Clinical context
Primary care is the level of health care that: acts as the patient's gateway into the healthcare system for all of their health-related problems and needs; provides care focused on the individual and their context (patient-oriented instead of just disease-oriented); offers care for all but the most uncommon or unusual conditions; ensures continuity of care; and monitors the coordination or integration of care provided at other levels of the system or by other professionals [
22]. Encompassing the widest possible spectrum of health conditions, primary care is by definition the forum where the greatest diversity of medical decisions takes place. For example, in a study that assessed for 903 consultations how comfortable family physicians and their patients were regarding a decision that had been made (
i.e., personal uncertainty), 43% dealt with treatment decisions, 27% with diagnostic and screening tests, 12% with follow up and continuity of care, 6% with lifestyle issues, 5% with work-related issues, 4% with birth control, and 2% with vaccination [
23]. Furthermore, on average, 90% of all monthly healthcare interactions occur in ambulatory clinical settings and 10% occur in hospital-based outpatient settings [
24]. Together, these results emphasize that it is important to study decision-making in primary health care contexts because of the potential benefit to patient outcomes and ultimately to population health [
14].
Data collection procedures
Pairs of physicians and patients will be recruited using a procedure that we have successfully used in the past [
25]. We will begin by enrolling physicians, asking them to complete a consent form, a socio-demographic questionnaire, and the physicians' reactions to uncertainty scale (PRU) [
26]. During participating physicians' appointment hours, a research assistant will recruit patients in the waiting room at a randomly pre-determined time. Patients will be recruited according to the following criteria: ≥18 years old, able to read French or English according to the recruitment site, able to provide informed consent, not suffering from an acute condition that requires immediate medical intervention (
i.e., transfer to emergency department), and able to report on a decision that they have made with the physician. Given that the recruitment procedures will be independent of the family physician and the time of recruitment randomly selected, we aim to protect against selection bias. As only one patient per physician will be recruited, patients who have already participated in the study once with a physician will be excluded. The goal of recruitment is to find one eligible patient per physician.
Once the subjects have been recruited, participating physicians will audio-tape one consultation with their consenting patient by using a digital audio recorder. Following the consultation, eligible patients (patients that have experienced that a decision was made) and physicians will be independently asked to complete a set of relationship-centered questions that assess their interaction. Based on prior projects that have shown this information to be valuable [
25,
27], we will ask each patient to describe the decision (
i.e., the index decision) they have made with the physician in their own words. Following the patient's description of their decision, the patient will answer the questionnaire referring to the index decision immediately after the consultation. Patients' socio-demographics will also be assessed. Once the patient has completed their questionnaire, the research assistant will enter the decision identified by the patient on the physician's post-consultation questionnaire. The research assistant will then give the physician the post-consultation questionnaire to complete. The physician will be blinded to the patient's questionnaire. All audiotapes will be transcribed.
Variables and measures
Using five published systematic reviews of instruments relevant to SDM research, two of which were performed by team members [
28‐
32], we identified several questionnaires that map the various dimensions of EXACKTE
2. These questionnaires have the potential of unraveling the relationship phenomena that underlie effective KTE between physicians and patients and can be administered to both parties alike. The same 'uncertainty' subscale of the decisional conflict scale (DCS) [
17], for example, can determine how comfortable either physicians or patients are with the decision made [
25]. Our review identified six measures of physicians' perceptions of the decision-making process [
31] that have corresponding patient versions [
18,
33‐
36]. All are standardized measures that have been pilot-tested with physician-patient pairs.
Relationship-centered explanatory variables
The definition and explanation of the problem, the presentation of the options, and the discussion of the pros and cons (
i.e., benefits, risks, and costs) will be assessed with the 'information-giving' construct of the medical communication competence scale (MCCS). This construct is comprised of nine items (Cronbach's alpha = 0.86) [
34]. It will be administered to both physicians and patients.
