Evidence-Based Behavioral Medicine: What Is It and How Do We Achieve It?

Karina W. Davidson, Ph.D.

Mt. Sinai School of Medicine

Michael Goldstein, M.D., Ph.D.

Bayer Institute for Health Care Communication

Robert M. Kaplan, Ph.D.

University of California, San Diego

Peter G. Kaufmann, Ph.D.

National Heart, Lung, Blood Institute

Genell L. Knatterud, Ph.D.

Maryland Medical Research Institute

C. Tracy Orleans, Ph.D.

The Robert Wood Johnson Foundation

Bonnie Spring, Ph.D.

University of Illinois at Chicago

Kimberlee J. Trudeau, M.A.

City University of New York Graduate Center

Evelyn P. Whitlock, M.D., M.P.H.

Kaiser Permanente Center for Health Research

ABSTRACT efforts to apply behavioral medicine research to improve the pro-

The goal of evidence-based medicine is ultimately to improve
patient outcomes and quality of care. Systematic reviews of the
available published evidence are required to identify interven-
tions that lead to improvements in behavior, health, and well-
being. Authoritative literature reviews depend on the quality of
published research and research reports. The Consolidated Stan-
dards for Reporting Trials (CONSORT) Statement (www.con-
sort-statement.org) was developed to improve the design and re-
porting of interventions involving randomized clinical trials
(RCTs) in medical journals. We describe the 22 CONSORT
guidelines and explain their application to behavioral medicine
research and to evidence-based practice. Additional behavioral
medicine–specific guidelines (e.g., treatment adherence) are also
presented. Use of these guidelines by clinicians, educators,
policymakers, and researchers who design, report, and evaluate
or review RCTs will strengthen the research itself and accelerate
Authors (members of the Evidence-Based Behavioral Medicine Com-
mittee) are listed in alphabetical order. Each contributed equally to
the process of thinking and writing that produced this document. This
work was supported by the National Institutes of Health Office of Be-
havioral and Social Science Research (OBSSR) Contract No. NLM
00–158/LTN. Karina W. Davidson is now at Columbia University Col-
lege of Physicians and Surgeons.
Reprint Address: K. W. Davidson, Ph.D., Behavioral Cardiovascular
Health & Hypertension Program, Columbia University, College of
Physicians & Surgeons, 622 West 168 Street, PH9 Center, Room 948,
New York, NY 10032. E-mail: kd2124@columbia.ed
© 2003 by The Society of Behavioral Medicine.
cesses and outcomes of behavioral medicine practice.
(Ann Behav Med 2003, 26(3):161–171)
INTRODUCTION
Evidence-based behavioral medicine consists of interven-
tions for which there is accepted evidence of clinical efficacy or
effectiveness. Interventions that fall in the domain of behavioral
medicine include those that promote health and prevent disease
(e.g., physical activity promotion, treatment of tobacco depend-
ence), those that promote adherence to evidence-based medi-
cine protocols from prevention to acute and chronic illness care
to palliative care, and those that alter biobehavioral determi-
nants of acute or chronic diseases and conditions. The purpose
of this article is to explain and demonstrate how the revised Con-
solidated Standards for Reporting Trials (CONSORT) State-
ment criteria (1) can be applied to the design, reporting, and
review of research testing behavioral medicine interventions.
The CONSORT guidelines have been adopted by most major
biomedical journals, including the Annals of Behavioral Medi-
cine. Because behavioral medicine research interventions have
several unique characteristics, additions to the CONSORT
guidelines are proposed to help researchers and reviewers effec-
tively plan, analyze, and use behavioral medicine intervention
research.
EVIDENCE-BASED MEDICINE
Evidence-based medicine is “the conscientious, explicit
and judicious use of current best evidence in making clinical de-

