the calgary audit and feedback framework: a practical
TRANSCRIPT
RESEARCH Open Access
The Calgary Audit and FeedbackFramework: a practical, evidence-informedapproach for the design andimplementation of socially constructedlearning interventions using audit andgroup feedbackLara J. Cooke1* , Diane Duncan2, Laura Rivera2, Shawn K. Dowling3, Christopher Symonds4 and Heather Armson5
Abstract
Background: Audit and feedback interventions may be strengthened using social interaction. The Calgary office ofthe Alberta Physician Learning Program (CPLP) developed a process for audit and group feedback for physicians.This paper extends previous work in which we developed a conceptual model of physician responses to audit andgroup feedback based on a qualitative analysis of six audit and group feedback sessions. The present studyexplored the mediating factors for successfully engaging physician groups in change planning through auditand group feedback.
Methods: To understand why some groups were more interactive than others, we completed a comparativecase analysis of the six audit and group feedback projects from the prior study. We used framework analysisto build the case studies, triangulated our observations across data sources to validate findings, compared thecase studies for similarities and differences that influenced social interaction (mediating factors), and thematicallycategorized mediating factors into an organizing framework.
Results: Mediating factors for socially interactive AGFS were a pre-existing relationship between the program teamand the physician group, projects addressing important, actionable questions, easily interpretable data visualization inthe reports, and facilitation of the groups that included reflective questioning. When these factors were in place (cases1, 2A, 3), the audit and group feedback sessions were dynamic, with physicians sharing and comparing practices, andraising change cues (such as declaring commitments to de-prescribing, planning educational interventions, andimproving documentation). In cases 2C–D, the mediating factors were less well established and in these cases, thesessions showed little physician reflection or change planning. We organized the mediating factors into a frameworklinking the factors for successful sessions to the conceptual model of physician behaviors which these mediatingfactors drive.
(Continued on next page)
* Correspondence: [email protected] of Clinical Neurosciences, Cumming School of Medicine,University of Calgary, UCMC Area 3, 3350 Hospital Drive NW, Calgary, AB T2N4N1, CanadaFull list of author information is available at the end of the article
© The Author(s). 2018 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, andreproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link tothe Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
Cooke et al. Implementation Science (2018) 13:136 https://doi.org/10.1186/s13012-018-0829-3
(Continued from previous page)
Conclusions: We propose the Calgary Audit and Feedback Framework as a practical tool to help foster sociallyconstructed learning in audit and group feedback sessions. Ensuring that the four factors, relationship, questionchoice, data visualization, and facilitation, are considered for design and implementation of audit and groupfeedback will help physicians move from reactions to their data towards planning for change.
Keywords: Audit and feedback, Feedback, Social learning theory, Framework, Practice improvement, Professionaldevelopment, Comparative case study, Physician learning, Implementation, Knowledge translation
BackgroundAudit and feedback (AF) is a widely published method ofproviding performance data to physicians to help themtranslate knowledge into practice [1]. It has been shownto be more effective in helping physicians change their be-havior than many traditional models of professional devel-opment [2]. However, the effectiveness of published AFinterventions is variable [1]. Several authors have called at-tention to this issue, citing the need for further study sothat the reasons for varied effectiveness of AF can be morefully understood and addressed [3–7].Here, we extend our previous work, which examined
physician responses to a novel type of audit and groupfeedback (AGF) and presented a conceptual model ofphysician responses to AGF sessions (AGFS) [8]. We ob-served that physicians engaged in planning for changemore robustly in some AGFS than in others. Thepresent study explores factors that influenced the socialinteractions in those AGFS.Our understanding of what influences AF effectiveness
is informed by three main areas of literature: implemen-tation science, motivational and behavior change theory,and the educational feedback literature [5]. Colquhounet al. have emphasized the need to draw from thesedifferent domains in order to optimize AF design andimplementation [5].Frameworks and theories from these fields can help us
to understand the factors that influence implementation[9–19]. One widely cited, evidence-based framework isiPARIHS [17], which identifies four key domains thatcan be used to determine why an intervention may ormay not be successful: the innovation, the recipient, thecontext, and the facilitation [17].The authors of iPARIHS described facilitation as the
“active ingredient” for implementation [17]. It was de-fined as “the construct that activates implementationthrough assessing and responding to characteristics ofthe innovation and the recipients” in context ([17], p 8).iPARIHS situates the success of implementation uponwhether the facilitator can enable the recipients to makethe desired change [17].Because many published AF interventions describe
passive, non-facilitated feedback in the form of physician‘report cards’, the lack of attention to facilitation of
feedback in AF is a potential criticism and may explainsome of the variation in AF interventions [1, 4, 5, 7].Brehaut et al. published 15 recommendations to en-
hance effectiveness of “practice feedback” [7]. Thoserelevant to the current study include choosing the rightitems on which to provide feedback (aligned with localpriorities, actionable, specific), providing individualizeddata with relevant comparators, integrating summarymessages and data visualization, using social interactionto construct feedback, managing cognitive load, and ad-dressing barriers to change [7].The use of social interaction to construct feedback un-
derpins the design of AGFS described in the presentstudy and stems from Social Learning Theory [20] andthe work of Vygotsky and Ajzen, who emphasized thatefficient learning can occur through observation ofothers’ behaviors, the rewards or consequences of others’behaviors, and the social norms that develop in groupsettings [7, 20–22].Similarly, the Theoretical Domains Framework
(TDF) explores factors that impact behaviors [23].The TDF identifies 14 domains to consider in imple-mentation [23]. These include knowledge and skills,beliefs about capability for change, goal setting, theenvironmental context and exploration of social influ-ences on implementation [23].The medical education literature about feedback up-
take highlights several aspects of feedback delivery.Ideally, feedback occurs in an in-person, facilitated,coaching-oriented manner within the setting of a trust-ing, respectful relationship between the provider and re-cipient of feedback [24–27].In the R2C2 model, Sargeant et al. emphasize the pri-
macy of the relationship between feedback providers andrecipients [25]. Likewise, the R2C2 model focuses on un-derstanding and accepting feedback prior to coachingfor change [25, 28, 29]. Watling et al. highlight the valueof the credibility of the feedback provider as perceivedby the recipient [24, 26]. In an overview of the educa-tional feedback literature, Telio et al. proposed a con-struct of “educational alliance,” which may influencefeedback uptake [27].The Calgary office of the Alberta Physician Learning
Program (CPLP) delivers AF to groups of physicians.
Cooke et al. Implementation Science (2018) 13:136 Page 2 of 18
This team has delivered over 30 AF projects on variousclinical topics, locally and provincially, addressing prac-tice variation and appropriateness. Some projects en-gaged physicians more than others and we wished toexplore why.In a previous study, we described our approach to AF:
audit and group feedback sessions (AGFS) [8]. These AGFSwere designed based upon the principles of social learningtheory and best practices from the education feedback andimplementation science literature [7, 17, 23–27]. In theAGFS, physician groups participated in face-to-face, [25,27] facilitated group feedback sessions with peers, duringwhich they reviewed reports containing their own individu-alized performance data (along with anonymized peer com-parators) in order to identify opportunities, barriers, andenablers for making change [7, 17, 23, 25–27].We investigated the behaviors of physicians in AGFS in
order to capture their reactions and engagement with thedata and how the presence of peers influenced the direc-tion of group discussions [8]. Through a qualitative ana-lysis, we developed a conceptual model of how physiciansreact and interact in AGFS. Physicians expressed initial re-actions to the data (skepticism, interest) and then transi-tioned through several discrete behaviors: understandingand questioning, justifying and contextualizing the data,and reflection and sharing of practices, before beginningto raise ‘change cues’ and make change plans. Change cueswere defined as “turning points in the group discussion,initiated by a brief comment highlighting the importanceof a performance gap revealed by the data reports,” andwere usually raised by a group participant rather than thefacilitator, who was not a group member [8].Qualitative analysis of the AGFS transcripts showed
that the degree of interaction and engagement of thephysicians varied between AGFS, as did the groups’ ten-dency to raise change cues and plan for change [8]. Thisfollow-up study sought to explore the factors mediatingsocial interaction during the six AGFS.We present the results of a comparative case study of
the AGFS described in the previous study [8]. The aimwas to understand what factors contribute to the suc-cessful engagement of physicians in change planningand to develop a practical, evidence-informed tool toguide AGF design and implementation.
MethodsEthics approval for this work was received for each casefrom the Conjoint Health Research Ethics Board:REB13-0075 (case 1); REB14-0484 (cases 2a, 2b, 2c, 2d);REB13-0459 (case 3).
