We conducted a randomized controlled post-only educational intervention with patients clustered within physicians to examine the qualitative and quantitative outcomes of EF training and usage (Fig. ).
Profile of randomized establishing focus upfront agenda setting intervention.
This design randomly assigns subject to condition, but collects response data only at follow-up. The design is less costly and intrusive to its subjects while providing better control of confounds including history, maturation, and instrumentation. The intervention consisted of two phases informed by previous successful randomized trials to improve communication skills15-17
: 1) a two-hour group training session led by Mr. Mauksch, which included an overview of the protocol, a videotape demonstration, role-play practice, and moderated group discussion and 2) coaching by trained behavioral scientists, who shadowed the physicians for two hours per week over four weeks. At the training session, physicians received an EF handbook. After the first coaching session they received a video and a cue card detailing EF behaviors. To avoid biasing our conclusions, Mr. Mauksch’s role in developing EF and the training of study participants, precluded him from participating in data coding activities.
The trained sequence of EF skills included: 1) orienting the patient to the EF process, 2) asking the patient to list concerns, 3) making space for pressing patient stories early on, when necessary, 4) avoiding premature ‘diving’ into diagnostic sequences or patient story telling before a full agenda has been set, 5) asking the patient to prioritize their concerns, 6) when necessary, negotiating priorities with the patient, and 7) seeking confirmation and commitment from the patient. Cognitive cues, after skills 1 and 4, reminded physicians to “askyourself whether you feel able to address all the patient's concerns” and if not, to use prioritization, negotiation, and a follow-up appointment for deferred issues.
Between July 2003 and October 2004 we invited all physicians (n
75) in a convenience sample of 12 community-based primary care clinics serving the Puget Sound region to participate in this study. A total of 59 (79%) physicians consented to participate. Physicians were randomly assigned to the intervention or the control group stratified by clinic and gender. Forty-eight physicians participated in all aspects of the study (Table ). Thirty-one worked in a university-affiliated primary care network consisting of eight neighborhood clinics. Seventeen physicians worked in a consumer-governed, nonprofit health care system. Physicians received CME credits and a $150 payment for participation. Institutional review boards at each institution approved the study.
Patients were recruited approximately 6 months following completion of physician EF training (March 2004 – March 2005). Eligibility criteria included: being 18 years or older, acting as their own legal guardian, having seen the physician at least twice in the previous two years, having no serious cognitive impairment, and fluency in English. Clinic staff advised study coordinators when eligible patients arrived. The majority (71%) of patients approached agreed to participate. Most (98%) participants completed the study questionnaires following the visit. Concerns about the burden of completing the survey were uncommon (<5%). Visits were audio recorded and patients were paid $20. Prior to analysis, scheduled health maintenance exams were removed from the data set because the physician view of agenda setting in these visits is heavily influenced by quality of care criteria. Thirteen patients were removed because their encounters were shorter than three minutes. Our final data set included an average of 30 visits per physician (n
1460) (Table ).
Questionnaires were established research instruments selected to represent a wide range of variables important to describing patient physical and mental status, satisfaction, trust, and perceptions of the patient/physician relationship. Patient self-report instruments included the: 1) SF-8 (24-hour version), a functional status measure with sub-scales for physical and mental health, 2) Primary Care Evaluation of Mental Disorders (PRIME-MD)18
depression sub-scale, 3) Patient Health Questionnaire (PHQ-15)19
to assess somatization, 4) Medical Outcomes Study Participatory Decision-Making Scale (PDM)20
assessing perceptions of physician decision-making style, 5) Health Care Climate Questionnaire (HCCQ)21
assessing beliefs regarding physician supported autonomy, 6) trust sub-scale of the Primary Care Assessment Survey (PCAS)22
assessing confidence in physician integrity and competence, and 7) Mauksch et al.8
Scale assessing patient satisfaction within the EF pilot study. We hypothesized these scales would be positively impacted when a physician adopted EF skills The PDM, HCCQ, PCAS trust, and Mauksch scales provide a multidimensional view of patient satisfaction.
Immediately following each patient encounter, physicians completed a self-report questionnaire assessing satisfaction with the visit and perceptions of difficulty experienced with the patient. This questionnaire included a subset of six items developed from the Difficult Doctor Patient Relationship Questionnaire (DDPRQ)23
used to assess physicians’ perceptions of the relationship they held with their patients.
Trained coders coded each encounter for the presence of key linguistic and quantitative data including patient and physician raised concerns and EF behaviors All coding achieved acceptable inter-rater reliabilities (kappas
0.70 for patient and physician raised concerns; kappas
0.60 for EF behaviors). Random sampling was used to select a sufficiently powerful sample of encounters for estimating patient voiced concerns. Purposeful sampling was used for the qualitative linguistic assessments. Funding constraints did not allow coding of all audios.
Establishing Focus Protocols Four trained raters, blinded to condition, listened to 936 audio files selected using purposeful sampling and coded for the presence of the EF behaviors taught in training. Since the full protocol was rarely demonstrated, but specific behaviors were present in many encounters, we defined adoption of collaborative upfront agenda setting as exhibiting or not exhibiting one of three behavioral combinations: 1) physician requested a list of concerns OR initiated an additional elicitation AND the patient indicated that they had completed listing their concerns; 2) physician asked for a list of concerns OR initiated an additional elicitation AND demonstrated negotiation or prioritization; 3) physician made multiple additional elicitations OR asked for a list of concerns multiple times.
Physician and Patient Raised Concerns
Five trained raters, blinded to provider condition, coded randomly selected audio files (n
746) for patient and provider raised concerns.
Time Spent with Physician
Physician entry and exit times were recorded for every audible file longer than three minutes (n
1282). Total face-to-face time spent with the physician was calculated.
Skill Use and Concerns by Phase of Encounter
We assigned each behavior and concern to one of three encounter phases: the first third (up to 300 seconds from the start), the last third (the last third of the total encounter time up to 300 seconds), or the middle of the encounter (the remainder of the encounter time after removal of the first and last thirds). We coded each demonstrated EF behavior and each physician or patient raised concern as occurring in one of these phases. This coding was completed for a random subset of encounters (n
The patient/physician encounter represented our primary unit of analysis. Exploratory analysis of unadjusted comparisons between the intervention and the control groups were first conducted to assess differences using SPSS 15.0 and included t-tests and analysis of covariance (ANCOVA) tests controlling for attitudes and functional status in delineating the relationship between patient variables and coded behaviors. When significant associations were not revealed we elected to report bivariate relationships examining mean differences. Because encounters were clustered within physicians possibly resulting in significant intra-class correlations, and a violation of independence among study subjects, hierarchical linear model analyses were used to generate unbiased estimators. Analyses were conducted using the SPSS 15.0 MIXED procedure. Means, confidence intervals, and statistical tests reported for patient level data reflect these unbiased estimates.
Different sampling techniques were used to address questions to be answered by the data. For example, random sampling was employed for counting concerns and purposeful sampling was used for coding EF behaviors. These different sampling techniques represent the authors’ attempts to best sample the data for the specific analyses being conducted.