161
162 Davidson et al.
cisions about the care of patients … integrating individual clini-
cal care with the best available clinical evidence from system-
atic research” (2). This movement is a powerful force in today’s
health care environment and has grown in the last 20 years to
prominence in the development of clinical standards and guide-
lines to improve quality of care (3,4). In addition to guiding
quality improvement, evidence-based medicine has provided
objective criteria needed for decision making concerning the al-
location of health care resources.
Evidence-based medicine can contribute substantially to
improvements in the ability of clinicians to care for their pa-
tients if results of pertinent studies can be assessed and dissemi-
nated rapidly. Systematic literature reviews, including meta-
analyses and narrative reviews, provide a manageable and reli-
able approach to synthesizing a burgeoning clinical literature,
although meta-analytic reviews themselves are not without limi-
tations (5). The methods and results of these reviews are be-
coming more accepted in the scientific, clinical, and policy-
maker communities, in part because of the phenomenal growth
in their publication rate (6) and the increasing visibility of sys-
tematic review organizations such as the Cochrane Collabora-
tion (http://www.cochrane.org) and Campbell Collaboration
(http://campbell.gse.upenn.edu/). The transparency of methods
and standardization of these reviews enhance their credibility
with clinicians, patients, health care administrators, educators,
and policymakers. Thus, it is important that as behavioral medi-
cine researchers and practitioners we join this growing move-
ment, particularly given the strong empirical tradition upon
which our area is based.
EVIDENCE-BASED
BEHAVIORAL MEDICINE
Although medicine has moved forward vigorously in the
adoption and evolution of evidence-based principles, behavioral
medicine has not established a mechanism or forum to identify
and discuss the ways in which these principles should be incor-
porated into our discipline. Recently, a committee was formed
to identify the current status of evidence-based principles within
behavioral medicine and to identify the steps required to accel-
erate their use and dissemination.
Under the sponsorship of the Office of Behavioral and So-
cial Sciences Research, National Institutes of Health, a commit-
tee of the Society of Behavioral Medicine members with exper-
tise in evidence-based approaches was formed. Upon reviewing
the current status of evidence-based medicine in our discipline,
the committee concluded that the reporting of randomized con-
trolled trials (RCTs), an important source of knowledge guiding
evidence-based practice, is currently neither uniform nor ade-
quate. Behavioral medicine could benefit from adopting stan-
dardized reporting requirements for RCTs as has been done by
many medical journals and editors (7). We recognize that many
other research designs contribute valuable evidence to our field,
but for this article we focus on RCTs. Thus, we explain and aug-
ment recently updated requirements for the reporting, review-
ing, and evaluating of two-group parallel design RCTs based on
criteria from the CONSORT (8).
Annals of Behavioral Medicine
THE CONSORT STATEMENT:
REPORTING AND REVIEWING RCTs
The CONSORT approach was developed and revised by an
international group of clinical interventionists, statisticians,
epidemiologists, and medical editors within the last 7 years (1).
It was designed to facilitate uniform reporting of clinical inter-
vention results and to allow readers to determine efficiently
whether sources of bias threaten internal or external validity of
findings (8,9). Notably, the CONSORT guidelines do not make
a distinction between efficacy and effectiveness studies. Rather,
the focus is on components of the reporting for all clinical inter-
ventions, regardless of their purpose. In the following, we list
each of the 22 items from the CONSORT checklist and describe
the manner in which this information should be included in the
report of an RCT, noting some special considerations for behav-
ioral medicine intervention research (8).1
“Item 1: How participants were allocated to interven-
tions (e.g., ‘random allocation,’ ‘randomized,’ or ‘ran-
domly assigned’).”2
Information regarding this item should be stated in the Title
and the Abstract of the manuscript; see Figure 1 (8).
Use of the term randomized in both the title and the abstract
of an RCT report is necessary so that the study will be identified
in a literature search. Although behavioral medicine researchers
frequently use this term in the abstract, lack of standard place-
ment of this term in both the title and the abstract can impede the
retrieval, review, and evaluation of behavioral medicine evi-
dence. Although placement in both places may seem redundant,
hand searching of citations within meta-analyses or other re-
views of RCTs is facilitated by the inclusion of the term ran-
domized in the article title.
“Item 2: Scientific background and explanation of ra-
tionale.” Information regarding this item should be
stated in the Introduction section of the report.
Excellent reporting of an RCT begins with a persuasive ra-
tionale. A number of issues should be addressed. First, the need
for intervention should be explained. Other relevant interven-
tions (medical, surgical, and behavioral) should be described
and previous results of these interventions delineated. Limita-
tions of previous research, including consideration of harm or
adverse events that may have arisen, should be presented. If the
intervention to be tested is new, its theoretical plausibility and
any pilot study data should be presented to support the rationale
for conducting the intervention. Of note, previously published
data and preliminary data suggesting that a variable is a risk fac-
1An example of a CONSORT review of a behavioral medicine
study is available via the Internet (http://www.sbm.org/publica-
tions/outlook/2002winter/2002_winter.pdf). The citation for this news-
letter article is Trudeau, KJ, Davidson, K, for the Evidence-based Be-
havioral Medicine Committee: A CONSORT primer. Outlook: A
Quarterly Newsletter of the Society of Behavioral Medicine.
2001–2002, Winter:5–8.
2The items are verbatim from the CONSORT Statement and may
not be changed.
FIGURE 1 The Consolidated Standards for Reporting Trials checklist (http://www.consort-statement.org/revisedstatement.htm). The information
shown should be of use to readers both in designing studies and in reviewing manuscripts.

163
164 Davidson et al. Annals of Behavioral Medicine

tor for morbidity or mortality do n ot constitute a sufficient ra-
tionale for exposing a patient to a new intervention. Rather, care-
ful consideration of possible risks and benefits of the
intervention itself should be presented.
The importance of clearly defining selected outcomes is de-
scribed in Item 6.
Information regarding Items 3 through 12 should be re-
ported in the Methods section of the report.
“Item 3: [a] Eligibility criteria for participants and [b]
the settings and locations where the data were col-
lected.”
Where the data were collected and those who were eligible
to participate in the study are details that enable a reader to com-
pare the characteristics of the study population with a specific
population of interest to the reader. For example, a clinician who
is investigating interventions to increase psychotropic medica-
tion adherence for a patient with substance dependency and
Axis I disorder requires information about whether the con-
sulted study excluded persons with substance dependency.
Of note, the purpose of randomization in clinical trials is to
control for personal characteristics so that the independent vari-
able of interest can be isolated; therefore, individual-differences
variables often are not analyzed in RCTs. One of the advantages
of the RCT methodology is that groups created by random as-
signment are expected to be equivalent for participant character-
istics. For example, a treatment and control group created by
random assignment would be expected to have equal representa-
tion by gender, ethnicity, and age. Typically the first table in the
report confirms that the personal characteristics of patients were
not significantly different at baseline. Although studies of indi-
vidual differences in response to treatment are important, they
are usually not the central scientific question in an RCT. Studies
seeking to identify Patient × Treatment interactions must be pro-
spectively planned and powered accordingly.
However, understanding to whom the treatment reasonably
generalizes rests with both the reader and the writer of the RCT
report. The reader must consider whether there are RCT individ-
ual differences—gender of treatment provider, sexual orienta-
tion of the patients, education level of the patients—from the re-
ported eligibility criteria and baseline characteristics that render
the treatment effects not useful for the reader’s current patient
population. Similarly, the clinical trialist reporting the results
must consider both in the design of the eligibility criteria and in
the limitations section of the article the characteristics of the pa-
tients studied that may reasonably limit the treatment findings.
In addition to describing the eligibility criteria of potential
participants, it is recommended that reports include details re-
garding the setting and locations of data collection as such de-
tails affect generalizability of study outcomes as well. (See also
Item 21 on generalizability.)
Although not explicitly included in CONSORT, recruit-
ment methods should also be described, including whether the
study accrued patients reactively (because they volunteered to
participate) or proactively (via outreach to randomly identified
candidates). Results derived from studies using these two re-
cruitment methods may differ, making it especially important to
characterize ways in which research volunteers and proactively
recruited patients may differ on various medical and demo-
graphic characteristics.
“Item 4: Precise details of the interventions intended
for each group and how and when they were actually
administered.”
Key elements of the intervention must be provided in suffi-
cient detail to allow it to be clearly understood and replicated by
other researchers and considered for inclusion in systematic re-
views (10). To aid in the preparation of this information, we pres-
ent Table 1, which includes questions that should be answered in
the section of the report that describes the intervention. In several
key areas of behavioral medicine intervention research recently
reviewed by the United States Preventive Services Task Force
(USPSTF), insufficient intervention detail was cited as a barrier
to evidence review and guideline development (e.g., 11,12).
The report of the RCT should state whether the intervention
was delivered in a standardized manner to all patients or was indi-
vidually tailored to either treatment response or other factors. If
the research design is one that uses a contact control intervention

TABLE 1
Minimal Intervention Detail to Be Described in Research Reports

Intervention Component Intervention Question Addressed

Content/elements What was the content of the intervention and how was it delivered (e.g., oral communication, written material,
videos, interactive computer programs, other)?
Provider Who delivered it?
Format What were the method(s) of intervention administration? (e.g., self-help, individual, group, telephone, other)?
Setting Where and when was the intervention delivered?
Recipient To whom was the intervention delivered? Was the recipient also the target of the intervention?
Intensity How many different patient contacts & how much total contact time was involved?
Duration Over what time period were intervention contacts conducted and how were they spaced?
Fidelity Was the intervention delivered as intended? How was this monitored and measured?