SettingThis analysis of the work of the CPLP between 2014 and2016 took place at the Cumming School of Medicine at
the University of Calgary. The CPLP is funded by theprovincial medical association, to deliver audit and feed-back reports about individual practice performance toAlberta physicians. Most projects are initiated when aphysician group (such as a department or clinic) ap-proaches the program with a clinical question. Programstaff clarify the question; if it is amenable to AF, an algo-rithm for data extraction is developed and data isaccessed from relevant provincial repositories to createthe AF report. Each project has a unique, multi-pageaudit and feedback report reflecting the amount of dataand information needed to answer the physician groups’questions. Confidential reports for participating physi-cians contain individual data, anonymous group compar-ators, and relevant references reflecting best practices.The project culminates in a facilitated AGFS in which
consenting physicians have their AF reports (provided atleast 1 week before the AGFS), work as a group with aCPLP facilitator to review them, and identify opportun-ities, barriers and enablers for change [10, 23]. In thisstudy, across all AGFS, the facilitator was the CPLPmedical director, who was not a member of any of thephysician groups. The sessions were attended by CPLPstaff who observed the process and served as projectmanagers for each case from conceptualization todelivery.The process is depicted in Fig. 1.
Study designWe wished to understand why social interaction be-tween physicians in AGFS varied across cases [8]. Com-parative case analysis is an appropriate approach whencontext, culture, and system factors may influence a pro-gram [30, 31]. We used framework analysis to build theindividual cases for this research [32–35]. The overalldesign of the project was as follows: (1) identification ofdata sources, (2) familiarization with data sources, (3)development of a program model or “change theory,” (4)creation of a framework table from the program model,(5) framework analysis to extract and index data to buildindividual case studies, (6) comparisons across cases toidentify similarities and differences believed to influ-ence social interaction, (7) thematic organization ofsimilar findings into key “mediating factors,” (8) cre-ation of an organizing framework linking the mediat-ing implementation and design factors for socialinteraction in AGFS to the conceptual model of phys-ician behaviors [8].
ParticipantsParticipants included the physicians who participated inthe AGFS and the CPLP staff who created the AF re-ports and delivered the AGFS through collaborative
Cooke et al. Implementation Science (2018) 13:136 Page 3 of 18
relationships with the physicians. AF participants pro-vided written informed consent to allow access to theiradministrative health data for purposes of creating AFreports and to record, evaluate, and study the AF ses-sions in which they participated.
Characteristics of staff who contributed are describedin Table 1. The staff worked closely with the physiciangroups to develop the AGFS and observed the sessions.A typical project would take 1 year to complete. TheCPLP staff had extensive longitudinal contact with
Fig. 1 The CPLP process from clinical question to AGFS. Physician groups bring clinical questions of interest for review by the CPLP. The CPLPteam reviews the questions for appropriateness for audit and feedback. Consideration is given to impact, reach, actionability, and accessibilityof the data. CPLP collaborates with data custodians to make individualized AF reports for consenting doctors. The confidential reports includeindividual data with anonymous peer comparators and relevant best practice information. Consenting physicians then participate in a facilitatedgroup feedback session with their peers, led by a CPLP and/or participant facilitator. As a group, the physician peers review each aggregate datapoint, along with their own performance reports and seek opportunities for practice improvement
Table 1 Descriptions of training and roles of CPLP staff and the research team
CPLP staff participants Description
Project managers (2) The two CPLP project managers were experienced in audit and feedback project development and hadformal project management training.
CPLP facilitator (CS) A physician with experience in education and clinical research, who worked at the CPLP for 3 years as theprogram medical director.
Research team members
LC An academic clinician educator with training in medical education research and experience in knowledgetranslation and audit and feedback, who oversaw the CPLP program at the time when these AGFS occurred.
CS See above. CS was the medical director at the time the AGFS were completed.
DD CPLP project and program manager during the time that these AGFS took place.Has training in education and knowledge translation
LR Research associate in CPLP when the AGFS were conducted. Training in epidemiology, quantitative andqualitative methodologies.
SD A physician with training in epidemiology and knowledge translation who became the program medicaldirector after the AGFS were developed.
HA An academic professor of family medicine with training in knowledge translation, medical education andqualitative methodologies with extensive research experience with physician learning and feedback.
Cooke et al. Implementation Science (2018) 13:136 Page 4 of 18
members of the physician groups and familiarity withprocesses and contextual and cultural elements thatwere observed over the course of the projects.Descriptions of the research team are also included in
Table 1 in order to acknowledge their orientation, posi-tions, and perspectives at the outset of this research asthese perspectives likely inform our analysis.It is important to note that while CS, LR, DD, and LC
participated in the development and evolution of theCPLP processes over time, HA and SD were not in-volved in the building of the original projects or evolu-tion of the processes described in this study, but ratherjoined the research team afterwards. Their perspectiveswere routinely sought in an effort to balance and miti-gate any biases or pre-conceived ideas of other teammembers more intimately involved with the cases.
Case definitionFor purposes of this study, a “case” was defined as anAGFS with an individual physician group, including theprocesses of working with the group to refine the ques-tion, prepare the report and coordinate, and facilitatethe AGFS.Cases selected for this study included all AGFS that
took place through the CPLP between January 2015 and
January 2016. These cases were selected because theywere offered in short succession, such that researcherscould begin to understand the respective cultures, pat-terns, and influences within each group.
Data collectionThe research team began by identifying possible datasources from which to derive the six case studies. Datasources are listed and described in detail in Table 2.These included sample AF reports from each case, thetranscripts and qualitative analysis from the prior study[8], basic data about the projects from a CPLP trackingdatabase (number of reports, dates, ethics approvals,etc.), a process evaluation document that was written forcase 1, field notes that were collected directly into ourframework table to capture the observations of theresearch team as they explored the cases, and notesfrom structured interviews with CPLP staff who werepresent at the AGFS to corroborate and validate thefindings of the research team (detailed later in themethods section).Next, the research team built a program model
comprised of possible elements influencing the AGFprojects [30, 31].
Table 2 Description of data sources for the framework analysis and how they were used by the research team
Data sources Description of how data was collected/used
Sample anonymous AF reports for each project The research team reviewed the AF reports and described the quality of data visualization foreach case. Through familiarization with the data from the first study, the team capturedparticipant responses to the reports which could support or refute the team observations aboutthe reports. Observations were noted in the framework table (Exemplar graphs from AF reportsare shown in Figs. 3 and 4).
Process evaluation for case 1 A formal process evaluation of case 1 was conducted by a senior CPLP team member at thetermination of that project. This was a seven-page document outlining processes, procedures,stakeholders, and lessons learned for this project. This report was reviewed by the researchers(LC, DD) information from the report that provided information about influencing socialinteraction in case 1 was added to the case 1 description in the framework table.
Transcripts and qualitative analysis of AGFS An inductive thematic analysis of the transcripts for the six AGFS was conducted in a priorstudy [8]. Team members who reviewed these transcripts repeatedly for the first study madeobservations about the interactivity, collegiality, and change orientation of the groups whichwere included in the case analysis. These observations were collected as “field notes” recordeddirectly into the framework table during research team meetings to discuss the case studies.They were corroborated by returning to review the coding in the transcripts and in interviewswith program staff who were present at those AGFS.
Structured interviews of CPLP staff A staff member who observed each AGFS was interviewed using a the framework table as astructured guide. They were asked to comment about each element in the framework for eachcase. Their responses were captured in notes entered directly into the framework analysis table.Likewise, the facilitator of the six sessions was interviewed and all responses were captured inthe same document.
CPLP tracking document Basic information about each AGFS was captured by the CPLP staff in a tracking documentmaintained by the program. These included key performance indicators such as numbers ofreports distributed, timing of AGFS.
Observations of the research team A consensus meeting of members of the research team (LC, HA, LR, DD) was held to shareand compare observations of the AGFS and AGF projects. These observations were capturedand noted directly into the framework table. This content was reviewed iteratively during thecase analysis and during the development of the CAFF to ensure accuracy and consensus aboutthe findings for each case that was analyzed. Observations of the research team were triangulatedwith the other data sources for corroboration.
Cooke et al. Implementation Science (2018) 13:136 Page 5 of 18
Building the program modelDeveloping a program model is an important early stepin a comparative case analysis [30, 31]. The model iscomprised of elements that are expected to influence thecases [30, 31]. The research team worked collaborativelyover several meetings to diagram the program modelbased on literature that describe factors that influenceimplementation success and acceptance of feedback [7,17, 23–27]. Based on the team’s tacit knowledge derivedfrom their collected experience in developing and deliv-ering audit and feedback and educational feedback aswell as their familiarity with the literature, additional ele-ments not specified in published frameworks were addedto the program model. The team met, drafted, andre-drafted the model iteratively until there was consen-sus on the likely factors influencing social interaction inAGFS. This program model, described in the “Results”section, was used as the a priori framework for collect-ing and organizing information from all of the datasources (Table 3).The program model became the a priori framework
into which all data was later indexed in order to buildeach case study. The framework was put into a table(the “framework table”) with the elements of the frame-work on the vertical axis and the cases on the horizontalaxis.Three 60-min interviews structured upon the a priori
framework comprised one source of data about thecases. The interview participants (the CPLP medical dir-ector and two CPLP project managers, detailed inTable 1) were asked to describe their observations of thecases with respect to each of the framework compo-nents. Their responses were recorded directly into theframework table. They were used to corroborate and/oradd to the observations and findings of the research
team. The notes from the interviews were reviewed, con-densed, and reviewed again by the research team andsummarized in the case analysis table for purposes ofthe publication.
