Background

Practising medicine involves making decisions at all stages of the clinical process [1]. Although a great deal of varied terminology is used to describe doctors' thinking, the term "decision-making process" is used extensively in the medical and healthcare literature [2]. The decision-making process is broadly defined as global judgements by a clinician about the appropriate course of action and is said to be unspecified, as a number of processes may produce a decision [3]. In clinical settings, it is also understood as the use of diverse strategies to generate and test potential solutions to problems that are presented by patients and involves using, acquiring and interpreting the indicators and then generating and evaluating hypotheses [4]. Processes or strategies that will be used may be based on what the clinician was taught, his or her own representation of the evidence supporting each course of action, or the prevailing practice in a given institution [4].

In recent years, there has been a growing interest in new representations of the clinical decision-making process that better address its complexity within specific clinical encounters. Indeed, providing medical care to a patient is now increasingly considered a dynamic and interactive process known as "shared decision-making" [5–7]. Characteristics of shared decision-making include that at least two participants, clinician and patient, be involved; that there be a two-way exchange not only of information but also of treatment preferences; that both parties take steps to build a consensus about the preferred treatment; and that an agreement be reached on the treatment to be implemented [5]. Shared decision-making includes the following components: establishing a context in which patients' views about treatment options are valued and deemed necessary, transferring technical information, making sure patients understand this information, helping patients base their preference on the best evidence; eliciting patients' preferences, sharing treatment recommendations, and making explicit the component of uncertainty in the clinical decision-making process [8].

Shared decision-making does not exclude a consideration of the values and preferences of the physician and occurs through a partnership in which the responsibilities and rights of each of the parties and the benefits for each party are made clear [9]. Given the recognition that patient-physician interactions and by extension, clinical decision-making processes, are dynamic and reciprocal in their nature, it is surprising to find little systematic evaluation of the physicians' perspective of this entity [10]. Consequently, there has been a renewed interest in capturing the perspective of physicians of the decision-making process within specific clinical encounters. Therefore, the aim of this systematic review was to identify instruments that assess the perception of physicians of the decision-making process within specific clinical encounters.

Methods

Search strategy

Covering all years available (to April 2007), we conducted an electronic literature search of the following databases: PubMed, PsycINFO, Current Contents, Dissertation Abstracts and Sociological Abstracts. Three information specialists were consulted to help develop, update and run the search strategy. The following MeSH terms and free text words were used to create specific search strategies for each database: "decision making", "physicians", "health personnel", "doctors", "practitioners", "health personnel attitudes", "measurement", "questionnaire", "psychometrics" and "psychological tests". We included titles of publications and their respective abstract in English or French that potentially included an eligible instrument. Initially, if a dissertation abstract was found along a publication, both were kept. We also contacted 10 experts in the field (list available from authors) and contacted corresponding authors of included instruments. Lastly, we reviewed bibliographies of the included instruments. Once we included an instrument, we conducted an electronic search of the first author.

Selection criteria

All of the searches were downloaded to a reference database for initial screening of titles and abstracts by a single member of the review team. Prior to screening, duplicates were removed from the database. Titles of publications and their respective abstract reporting editorials, letters, surveys, clinical vignettes or the completion of an Objective Structured Clinical Examination or the evaluation of a simulated patient were excluded. After the initial screening, if detailed information about the titles of publications and their respective abstract was questionable, the full text of these publications was sought. Then, two reviewers independently appraised these publications to identify ones that reported on the use or development an eligible instrument. Discrepancies between the two reviewers were resolved through discussion.

Identification of eligible instruments

The following inclusion criteria were applied: 1) a self-administered instrument was presented; 2) the instrument evaluated the perspective of physicians, including residents, of the decision-making process within specific clinical encounters, 3) the collection of data occurred after a specific clinical encounter in a 'real' clinical setting; 4) the report included sufficient description to permit critical appraisal of the instrument (for example, the instrument was provided as an appendix or we were able to get a copy from the author); and 5) there were quantitative results following the administration of the instrument. An instrument was defined as a systematic procedure for the assignment of numbers to aspects of objects, events or persons as indicated by its construction, administration and scoring procedure according to prescribed rules [11].

The outcomes of interest included the perception of physicians of the decision-making process within specific clinical encounters as well as the outcome of the decision itself such as satisfaction with the decision. The decision-making process was defined in an inclusive manner as global judgements by a physician about the appropriate course of action [3]. An instrument was deemed eligible if one of its sub-scales or some of its items tapped into the outcomes of interest.

Data extraction

The data extraction form, derived from McDowell (1987) [12], covered characteristics of the source of information and characteristics of the instrument itself, such as name of the instrument, origin of first author, main purpose, description of the instrument, characteristics of the response scale, presence of a corresponding patient instrument, development procedures, conceptual/theoretical foundation, validity, reliability (e.g. internal consistency) and responsiveness of the instrument.

A conceptual framework was considered to be used if the author referred to a set of concepts and the propositions that integrate them into a meaningful configuration [13]. A theory was deemed to be used if the author referred to a theory, defined as a series of statements that purport to account for or characterize some phenomenon with a much greater specificity that a conceptual framework [13]. Otherwise, the nature of the source of references used by the author was used to identify a broad conceptual basis.

Content validity (i.e. the extent to which all relevant aspects of the domain or area that is being measured are represented in the instrument), construct validity (i.e. the extent to which the instrument relates to other tests or constructs in the way that was expected) and criterion validity (i.e. the extent to which the instrument relates to a gold standard to which it is compared) were also assessed [14]. Responsiveness (i.e. the extent to which the instrument measured change within persons over time) was also assessed [15].