Note. This table is based on information from Whitlock EP, Orleans CT, Pender N, Allan J: Evaluating primary care behavioral counseling interventions:
An evidence-based approach. American Journal of Preventive Medicine. 2002, 22(4):267–284.
Volume 26, Number 3, 2003
(to control for attention effects), the report should also describe
the control intervention in a fashion analogous to that described
for the intervention. If the control condition is “usual care,” efforts
should be made to specify the therapeutic elements that constituted
usual care so that treatment process analyses can establish whether
these factors were, in fact, offered, and readers can compare the in-
tensity of usual care with the treatment intervention. Additional oc-
casions of provider contact at assessment and follow-up sessions
should also be described, along with discussion of whether these
meetings had therapeutic or neutral intent and content.
There should also be a description of the patient contact
procedures that were implemented during any run-in phase, and
whether (and how) these run-in contacts were intended to en-
courage the withdrawal prior to randomization of patients who
exhibited low motivation to adhere to treatment. For example,
the Diabetes Prevention Program (13) used diet records to docu-
ment adherence to the study protocol during the run-in phase.
Any monetary compensation offered for participating in an in-
tervention or adhering to treatment recommendations should be re-
ported. A description of the procedures used to determine whether
the patient experienced extra or concurrent treatments for the con-
dition under investigation beyond the targeted intervention should
also be described. This is especially important when pharmacolog-
ical treatments are used as adjuncts to behavioral treatments (e.g.,
lipid lowering drugs adjunctive to diet counseling, nicotine re-
placement therapy adjunctive to smoking cessation counseling).
Often, behavioral medicine research reports fail to describe
the actual behavioral intervention techniques used; instead, they
provide details regarding treatment format (e.g., number of ses-
sions, type of treatment). For example, counseling to promote
smoking cessation might be described with reference to the
number of counseling sessions delivered rather than detailing
the methods or approaches used to enhance smokers’ quitting
motivations, skills, or supports or to bring about aversive condi-
tioning. This omission of pertinent information (see Table 1) is
an obstacle not only to replication but also to the credibility and
understanding of core, science-based behavioral medicine inter-
vention technology (e.g., 10).
Treatments meeting the highest standard of evidence will
offer access to written manuals via the Internet, publication, or
contact with the author, specifying how to train practitioners and
how to conduct the intervention so that other investigators can
replicate or review the intervention procedure. Ideally, details of
the intervention itself and copies of intervention materials and
guidelines should be available on request.
“Item 5: Specific objectives and hypotheses.”
The study’s objectives and hypotheses have to be defined
and reported in a clear, unambiguous way. Although this item
seems self-evident, authors of behavioral medicine trial reports
must ensure that these statements are explicitly, rather than im-
plicitly, presented in the report.
“Item 6: (a) Clearly defined primary and secondary
outcome measures and, (b) when applicable, any meth-
ods used to enhance the quality of measurements (e.g.,
multiple observations, training of assessors).”
Evidence-Based Behavioral Medicine 165
A clear distinction between primary and secondary out-
comes and mediating variables has not yet become standard
practice in many scientific areas, including behavioral medi-
cine. Primary outcomes are the main measure(s) or construct(s)
hypothesized to be affected by the interventions under study and
are the main focus of the rationale and objectives of the interven-
tion. For example, if prevention of childhood obesity is the pri-
mary outcome, then authors should describe the manner in
which this outcome will be operationalized. Secondary out-
comes are measures or constructs that, by theory or speculation,
should in addition be altered by the intervention. For example,
remission and recurrence of clinically significant depression
and recurrence of cardiac events may both be primary outcomes
hypothesized to be altered in an RCT of a treatment to amelio-
rate depression after myocardial infarction.
Secondary outcomes such as anxiety or cardiovascular re-
activity may be of interest as mediators. Mediating variables are
those hypothesized to precede or pave the way for changes in
primary or secondary outcomes. In behavioral medicine re-
search, mediators are informative about the mechanisms of
treatment action. Mediating variables can be physiological, cog-
nitive, or behavioral and often include measures of treatment
use or adherence (see Item 27).
The rationale for the primary and secondary outcomes to be
used in the RCT also must be presented. Many variables that
might be considered as intermediate health outcomes are fre-
quently selected as outcome variables, given their role as pu-
tative etiologic factors in the targeted disease. For example, ele-
vated blood pressure and high serum cholesterol are important
risk factors for a variety of health outcomes (e.g., stroke, athero-
sclerosis, heart attack). Behavioral medicine intervention re-
searchers must provide a clear justification for the key health
variables selected for outcome measures as primary.
The USPSTF and the Institute of Medicine of the National
Academy of Sciences, among others, have distinguished be-
tween such intermediate outcome measures and more distal
health outcomes including morbidity, mortality, quality of life,
pain, and functional capacity (14–18). Evidence supporting ab-
solute improvements in these primary health outcomes is given
special weight by the USPSTF in assessing relative benefit and
harm of a clinical preventive service and in developing evi-
dence-based recommendations (18).
In addition, the type of measurement used should be re-
ported for all forms of outcome collection. For example, in re-
porting outcome data, researchers should report the source of
the information, whether or not an event review committee was
used, and how differences of judgment or ambiguities were ad-
judicated. If mortality information is obtained from death certif-
icates or patient charts, the process of abstracting the informa-
tion should be described. Measures of self-reported outcome or
behavioral observation must be clearly identified as such and, at