Data analysisFramework analysis was the primary mode of data ana-lysis in this project [32–35]. This is a method of organiz-ing textual data according to an a priori “coding”framework and is commonly used in social sciences re-search [32–35]. The first step involves familiarizationwith the data through iterative reading and discussion.Familiarization with the AF reports and the qualitativedata from the prior study took place during project de-velopment within the CPLP and in conducting the priorstudy [8]. Next was developing the program model and apriori framework as outlined in the previous section.Data were indexed in the framework table by searchingthe data sources for evidence for each element of the apriori framework in each case and populating the tablewith findings. If observations from the data sources didnot align with the framework, there was opportunity toadd additional elements.The research team then met to review the case studies
and develop consensus on the case descriptions. Corrob-orating evidence for the observations were sought by tri-angulating the evidence from different data sources.The next step in data analysis was the case compari-
son. Once consensus was reached on the case descrip-tions, the research team worked collaboratively toidentify similarities and differences across the cases.Each of these was considered by the team as to whetheror not they influenced the social interactions in the re-spective AGFS.
Table 3 Program model for the comparative case analysis
Program model elements Components
Innovation [17] The clinical question upon which the project was based
Style of report [7] Design of the report, co-creation of report design with CPLP and physician leadsfrom participating groups
Available gold standard/benchmark for the clinical question [7]
Recipients [17] Who participants were and how they interacted with one another [7, 17, 23]
Context [17] Project origin/historyGroup dynamic (collegiality/culture) and leadership involvement in the sessions [23]Pre-work/relationship building between PLP and participant group [25]Co-creation of data/metrics [7]Proximity of doctor to data/patient (perceived control) [23]
Facilitation [17, 24–26] CPLP led or co-facilitated with a participant physician-leadCoaching-oriented facilitation [25]
Physician engagement/change talk/change planning [8] Interactivity of the AGFS, extent of change talk
Outputs Further project development with CPLPEmergence of autonomous data reporting by the physician group (dashboards)Measured/reported behavior change by the physicians
Cooke et al. Implementation Science (2018) 13:136 Page 6 of 18
The final step in the analysis was to create an organizingframework. The research team iteratively diagramed aframework that captured and organized factors that wereidentified as likely influences on physician engagement inthe AGFS. Similar elements were grouped thematicallyunder a single “mediating factor” for AGFS success. Eachmediating factor was then ordered to reflect the CPLP pro-cesses and the role these processes might play in the phys-ician behaviors identified in the previous study [8].
ResultsParticipantsA total of 99 AF reports were distributed and 50 physi-cians participated across the six AGFS. Two CPLP pro-ject managers and the CPLP Medical Director wereinterviewed for data collection.
Program modelThe program model, shown in Table 3, is comprised ofeight elements. These elements were derived from twowidely cited frameworks that describe influencing factorsin implementation [17, 23], Brehaut’s recommendationsfor enhancing “practice feedback”, and elements fromseveral models of educational feedback [7, 24–27]. Inaddition to the factors derived from the literature, theteam added these elements: AGFS interactivity andchange talk and group dynamic.
Case descriptionsCase 1 was a project exploring practice variation withspecialists involved in a specific surgical procedure. Thegroup demonstrated a collegial relationship during theAGFS with a high frequency of sharing and reflecting oncommon practices and raising change cues which led tochange talk between the participants during the AGFsession. Representative examples are listed in Table 4.Participants in case 1 were familiar with the CPLP ap-
proach and staff from a prior project with the program.Their AF reports were co-created with input from thephysician lead and three other group members. Thephysician lead was a highly respected leader in thegroup. These “group champions” were involved with allaspects of project design from the clinical question tothe AGFS. The project lead shared their individual datawith the group while co-facilitating the AGFS. Thisgroups’ clinical question related to treatment decisionsthat were largely under their direct control.Cases 2A–D addressed a de-prescribing question de-
rived from Choosing Wisely Canada recommendations.These cases took place at four adult hospitals, each withseparate groups of physicians, leaders, institutional cul-tures, and serving unique patient populations in differ-ent parts of the city. For these reasons, these AGFS aretreated as separate cases. These physician groups were
reviewing baseline data. While there were varying de-grees of change talk and interactivity across these fourAGFS, it was uncommon across all four AGFS for physi-cians to compare and share their practices with oneanother.Case 2A occurred at a hospital serving an area with a
large immigrant population. This physician group alsoprovided palliative care. The AGFS focused heavily ondiscussion about the uniqueness of the patient popula-tion. There was skepticism about the data on the basisof the uniqueness of this group’s patient population ac-counting for the variations in prescribing. The projectlead was from this group but was unable to attend thissession. The group discussion was interactive and somechange cues arose (see Table 4).Case 2B took place in a very large tertiary academic
health center serving a complex, high-acuity population.The group was comprised of physicians who worked to-gether for many years. The session was interactive andled to change planning. A nurse with a formal qualityimprovement role attended this session. A physician par-ticipant prompted several interactions around changeplanning with the allied health staff who sharedin-patient care.Case 2C took place in a smaller, community-based
hospital serving a predominantly elderly population.One of the physicians for the project was based withinthis group, attended the session, helped facilitate, andshared their data with the group. This session was lessdynamic than cases 2A and 2B. During the AGFS, therewere few instances of comparing and sharing practiceshowever, two change cues were raised and led to changetalk around improving documentation and dischargesummaries as a way to improve community prescribing.Case 2D took place in a newer, smaller hospital on the
outskirts of the city. Because of its short history, thishospital has a unique physician culture and medical cul-ture with a focus on patient and community-centeredcare. Participants expressed that they believed that themake-up of their group and their consultant colleaguesinfluenced their prescribing behaviors (Table 4).This session was the least interactive of our six cases;
physicians did not raise change cues, share their practices,or discuss making changes. The facilitator posed few ques-tions to the group in this session, and few questions wereraised beyond trying to understand the data that was pre-sented. Participants pointed out that the AF report wasquite difficult to interpret; something that was reflectedacross cases 2A–D in the amount of time spent question-ing the facilitator to understand the data during theseAGFS (see Table 4). The facilitator expressed that thisgroup did not seem “interested” in the AGFS.Case 3 was a project with a group of specialists ad-
dressing broad questions about practice variation in
Cooke et al. Implementation Science (2018) 13:136 Page 7 of 18
Table
4Represen
tativeexam
ples
ofdata
that
wereused
inbu
ildingthecase
stud
ies.In
thistable,theway
inwhich
raw
data
was
inde
xedto
thecompo
nentsof
theprog
ram
mod
elisshow
n.Whe
rerelevant,h
owthisdata
was
linkedto
thekeyelem
entsof
theCAFF
mod
elisalso
show
nin
orde
rto
demon
strate
theway
inwhich
inferences
were
draw
nbe
tweentheoriginalraw
data,the
prog
ram
mod
el,and
,ultimately,theframew
ork
CAFF
mod
elelem
ent(overarching
them
e)Com
pone
ntof
prog
ram
mod
elExam
plefinding
sfro
mcase
analysisextractedfro
mtranscrip
tsRepresen
tativequ
otations
Facilitation:
Facilitator
encouraged
andcreatedspacefor
thesechange
cues
tobe
raised
anddiscussed
bygrou
p.
Physicianen
gage
men
t/change
talk/chang
eplanning
[8]
Ahigh
degree
ofsocialinteractionin
case
1.Data
captured
from
originalAGFS
transcrip
tsReflectingandsharingof
practices
arou
ndusingintraven
ous
anesthesia:
“Ifindthat
thedistinctionbe
tweende
epandaw
akehas
becomeblurredwith
TIVA
.Ijustturn
downtheTIVA
andjust
extubate
them
andthey’re
almostaw
akebu
tthey’re
notas
deep
asforgasextubatio
n.Iu
suallycallitde
epbu
tit’sno
treallyade
epextubatio
nin
thetruestsense…
itkind
ofis
halfw
ayin
betw
een.”
Achange
cuefro
manothe
rparticipant:
“Our
bigg
estprob
lem
ispo
st-oppain”
Thischange
cueledto
adiscussion
betw
eentheparticipants
abou
tstrategies
forim
provingpain
managem
entwith
out
worsening
nausea
andvomiting
.
Facilitation:
Facilitator
encouraged
andcreatedspacefor
thisdiscussion
tooccur.
