Some but not all dyadic measures in shared decision making research have satisfactory psychometric properties France Legare a,b, * , St ephane Turcotte a , Hubert Robitaille a , Moira Stewart c , Dominick Frosch d , Jeremy Grimshaw e,f , Michel Labrecque b , Mathieu Ouimet g , Michel Rousseau a , Dawn Stacey e,h , Trudy van der Weijden i , Glyn Elwyn j a Centre Hospitalier Universitaire de Quebec Research Centre, H^ opital St-Franc ¸ois d’Assise, 10, rue de l’Espinay, Quebec City, Quebec, Canada G1L 3L5 b Department of Family and Emergency Medicine, Universite Laval, 1050, avenue Ferdinand-Vandry, Quebec City, Quebec, Canada G1V 0A6 c Department of Family Medicine, The University of Western Ontario, 1151, Richmond Street North, London, Ontario, Canada N6A 5C1 d Department of Medicine, University of California, 10920 Wilshire Blvd., Los Angeles, Los Angeles, CA 90024-6512, USA e Ottawa Hospital Research Institute, 501, Smyth Road, Ottawa, Ontario, Canada K1H 8L6 f Department of Medicine, University of Ottawa, 451, Smyth Road, Ottawa, Ontario, Canada K1H 8M5 g Department of Political Science, Universite Laval, 1030, avenue des sciences humaines, Quebec City, Quebec, Canada G1V 0A6 h Faculty of Health Sciences, School of Nursing, University of Ottawa, 451, Smyth Road, Ottawa, Ontario, Canada K1H 8M5 i Department of General Practice, School of Public Health and Primary Care (Caphri), Maastricht University, Universiteitssingel 40, 6229 ER Maastricht, The Netherlands j Department of Primary Care and Public Health, School of Medicine, Cardiff University, Neuadd Meirionnydd, Heath Park, Cardiff, UK CF14 4YS Accepted 8 June 2012; Published online 13 September 2012 Abstract Objective: To assess the psychometric properties of dyadic measures for shared decision making (SDM) research. Study Design and Setting: We conducted an observational cross-sectional study in 17 primary care clinics with physician-patient dyads. We used seven subscales to measure six elements of SDM: (1) defining the problem, presenting options, and discussing pros and cons; (2) clarifying the patient’s values and preferences; (3) discussing the patient’s self-efficacy; (4) drawing on the doctor’s knowledge; (5) verifying the patient’s understanding; and (6) assessing the patient’s uncertainty. We assessed the reliability and invariance of the fac- torial structure and considered a measure to be dyadic if the factorial structure of the patient version was similar to that of the physician version and if there was equality of loading (no significant chi-square). Results: We analyzed data for 264 physicians and 269 patients. All measures except one showed adequate reliability (Cronbach alpha, 0.70e0.93) and factorial validity (root mean square error of approximation, 0.000e0.06). However, we found only four measures to be dyadic (P O 0.05): the values clarification subscale, perceived behavioral subscale, information-verifying subscale, and uncertainty subscale. Conclusion: The subscales for values clarification, perceived behavioral control, information verifying, and uncertainty are appropriate dyadic measures for SDM research and can be used to derive dyadic indices. Ó 2012 Elsevier Inc. All rights reserved. Keywords: Shared decision making; Physician-patient relationship; Psychometrics; Dyadic data analysis; Measurement; Factorial analysis 1. Introduction In clinical settings, many health-related decisions are made in the context of physician-patient dyads. The dyad, a group of two parties, is arguably the fundamental unit of all interpersonal interactions, including interactions un- derlying decision making in health care [1]. Although the relationship-centered research in the fields of marital ther- apy, adult attachment, and how couples cope with chronic disease [2e4] has used a dyad-based perspective for years, the approach is only now beginning to attract the attention of health services researchers interested in clinical decision making [5]. To date, most research on how physicians and Conflicts of interest/Financial disclosure: The authors declare that they have no competing interests. This study was funded by the Canadian Institutes of Health Research (CIHR 2008-2011; grant number 185649-KTE). Ethical considerations: All participants were asked to complete consent forms. Ethical approval for the project was obtained from the Research Ethics Board of the Centre de Sante et de Services Sociaux de la Vieille Capitale in Quebec City, Canada (final approval 2008/11/25; ethics number #2008-2009-23), and the Office of Research Ethics of The University of Western Ontario (final approval 2009/07/10; ethics number #15712E). Physicians and patients were not financially remunerated for their participation. * Corresponding author. Tel.: 418-525-4437; fax: 418-525-4194. E-mail address: [email protected](F. Legare). 0895-4356/$ - see front matter Ó 2012 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.jclinepi.2012.06.019 Journal of Clinical Epidemiology 65 (2012) 1310e1320
Transcript
Journal of Clinical Epidemiology 65 (2012) 1310e1320
Some but not all dyadic measures in shared decision making researchhave satisfactory psychometric properties
France L�egar�ea,b,*, St�ephane Turcottea, Hubert Robitaillea, Moira Stewartc, Dominick Froschd,Jeremy Grimshawe,f, Michel Labrecqueb, Mathieu Ouimetg, Michel Rousseaua, Dawn Staceye,h,
Trudy van der Weijdeni, Glyn ElwynjaCentre Hospitalier Universitaire de Qu�ebec Research Centre, Hopital St-Francois d’Assise, 10, rue de l’Espinay, Quebec City, Quebec, Canada G1L 3L5