a minimum, references for their reliability and validity should
be provided. For behavioral medicine interventions, measure-
ment demand characteristics of the outcome measure can be
considered. For example, potential for self-report bias or differ-
ential bias by treatment arm exists (e.g., patients receiving more
166 Davidson et al.
intensive interventions may be motivated to report greater be-
havioral or symptom changes). Staff collecting self-report inter-
view outcome data, if possible, should not be the same staff as
those delivering treatment. (See also Item 11.)
“Item 7: (a) How sample size was determined and, (b)
when applicable, explanation of any interim analyses
and stopping rules.”
Sample size estimation and stopping rules are aspects of the
study design that are important factors in assessing the validity
of an intervention. Although included in most research funding
applications, this information is not routinely presented in pub-
lished research reports. The description of how the number of
patients for the study was estimated should include the expected
value for the control arm, the expected value for the treated arm,
the alpha level to be used for inferential statistics, and the power
to detect the expected difference between the control and treated
arms for the primary outcome. If there are dropouts, crossovers,
or interim statistical testing, these should be reported. Numer-
ous authors have pointed out that many small studies have large
confidence intervals around the reported treatment effect, reduc-
ing the likelihood that the previously significant treatment dif-
ference will be replicated in future studies (19–21). When a
study has been terminated early due to a positive effect, the
treatment difference may be exaggerated (20–22). Finally, and
of importance, the rules for early termination for benefit or harm
should be stipulated in advance (20,23,24).
“Item 8: (a) Method used to generate the random allo-
cation sequence, including (b) details of any restriction
(e.g., blocking, stratification).”
Many methods exist for randomly assigning patients to
treatment conditions (i.e., the random allocation sequence).
However, some of these methods impose restrictions on the ran-
domization, such as fixed or variable blocked randomization. In
fixed blocked randomization, the process is started anew each
time a fixed number of patients have been randomized. For ex-
ample, if the block size is 10 and there are two arms in the inter-
vention, 5 out of each sequence of 10 patients will be assigned to
each arm. In contrast to simple randomization, in which it is
possible to have arms with different numbers of patients,
blocked randomization is more likely to generate arms of equal
size. Stratification in randomization requires that the random-
ization be done separately for different defined arms. For exam-
ple, stratification by gender would require a separate random-
ization for women and for men. The goal is to assure that the
proportions of men and women are about equal in each treat-
ment arm. There is considerable debate among statisticians
about the advantages and disadvantages of stratification
(19,21,25). Because of the potential impact of the method of
randomization, the exact procedures must be reported.
“Item 9: Method used to implement the random alloca-
tion sequence (e.g., numbered containers or central
telephone), clarifying whether the sequence was con-
cealed until interventions were assigned.”
Annals of Behavioral Medicine
The mechanisms by which patients are assigned to treat-
ment arms can lead to serious bias and distortion of treatment
comparisons (26). Ideally, the allocation process for a random-
ized clinical intervention should not provide any opportunity for
researchers making or receiving the allocation to influence the
process (27, pp. 48–49). For these reasons, details on steps in the
randomization process (process of random assignment genera-
tion, steps to ensure concealment of allocation) are crucial to un-
derstanding whether selection bias may have occurred. Studies
also differ in how the randomization was concealed. Sometimes
someone places the assignments in sealed envelopes in a cen-
tralized location away from the intervention site; sometimes a
computer delivers the assignment after initial information and
consent are logged in. Method(s) of concealment should be
planned in advance to assure that neither the investigator nor the
patient, if possible, could know the assignment in advance. Re-
ports should describe the exact method of concealment or lack
of concealment.
“Item 10: Who generated the allocation sequence, who
enrolled participants, and who assigned participants to
their groups.”
The role of different staff members, from the generation of
the randomization sequence to the assignment of patients to
treatment arm, should be described. Of note, it is preferable that
those who generate the randomization sequence, and those who
later assign patients to the randomized arm, are not the same re-
search staff as those who have enrolled the patient. Staff mem-
bers who enroll patients likely know some of the clinical charac-
teristics of the patients, particularly if they have conducted the
screening examinations. Therefore, division of labor to different
research staff can aid in the preservation of blinding and avoid
selection bias.
“Item 11: (a) Whether or not participants, those admin-
istering the interventions and those assessing the out-
comes were blinded to group assignment. (b) If done,
how the success of blinding was evaluated.”
Blinding or masking refers to withholding certain informa-
tion from patients or members of the research team. In studies
evaluating drug therapy it is usually possible to prepare placebos
that are identical in appearance to the medication under study so
that the patient does not know which treatment he or she is re-
ceiving. Also, the treating physician does not know which treat-
ment has been assigned to the patient he or she is treating or ex-
amining. This approach is referred to as double-blind or
double-masked. Some studies of surgical procedures have used
sham procedures so that the patient would not know the treat-
ment assignment (e.g., whether the surgery was conducted or
not).
In contrast to drug interventions in which active drugs and
placebos look exactly the same, it is difficult to achieve a behav-
ioral placebo that has the same appearance and credibility as the
active treatment. Moreover, in studies testing counseling or
other behavioral interventions that are delivered to patients in
health care settings, it may be impossible to blind the providers
Volume 26, Number 3, 2003 Evidence-Based Behavioral Medicine 167