Physicianen
gage
men
t/change
talk/chang
eplanning
[8]
Case2A
Datacaptured
from
originalAGFS
transcrip
tsAparticipantexpressedtheird
esire
toimprovetheirp
rescrib
ing
habits:
“…Letus
say3calls
that
Iget
atnigh
tarerequ
ests,thispatient
wantsasleeping
pill.Im
ight
besuccessful
in1in
3in
convincing
thepatient
that
nothey
dono
treallyne
edthis.O
rconvincing
thenu
rse,no
they
dono
treallyne
edthis.M
aybe
I’llb
esuccessful
with
1on
3.Bu
tits’w
orth
trying
.”Thischange
cueledto
adiscussion
arou
ndthene
edforthe
grou
pto
developconsen
suson
de-prescrib
ingthecompo
unds
inqu
estio
n.
Recipien
tsandcontext
Case2D
:Theph
ysiciangrou
pin
thisho
spitalw
asyoun
gand
they
believedtheirgrou
pmake-up
andrelatio
nships
with
consultantsat
that
site
influen
cedtheir
prescribinghabits.
Datafro
moriginalAGFS
transcrip
ts
“Wearearelativelyyoun
gergrou
pcomparedto
some...And
Ithinkalotof
ustryto
utilize…
restraintin
antip
sychotic
prescriptio
ns.W
ehave
hadgo
odteam
sinvolved
with
alotof
our
patientsandwetryandavoidalotof
that.”
Datarepresen
tatio
nStyleof
repo
rtCase2D
:Participantsraised
concerns
abou
tthe
complexity
andcogn
itive
load
intheirAFrepo
rts.
Datafro
moriginalAGFS
transcrip
ts
“But
whe
nyoufirstlook
atthedo
cumen
t…itisvery
difficultto
unde
rstand
alotof
it,be
causethere’salotof
inform
ation.”
Exam
ples
ofdata
extractedfro
minterviewswith
CPLPStaff
Represen
tativequ
otations
Relatio
nshipbu
ildingandqu
estio
nchoice
Recipien
ts,inn
ovation,and
context
Whe
naskedabou
tho
wtheprojectoriginated
and
nature
oftheinno
vatio
n(case1):
CPLPfacilitator:“Manyof
thesedo
cshadbe
eninvolved
in[our
first]project,bu
tthereweresomeyoun
gerrecent
grads
who
weremoreinvolved
inchoo
sing
questio
ns—NAMEwas
seen
asastrong
leader,g
roup
hadbo
ught
into
hisvision
ofed
ucationin
gene
raland
theidea
ofauditandfeed
back.M
aybe
thegrou
pge
nerally
hasacultu
reof
mon
itorin
ganddata
collection—
…no
tabigsellto
thisgrou
p–they
werekeen
,thou
ghtful,interested,
andacadem
icin
theirthinking
….they
hadtheadvantageof
having
alreadybe
enthroug
htheprocess.
Lotsof
grou
ndworkdo
neby
ourteam
working
with
theclinic
Cooke et al. Implementation Science (2018) 13:136 Page 8 of 18
Table
4Represen
tativeexam
ples
ofdata
that
wereused
inbu
ildingthecase
stud
ies.In
thistable,theway
inwhich
raw
data
was
inde
xedto
thecompo
nentsof
theprog
ram
mod
elisshow
n.Whe
rerelevant,h
owthisdata
was
linkedto
thekeyelem
entsof
theCAFF
mod
elisalso
show
nin
orde
rto
demon
strate
theway
inwhich
inferences
were
draw
nbe
tweentheoriginalraw
data,the
prog
ram
mod
el,and
,ultimately,theframew
ork(Con
tinued)
CAFF
mod
elelem
ent(overarching
them
e)Com
pone
ntof
prog
ram
mod
elExam
plefinding
sfro
mcase
analysisextractedfro
mtranscrip
tsRepresen
tativequ
otations
staff,daycarestaff,data
analyst,etc...lotsof
face
timefro
mPLP
onsite—seem
edto
behe
lpfulw
ithbu
yin.”
Relatio
nshipBu
ilding&QuestionCho
ice
(thisgrou
pwas
notdirectlyinvolved
inco-creationof
questio
n)
PhysicianEngage
men
t/Chang
eTalk/Chang
ePlanning
[8]
Whe
naskedabou
tgrou
pdynamic(case2D
):Facilitator:“Po
orparticipationfro
mthegrou
p.Thegrou
pcultu
reat
thisho
spitalseemed
different
from
theothe
rgrou
ps…very
silent
durin
gthesession…
notvery
manyshow
edup
..for
their
repo
rts”.Project
manager
1:“[T
heparticipantswere]
youn
g.Did
they
feelun
comfortablespeaking
up?”
Questionchoice
Inno
vatio
nWhe
naskedabou
tnature
oftheprojectcase
3:Projectmanager
2:“The
goalwas
hazy…very
broad—
things
that
werecommon
”.
Relatio
nshipbu
ilding
Recipien
tsandcontext
Describingtherecipien
ts(case3):
Facilitator:“Thisgrou
pissupe
ren
gage
d.Abu
nchof
XXXreside
ntswho
may
have
been
frien
dsbe
fore,or
becamefrien
dsafterw
ards—they
areallabo
utthesameage,
early
millen
ialsmostly
allo
nthesamepage
.The
grou
pwas
onthesamepage
.NAMEisacompe
tent,q
uiet
leader
…very
practical,d
ownto
earth,that’swhy
NAMEwas
askedto
dothis.
Thegrou
pistig
htbe
causethey
coverforon
eanothe
rand
cross-cover.Themeanageof
thisgrou
pwas
particularly
youn
g–senseof
buyin
todata,com
puters,d
igitaln
atives.H
admanysugg
estio
ns…cameup
with
atleasttenpo
tentialp
rojects.
They
werekeen
–lotsof
sharingof
oneanothe
r’sdata
inthe
sessions”.
Cooke et al. Implementation Science (2018) 13:136 Page 9 of 18
relation to anesthesia for various surgical procedures. Inthis AGFS, there was a high degree of interactivityaround change cues which led to change planning, andalso comparing and sharing of practices. Participantsraised ideas about additional rounds and education, im-provements in charting, and working with other system“actors” to foster improvements in care. The participantswere young and very collegial with one another. Severalof them had reportedly trained together. The medicallead for the project co-created the questions with theCPLP medical director, helped with report design,pre-circulated relevant journal articles to the participantgroup, and shared their data with the group during thepresentation.
Comparative case analysisThe results of the comparative case analysis are summa-rized in Table 5. While the authors recognize that phys-ician behavior change was the goal of undertaking theAGF projects, the focus of this study was the factors in-fluencing whether physicians engaged with the data andchange planning during the AGFS. Thus, the “success”of the individual sessions was gauged based on the fol-lowing three criteria: (1) The physicians engaged in thegroup discussion about the data, (2) Change cues thatarose were followed by change talk, and (3) Further ac-tion was taken based on the AGFS to address identifiedgaps. There were numerous additional outputs thatarose following the AGFS studied, and these are cap-tured in Table 3 but not elaborated here as they are out-side the scope of this paper.In comparing across cases, the most change talk and
planning occurred during cases 1, 2A, and 3. Compara-tive case analysis highlighted several key elements foster-ing physician engagement with the data in these groups.These included the pre-existing relationship between theCPLP and the physician group (case 1), the active in-volvement of a physician leader from within each AGFgroup (cases 1, 2A, and 3), co-creation (between groupmembers and CPLP) of the clinical questions and AF re-port designs (cases 1 and 3), perceived control over thebehaviors being measured in the AF report (cases 1 and3), intrinsic group dynamic or cohesiveness (cases 1, 2A,2B, and 3), and the approach of the facilitator leadingthe sessions (co-facilitation in cases 1 and 3 andcoaching-oriented facilitation with prompts and ques-tions in cases 1, 2A, 2B, and 3).The physicians’ tendency towards sharing and compar-
ing practices during AGFS were greater in AGFS 1 and3, which were based on clinical questions about practicevariation, than they were in cases 2A–D. The datavisualization for the AF reports in cases 2A–D wasoverly complex. During AGFS 2A–D, a disproportionateamount of time was spent by the participants on
understanding the reports. Participants in AGFS 2Dcommented specifically on this finding. In addition, theextent of interactivity in AGFS 2A-D diminished witheach successive case. In reviewing the transcripts, and ininterviewing the facilitator and CPLP staff who observedthe sessions, it became apparent that the facilitation stylechanged across those four cases. In cases 2A and 2B, thefacilitator asked many questions of the participants,prompting them to consider their data. In contrast, incase 2D, the facilitator asked very few questions and theparticipants made few inquiries about the data. Indeedthe transcriptionist for the AGFS described the AGFSfor 2D as a “lecture rather than a focus group” aftertranscribing the session.