bDepartment of Family and Emergency Medicine, Universit�e Laval, 1050, avenue Ferdinand-Vandry, Quebec City, Quebec, Canada G1V 0A6cDepartment of Family Medicine, The University of Western Ontario, 1151, Richmond Street North, London, Ontario, Canada N6A 5C1
dDepartment of Medicine, University of California, 10920 Wilshire Blvd., Los Angeles, Los Angeles, CA 90024-6512, USAeOttawa Hospital Research Institute, 501, Smyth Road, Ottawa, Ontario, Canada K1H 8L6
fDepartment of Medicine, University of Ottawa, 451, Smyth Road, Ottawa, Ontario, Canada K1H 8M5gDepartment of Political Science, Universit�e Laval, 1030, avenue des sciences humaines, Quebec City, Quebec, Canada G1V 0A6
hFaculty of Health Sciences, School of Nursing, University of Ottawa, 451, Smyth Road, Ottawa, Ontario, Canada K1H 8M5iDepartment of General Practice, School of Public Health and Primary Care (Caphri), Maastricht University, Universiteitssingel 40, 6229 ER Maastricht,
The NetherlandsjDepartment of Primary Care and Public Health, School of Medicine, Cardiff University, Neuadd Meirionnydd, Heath Park, Cardiff, UK CF14 4YS
Accepted 8 June 2012; Published online 13 September 2012
Abstract
Objective: To assess the psychometric properties of dyadic measures for shared decision making (SDM) research.Study Design and Setting: We conducted an observational cross-sectional study in 17 primary care clinics with physician-patient
dyads. We used seven subscales to measure six elements of SDM: (1) defining the problem, presenting options, and discussing pros andcons; (2) clarifying the patient’s values and preferences; (3) discussing the patient’s self-efficacy; (4) drawing on the doctor’s knowledge;(5) verifying the patient’s understanding; and (6) assessing the patient’s uncertainty. We assessed the reliability and invariance of the fac-torial structure and considered a measure to be dyadic if the factorial structure of the patient version was similar to that of the physicianversion and if there was equality of loading (no significant chi-square).
Results: We analyzed data for 264 physicians and 269 patients. All measures except one showed adequate reliability (Cronbach alpha,0.70e0.93) and factorial validity (root mean square error of approximation, 0.000e0.06). However, we found only four measures to bedyadic (PO 0.05): the values clarification subscale, perceived behavioral subscale, information-verifying subscale, and uncertaintysubscale.
Conclusion: The subscales for values clarification, perceived behavioral control, information verifying, and uncertainty are appropriatedyadic measures for SDM research and can be used to derive dyadic indices. � 2012 Elsevier Inc. All rights reserved.
0895-4356/$ - see front matter � 2012 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.jclinepi.2012.06.019
1. Introduction
In clinical settings, many health-related decisions aremade in the context of physician-patient dyads. The dyad,a group of two parties, is arguably the fundamental unitof all interpersonal interactions, including interactions un-derlying decision making in health care [1]. Although therelationship-centered research in the fields of marital ther-apy, adult attachment, and how couples cope with chronicdisease [2e4] has used a dyad-based perspective for years,the approach is only now beginning to attract the attentionof health services researchers interested in clinical decisionmaking [5]. To date, most research on how physicians and
1311F. L�egar�e et al. / Journal of Clinical Epidemiology 65 (2012) 1310e1320
What is new?
� Although shared decision making (SDM) is by na-ture a dyadic process, there is a lack of valid reli-able dyadic measures.
� This study confirmed the validity and reliability ofdyadic measures for four elements of SDM: (1)clarifying the patient’s values and preferences,(2) discussing the patient’s self-efficacy, (3) verify-ing the patient’s understanding, and (4) assessingthe patient’s uncertainty.
� Our results suggest that dyadic measures withitems of a similar nature (i.e., all items includedin the measure assess either feelings or cognitionsbut do not mix the two) tend to be more valid andreliable than those that mix items.
� The use of valid and reliable dyadic measurescould be used to derive dyadic indices and unravelthe complex relationship that develops betweenphysicians and patients in the decision-makingprocess.
patients relate to each other has examined the parties sepa-rately, as if each member of a dyad were entirely indepen-dent from the other [6]. In truth, however, not only dophysicians and patients have much in common but previousstudies show their separate outcome scores to be more sim-ilar than those of two individuals who are not members ofa dyad [7e12]. Compounding the problem facing the re-search community today is the fact that many studies thathave used dyadic indices (an agreement score and differ-ences in score) failed to consider the basis on which the in-dices in question were developed. Had they considered this,they would have found that many measures used for the in-dices assessed different items in physicians from in pa-tients, which calls into question the use of such measuresfor dyadic research. In summary, examining the dyad’s con-tribution to clinical decision making holds great potential,but more valid and reliable dyadic measures are needed,with particular attention focused on how the measures aredeveloped.