Presentation of the doctor's knowledge and recommendations will be assessed using an instrument derived from the work by Sewitch and colleagues on patient-physician interactions [
10]. This instrument assesses physician-recommended interventions from both the physician and the patient perspective according to four binary yes or no indicators: the prescription of medication; the scheduling of a further appointment; the consultation of another healthcare professional; and the conduct of further medical investigation [
10]. It will be administered to both physicians and patients. Presentation of the doctor's knowledge and recommendations will be also assessed using the 'perception that an ineffective decision has been made' subscale of the DCS which is comprised of four items (Cronbach's alpha = 0.70 in physicians and 0.65 in patients) [
23]. It will be administered to both physicians and patients.
Checking and clarifying the patient's understanding will be assessed with the 'information verifying' construct of the MCCS, which is comprised of four items (Cronbach's alpha = 0.78) [
34] and with the 'feeling uninformed' subscale of the DCS, comprised of three items (Cronbach's alpha = 0.71) [
23]. Both measures will be administered to both physicians and patients.
Exploration of values and preferences will be assessed with the 'value clarification' subscale of the DCS which is comprised of three items (Cronbach's alpha = 0.72) [
23].
Discussion of the patient's ability and self-efficacy to act upon their choice will be assessed with the 'perceived behavioral control' construct of the Theory of Planned Behavior, which is comprised of three items [
37]. Perceived behavioral control is a measure of the amount of control the individual perceives he or she has over the behavior in question, and is referred to as a measure of self-efficacy. As stated by Makoul and Clayman, 'the rationale for incorporating a patient's efficacy expectation parallels the argument for discussing patient preferences and values: both provide important perspective regarding acceptability of the options at hand' [
14]. Team members have extensive expertise in the use of this scale in both patients and healthcare professionals in the context of SDM studies [
27,
38].
Statistical analyses
To further assess evidence of the validity and reliability of the relationship-centered measures, internal consistency,
i.e., how consistently subjects' scores on a measurement tool can be generalized to the domain of items that could be asked [
44], will be used to estimate the reliability of the measures. Cronbach's alpha will be computed independently for each of the four subgroups of subjects (French and English, physicians and patients) and then for the overall groups of physicians and of patients [
45].
Construct validity will first be assessed by confirmatory factor analysis (CFA) for each scale. This statistical method will be used to test for unidimensionality in each one of the relationship-centered measures. Results will help us determine if the empirical factor structure corresponds to the hypothesized theoretical unidimensional factor structure of each relationship-centered measure. CFA will be conducted with AMOS software. Since we will be recruiting unique physician-patient pairs, there will be no need to take the clustering of patients under physicians into account. However, clustering of the clinic from which physician-patient pairs will be assessed by computing an intra-class correlation coefficient for each measured outcome.
Second, construct validity will also be assessed by correlating the relationship-centered measure scores with OPTION, a validated third observer instrument (Cronbach's alpha = 0.79) that assesses SDM (convergent validity) [
46]. Based on audio ratings of the consultations, two assessors will independently rate the encounters using observer OPTION (inter-rater reliability k = 0.71). Our
a priori hypothesis is that the relationship-centered measures will correlate with observer OPTION in the expected direction (
e.g., for the measure of personal uncertainty, we should be able to observe a positive correlation). Author GE and colleagues recently developed a dyadic reported version of OPTION with six family physicians in Cardiff, Wales. Using the results of this analysis, our team will triangulate the measurements of the consultation process using observer OPTION and the patient-physician version of dyadic OPTION.
