigarettes smoked
because the cohort was defined and recruited by the participant is monitored as opposed to
into the study prior to data collection. family members’ retrospective report of the
■ Practice-based evidence (PBE) study design: subject’s cigarette use.
● This is a prospective observational study design ■ Both retrospective and prospective observational
TABLE 5.1 Comparison of Retrospective and Prospective Observational Studies With Regard to Timing of Data
Collection, Cohort Creation, and Data Analysis
Time
DEFINITIONS AND DESCRIPTIONS ■ However, historical controls are often used inap-
propriately. Researchers should give careful thought
■ Historical control group: to the inclusion of this type of control group to ensure
● An external control group made up of persons the suitability of this method for their project.
who have been previously studied and who are
similar in characteristics to the present sample.
● Nonconcurrent and nonrandomized control.
STRENGTHS
■ Rapid
INTRODUCTION ■ Inexpensive
■ Avoids need to withhold potentially beneficial treat-
■ This chapter reviews the appropriate use of histori- ment, which may be unethical
cal controls in the research setting, identifies the limi- ■ Improved recruitment due to all participants receiv-
tations of such controls, and discusses strategies for ing potentially beneficial therapy
using them successfully.
WEAKNESSES
IMPLICATIONS
■ Tend to exaggerate the value of a new treatment
■ Historical control groups provide unique opportuni- ■ Vulnerable to bias
ties to examine research questions when more stan- ■ Large differences on covariates (eg, age and gen-
dard control groups are unavailable. der) could lead to biased estimates of treatment effect
■ Understanding the proper application of this type of between the treatment and control groups
control will permit more effective study design and ■ Changes in outcome over time may be due to
data interpretation. changes in various factors other than the introduction
of the intervention being studied
● Underlying patient populations
BACKGROUND ■ Recruitment of patients, for example, from
for example, in disease severity, comorbidities, Statistical Adjustments Used to Limit Bias
or concurrent treatments. or Overestimation of Effects
● Patient care and management peripheral to
treatment ■ Matching controls to subjects:
■ Changes in treatment of the primary condi- ● Example: Proposed study on the effect of
tion or comorbidities may incidentally affect robotic-assisted gait training on the neurological/
outcomes. functional outcomes of patients with subacute spi-
● Diagnostic or evaluating criteria nal cord injury
■ Using different assessment tools to measure ■ Historical controls: Patients treated in the
independent/dependent variables or outcomes same department in previous years, matched to
may lead to an inability to accurately compare current patients based on age, severity of injury,
groups. level of injury, and cause
● Quality of data available ■ Covariate adjustment:
■ Variations in how or when data are recorded ● Limited in the number of variables that can be
may impact the ability to accurately compare included, although propensity analysis provides
groups. method to include a large number of covariates in
● Example: A proposed study of the effect of spi- studies with a large sample (see below)
nal cord injury rehabilitation on functional out- ■ Interpretation can be challenging
comes of paraplegic veterans ■ Bayesian methods
■ Historical controls: 30 paraplegic veterans ■ Propensity scores (see D’Agostino, 1998):
who had recently completed a different study ● Provide an estimate of the effect of an interven-
conducted at the same facility who only received tion by accounting for the covariates that predict
physical therapy. receiving the treatment, thereby reducing selection
■ Weakness: No other obvious systematic bias by equating groups based on these covariates
changes in patient care between times in which ■ Sensitivity analysis (see Greenland, 1996):
patient groups were evaluated, but there may ● Used to determine what impact any unmeasured
have been unrecognized differences. covariates would have to have on the data in order
to alter the conclusions of a study
STRATEGIES
When to Use
Requirements for a Valid Historical
Control Group ■ Risk of withholding treatment is too high:
● If an effective treatment exists, use of a placebo
■ A relatively recent study or data are available for control may not be ethical
same indication ● Example: Proposed study of prevention of aspi-
■ Subject samples with similar demographics and ration pneumonia in stroke patients at an inpatient
prognostic factors rehabilitation facility using a newly implemented
■ Similar inclusion/exclusion criteria clinical pathway for the evaluation and treatment
■ Similar study procedures or standard of care of dysphagia
■ Similar outcome measures ■ Historical controls: Patients treated at same
■ Outcomes of a standard treatment (or no treatment) ■ Alternative therapies could serve as a suitable pro-
are well known and vary little for a given population. spective control treatment.
■ The experimental therapy is expected to have a very ■ No appropriate historical sample exists.
large beneficial effect. ■ Use of placebo will not negatively impact subjects.
● Use of a historical control group avoids with-
holding of potentially effective treatment.
● Limits concern regarding exposure of all sub- SUGGESTED READINGS
jects to experimental treatment while allowing the
focus of study to remain on primary outcome. Baker SG, Lindeman KS. Rethinking historical controls.
Biostatistics. 2001;2(4):383–396.
D’Agostino RB Jr. Propensity score methods for bias reduc-
When Not to Use tion in the comparison of a treatment to a non-randomized
control group. Stat Med. 1998;17(19):2265–2281.
Friedman LM, Furberg CD, DeMets DL. Fundamentals of
■ No undue risk associated with receiving placebo Clinical Trials. 4th ed. New York: Springer; 2010.
versus experimental therapy Greenland S. Basic methods for sensitivity analysis of biases.
● Unless withholding an experimental treatment is Int J Epidemiol. 1996;25(6):1107–1116.
felt to be unethical due to risks associated with not Yoshimura I, Matsumoto K. Notes on the use of histori-
receiving the treatment, a more rigorous placebo- cal controls. Environ Health Perspect. 1994;102(Suppl
controlled study is preferable. 1):19–23.