Shared decision making (SDM) is essentially a relation-ship-centered process, in which a choice of health care ismade jointly by the practitioner and the patient [13]. Morespecifically, SDM is both an interpersonal process, bywhich the parties relate to each other, and an interdepen-dent process, by which the parties influence each other[14e16]. Concretely, SDM holds that one party’s percep-tions can influence those of the other and that each party’sperceptions have several layers. SDM research requiresvalid and reliable dyadic measures, that is, standardizedmeasures that can be administered to clinicians and patients
concurrently and can be used to derive dyad-level indices.In turn, dyadic indices may provide valuable informationon the unique contribution of the dyad level to thedecision-making process. It can also help unravel the com-plex relationship that develops between physicians and pa-tients in the decision-making process. Consequently, thisstudy sought to assess the psychometric attributes of dyadicmeasures for SDM research.
2. Methods
2.1. Study design and data collection procedures
We have published details about our study protocolelsewhere [17]. Briefly, between January 2009 and April2010, we conducted a cross-sectional study in 17 primarycare clinics. We recruited physicians through family prac-tice teaching units in Quebec City, Canada, and throughthe Thames Valley Family Practice Research Unit of theCentre for Studies in Family Medicine in London, Ontario.We asked physicians to complete a consent form and a so-ciodemographic questionnaire. During doctors’ visitinghours, a research assistant recruited patients in the waitingroom at random moments. To participate, patients had tomeet the following criteria: aged 18 years or older; ableto read in French if they were recruited in Quebec and inEnglish if they were recruited in Ontario; able to provideinformed consent; not suffering from an acute conditionthat required immediate medical intervention; and willingto report on a decision that they would make with theirphysician. After recruitment, encounters between physi-cians and consenting patients were audiotaped usinga digital recorder [18]. At the end of the encounter, the pa-tient and the physician independently completed a question-naire whose dyadic measures assessed elements of theirSDM experience with the corresponding subscales. Thequestionnaire also collected patients’ sociodemographicinformation.
2.2. Dyadic measures assessed
First, we consulted a list of essential elements of SDMcompiled from a systematic review [19] and designed a dy-adic SDM model that conceptualized the interpersonal andinterdependent elements of the relationship between physi-cians and patients. Second, referring to systematic reviewsof measures relevant to SDM research [20e24], we identi-fied several instruments that had been tested on both physi-cians and patients [17]. From this list of instruments, weidentified subscales that mapped the essential elements ofSDM included in our dyadic model. Third, we performedback-to-back translation of the subscales included in thisstudy following guidelines for the cross-cultural adaptationof self-reported measures [25].
We used seven subscales to assess the six essentialelements of our model (see Appendix A on the journal’s
1312 F. L�egar�e et al. / Journal of Clinical Epidemiology 65 (2012) 1310e1320
website at www.jclinepi.com for a detailed description ofthe subscale items). To assess the first element of SDM, thatis, defining the problem, presenting options, and discussingpros and cons, we used the nine items of the information-giving subscale of the Medical Communication Compe-tence Scale (MCCS) [26]. We assessed the second element,clarifying the patient’s values and preferences, with thevalues clarification subscale of the Decisional ConflictScale (DCS), which has three items. We assessed the thirdelement, discussing the patient’s self-efficacy to act on his/her choice, with the perceived behavioral control subscaleof the Theory of Planned Behavior (TPB), which has threeitems [27]. Perceived behavioral control is a measure of thecontrol the individual perceives himself/herself to have overthe behavior in question and is considered a measure ofself-efficacy [28e30]. We assessed the fourth element,drawing on the doctor’s knowledge and making recommen-dations, with the physician’s recommendations subscaleof the Patient-Physician Discordance Scale (PPDS) [31].This instrument uses five items to assess physician-recommended interventions from both the physician’s andpatient’s perspectives [10]. We assessed the fifth element,verifying the patient’s understanding, with two measures:the four-item information-verifying subscale of the MCCS[26] and the three-item feeling uninformed subscale of theDCS, a measure of the perception of feeling informed [28].We assessed the last element, the patient’s uncertainty, withthe uncertainty subscale of the DCS [28], which has threeitems.
Fig. 1. The study flow chart.