Developing an organizing frameworkIn reviewing the similarities and differences across thecases, and grouping similar items, four overarchingthemes emerged: relationship building, question choice,data visualization, and facilitation. In organizing these ina way that reflected both the processes used by theCPLP and the observed physician behaviors in AGFS, itwas possible to link the design/implementation factorswith the cycle of predictable physician behaviors [8] tocreate a practical framework for AGF project planningand design. The Calgary Audit and Feedback Framework(CAFF) is shown in Fig. 2.
DiscussionThe audit and feedback literature consists of many pub-lished interventions demonstrating that AF can be a use-ful means to change physician behavior [1]. However, ithas been pointed out that there are gaps in the designand implementation of AF and that these may accountfor some of the variability in the effectiveness of pub-lished interventions [1, 3–7].The CPLP developed a novel way to deliver AF that
aimed to address some of the design and implementa-tion elements emerging from the AF and educationalfeedback literature: AGFS [4–7, 9, 24–27]. Recognizingthat social learning was a key ingredient in pivotingAGFS towards change planning, we wished to explorewhy some AGFS were more socially interactive thanothers [8, 20].In this study, we present the results from a compara-
tive case analysis of six AGFS and based on thesefindings, we propose a practical, evidence-informedframework for the development, and implementation ofAGFS. We have termed the product of this research a“framework” based on Nilson’s definition:“Frameworks in implementation science often have
a descriptive purpose by pointing to factors believedor found to influence implementation outcomes.”([36], p 2).
Cooke et al. Implementation Science (2018) 13:136 Page 10 of 18
Table
5Summaryof
results
ofcomparativecase
analysisforsixauditandfeed
back
sessions
with
different
physiciangrou
ps.The
prog
ram
mod
elelem
ents,w
hich
form
edthe
framew
orktableused
fordata
analysisandinde
areon
theverticalaxisof
thistableandthecasesarelistedacross
theho
rizon
tal.Theresults
presen
tedbe
low
are
summarized
,for
purposes
ofthepu
blication,
from
thede
tailedno
tatio
nscaptured
intheoriginalframew
orktables.The
data
sourcesfro
mwhich
thedata
wereextractedare
describ
edin
Table1
Prog
ram
mod
elelem
ents
Case1
Case2a
Case2b
Case2c
Case2d
Case3
Inno
vatio
nReview
ofpractice
variatio
nandou
tcom
esof
anesthesiafor
proced
urex
Review
ofprescribingpractice
variatio
nformed
ications
inpo
pulatio
nxat
hospitalA
Review
ofprescribing
practicevariatio
nfor
med
ications
inpo
pulatio
nxat
hospital
B
Review
ofprescribing
practicevariatio
nfor
med
ications
inpo
pulatio
nxat
hospitalC
Review
ofprescribing
practicevariatio
nfor
med
ications
inpo
pulatio
nxat
hospitalD
Survey
ofpracticevariatio
nin
anesthesiafor5proced
ures
atHospitalA
.Asurvey/overview
ofpracticetype
ofprojectin
prep
arationforamore
subseq
uent
specificproject.
Intent
tointrod
ucethegrou
pto
A&F
Forthisproject,A&F
was
onecompo
nent
ofamultifaceted
interven
tion:Itwas
preced
edby
adidacticinter-
profession
al(doctors,nurses,ph
armacists)con
tinuing
med
icaled
ucationsession,andsubseq
uentlyinclud
eda
commun
icationcampaignandchange
sto
physicianorde
ren
trysets,w
hich
stem
med
from
thisseriesof
AGFS
(cases
2A–D
)
Styleof
auditrepo
rtHighlyrefined
with
heavygrou
pinpu
tTherepo
rtsacross
cases2A
–Dpresen
tedthesameaggreg
atedata,b
uthadindividu
alph
ysiciandata
forthe
participatingph
ysicians
ateach
site.Ineach
ofthesefour
AGFS,a
sign
ificant
amou
ntof
timewas
spen
twith
the
participantsasking
questio
nsto
tryto
unde
rstand
thedata
asitwas
presen
ted
Engage
dgrou
plead
contrib
uted
substantially
torepo
rtde
sign
One
ofthetw
oprojectleads
was
from
thissite
andhe
lped
torefinethequ
estio
nforthe
AFrepo
rts
One
ofthetw
oproject
leadsforthisworkcame
from
thissite
andhe
lped
toco-createandrefine
thequ
estio
nfortheAF
repo
rts
2DParticipantscommen
ted
durin
gtheAGFS
that
the
repo
rtwas
difficultto
interpretdu
ringtheAGFS
Goldstandard
for
prop
osed
best
practice
Non
eNo,bu
tseveralexistinggu
idelines
andrecommen
datio
nsNo,bu
texistin
geviden
ceand
guidelines.
Recipien
tsSpecialistsat
hospital1.
18ph
ysicianrepo
rts,13
physicians
attend
edfeed
back
sessions
Gen
eralistph
ysiciangrou
p,nu
rsemanagers,QIleadat
hospitalsA,B,C
,and
Drespectively(64ph
ysicianrepo
rts,28
physicians
attend
edfeed
back
sessions)
Specialistgrou
p,ho
spitalD
17ph
ysicianrepo
rts,9
physicians
attend
edfeed
back
session
Participantsin
thisgrou
piden
tifiedthat
theirpatient
popu
latio
ninclud
edho
spice
patientsandfeltthat
this
influen
cedprescribing
behaviors
Participantsin
this
grou
pcaredforvery
high
volumeservices
with
high
degreesof
acuity
andcomplexity
Participantsat
thissite
iden
tifiedthat
they
care
foran
olde
rpatient
popu
latio
n.Thissite
had
access
toage
riatric
psychiatry
servicewhich
may
have
influen
ced
prescribingpatterns.The
ywerepo
sitivelyoriented
,cohe
sive,non
-judg
men
tal
Participantsat
thissite
self-
iden
tifiedas
being
“you
nger”andfeltthat
their
prescribingbe
haviorswere
influen
cedhe
avily
bytheir
training
andby
thelocal
consultantswith
who
mthey
collabo
rated
Con
text
Project
origin
Asecond
projectwith
CPLPforthisgrou
pAqu
estio
nraised
bycity-w
ideInno
vatio
nCom
mittee
who
invitedCPLPto
developtheAFrepo
rtsandde
liver
AGFS
across
the4sites
CPLPinvitedby
thisgrou
pafterthey
learne
dabou
tcase
1
Group
dynamicand
leadership
involvem
ent
fortheA&F
Expe
rienced
,respe
cted
senior
physicianwith
strong
commitm
entto
profession
alde
velopm
ent.
One
projectlead
was
amem
berof
thisgrou
pbu
twas
notableto
attend
thissession.
Thesite
lead
was
presen
tbu
tdidno
thave
facilitator
rolein
Site
lead
presen
t.Lotsof
questio
ning
ofdata
and
discussion
abou
tho
wthedata
was
analyzed
.Site
lead
expressed
One
ofthetw
oproject
leadswas
from
thissite.
Somego
oddiscussion
insession,bu
tCPLPstaff
notedthat
participants
Youn
gph
ysiciangrou
pin
ane
who
spitalw
orking
in2weekshiftswith
lotsof
hand
over.Poo
rattend
ance,
few
questio
nsat
session.
Very
motivated
,dynam
icsite
projectlead
with
anadmin
positio
nfored
ucationwith
inthegrou
p.Ayoun
gph
ysician
coho
rt,d
igitally
literate,op
en
Cooke et al. Implementation Science (2018) 13:136 Page 11 of 18
Table
5Summaryof
results
ofcomparativecase
analysisforsixauditandfeed
back
sessions
with
different
physiciangrou
ps.The
prog
ram
mod
elelem
ents,w
hich
form
edthe
framew
orktableused
fordata
analysisandinde
areon
theverticalaxisof
thistableandthecasesarelistedacross
theho
rizon
tal.Theresults
presen
tedbe
low
are
summarized
,for
purposes
ofthepu
blication,
from
thede
tailedno
tatio
nscaptured
intheoriginalframew
orktables.The
data
sourcesfro
mwhich
thedata
wereextractedare
describ
edin
Table1(Con
tinued)
Prog
ram
mod
elelem
ents
Case1
Case2a
Case2b
Case2c
Case2d
Case3
sessions
Senior
physicianand3
colleaguesinvolved
inallaspectsof
project.
Coh
esive,collegial
grou
p.Senior
physicianand1
colleague
ledsession
andshared
owndata.
session.Non
ethe
less,C
PLP
staffiden
tifiedthat
thesite
lead
was
“astrong
lead”and
contrib
uted
tothesession
meaning
fully.Lotsof
questio
ning
anddiscussion
abou
twhy
thissite
might
bedifferent
vsothe
rs,skepticism
regardingdata,req
uestsfor
data
that
wereno
tinclud
edin
originalproject.
CPLPmed
icaldirector-led
session
concernre
potential
defensiven
essof
grou
ppriorto
AGFS.G
ood
attend
ance.