to the patient’s assigned arm, especially when they are deliver- ter a particular treatment and all of her or his patients then re-
ing a component of the intervention. However, there are meth- ceive the same therapy. If the randomization assigned providers,
ods that one can use to preserve blinding during data collection rather than individual patients, to a particular treatment, then the
and during the assessment of behavioral characteristics. For ex- analysis must use the provider as the unit of analysis or statisti-
ample, patients can be prompted to not mention their treatment cally adjust for the intraclass correlation introduced by differ-
assignment when symptoms and adverse events are solicited. ences in the unit of randomization and the unit of analysis. Sev-
Videotapes or other assessment data can be coded by research eral different statistical methods are available to take account of
staff who do not know treatment assignment. cluster randomization (31). Often the loss in statistical power
Investigators should always report whether the assessors due to such adjustments is not as severe as expected.
for primary and secondary outcomes were aware of treatment
“Item 13: (a) Flow of participants through each stage (a
assignment at the time patients were evaluated. In some studies
diagram is strongly recommended). Specifically, for
research staff who lack information on the assigned treatment
each group report the numbers of participants ran-
may classify outcome events (such as myocardial infarction) by
domly assigned, receiving intended treatment, com-
review of appropriate documentation. In other studies the pri-
pleting the study protocol, and analyzed for the pri-
mary dependent variable may be based on readings of materials
mary outcome. (b) Describe protocol deviations from
such as ambulatory EKGs or coronary angiograms made in a
study as planned, together with reasons.”
central reading center by personnel who are blinded to the pa-
tients’ assigned treatment and clinical outcomes. Coders who Figure 2 (8) shows the preferred method to describe the
are blind to treatment status may rate depression level or adher- flow of patients through each stage of the design. Protocol devi-
ence to medication regimen. ations should then be presented, and their ramifications for
The Sertraline Antidepressant Heart Attack Randomized generalizability and bias should be discussed. Studies often in-
Trial (28) is an example in which masking of the assessors of clude the presentation of numbers of patients assigned to treat-
treatment outcome was not reported, and so it is unclear if those ment and those that completed the study protocol, but the use of
conducting the depression improvement rating were masked to the flowchart would enhance reporting of patient flow in other
treatment arm. If it is not possible to administer study treatments important stages as well.
in a double-masked fashion, other steps should be taken to avoid
“Item 14: Dates defining the periods of recruitment
bias in the assessment of the primary outcome variable. For ex-
and follow-up.”
ample, the Enhancing Recovery in Coronary Heart Disease Pa-
tients intervention (29) used centralized ratings of outcome and Although it is simple to include the beginning and ending
therapy quality control and also kept investigators and health dates of an RCT, this is rarely reported for behavioral medicine
care providers masked to the outcome. The research literature
includes many excellent examples of successful masking of as-
sessment in clinical interventions (20,30). Although rarely done
in behavioral medicine, ideally there should be postintervention
evaluations of the degree to which blinding was successful. This
can be measured by the accuracy with which research staff
thought to be blinded to treatment condition are able to guess
patients’ treatment assignments.