2.3. Statistical analyses
First, we compiled descriptive statistics to summarizeparticipants’ characteristics and their scores on each ele-ment of SDM measured. Second, we analyzed the reliabil-ity of the measures with Cronbach alpha. Third, we testedthe invariance of the factorial structure by performing ex-ploratory factor analysis using both the Varimax and Pro-max methods to determine whether the empirical datafactor structure corresponded to the hypothesized factorstructure. We then performed confirmatory factor analysiswith a maximum likelihood estimator. Statistical test valuesincluded chi-square, the comparative fit index (CFI), andthe root mean square error of approximation (RMSEA).A nonsignificant chi-square value, CFI �0.95, and anRMSEAvalue of �0.06 indicated a good fit with the model.Fourth, we conducted more construct validity (concomitantvalidity) analyses by correlating each one of the dyadicmeasure scores with the Observing Patient Involvement inDecision Making (OPTION) scale, a validated third-observer instrument (raters assess verbatim transcribed au-dio-recorded consultations) that evaluates SDM-specificbehaviors of physicians during consultation on a scale of0e100% [18,32,33]. Last, based on the analytical frame-work for dyadic data analysis proposed by Kenny et al.[6], we assessed (1) equality of correlation, (2) equality
of item variances, (3) reliability, (4) equality of loading,and (5) equality of variance for the latent factor in bothphysicians’ and patients’ measures. Our minimal criteriafor declaring a measure fit for use in dyadic research wastwofold: (1) similar factorial structures for the physicianand patient measures and (2) equality of loading. We usedSAS version 9.2 (SAS Institute Inc., Cary, NC, USA) tocalculate descriptive statistics, measure reliability, and per-form exploratory factor analyses. We used SPSS version17.0 and AMOS version 6.0 to perform confirmatory factoranalyses and dyadic data analysis.
3. Results
3.1. Participants’ characteristics
Fig. 1 shows the participant flow through the study flowchart. Of 382 eligible physicians, 274 (72%) agreed toparticipate. Of 430 eligible patients, 276 (64%) agreed toparticipate. After loss to follow-up, we collected 264post-encounter questionnaires for the physicians and 269for the patients. A total of 259 unique complete dyads wereanalyzed. Tables 1 and 2 detail the characteristics of theparticipating physicians and patients, respectively.
Abbreviations: SD, standard deviation; FPTU, family practice teaching unit.
1313F. L�egar�e et al. / Journal of Clinical Epidemiology 65 (2012) 1310e1320
3.2. Means and standard deviations of the measuresassessed
Table 3 indicates the means of physician and patientscores for the seven measures. Although the doctor’s rec-ommendation subscale of the PPDS showed a lack of reli-ability as shown by inadequate Cronbach alpha (0.36 forthe physician subscale and 0.46 for the patient subscale)its means and standard deviations are presented as well.
Table 2. Characteristics of the participating patients
Characteristics English (n[ 10
Women, n (%) 65 (60)Age (year, mean6 SD) 526 17 (n5 10Marital status, n (%)Single 18 (17)Married/living with partner 67 (62)Separated/divorced 15 (14)Widowed 7 (7)No answer 1 (1)
Employment status, n (%)Working full-time 34 (32)Working part-time 11 (10)Unemployed or laid off 9 (8)Looking for work 3 (3)Keeping house or raising children full-time 7 (7)Retired 32 (30)Other 11 (10)No answer 1 (1)
Schooling, n (%)None 12 (11)High school diploma or equivalent 39 (36)College degree 14 (13)Bachelor’s degree 12 (11)Master’s degree 10 (9)Doctorate 2 (2)Professional degree 13 (12)Other 3 (3)No answer 3 (3)
Household members (n6 SD) 36 1 (n5 106Family income ($CAD), n (%)Less than 50,000 41 (38)50,000e59,999 19 (18)60,000e79,999 14 (13)80,000e99,999 13 (12)100,000 or more 13 (12)No answer 8 (8)
Abbreviation: SD, standard deviation.
3.3. Reliability and factorial validity of the measuresassessed
Table 4 presents the results from the exploratory factoranalysis for all the subscales used to assess each one ofthe six elements of SDM. Both Varimax and Promax rota-tion methods were used and have displayed very similar re-sults. However, Table 4 presents only the results obtainedwith the Varimax method. None of the measures assessed
1314 F. L�egar�e et al. / Journal of Clinical Epidemiology 65 (2012) 1310e1320
in physician and patient demonstrated a statistically signif-icant relationship to OPTION (Cronbach alpha, 0.76).
3.3.1. Defining the problem, presenting options, anddiscussing pros and cons
Using the information-giving subscale of the MCCS, theresults of exploratory factor analysis suggested retainingtwo factors to explain the physician scores, namely, definingthe problem and presenting options (items Q1a, Q1bmQ1cm Q1e and Q1f), and discussing pros and cons (itemsQ1d, Q1g, Q1h and Q1i), as the internal consistency of boththese factors was very good. Table 5 shows the results of theconfirmatory factor analyses: the suggested model for thistwo-factor subscale, once modified based on the followingdiagnostic methods (modification indices and standardizedresiduals), was adequate (RMSEA, 0.055 and CFI, 0.980).
Exploratory factor analysis suggested retaining two fac-tors to explain the patient’s measure. Few items load underboth factors (1c, 1e, 1g, and 1h), indicating that in spite ofthe similarity of the questionnaires provided to the patientand the physician, the factors were not equally definedfor the two parties. The internal consistency of the two-factor patient measure was very good (Cronbach alpha,0.90 and 0.92), and confirmatory factor analysis (Table 5)revealed that after modification (using modification indicesand standardized residuals as diagnostic methods), themodel had a good fit (RMSEA, 0.056 and CFI, 0.988).