CPLPmed
icaldirector-
ledsession
begansharingand
comparin
gdata
asthe
CPLPteam
lefttheroom
.Goo
dattend
ance.Project
lead
shared
owndata.
CPLPmed
icaldirector-led
session
CPLPmed
icaldirector-led
session
toself-reflection,keen
todiscusschange
andun
derstand
finding
s.Group
cohesio
nand
collegiality
high
,inpartbecause
ofcallstructure.
CPLPmed
icaldirector-led
session.Ph
ysicianproject
helped
prom
ptgrou
pdiscussion
andleddiscussion
arou
ndeviden
ce
Pre-work,
relatio
nship-
building
Thiswas
asecond
projectforthisgrou
p.Lotsof
interactionwith
CPLPteam
inproject
developm
ent
Thiswas
afirstprojectforeach
ofthefour
hospitalg
roup
s.Therewereseveralm
eetin
gswith
thetw
oprojectleads
(From
cases2A
and2C
)and
ameetin
gwith
thecity-w
ideQIg
roup
(atsite
2B),bu
tno
twith
grou
pmem
bers
Firstcontactwas
byinvitatio
n:CPLPpresen
tedto
theen
tire
physiciangrou
pto
introd
uce
concep
tof
A&F
prog
ram.
Projectlead
met
frequ
ently
with
MDgrou
pandCPLP
team
Co-creatio
nof
data/
metrics
Aworking
grou
pwith
astrong
lead
andthree
departmen
tmem
bers
andinpu
tfro
mMD
grou
pde
velope
dmetricsandrepo
rt
Thiswas
onepartof
alarger
project.Individu
alph
ysicians
ateach
site
didno
tchoo
sewhatqu
estio
nwou
ldbe
addressed;
however,the
questio
nwas
brou
ghtto
theCPLPby
thetw
oprojectleads(From
sites2A
and2C
)and
was
aligne
dwith
aprovincialinitiativeto
de-prescrib
ein
thispo
pulatio
n.Theprojectleadshe
lped
theCPLPteam
tode
term
inewhich
data
pointsto
build
into
therepo
rts
Strong
leadership
from
site
lead
andcollabo
rative
developm
entof
project
betw
eensite
lead
andMD
grou
pmem
bers
Proxim
ityof
doctor
todata
and
patient
Direct
physicians
administerin
gorde
rsat
bedside
In-direct:A
dmittingph
ysicianwas
notalwaysattend
ing,
manyorde
rswerePRNandat
discretio
nof
nursingstaff.
MDsdidno
tfeelthey
hadgo
odcontrolo
vernu
rsediscretio
nno
rorde
ren
trysystem
Direct,asin
case
1,andMDs
haddirect
access
todatabase
design
ersfordata
capture/
metrics
Facilitationof
Session
Co-ledby
CPLPand
grou
pcham
pion
who
presen
tedow
ndata
togrou
p.Prom
ptingand
questio
nsfro
mCPLP
andfro
mco-facilitator
CPLPfacilitator,w
howas
aph
ysicianbu
tno
tagrou
pmem
ber,ledeach
session.Facilitationfollowed
structureof
therepo
rt,tableby
table.Forcase
2C,exemplar
data
from
thesite
lead’srepo
rtwas
presen
tedbu
tthesessionwas
stillbasedon
structureof
therepo
rt.Som
ede
gree
ofinpu
tfro
msite
lead
occurred
atsites2A
,B,b
utno
tat
site
Cor
D.The
facilitationstyleat
siteDwas
different
from
theprevious
sessions.Thissessionwas
largelyadidactic
presen
tatio
nfro
mthefacilitator
tothegrou
p.Thiswas
thelastof
four
sessions
inthisproject.Thefacilitator
asked
fewer
questio
nsthan
inothe
rgrou
ps.W
heninterviewed
,the
facilitator
describ
edthefeelingthat
thisgrou
pwas
‘not
asinterested
’intherepo
rtandthat
attend
ance
atthissessionwas
poor
Ledby
CPLPfacilitator
with
substantiveinpu
tfro
mproject
lead.Promptingqu
estio
nsfro
mbo
thCPLPfacilitator
and
projectlead
Engage
men
t/change
talk/actionplanning
Very
interactive.Lotsof
sharingof
person
alpractices
andplanning
arou
ndadditio
nal
change
activities
Mod
erateinteraction.Some
change
cues
wereraised
,leadingto
plansforchange
,andsomesharingof
practices
occurred
Mod
erate.Aparticipant
spon
tane
ouslyraised
need
forchange
and
thisledto
considerable
discussion
ofho
wthis
couldbe
achieved
Minim
al.The
rewas
little
change
discussion
durin
gthisAGFS,b
utparticipants
didshareandcompare
practices
tosomeextent
Minim
al.The
grou
pasked
questio
nsrelatedon
lyto
unde
rstand
ingthedata.N
ochange
cues
arosein
this
AGFS
Very
dynamicsessionwith
lots
ofqu
estio
nson
eviden
cedirected
totheprojectlead
anddiscussion
betw
eengrou
pmem
bers
Cooke et al. Implementation Science (2018) 13:136 Page 12 of 18
Table
5Summaryof
results
ofcomparativecase
analysisforsixauditandfeed
back
sessions
with
different
physiciangrou
ps.The
prog
ram
mod
elelem
ents,w
hich
form
edthe
framew
orktableused
fordata
analysisandinde
areon
theverticalaxisof
thistableandthecasesarelistedacross
theho
rizon
tal.Theresults
presen
tedbe
low
are
summarized
,for
purposes
ofthepu
blication,
from
thede
tailedno
tatio
nscaptured
intheoriginalframew
orktables.The
data
sourcesfro
mwhich
thedata
wereextractedare
describ
edin
Table1(Con
tinued)
Prog
ram
mod
elelem
ents
Case1
Case2a
Case2b
Case2c
Case2d
Case3
includ
ingen
gage
men
tof
nursingstaffand
makingacollective
commitm
entto
de-
prescribing
Other
outputs
Ledto
7subseq
uent
projectswith
same
specialty
indifferent
settings
with
different
inno
vatio
nsandto
developm
entof
continuo
usrepo
rting
platform
foralld
octors
inthisspecialty
ThisAGFS
occurred
aftertheed
ucationcampaignthat
was
apartof
thelarger
initiative.Therewas
nochange
inprescribingfro
mbaselineto
thisAGFS.H
owever,1
year
afterthisAGFS,the
rewas
arelativeredu
ctionin
prescribingover
allfou
rsitesof
39%
durin
gstud
ype
riod.
Engage
men
tof
grou
pin
afollow-upprojectlookingat
anothe
rinno
vatio
narou
ndinsulin
prescribing,
also
inconjun
ctionwith
alarger,health
-autho
rityinitiated
educationalintervention.
Chang
esin
orde
r-sets,eng
agem
entof
multi-disciplinaryteam
s
Specificfeed
back
infollow-up
from
onegrou
pmem
berwho
change
dop
ioid
practice
subseq
uentlyafterlearning
abou
tpracticevariatio
nvs
othe
rmem
bers.Led
tofollow-
upprojectswith
samegrou
plookingat
othe
rclinical
questio
nsandultim
ately
additio
nalp
rojectsin
city-w
ide
departmen
t(see
case
1)
Cooke et al. Implementation Science (2018) 13:136 Page 13 of 18
In the Calgary Audit and Feedback Framework(CAFF),the key findings from the case analyses were groupedinto four “factors” and aligned with the conceptualmodel of physician behaviors to demonstrate linkagesbetween the design and implementation elements andthe desired progression of the physicians towards changeplanning in AGFS (Fig. 2).
The Calgary Audit and Feedback FrameworkThe intent in developing the CAFF was to provide aconcise way to organize our findings and to delineatehow the various design elements could drive physicianstowards planning for change in a socially constructedlearning environment (Fig. 2). The first two factors ofthe framework, relationship building and questionchoice, help physicians in AGFS overcome potential bar-riers to acceptability of the feedback (for example:skepticism, mistrust, non-actionable feedback). Qualitydata representation is needed to facilitate understandingand interpretation of the data. Facilitation as a mediatingfactor is dependent on the trusting collaborative relation-ships between the feedback providers and recipients andhelps participants move from understanding the data tochange planning. Each of the identified mediating factorsare supported by existing literature and can be used to
help us understand how social interaction drives physi-cians towards change planning.