“Item 12: (a) Statistical methods used to compare

groups for primary outcome(s); (b) methods for addi-
tional analyses, such as subgroup analyses and ad-
justed analyses.”

Reports should include sufficient detail, references to stan-

dard works (citing specific pages), and other documentation so
that readers can repeat the analyses for at least the primary and
major secondary reported outcomes and verify that the results
are correct. In addition, the statistical reporting should enable
the reader to understand the study design’s strengths and weak-
nesses in sufficient detail to form a clear and accurate impres-
sion of the reliability of the data.
Typically, individual patients are randomized to treatments.
However, other units of randomization are common in clinical
research, and this has important implications for the statistical
analysis of intervention studies. For example, some interven- FIGURE 2 The Consolidated Standards for Reporting Trials flow-
tions involve nesting, in which a provider is assigned to adminis- chart (http://www.consort-statement.org/revisedstatement.htm).
168 Davidson et al.
interventions. Providing this information allows a reader to con-
sider secular trends that may have occurred for that particular
health outcome or other contextual factors that may influence
the usefulness or generalizability of the reported outcomes. For
example, cardiac mortality has steadily declined over the last 30
years, and the reader of reports about cardiac mortality re-
ductions by behavioral or pharmaceutical interventions in a car-
diac population must carefully consider the typical treatments
and mortality rates that exist currently compared with those in
existence during the enrollment and follow-up dates of the
intervention.
“Item 15: Baseline demographic and clinical charac-
teristics of each group.”
Distributions and summary statistics of relevant character-
istics should be presented, usually in table form, to make it pos-
sible to determine the baseline comparability of the patients who
were originally assigned to each arm. To assess comparability
between arms, researchers should present data separately for
each arm, not merely for the entire study group as in standard
practice in behavioral medicine literature.
“Item 16: Number of participants (denominator) in
each group included in each analysis and whether the
analysis was by ‘intention to treat.’ State the results in
absolute numbers when feasible (e.g., 10 of 20, not
50%).”
An intention-to-treat analysis keeps all patients in the arm
to which they were randomized, whether or not they received the
treatment designated for that arm. To illustrate this point, we use
the following example. Patients were randomly assigned to ei-
ther a cognitive behavior modification treatment arm or a phar-
macological treatment arm. Later it is learned that 25% of those
assigned to the cognitive arm failed to complete treatment and
instead were switched to the pharmacological treatment. Never-
theless, in the intention-to-treat analysis, these patients would
remain in the cognitive treatment condition. They should neither
be analyzed as part of the pharmacological arm nor be dropped
from the analysis. Analysis of data that does not use the inten-
tion-to-treat principles may contribute to a biased treatment ef-
fect, making it difficult to assess whether bias due to differential
loss to follow-up may have influenced the result (32).
If any of the randomized patients were excluded from the
analysis (which may not be acceptable to many peer-reviewed
journals), the report of the RCT should explain why these patients
were excluded. Sufficient information must be provided to de-
scribe the outcome for each patient, including number lost to fol-
low-up, number for whom there is a known vital status, and num-
ber who completed or participated in certain study procedures,
particularly if vital status (an incontrovertible outcome) is not the
primary outcome. The number of patients described in each stage
here should be consistent with the flowchart (see Item 13).
“Item 17: For each primary and secondary outcome, a
summary of results for each group and the estimated
effect size and its precision (e.g., 95% confidence in-
terval).”
Annals of Behavioral Medicine
For each statistical test, authors should provide the mean,
standard deviation, whether one-sided or two-sided tests were
performed, and the associated effect size.
“Item 18: Address multiplicity by reporting any other
analyses performed, including subgroup analyses and
adjusted analyses, indicating those pre-specified and
those exploratory.”
A common problem for all interventions is the use of multi-
ple outcome measures. When too many measures are used, the
probability of falsely concluding that a treatment is effective is in-
creased. Usually, there is a statistical penalty for making multiple
comparisons, but in practice, adjustments for multiple compari-
sons are not applied as often as they should be. Investigators may
administer multiple measures but report only the results that are
statistically significant. To avoid these problems, primary and
secondary outcomes should be selected in advance and clearly re-
ported as such in study publications. The interpretation of the
probability value and effect size also depends on whether the test
was prespecified, on whether interim data analyses were per-
formed, and on the number of other outcome variables examined.
If no other analyses were performed, authors should state this.
Otherwise, it is unclear whether analyses that produced positive
findings were the only analyses conducted and reported.
“Item 19: All important adverse events or side effects
in each intervention group.”
Sufficient information about side effects, treatment compli-
cations, or adverse or unexpected events should be provided to
enable a complete assessment of the risks and benefits of treat-
ment. USPSTF recommendations for counseling to promote
physical activity and healthy diet were handicapped by the ab-
sence of any assessment of potential intervention harms or ad-
verse efforts in all but one or two interventions (11,12).
“Item 20: Interpretation of the results, taking into ac-
count study hypotheses, sources of potential bias or
imprecision, and the dangers associated with multi-
plicity of analyses and outcomes.”
The interpretation of results from clinical interventions is
typically reported in the Comment section or Discussion section.
This section of the report should include comments on several is-
sues. First, the results should be interpreted within the context of
other relevant clinical interventions and observational studies.
Beyond a restatement of the results, the report should include the
limitations of the study with respect to internal and external valid-
ity, including any potential sources of bias. For example, the re-
port should specify any deviations from the planned procedures
or any other factors that may have influenced the results unexpect-
edly. If multiple outcome measures were tested, the report should
note the potential problems that might occur because of multiple
comparisons. Whether or not the intervention produced the ex-
pected result, biases associated with potential measurement error
or other design limitations should be acknowledged explicitly.
“Item 21: Generalizability (external validity) of the in-
tervention findings.”
Volume 26, Number 3, 2003
The Comment or Discussion section should also describe
the external validity or generalizability of the results. In particu-
lar, the report should specify factors that might limit gener-
alizability so as to guide readers about the applicability of these
findings to their own populations and settings.
Eligibility criteria used in the study influence generaliza-
bility because they restrict the range of patients who could par-
ticipate in the study. Therefore, the report of a clinical inter-
vention should specify the criteria used to select the study
population (33). Of importance, the study rationale and objec-
tives should dictate the appropriate population and, thus, the eli-
gibility criteria.
In addition, settings, locations, and other contextual vari-
ables that could reasonably be expected to have influenced data
collection and results should also be described and, if possible,
examined through appropriate analyses. For instance, the USPSTF
systematically reviews published studies to assess feasibility of
interventions in primary care settings (18). Such influences in-
clude the number of settings (e.g., single or multicenter) and the
type of setting (e.g., inpatient vs. outpatient), geographic loca-
tion, and historical or legislative events. For example, the find-
ings of a longitudinal study of an intervention for depression and