3.3.2. Clarifying the patient’s values and preferencesUsing the values clarification subscale of the DCS, the
results of exploratory factor analysis suggested only onefactor to explain the physician’s and the patient’s measures.The internal consistency of this measure was very good for
both parties (Cronbach alpha of 0.84 for the physician sub-scale and 0.85 for the patient subscale). The model for theconfirmatory analysis was saturated for both measures, withno degree of freedom left available for testing the adjust-ment of the model.
3.3.3. Discussing the patient’s self-efficacyUsing the perceived behavioral control subscale of the
TPB, the analysis revealed one factor in both the physicianand patient (Table 4). With Cronbach alpha at 0.70 for bothparties’ scores, the internal consistency was good. Themodel for the confirmatory factor analysis was saturated,meaning that no degree of freedom was available to testthe adjustment of the model.
3.3.4. Drawing on the doctor’s knowledgeUsing the physicians’ recommendations subscale from
the PPDS, the exploratory factor analysis of the physician’sand the patient’s measures indicated that it was not possibleto identify a clear underlying structure for this measure inboth members of the dyad. The internal consistency ofthe measures was inadequate (Cronbach alpha of 0.36 forthe physician subscale and 0.46 for the patient subscale).
3.3.5. Verifying understandingUsing the feeling-uninformed subscale of the DCS, the
analysis revealed one factor in both the physician and pa-tient (Table 4). The internal consistency of the subscalewas adequate (Cronbach alpha of 0.72 for the physiciansand 0.86 for the patients). The model for the confirmatoryfactor analysis was saturated.
Using the informationverifying subscale of theMCCS, theanalysis revealed one factor in both the physician and patient
Table 4. Factor loadings of the exploratory factor analysis and relationship (concomitant validity; Spearman correlation) between OPTION third-observer instrument and relationship-centered measures
Subscales Item number
Exploratory factor analysisa for thephysician subscales (factor loadings)
Exploratory factor analysisa for thepatient subscales (factor loadings)
NOTE. Values presented in bold indicate a factor loading O.05.a The rotation method used is Varimax.
1315F. L�egar�e et al. / Journal of Clinical Epidemiology 65 (2012) 1310e1320
(Table 4). The internal consistency of the subscale was verygood (Cronbach alpha of 0.86 for the physicians and 0.92for the patients). The results from the confirmatory factoranalysis (Table 5) showed that the model for the physician’ssubscale had a good fit (RMSEA, 0.000 andCFI, 1.000). Sim-ilar results for the model for the patient’s subscale were ob-tained after modification (RMSEA, 0.000 and CFI, 1.000).
3.3.6. Assessing the patient’s uncertaintyUsing the uncertainty subscale of the DCS, the analysis
revealed one factor for both the physician and patient(Table 4). The internal consistency of the subscale was verygood (Cronbach alpha of 0.88 for the physicians and 0.82 forthe patients). The model for the confirmatory factor analysiswas saturated.
3.4. The dyadic nature of the measures
Because the results show that physician measures and pa-tient measures for the information giving subscale and thephysicians’ recommendations subscale do not have a similarfactorial structure, we did not evaluate the dyadic nature of
these two subscales further. Table 6 gives the results of ourdyadic data analysis of the remaining five measures. Fourmeasures showed equality of loading: the values clarificationsubscale (P5 0.95), self-efficacy subscale (P5 0.83),information verifying subscale (P5 0.46), and the uncer-tainty subscale (P5 0.72). This suggests that these measuresmay have the same meaning for both members of the dyad(physicians and patients). Three of the four measures showedgood fit for the model to test the equality of the correlationstructure, suggesting similar intrapersonal and interpersonalcorrelations between the physician model and the patientmodel (i.e., CFI,�0.95 and an RMSEAvalue,�0.06). How-ever, the item variance and the variance for the latent factorbetween the two models were not similar (all P! 0.05).
4. Discussion
To the best of our knowledge, this study is among the firstto assess the psychometric attributes of dyadic measures forSDM research based on an existing SDM model. It is alsoamong the first to shed light on the methods needed to
Table 5. Factor loading from the confirmatory factor analysis
Subscales Item
Confirmatory analysis for the physician subscales Confirmatory analysis for the patient subscales
1316 F. L�egar�e et al. / Journal of Clinical Epidemiology 65 (2012) 1310e1320
develop valid and reliable dyadic measures. Overall, usingcriteria of similar factorial structures and equal loading, wefound four measures with adequate psychometrics to be fitfor SDM research using a dyadic approach: the values clari-fication subscale of the DCS, the perception of control sub-scale of the TPB, the information verifying subscale of theMCCS, and the uncertainty subscale of the DCS. These fourmeasures can be used to derive dyadic indices in the contextof SDM research. Consideration of these findings leads us tomake three principal observations.