Building relationshipsBing-You et al. identified that feedback recipients in-corporate feedback into their learning inconsistently[37]. This may be mediated by failure to create an “edu-cational alliance” between the providers and recipientsof feedback, the credibility and constructiveness of thefeedback, and/or the nature of the relationship and trustbetween the providers and the feedback recipients [25–28]. To mitigate skepticism and enhance feedback ac-ceptability, the primacy of establishing respect and trustbetween the “provider” and “recipient” of in-person feed-back cannot be over-emphasized [25, 27–29].We found that when present, skepticism about the
data was a barrier to moving physicians in AGFS to-wards change talk. Groups who had prior experiencewith the CPLP program had a working relationship withour team and appeared to engage with their data andthe facilitator readily.Key elements of relationship building in our program
include empowering a group member to co-facilitate theAGFS, clarifying use of and the confidential nature ofthe AF reports, and the data limitations. These elements
Fig. 2 The Calgary Audit and Feedback Framework (CAFF) for the design and delivery of AGF. This model organizes the key findings from ourcase analysis that were identified as important drivers in moving physicians through the cycle of discrete behaviors that occur in AGFS towardsthe end goal of planning for change. Under each “mediating factor” are listed the distinguishing elements between the cases with more or lesssocial interaction that emerged from the comparative case analysis. The framework is linked to the conceptual model of physician behaviors inAGFs to show how the different mediating factors drive the behaviors towards change
Cooke et al. Implementation Science (2018) 13:136 Page 14 of 18
necessitate advance planning and direct contact betweenthe individuals developing the AGF project and the pro-spective AGF recipients in order to forge an “educationalalliance” [27]. We suggest the successful creation of thisalliance is reflected in the subsequent engagement of theparticipating groups in additional projects with theCPLP. Co-creation also appeared to be a critical compo-nent in the development of the educational alliance:Cases where co-creation was emphasized in project de-sign had the most interactive AGFS.
Question choiceA key contributing factor to intervention effectiveness isensuring that the intervention selected is appropriate tothe desired behavioral change [10, 11, 19]. TheKnowledge-to-Action Cycle and the “Behavioural ChangeWheel” provide guidance for the choice of interventionfor fostering change depending on the nature of the de-sired behavior change [18, 19].It follows that not all questions about physician per-
formance are appropriately dealt with by using AF inter-ventions [7, 10, 11, 19, 23]. Choosing metrics over whichphysicians have little control or for which there is no goldstandard is unlikely to result in feedback acceptance [7].Brehaut and others emphasize the importance of
actionability of the feedback provided in AF [7, 10, 19,26]. Watling et al. found that perceived “constructive-ness” of feedback mediates its acceptance by learners tosome extent, and we suggest that the perceived “action-ability” described in the AF literature [7] and perceivedconstructiveness of educational feedback are similar con-structs [7, 26]. Likewise, the availability of best practiceevidence or gold-standards in addition to anonymizedpeer data with which to compare an individual’s per-formance may be an important component of the per-ceived “constructiveness” of the feedback [7, 24, 26].In AGFS (cases 1 and 3) in which the physicians had dir-
ect control over the outcome in question, participants fo-cused readily on the evidence, reflection, and changeplanning. In contrast, in cases 2A–D, physicians expressedreservations about their ability to make change because ofother “actors” in the system who could influence the out-come (e.g., allied health). We observed that this perceived“actionability” seemed to be mediated through the prox-imity between the physicians’ behavior or decision and theoutcome being measured.
Representation of the dataHow the data is presented in an AF report affects howeasily the participants can understand the report andmove on to making meaning from the findings and look-ing for change opportunities. Our analysis revealed thatthe data reports for cases 1 and 3 were simpler in theirdesign than the others (Figs. 3 and 4); this may be
because the project leads were more intimately involvedin co-creation of the questions the reports. This high-lights the importance of managing cognitive load, build-ing “end-user” testing of the AF reports into the designof an AGF project, and the value of co-creation of thelearning [7, 38].
Facilitation of the AGF sessionThe approach to facilitating the AGFS is pivotal to en-sure that participants in the session move from inter-preting the data to planning for change. The iPARIHSframework identifies the facilitator as playing the centralrole of fostering interaction between the participant,their data, and their context [17]. The iPARIHS authorsemphasize that experienced facilitators can grasp the nu-ances of system and organizational factors (context) thatmay influence implementation of a change intervention[17]. The need for an understanding of context lendsfurther support for the role of co-facilitating feedbackwith a group member. A non-group member facilitator,while expert at interpreting the data risked missing rele-vant change cues [8]. Physician engagement was highestwhen a respected group member co-facilitated AGFS,and we speculate that this resulted in positive credibilityjudgements by the other participants [10, 11, 19, 23, 26].We suggest that training a member of the recipientgroup to co-facilitate or lead AGFS will enhance accept-ability of the feedback and that with their privilegedknowledge (as a group member) of the context and cul-ture of the group, they will be better able to capturechange cues and incorporate elements such as barriersand enablers of change to support action planning [17].Sargeant et al. highlight that the facilitator’s role in a
feedback session should be coaching-oriented [25]. Thecoach-facilitator helps participants to navigate throughtheir reactions to data, to understand their data, andthen plan for change [25]. The “coaching-oriented ap-proach,” with prompts and questioning is essential. Inour study, when prompts and questioning were not usedby the facilitator (perhaps because of facilitator fatigueover the course of four AGFS), there was very limitedsocial interaction in the AGFS [25].
Limitations and future researchThere are several limitations of this study. Our findingsof physician behaviors in AGFS represent the collectedobservations over the course of six AGFS, five hospitals,and three specialties, but included only consenting phy-sicians. The authors acknowledge the risk of selectionbias in our participants. Another potential limitation isthe perspectives of some of the research team members,who were intimately involved with the leadership ofCPLP when this work was carried out. The authors
Cooke et al. Implementation Science (2018) 13:136 Page 15 of 18
attempted to balance important tacit knowledge that in-formed the program development with the risk of biasfrom pre-conceived ideas by acknowledging their per-spectives, bringing on experienced research team mem-bers who were not intimately involved with the CPLP atthe time of the analyses and triangulating across data
sources to corroborate our observations. Finally, it maybe argued that cases 2A–D should be treated as one casebecause they all addressed the same clinical question.However, the authors were interested to explore context-ual and cultural factors that influenced social interactionin AGFS, and because cases 2A-D occurred in different
Fig. 4 This graph is an aggregate exemplar of how the data for cases 2A–D were initially presented. The goal was to show physicians whetherthey were discontinuing sedatives and anti-psychotics in patients who were admitted with a pre-existing prescription, whether the patients werestarted on these medications in hospital, and whether they remained on them after discharge. A desirable outcome for a patient on sedatives oranti-psychotics either before or during their hospital stay was considered to be discharge from hospital without a prescription for thesemedications. Physicians found these graphs challenging to interpret, resulting in disproportionate AGFS time being spent on clarifyingand questioning
Fig. 3 Representative example of a single data table from a generic report from the case 1 project. Physicians appeared to be readily able tointerpret these graphs during the AGFS, allowing more time for reflection and planning for change
Cooke et al. Implementation Science (2018) 13:136 Page 16 of 18
settings with different participants, it was felt that thesecase should be treated separately [17, 23].The framework we present is a synthesis of our find-
ings from analysis of AF projects designed and deliveredin a novel way to address recommendations for enhan-cing AF in the literature [7]. While it appears that designand implementation elements used by the CPLP to de-liver AGFS promote social interaction, prospectiveevaluation and refinement of the CAFF will be import-ant next steps.
ConclusionBased on the findings of our study, we present a prac-tical, evidence-informed approach for the design anddelivery of AGFS in a way that links design and imple-mentation elements (relationship building, choice ofquestion, quality of data representation, and facilitationstyle) to the anticipated behaviors of physicians partici-pating in AGFS in order to promote social learning andbehavior change.
AbbreviationsAF: Audit and feedback; AGF: Audit and group feedback; AGFS: Audit andgroup feedback sessions; CAFF: Calgary Audit and Feedback Framework;CPLP: Calgary Physician Learning Program
AcknowledgementsThe authors would like to acknowledge the tremendous work of Ms. IneldaGjata and Ms. Wenxin Chen in support of the six audit and feedback projectsthat were studied in the course of this research.
FundingThis research was not funded.
Availability of data and materialsThe qualitative datasets generated for and analyzed during the current studyare not publicly available due to the need for anonymity of participants inthe AGFS, which could be compromised by sharing of transcripts in theirentirety. The complete comparative case analysis document, which wassummarized for this publication, is available from the corresponding authoron reasonable request.
Authors’ contributionsLC led the study design, qualitative and comparative case analysis, andconceptual model of physician behaviors and wrote the draft of themanuscript. DD was a major contributor to the concept and design ofthe CAFF, qualitative and comparative case analysis, conceptual modelof physician behaviors, and editing of the manuscript. LR contributed tothe research methodology, design of the CAFF, program model for thecomparative case analysis, and editing of the manuscript. SKD contributed tothe design of the CAFF and editing of the manuscript. CS contributed to thecomparative case analysis, design of the CAFF, and editing of the manuscript.HA contributed to the research design, conceptual model of physicianbehaviors, comparative case analysis, the CAFF, and edited the manuscript. Allauthors read and approved the final manuscript.