anxiety in New York City would have been influenced by the
events of 9/11. Knowledge of contextual factors will help the
reader to assess whether the treatment results are applicable to
the reader’s setting.
“Item 22: General interpretation of the results in the
context of current evidence.”
The final paragraph of the article should provide a general
interpretation of the results. Given the strengths and weaknesses
of the study, and considered within the context of other com-
pleted research, the concluding statement should offer an objec-
tive appraisal of the scientific and practical contribution of this
particular intervention.
ADDITIONAL EVIDENCE-BASED
BEHAVIORAL MEDICINE–SPECIFIC
GUIDELINES
Five additional requirements should be reported on for be-
havioral medicine RCTs. They are as follows:
Item 23: Training of treatment provider(s).
Because behavioral medicine treatments differ in their
complexity, reporting should describe the background training
and professional credentials of the study providers, as well as
the specific procedures that were used to train providers to uni-
formly conduct the study treatments.
Item 24: Supervision of treatment provider(s).
All that is needed to fulfill these reporting criteria is a few
sentences describing the type, duration, and form that supervi-
sion, if any, required. For example, if weekly supervision meet-
ings occurred with a group of providers, and videos of random
samples of therapy were observed and commented on, this pro-
cess should be described succinctly in the report.
Evidence-Based Behavioral Medicine 169
Item 25: Patient and provider treatment allegiance or
preference.
The treatment preference of the investigators, the providers,
and the patients should all be recorded and reported. This allows
an examination of a source of bias that may be more relevant to
behavioral medicine interventions than to medical interventions
in which investigators can be more easily masked to treatment
conditions. Reviews of psychotherapy research indicate that
treatment provider allegiance has influenced treatment effec-
tiveness outcomes (34,35); however, a more recent analysis of
cognitive therapy effectiveness suggests that research allegiance
either is a historical phenomenon or is simply more difficult to
detect in recent work (36).
A patient’s preference for one type of treatment over another
may also bear on an outcome in a behavioral medicine interven-
tion in a way that may not occur with a surgical or medical inter-
vention. This is because quality of life, symptom–pain report, or
report of improvement of function often have subjective compo-
nents that influence measurement and, ultimately, interpretation
of results. These issues are especially relevant when one is exam-
ining early dropout data and when adherence is compromised in
an intervention arm, potentially by those assigned to an interven-
tion that was not originally preferred by the participant.
Item 26: Manner of testing, and success of, treatment
delivery by the provider.
Although health care providers may have been trained ade-
quately, it cannot be assumed that they implemented the study
interventions as the investigator intended. Treatment delivery
(also called treatment integrity) pertains to whether a given in-
tervention was administered according to plan and to whether it
was inadvertently delivered to the study’s control or comparison
group(s) as well (i.e., contamination) (37,38).
Treatment delivery is compromised when intended ele-
ments of the intervention are omitted or other treatment ele-
ments are added that could influence study outcomes. Providing
standardized training and ongoing supervision of the study and
implementing periodic assessments to determine drift over time
in treatment delivery should facilitate evaluation of treatment
delivery. In multicenter interventions, it is crucial to establish
methods to assure that the intervention is delivered identically at
all sites if identical treatment is warranted and appropriate for
the intent of the treatment and the type of site.
There is also a need to guard against contamination across
the alternative study treatments, either because project providers
begin to blend elements of multiple treatment approaches or be-
cause experimental and control patients share the same environ-
ment and get exposed to elements of both interventions. Ideally,
different treatments should be delivered by different research
staff so that treatment contamination can be minimized.
If different treatments are provided by different treatment
staff, this design element must be considered as a nesting factor
within the statistical plan. The primary consideration in assess-
ing treatment delivery is whether the intervention to which each
patient was randomized was the one that was delivered. Whether
or not adequate treatment delivery has been achieved is best
170 Davidson et al.
evaluated by having trained persons blind to treatment condition
observe or review videotapes or audiotapes of a random and rep-
resentative sample of intervention sessions. Regular confer-
ences between relevant staff, providers, as well as supervisors,
can also improve treatment delivery.
Item 27: Treatment adherence should also be moni-
tored and reported.
Determining whether an adequate “dose” of treatment was
received is a judgment that also requires evaluation of the pa-
tient’s adherence to treatment. Several levels of adherence can
be differentiated and should be described. The most rudimen-
tary of these is whether or not patients attended treatment ses-
sions and were, therefore, present to receive the intervention as
delivered. A higher level of assessment of adherence is obtained
by measuring whether or not patients enacted the treatment rec-
ommendations. For example, did they fill out the exercise club
registration forms? Did they attend the exercise class, as evi-
denced by fitness counselor report or by actigraphy? Did they
implement a recommended increase in exercise, as evidenced by
self-report diaries or changes in aerobic capacity? Did they read
or complete homework assignments in self-help materials?
When assessing adherence to treatment, it is recommended that
investigators use both self-reported and objectively measured
evidence of adherence with treatment recommendations and,
further, that they report the decision rules, if any, whereby these
adherence measures were combined.
It should also be noted that behavioral adherence and health
outcomes may mistakenly be assumed to be interchangeable.
For example, occurrence of weight loss in a patient enrolled in a
dietary intervention is often taken to signify that the patient ad-
hered to the prescribed regimen of caloric restriction. He or she
may have done so or may have implemented a different eating or
activity program from the one prescribed. He or she may have
lost weight due to illness or may have initiated treatment with an
anorectic agent. Thus, the patients’ adherence behaviors have to
be assessed accurately and reported rather than being inferred
from study outcomes.
SUMMARY AND CONCLUSION
Standardization of reporting in behavioral medicine re-
search will enhance the quality of research designs and the
understanding of research, ultimately improving the quality of
care provided and the outcomes achieved. In this article we
have highlighted the advantages of standardized reporting in
communicating rationale, hypotheses, methodology, results,
interpretation, and generalization of randomized controlled
trials. We have recommended use of the CONSORT (1)
guidelines developed to promote identifying evidence-based
medical interventions, with some additions to address special
issues arising in controlled behavioral medicine intervention
research.
As behavioral medicine researchers, we have concluded
that the field should pay particular attention to CONSORT items
focused on randomization and blinding procedures, multiplicity
of outcomes, and the clear identification of which outcome is
Annals of Behavioral Medicine
primary. We have selected these for particular attention because
they were identified as almost always absent in one recent re-
view of behavioral medicine randomized controlled trial reports
(39). These CONSORT-based guidelines hold immense value to
advance the identification and dissemination of evidence-based
behavioral medicine interventions.
Specifically, CONSORT guidelines will help behavioral
medicine researchers to become better at reporting on key
threats to internal and external validity. Because behavioral
medicine research often requires that the health care provider
participate in the study as an element of the intervention, we also
recommend the inclusion of provider-specific items such as
training and supervision of treatment providers, treatment alle-