First, in this study, we observe that although the items ofthe instruments we chose to map over the integrated SDMmodel had been developed with care, some items assesseddifferent types of interpersonal perceptions for each party(e.g., the DCS asks patients whether they ‘‘know optionsavailable for me’’ and practitioners whether they ‘‘ knowthe benefits of each option for this patient’’). We also ob-serve that some measures mix items assessing cognitionwith items assessing feelings (e.g., the doctor’s recommen-dations subscale of PPDS asks ‘‘to what extent did you dis-cuss personal issues that might affect your patient’s medicalcondition?’’ and ‘‘to what extent was your patient satis-fied?’’). A model by Kenny [34] for conceptualizing dyadicindices comprises the parties’ cognitions (perceptions andbeliefs), their feelings, and their behaviors. Based on
a taxonomy of interpersonal perceptions [34], we can groupcognitions into four categories: perceptions and beliefsabout the self (I listened to the doctor), perceptions and be-liefs about the other (the doctor listened to me), perceptionsand beliefs about the relationship (we listened to eachother), and meta-perceptions and beliefs (I think the doctorbelieves that I listened to him or her). Feelings are alsolikely to influence the patientedoctor interaction and canbe grouped into the same four categories: for example, atthe level of the self, the patient could have felt comfortableduring the consultation, whereas at the level of meta-perceptions, the patient could think that the doctor felt thats/he (the patient) had felt comfortable during the consulta-tion. Our results suggest that mixing items of a different na-ture (e.g., cognition and feeling, assessing the self, andassessing the other) may lower the potential for valid andreliable dyadic measures. Thus, by revealing that instru-ments that use items of a similar nature are more reliablethan measures that do not, our results lay the groundworkfor the design of valid and reliable dyadic indices. In turn,dyadic indices may provide valuable information on theunique contribution of the dyad level to the decision-making process.
Second, our results have the potential to bring more clar-ity to SDM research methods, especially as regards the
Table 6. Dyadic data analysis
Model c2 df RMSEA CFI c2 difference df difference P difference
Values clarification subscale of the Decisional Conflict ScaleM1 6.18 6 0.01 1.00 d d d
M2 31.59 9 0.10 0.97 25.41 3 !0.01M3 3.36 3 0.02 1.00 2.82 3 0.42M4 5.09 5 0.01 1.00 d d d
Uncertainty subscale of the Decisional Conflict ScaleM1 9.74 6 0.05 1.00 d d dM2 25.18 9 0.08 0.98 15.44 3 !0.01M3 0.16 3 0.00 1.00 9.58 3 0.02M4 12.24 5 0.07 0.99 d d d
Abbreviations: df, degrees of freedom; RMSEA, root mean square error of approximation; CFI, comparative fit index.M1, model to test the equality of correlation: Model fit in which all intrapersonal and interpersonal correlations between physician and patient
items were forced to be equal. A good model fit means that the correlation structure for the items scale is similar between physician and patient.The representation ‘‘d’’ indicates no test available. We use the baseline model to test the equality of item variances and reliability; M2, model totest the equality of variances: Model in which variance items were forced to be equal between the physician and patient. A significant P-value forthe difference in chi-square between M1 and M2 means that variance items between the physician and patient are not the same. N/A denotes notpossible to test this hypothesis because of inadequate M1 model fit; M3, model to test the reliability: Model in which we free the latent variablevariances of one of the member. A significant P-value for the difference between the M1 and M3 means that the variability of physician’s and pa-tient’s variance is not similar for each item. (N/A) denotes not possible to test this hypothesis because of inadequate M1 model fit; M4, baselinelatent variable model: Model in which errors of the same items across dyad members were correlated and a correlation between the two latent vari-ables across the physician and patient was added. This correlation can be viewed as a measure of latent dependence. A good model fit means thatthe baseline model to test the equality of loading is adequate. The representation ‘‘d’’ indicates no test available. We use the baseline model totest the equality of loading; M5, model to test the equality of loading: Model in which loadings between items and latent variable were forced to bethe same for the physician and patient. A significant P-value for the difference between the M4 and M5 means that the constructs have the samemeaning for both members; M6, model to test the equality of variance for the latent factor: Model in which variances of latent variable were forcedto be equal for the physician and patient. A significant P-value for the difference between M5 and M6 means that variances for latent variablebetween the physician and patient are not the same. N/A denotes not possible to test this hypothesis because of inadequate M5 model fit.
1317F. L�egar�e et al. / Journal of Clinical Epidemiology 65 (2012) 1310e1320
issue of ‘‘finding common ground’’ or the process by whichthe physician and patient reach agreement on a course ofaction. Taking his model further, Kenny [34] introducesthe ideas of similarity (the parties feel the same way), accu-racy (the health professional’s perception of the patient’scognition or feelings is accurate), assumed similarity (eachparty assumes that the other party is feeling or thinking thesame way), meta-accuracy (each party knows how the otherparty sees him/her), assumed reciprocity (the physician
thinks that the patient is comfortable and thinks that the pa-tient perceives the doctor as comfortable), and assumedmeta-similarity (the doctor is comfortable and thinks thatthe patient also perceives that the doctor is comfortable)[34]. The significance of each of these types of agreementbetween the parties’ perceptions and beliefs about the en-counter has yet to be determined, but a study publishedin 2009 [35] suggests that patients want their physician tohave the same beliefs as themselves and that this type of
1318 F. L�egar�e et al. / Journal of Clinical Epidemiology 65 (2012) 1310e1320
agreement enhances patients’ satisfaction and increasestheir intent to adhere to their physician’s recommendations[36,37]. Interestingly, our results indicate that two of thefour measures found to be valid and reliable dyadic mea-sures were based on similarity: the values clarification sub-scale and uncertainty subscale. The other two subscaleswere based on accuracy: the perception of control subscale(the health professional’s perception that the patient’s cog-nition is accurate) and verification of understanding sub-scale (the patient’s perception of the physician’s behavioris accurate). Future research needs to delve further intothe extent to which the nature of the items of a measuredwhether the items measure accuracy, similarity or otherattributesdmake the measure more valid and reliable.