Ethics approval and consent to participateEthics approval for this work was received for each case from the ConjointHealth Research Ethics Board: REB13-0075 (case 1); REB14-0484 (cases 2a, 2b,2c, 2d); and REB13-0459 (case 3).
Consent for publicationNot applicable
Competing interestsThe authors declare that they have no competing interests.
Publisher’s NoteSpringer Nature remains neutral with regard to jurisdictional claims in publishedmaps and institutional affiliations.
Author details1Department of Clinical Neurosciences, Cumming School of Medicine,University of Calgary, UCMC Area 3, 3350 Hospital Drive NW, Calgary, AB T2N4N1, Canada. 2Cumming School of Medicine, University of Calgary, 3330Hospital Dr NW, Calgary, AB T2N 4N1, Canada. 3Physician Learning Program,Cumming School of Medicine, University of Calgary, HSC G302, 3330 HospitalDr. NW, Calgary, AB T2N 4N1, Canada. 4Division of Endocrinology &Metabolism, Cumming School of Medicine, University of Calgary, 3330Hospital Dr NW, Calgary, AB T2N 4N1, Canada. 5Department of FamilyMedicine, Cumming School of Medicine, University of Calgary, 3330 HospitalDr NW, Calgary, AB T2N 4N1, Canada.
Received: 7 May 2018 Accepted: 18 October 2018
References1. Ivers N, Jamtvedt G, Flottorp S, Young JM, Odgaard-Jensen J, French SD, et
al. Audit and feedback: effects on professional practice and healthcareoutcomes. Cochrane Database Syst Rev. 2012;6:CD000259. https://doi.org/10.1002/14651858.CD000259.pub3.
2. Mansouri M, Lockyer J. Meta-analysis of continuing medical educationeffectiveness. JCEHP. 2007;27(1):6–15.
3. Ivers NM, Grimshaw JM, Jamtvedt G, Flottorp S, O’Brien MA, French SD, etal. Growing literature, stagnant science? Systematic review, meta-regressionand cumulative analysis of audit and feedback interventions in health care.J Gen Intern Med. 2014;29:1534–41.
4. Colquhoun HL, Brehaut JC, Sales A, Ivers N, Grimshaw J, Michie S, Carroll K,Chalifoux M, Eva KW. A systematic review of the use of theory in randomizedcontrolled trials of audit and feedback. Implement Sci. 2013;8:66.
5. Colquhoun HL, Carroll K, Eva KW, Grimshaw JM, Ivers N, Michie S, Sales A,Brehaut JC. Advancing the literature on designing audit and feedbackinterventions: identifying theory-informed hypotheses. Implement Sci. 2017;12:117.
6. Colquhoun H, Michie S, Sales A, Ivers N, Grimshaw JM, Carroll K, et al.Reporting and design elements of audit and feedback interventions: asecondary review. BMJ Qual Saf. 2017;26:54–60.
7. Brehaut JC, Colquhoun HL, Eva KW, Carroll K, Sales A, Michie S, et al.Practice feedback interventions: 15 suggestions for optimizing effectiveness.Ann Intern Med. 2016;164:435–41.
8. Cooke LJ, Duncan D, Rivera L, Dowling SK, Symonds C, Armson H. How dophysicians behave when they participate in audit and feedback activities ina group with their peers? Implement Sci. 2018;13:104.
9. Sales AE, Fraser K, Baylon MA, O’Rourke HM, Gao G, Bucknall T, et al.Understanding feedback report uptake: process evaluation findings from a13-month feedback intervention in long-term care settings. Implement Sci.2015;10:20.
10. French SD, Green SE, O’Connor DA, McKenzie JE, Francis JJ, et al.Developing theory-informed behaviour change interventions to implementevidence into practice: a systematic approach using the theoretical domainsframework. Implement Sci. 2012;7:38.
11. Cane J, O’Connor D, MIchie S. Validation of the theoretical domainsframework for use in behaviour change and implementation research.Implement Sci. 2012;7:37.
12. Ames C, Archer J. Achievement goals in the classroom: students’ learningstrategies and motivation processes. J Educ Psych. 1988;80:260–7.
13. Babenko O, Szafran O, Koppula S, Au L. Motivations for learning of familymedicine residents trained in competency-based education. Education forPrimary Care. 2018;29(2):86–93.
14. Anderman LH, Anderman EM. Oriented towards mastery: promotingpositive motivational goals for students. In: Gilman R, Huebner ES, FurlongMJ, editors. Handbook of positive psychology in schools. New York:Routledge/Taylor & Francis Group; 2009. p. 161–73.
Cooke et al. Implementation Science (2018) 13:136 Page 17 of 18
15. Urban TC. Achievement of goal theory: past results, future directions. In:Maehr ML, Pintrich PR, editors. Advances in motivation and achievement,vol. 10. Greenwich: JAI Press; 1997. p. 99–141.
16. Darnon C, Dompnier B, Poortvliet PM. Achievement goals in educationalcontexts: a social psychology perspective. Soc Personal Psychol Compass.2012;6(10):760–71.
17. Harvey G, Kitson A. PARIHS revisited: from heuristic to integrated frameworkfor the successful implementation of knowledge into practice. ImplementSci. 2016;11:33.
18. Graham I, Logan J, Harrison M, Straus S, Tetroe J, Caswell W, et al. Lost inknowledge translation: time for a map? JCEHP. 2006;26(1):13–24.
19. Michie S, Atkins L, West R. The behaviour change wheel: a guide todesigning interventions. Great Britain: Silverback Publishing; 2014.
20. Bandura A. Social learning theory. New Jersey: Prentice Hall; 1977.21. Vygotsky L. Mind in society: the development of higher psychological
processes. Cambridge: Harvard University Press; 1978.22. Ajzen I. The theory of planned behaviour. Organ Behav and Hum Decis
Process. 1991;50(2):179–211.23. Abraham C, Michie S. A taxonomy of behaviour change techniques used in
interventions. Health Psychol. 2008;27:379–87.24. Watling C, Driessen E, Van Der Vleuten CPM, Vanstone M, Lingard L. Beyond
individualism: professional culture and its influence on feedback. Med Educ.2013;47(6):585–94.
25. Sargeant J, Lockyer J, Mann K, Holmboe E, Silver I, Armson H, Driessen E,MacLeod T, Yen W, Ross K, Power M. Facilitated reflective performancefeedback: developing an evidence- and theory-based model that buildsrelationship, explores reactions and content, and coaches for performancechange (R2C2). Acad Med. 2015;90:1698–706.
26. Watling C, Driessen E, Van Der Vleuten CPM, Lingard L. Learning fromclinical work: the roles of learning cues and credibility judgements. MedEduc. 2012;46(2):192–200.
27. Telio S, Ajjawai R, Regehr G. The “Educational Alliance” as a framework forreconceptualising feedback in medical education. Acad Med. 2015;90(5):609–14.
28. Sargeant JM, Mann KV, van der Vleuten CP, Metsemakers JF. Reflection: alink between receiving and using assessment feedback. Adv Health Sci EducTheory Pract. 2009;14(3):399–410.
29. Eva KW, Armson H, Holmboe E, Lockyer J, Loney E, Mann K, et al. Factorsinfluencing responsiveness to feedback: on the interplay between fear,confidence, and reasoning processes. Adv Health Sci Educ Theory Pract.2012;17(1):15–26.
30. Yin RK. Case study research: design and methods. Los Angeles: Sage; 2014.31. Goodrick D. Comparative case studies: Methodological briefs – Impact
evaluation No. 9. 2014;36. Downloaded from: https://www.unicef-irc.org/publications/754-comparative-case-studies-methodological-briefs-impact-evaluation-no-9.html. Accessed 13 Aug 2018.
32. Srivastava A, Thomson SB. Framework analysis: a qualitative methodologyfor applied policy research. JOAAG. 2009;4(2):72–29.
33. Ritchie J, Spencer L. Qualitative analysis for applied policy research. In:Bryman A, Burgess RG, editors. Analyzing qualitative data. London:Routledge; 1994. p. 173–94.
34. Gale N, Heath G, Cameron E, Rashid S, Redwood S. Using the frameworkmethod for the analysis of qualitative data in multi-disciplinary healthresearch. 2013 13:117. BMC Med Res Methodol. 2013;13:117.
35. Ward D, Furber C, Tierney S, Swallow V. Using framework analysis in nursingresearch: a worked example. J Adv Nurs. 2013;69(11):2423–31.
36. Nilsen P. Making sense of implementation theories, models and frameworks.Implement Sci. 2016;10:53.
37. Bing-You RG, Trowbridge RL. Why medical educators may be failing atfeedback. JAMA. 2009;302(12):1330–1.
38. Noel G. Health design: mapping current situations, envisioning next steps.Design J. 2017;20(Sup1):S2304–14.
Cooke et al. Implementation Science (2018) 13:136 Page 18 of 18