Third, construct validity will be further assessed with a 'known groups' approach in physicians [
47]. At entry into the study, physicians will complete the 'anxiety due to uncertainty' subscale (five items, Cronbach's alpha = 0.86) and the 'reluctance to disclose uncertainty to patients' subscale (five items, Cronbach's alpha = 0.79) of the physicians' reactions to uncertainty scale (PRU) [
26]. Briefly, the PRU covered four areas of physicians' reactions to uncertainty derived from interviews with physicians: anxiety due to uncertainty; concern about bad outcomes; reluctance to disclose uncertainty to patients; and reluctance to disclose mistakes to physicians. The PRU assesses physicians' predisposition (
i.e., a trait) to the uncertainty that is inherent to patient care from all sources. Our
a priori hypothesis is that the relationship-centered measures will differentiate physicians with high scores on the 'anxiety due to uncertainty' as well on the 'reluctance to disclose uncertainty to patients' subscales from physicians with low scores (
e.g., personal uncertainty of physicians will be higher in physicians with high scores on the 'anxiety due to uncertainty' as well on the 'reluctance to disclose uncertainty to patients' subscales than in physicians with low scores).
The fourth way that we will assess construct validity is with a 'known groups' approach in patients [
47]. Based on a systematic review of patients' opinions on SDM, our
a priori hypothesis is that some of the relationship-centered measures will discriminate between patients with high levels of education and patients with low levels of education (
e.g., patients with high levels of education will have higher scores on the self-efficacy scale than patients with low levels of education). It will also discriminate between older patients and younger patients (
e.g., older patients will have lower scores on the self-efficacy scale than younger patients) [
2]. All of these analyses will first be performed independently for the physicians and the patients and subsequently for all subjects. This will help us to determine whether the different validity indices are adequate for physicians and patients in the relationship-centered approach to KTE.
To compare the physician and patient responses to the relationship-centered measures, invariance of the structural factor will be employed to verify that the factorial structure of the constructs is the same for patients and physicians. The invariance of the factorial structure will be assessed with CFA [
48]. In other words, we will assess and compare the number of items that load on the latent dimension as well as their loading value. We will assess possible item bias using the Mantel-Haenszel method [
49]. We will also estimate and test differences in variance and correlational structure within and across pairs using structural equation modeling [
15]. Finally, we plan to assess the equivalence of our tools between the French-speaking and English-speaking data that will be collected.
To assess the relationship phenomena between physicians and patients, the Actor Partner Interdependence Model (APIM) will serve as the analytical framework to assess the assumed relationship phenomena between physicians and patients as it takes into account the interdependence between observations without losing possibly valuable information about what each member contributes to the pair. Hence, statistical analysis will be performed by means of structural equation modeling (SEM) with a maximum likelihood estimator. The dependent variable (outcome) will be personal uncertainty about a course of action in both physicians and patients. The predictor variables will consist of the essential elements of SDM: definition and explanation of the problem; presentation of the options; discussion of the pros and cons (
i.e., the benefits, risks, and costs); clarification of patient values and preferences; discussion of patients ability and self-efficacy to act upon his or her treatment; presentation of doctor knowledge and recommendation; and checking and clarifying understanding as assessed in both physicians and patients. An initial APIM model will be constructed that allows all paths (effects) to be 'free'. Then, a second model will be constructed whereby all similar actor and partner paths are set to be equal, thereby assessing the similarities of effects between physicians and patients. Measures of model fit to be calculated include the chi-square, the comparative fit index (CFI) and the root mean square error of approximation (RMSEA). A non-significant chi-square value, CFI ≥ 0.95 and a RMSEA value of ≤ 0.06 will indicate good model fit [
50]. Statistical analyses will be performed using SPSS (version 17.0) and AMOS (version 6.0) software.
To assess the relationship between the agreement of physicians and patients on the uncertainty with patients' decisional regret, first, using the methods of dyadic analysis proposed by Kenny and colleagues, an agreement score for physician-patient pairs on the 'uncertainty' subscale will be computed [
15]. This agreement score will be entered into a general linear regression model as an explanatory variable of the decisional regret assessed in patients at two weeks. The relationship between patients' decisional regret and patients' QOL will be assessed by regressing the physical and the mental health component scores of the SF-12 on patients' decisional regret scores.