Third, our results indicate that although physicians’ andpatients’ perceptions of the clinical encounter and of eachother may differ [36,38e40], it is possible to identify validand reliable measures of process variables that mean thesame thing to both parties. This is important because itmay have some potential to reconcile the views of physi-cians and patients on specific components of SDM. In per-forming SDM research from a dyadic perspective, the mostfundamental conceptual question that needs answering iswhat processes or outcomes of interest should be assessedin the members of the dyad, and whether those processesor outcomes should be measured at the individual or dyadiclevel. With regard to processes, we refer in part to interac-tions between individuals.
Fourth, our results also confirm what previous studieshave demonstrated about using multiple measures of com-munication within the same study. While we observed norelationship between the third-observed OPTION scaleand our proposed dyadic measures, previous studies havereported that observer-coded measures may not be adequateto interpret the patient’s perception of the physician’s be-havior [41,42]. observer-coded measures should be usedto complement self-report measures, given that relyingsolely on patients’ perceptions has its limitations. One use-ful research strategy would be to use measures garneredfrom both observers and participants.
4.1. Strengths and limitations
This study has several strengths. It builds on publishedwork that mapped the essential components of an integratedmodel of SDM [19] that had been tested with both physi-cians and patients. Our sample of 264 family physiciansand 269 patients is large for this type of research, and toour knowledge, this is the first dyadic data set that can beused to explore the effect of certain variables on the per-sonal uncertainty of physicians and patients concomitantly[43]. Also, our recruitment of unique dyads avoided thebias of a physician completing the same questionnaire sev-eral times, possibly creating a learning effect.
In terms of limitations, we observed low scores on themeasures assessed, which may have affected our ability
to assess the dyadic nature of some of the scales. Further-more, our results are drawn from primary care consulta-tions, in which a wide variety of medical problems areencountered and diverse decisions must be made: this re-sults in heterogeneity [44,45]. This said, our results havethe advantage of reflecting the clinical context of primarycare, which is a highly accessed level of care [46]. Also,we thought that by using dyadic measures we could captureevidence of SDM rather than having to observe all encoun-ters. However, we observed no correlation between themeasures assessed and a third-observer instrument, OP-TION. It is possible that the elements of SDM that were as-sessed by both patients and their physicians in this study donot correspond to the 12 SDM-related doctor’s behaviorsthat are assessed by OPTION. In other words, OPTIONand the dyadic measures included in this study are assess-ing different constructs. Finally, we also need to recognizethat self-reported measures of perceptions have their limita-tions: many studies have reported a low degree of concor-dance between patients’ and practitioners’ perceptions ofthe same process [36,38].
5. Conclusion
Today’s health care observers acknowledge that a clinicalencounter in the 21st century is a meeting of two experts:the patient and the practitioner [47]. The research showsthat conversations are processes of mutual influence: the ac-tion of one communicator affects the action of the other[35]. In the clinical setting, this means that we shouldpay more attention to the interaction between the partiesthan to the individuals themselves [48].
For this reason, dyadic research is the perfect approachfor increasing our understanding of patient-physician inter-action [5,43]. Dyadic measures have the potential to cap-ture key mechanisms of reciprocity and mutual influencein clinical encounters. These instruments also lay thegroundwork for the design of valid and reliable dyadic in-dices. At the same time, dyadic methodology is relativelynew as applied to medicine. We believe that developingvalid and reliable dyadic measures in the context of SDMresearch will help researchers design or evaluate new typesof intervention for effective knowledge translation and ex-change and will surely steer SDM in new and exciting di-rections [49].
Acknowledgments
F.L. holds the Tier 2 Canada Research Chair in Imple-mentation of SDM in Primary Care. M.S. holds the DrBrian W. Gilbert Tier 1 Canada Research Chair in PrimaryHealth Care. J.G. holds the Tier 1 Canada Research Chairin Health Knowledge Transfer and Uptake. Louisa Blairedited this manuscript.
1319F. L�egar�e et al. / Journal of Clinical Epidemiology 65 (2012) 1310e1320
References
[1] Kelley HH, Thibaut JW. Interpersonal relations: a theory of interfer-
ence. Sons JW, editor. New York: John Wiley & Sons; 1978.
[2] Kenny D, Cook W. Partner effects in relationship research: concep-
tual issues, analytic difficulties, and illustrations. Personal Relation-
scope of and exposure to primary care physicians in Australia, New
Zealand, and the United States: cross sectional analysis of results
from three national surveys. BMJ 2007;334:1261.
[45] Braddock CH 3rd, Edwards KA, Hasenberg NM, Laidley TL,
Levinson W. Informed decision making in outpatient practice: time
to get back to basics. JAMA 1999;282:2313e20.