giance, treatment delivery, and treatment adherence. Complete
reporting of these details will provide additional information for
the reader to use in assessing the internal validity (e.g., was the
intended intervention actually conducted by the provider and re-
ceived by the patient?) and external validity (e.g., could we do
this in our setting with our providers?) of the particular behav-
ioral medicine intervention.
In closing, we note that the roots of behavioral medicine are
strongly empirical, and the continued research base of behav-
ioral medicine practice makes the leap to evidence-based behav-
ioral medicine a natural one. Renewed emphasis on evidence-
based care as the cornerstone of national health care quality im-
provement efforts (14) makes this a propitious time to apply rig-
orous research design and reporting standards to behavioral
medicine research. The CONSORT Statement offers a tool to
ground our behavioral medicine evidence base in the excellent
design, reporting, and review of randomized controlled trials
and will help to establish parity with the quality of reporting and
review of medical research evidence.
Behavioral medicine is on the verge of conducting large-
scale interventions that will test whether behavioral treatments
can reduce mortality and morbidity as well as whether they can
improve quality of life. These interventions have to be psycho-
logically sophisticated, culturally sensitive, and thoughtful, but
they must also adhere to good clinical intervention methodology
if they are to contribute meaningfully to an evidence base.3 We
believe that offering our behavioral medicine colleagues an
overview of excellent randomized controlled trial reporting will
further our goal of including all of us in the evidence-based
movement. Joining in the evidence-based movement represents a
first step toward ultimately achieving our professional goals re-
garding improved patient care.
REFERENCES
(1) Moher D, Schulz KF, Altman DG, for the CONSORT Group: The
CONSORT Statement: Revised recommendations for improving
the quality of reports of parallel-group randomized trials. Journal
of the American Medical Association. 2001, 285:1987–1991.
(2) Eddy DM: Evidence-based clinical improvements. Presentation
at Directions for Success: Evidence-Based Health Care Sympo-
sium sponsored by Group Health Cooperative, Tucson, AZ, 2001.
3We are grateful to an anonymous reviewer for contributing this
point to our article.
Volume 26, Number 3, 2003
(3) Sox Jr. HC, Woolf SH: Evidence-Based Practice Guidelines
from the U.S. Preventive Services Task Force. Journal of the
American Medical Association. 1993, 169(20):2678.
(4) Woolf SH, Atkins DA: The evolving role of prevention in health
care: Contributions of the U.S. Preventive Service Task Force.
American Journal of Preventive Medicine. 2001, 29(Supp.
3):13–20.
(5) Cook DJ, Greengold NL, Ellrodt G, Weingarten SR: The rela-
tion between systematic reviews and practice guidelines. An-
nals of Internal Medicine. 1997, 127:210–216.
(6) Cook DJ, Mulrow CD, Haynes RB: Systematic reviews: Syn-
thesis of best evidence for clinical decisions. Annals of Internal
Medicine. 1997, 126(5):376–380.
(7) Meade TW, Wald N, Collins R: CONSORT Statement on the
reporting standards of clinical trials. BMJ. 1997, 314:1126.
(8) Altman DG, Schulz KF, Moher D, et al.: The revised
CONSORT Statement for reporting randomized trials: Expla-
nation and elaboration. Annals of Internal Medicine. 2001,
134:663–694.
(9) Rennie D: CONSORT revised: Improving the reporting of ran-
domized trials. Journal of the American Medical Association.
2001, 285(15):2006–2007.
(10) Whitlock EP, Orleans CT, Pender N, Allan J: Evaluating pri-
mary care behavioral counseling interventions: An evi-
dence-based approach. American Journal of Preventive Medi-
cine. 2002, 22(4):267–284.
(11) Pigone MP, Ammerman A, Fernandez L, et al.: Counseling to
promote a healthy diet in adults: A summary of the U.S. Preven-
tive Services Task Force. American Journal of Preventive Medi-
cine. 2003, 24(1):75–92.
(12) U.S. Preventive Services Task Force: Behavioral counseling in
primary care to promote physical activity: Recommendation
and rationale. Annals of Internal Medicine. 2002, 137:205–207.
(13) Diabetes Prevention Program Research Group: The Diabetes Pre-
vention Program: Design and methods for clinical trial in the pre-
vention of Type 2 diabetes. Diabetes Care. 1999, 22:623–634.
(14) Institute of Medicine (U.S.) Committee on Health and Behav-
ior: Research Practice and Policy: Health and Behavior: The
Interplay of Biological, Behavioral, and Societal Influences.
Washington, DC: National Academy Press, 2001.
(15) Kaplan RM: The Ziggy theorem: Toward an outcomes-focused
health psychology. Health Psychology. 1994, 13(6):451–460.
(16) Kaplan RM: Two pathways to prevention. American Psycholo-
gist. 2000, 55(4):382–396.
(17) Weinstein MC, Siegel JE, Gold MR, Kamlet MS, Russell LB:
Recommendations of the Panel on Cost-Effectiveness in Health
and Medicine. Journal of the American Medical Association.
1996, 276(15):1253–1258.
(18) Harris RP, Helfand M, Woolf SH: Current methods of the U.S.
Preventive Services Task Force. American Journal of Preven-
tive Medicine. 2001, 29(Supp. 3):21–35.
(19) Peto R, Pike MC, Armitage P, et al.: Design and analysis of ran-
domized clinical trials requiring prolonged observation of each
patient: I. Introduction and design. British Journal of Cancer.
1976, 34(6):585–612.
(20) Friedman LM, Furberg CD, DeMets DL: Fundamentals of Clini-
cal Trials (3rd Ed.). New York: John Wright PSG Inc., 1998.
Evidence-Based Behavioral Medicine 171
(21) Pocock S: Clinical Trials: A Practical Approach. London:
Wiley, 1983.
(22) Coronary Drug Project Research Group: Practical aspects of
decision making in clinical trials: The Coronary Drug Project as
a case study. Controlled Clinical Trials. 1981, 1(4):363–376.
(23) Peto R, Collins R, Gray R: Large-scale randomized evidence:
Large, simple trials and overviews of trials. Journal of Clinical
Epidemiology. 1995, 48(1):23–40.
(24) Yusuf S, Collins R, Peto R, et al.: Intravenous and intracoronary
fibrinolytic therapy in acute infarction: Overview of results on
mortality, reinfarction and side-effects from 33 randomized
controlled trials. European Heart Journal. 1985, 6(7):556–585.
(25) Meinert CL: Clinical Trials: Design, Conduct and Analysis.
New York: Oxford University Press, 1983.
(26) Kunz R, Oxman AD: The unpredictability paradox: Review of
empirical comparisons of randomized and non-randomized
clinical trials. BMJ. 1998, 317(7167):1185–1190.
(27) Clarke M, Oxman AD (eds): Cochrane Reviewers Handbook
4.2.0 [updated March 2003]. In The Cochrane Library, Issue 2,
2003. Oxford: Update Software.
(28) Glassman AH, O’Connor CM, Califf RM, et al.: Sertraline
treatment of major depression in patients with acute MI or un-
stable angina. Journal of the American Medical Association.
2002, 288(6):701–709.
(29) Enhancing Recovery in Coronary Heart Disease Patients
(ENRICHD): Study design and methods. The ENRICHD in-
vestigators. American Heart Journal. 2000, 139(1, Pt 1):1–9.
(30) Glasgow RE, Whitlock EP, Eakin EG, Lichtenstein E: A brief
smoking cessation intervention for women in low-income
Planned Parenthood clinics. American Journal of Public
Health. 2000, 90(5):786–789.
(31) Green SB, Corle DK, Gail MH, et al.: Interplay between design
and analysis for behavioral intervention trials with community
as the unit of randomization. American Journal of Epidemiol-
ogy. 1995, 142(6):587–593.
(32) Hollis S, Campbell F: What is meant by intention to treat analy-
sis? Survey of published randomized controlled trials. BMJ.
1999, 319:670–674.
(33) Greenhalgh T: Assessing the methodological quality of pub-
lished papers. BMJ. 1997, 315:305–308.
(34) Robinson LA, Berman JS, Neimeyer RA: Psychotherapy for the
treatment of depression: A comprehensive review of controlled
outcome research. Psychological Bulletin. 1990, 108(1):30–49.
(35) Smith ML, Glass GV, Miller TI: The Benefits of Psychotherapy.
Baltimore, MD: The Johns Hopkins University Press, 1980.
(36) Gaffan EA, Tsaousis I, Kemp-Wheeler SM: Research alle-
giance and meta-analysis: The case of cognitive therapy for de-
pression. Journal of Consulting and Clinical Psychology. 1995,
63(6):966–980.
(37) Moncher FJ, Prinz RJ: Treatment fidelity in outcome studies.
Clinical Psychology Review. 1991, 11:247–266.
(38) Lichstein KL, Riedel BW, Grieve R: Fair tests of clinical trials:
A treatment implementation model. Advanced Behavioral Re-
search Therapy. 1994, 16:1–29.
(39) McGrath PJ, Stinson J, Davidson K: Commentary: The Journal
of Pediatric Psychology should adopt the CONSORT Statement
as a way of improving the evidence base in pediatric psychol-
ogy. Journal of Pediatric Psychology. 2003, 28(3):169–171.