[46] Green LA, Fryer GE Jr, Yawn BP, Lanier D, Dovey SM. The ecology
of medical care revisited. N Engl J Med 2001;344:2021e5.
[47] Coulter A. Paternalism or partnership? Patients have grown up-and
there’s no going back. BMJ 1999;319:719e20.
[48] Street RL Jr, Makoul G, Arora NK, Epstein RM. How does commu-
nication heal? Pathways linking clinician-patient communication to
health outcomes. Patient Educ Couns 2009;74:295e301.
[49] Cappelleri JC, Stecher VJ. Analysis of dyadic data: a guideline
applied to erectile dysfunction. J Eval Clin Pract 2010;16:
1340e4.
Appendix A. Items included in the measures assessed
SDM components Scale SubscaleItem
number Physician subscale items Patient subscale items
Defining the problem Medical CommunicationCompetence Scale(Cegala, 1998)
Information giving (9) 1a I provided goodexplanations of thefollowing to thepatient: the diagnosisof his or her medicalproblem
The doctor explainedthe following to mysatisfaction: what mymedical problem was
1b I provided goodexplanations of thefollowing to thepatient: the causes ofhis or her medicalproblem
The doctor explainedthe following to mysatisfaction: thecauses of my medicalproblem
1c I provided goodexplanations of thefollowing to thepatient: the treatmentfor his or her medicalproblem
The doctor explainedthe following to mysatisfaction: what Icould do to get better
Presenting the options 1d I provided goodexplanations of thefollowing to thepatient: theadvantages anddisadvantages oftreatment option(s)
The doctor explainedthe following to mysatisfaction: thebenefits anddisadvantages oftreatment choice(s)(i.e., choice(s) aboutwhat I could do to getbetter)
1e I provided goodexplanations of thefollowing to thepatient: the purposeof any tests that wereneeded
The doctor explainedthe following to mysatisfaction: thepurpose of any teststhat were needed
1f I provided goodexplanations of thefollowing to thepatient: how thetreatment plan (e.g.,medicine and lifestylechanges) would helphis or her problem
The doctor explainedthe following to mysatisfaction: how thetreatment plan (e.g.,medicine and lifestylechanges) would helpmy problem
Discussing the prosand cons
1g I provided goodexplanations of thefollowing to thepatient: how toperform the treatmentplan (e.g., medicineand lifestyle changes)
The doctor explainedthe following to mysatisfaction: how toperform the treatmentplan (e.g., medicineand lifestyle changes)
1h I provided goodexplanations of thefollowing to thepatient: the possibleside effects of thetreatment plan (e.g.,medicine and lifestylechanges)
The doctor explainedthe following to mysatisfaction: thepossible side effectsof the treatment plan(e.g., medicine andlifestyle changes)
1i I provided goodexplanations of thefollowing to thepatient: the long-termconsequences of hisor her medicalproblem
The doctor explainedthe following to mysatisfaction: the long-term consequences ofmy medical problem
(Continued )
1320.e1F. L�egar�e et al. / Journal of Clinical Epidemiology 65 (2012) 1310e1320
Appendix A. Continued
SDM components Scale SubscaleItem
number Physician subscale items Patient subscale items
Clarifying the patient’svalues andpreferences
Decisional ConflictScale (O’Connor,2005)
Values clarification (3) 2a I am clear about whichbenefits matter mostto this patient
I am clear about whichbenefits matter mostto me
2b I am clear about whichrisks and side effectsmatter most to thispatient
I am clear about whichrisks and side effectsmatter most to me
2c I am clear about whichis more important formy patient (thebenefits or the risksand side effects)
I am clear about whichis more important tome (the benefits orthe risks and sideeffects)
Drawing on the doctor’sknowledge
Patient-PhysicianDiscordance Scale(Sewitch, 2001)
Physician’srecommendations (5)
3a To what extent was yourpatient’s mainconcern/problemdiscussed?
To what extent was yourmain concern/problem discussed?
3b To what extent did youdiscuss personalissues that mightaffect your patient’smedical situation?
To what extent did youand your doctordiscuss personalissues that mightaffect your disease?
3c To what extent did yourpatient expect toreceive aprescription?
To what extent did youexpect to receive aprescription?
3d To what extent did yourpatient expect to besent for furthertesting?
To what extent did youexpect to be sent forfurther testing?
3e To what extent was yourpatient satisfied withthis visit?
To what extent were yousatisfied with thisvisit?
Discussing the patient’sself-efficacy
Theory of PlannedBehavior
Perception of control (3) 4a After this consultation,to what extent do yousee this patient ascapable of followingthrough with thedecision that wasmade?
After this consultation,to what extent do yousee yourself ascapable of followingthrough with thedecision that wasmade?
4b After this consultation,how confident are youthat this patient willbe able to followthrough with thedecision that wasmade?
After this consultation,how confident are youthat you will be ableto follow through withthe decision that wasmade?
4c After this consultation, Ibelieve that thispatient has the abilityto follow through withthe decision that wasmade
After this consultation, Ibelieve that I have theability to followthrough with thedecision that wasmade