Using the Science Citation Index, we assessed how many times the included instruments had been cited in subsequent published research in French or English. Lastly, for each instrument, we assigned one main evaluation domain defined as a subjective interpretation by the reviewers of the main construct that the instrument was assessing. Sources of disagreement were discussed and resolved by consensus and only consensus data was used. Data extraction was completed by two members of the team.

Quality assessment

The quality of reporting of the included studies was assessed by two reviewers independently, using a modified version of the following instrument, Standards for Reporting of Diagnostic Accuracy (STARD) [16–18]. The original STARD contains 25 items pertaining to study question, study participants, study design, test methods, reference standard, statistical methods, reporting of results and conclusions. However, because we were interested in instruments assessing the perception of physicians of the decision-making process within specific clinical encounters, we added one more item under the section "statistical methods." This new item assessed if the authors of the included instrument had taken into account that one physician could only contribute to one questionnaire for the statistical analyses used to provide evidence on its reliability and validity (i.e., the non-independence of data). For each instrument, we chose one main study. In instruments for which more than one report was included, we chose the one that reported the most details on the development and psychometrics of the instrument in its most recent version.

Results

Included instruments

The initial search resulted in 3431 records (Figure 1). From these, 192 records that were in a language other than French or English, and 138 duplicates were removed. After applying our eligibility criteria, 218 full text articles were retrieved for detailed evaluation. Twenty-six instruments (67 articles) were potentially eligible of which a further 15 (28 articles) were excluded because they were not designed to collect data for a specific clinical encounter [19–46]. Therefore, 11 instruments (39 articles) were included [47–85]. We were able to get access to a published version or a copy of all included instruments.

Figure 1
figure 1

Progress through the stages of the systematic review.

Full size image

Characteristics of the included instruments

Overall, the included instruments were published between 1986 and 2007 (Table 1) [47–85]. Nine instruments were developed in North America and available in English [47–67, 69–72, 74–85]. Among these, three were available in French [57, 58, 64–67, 69–71, 77–82]. Two instruments were developed in Europe [68, 73]. For most instruments, the first author was affiliated with a Faculty of Medicine or a medical organisation (n = 8) [54–59, 64–72, 74, 76–85], followed by a School of Nursing (n = 1) [47–53, 75], Department of Communication (n = 1) [60–63] and Research Group in Psychosomatic Rehabilitation (n = 1) [73]. Most instruments were developed for non-specific clinical problems (n = 9) [54–74, 76, 83–85]. One instrument was developed for intensive care unit-related problems [47–53, 75] and one for inflammatory bowel diseases [77–82].

Table 1 Characteristics of the 11 included instruments
Full size table

All instruments were multi-dimensional. The mean number of items per instrument was 16.7 (range: 6 to 37). Seven instruments used a Likert response scale [47–54, 57, 58, 60–67, 69–72, 75, 76, 83–85], two used a Visual Analog Scale [55, 56, 59, 74, 77–82] and one used a 5-point categorical response scale [73]. One instrument used a mix of response scales [68]. Five instruments offered a patient version [57, 58, 60–71, 77–82].

Based on the Science Citation Index, nine instruments had been cited at least once in subsequent research published in French or English with the older ones being more likely to be cited more often (Spearman r = -0.68; p = 0.03).

Development procedures and psychometrics of the included instruments

Authors of nine instruments reported on their explicit use of some conceptual framework or broad conceptual domain (Table 2) [47–53, 55–75, 77–82]. For seven instruments, we were able to find evidence of their validity and reliability [47–67, 69–71, 74–76, 83–85]. For two instruments, evidence of validity and reliability data was available only for the combined use of the physician's and patient's questionnaires [68, 77–82]. For one instrument, we could not find evidence of validity and reliability data published in French or English [73].

Table 2 Development and psychometric properties of the 11 included instruments
Full size table

None of the instruments provided data on their responsiveness (i.e. the extent to which it measures change within a physician over time). Lastly, the main evaluation domain that was assigned to each instrument were: satisfaction with the clinical encounter (n = 2) [54, 76, 83–85], mutual understanding between the health professional and the patient (n = 2) [68, 77–82], mental workload (n = 1) [55, 56, 59, 74], frustration with the clinical encounter (n = 1) [72], nurse-physician collaboration (n = 1) [47–53, 75], perceptions of communication competence (n = 2) [58, 60–63], degree of comfort with a decision (n = 1) [57, 64–67, 69–71] and information on medication (n = 1) [73].

Quality of the studies that reported on the included instruments

Overall, the mean number of items reported on the modified STARD was 12.4 (range: 2 to 18}(Table 3). During the development of four instruments, the authors used an analytical approach that took into account the non-independence of data [47–53, 55–57, 59, 64–67, 69–71, 74, 75, 84, 85]. For the Mental Work-Load Instrument, the authors used a mean score per physician to perform the correlation analyses [55, 56, 59, 74]. For the Physician Satisfaction Questionnaire, the authors used a bootstrapping approach to perform the factor analysis [84, 85]. For the Collaboration and Satisfaction about Care Decisions instrument, the authors restricted their sample size to one data entry per physician (n = 56) to perform most of their analyses [47–53, 75] and for the Provider Decision Process Assessment Instrument, the authors used a bootstrapping approach to perform their reliability analyses (i.e. Cronbach alpha) [57, 64–67, 69–71].

Table 3 Quality assessment of the studies that reported on the included instruments based on the modified version of STARD * For this instrument, only one publication in English was found. This publication reported on the study of physicians that had used the instrument. Other publications pertaining to this instrument were in German
Full size table

Discussion

We believe that the results of this systematic review are important. First, they indicate that there is an interest expressed by clinicians, health services researchers and educators in assessing the perspective of physicians about the processes leading to a decision within specific clinical encounters. This is congruent with the increasing number of randomized trials and systematic reviews examining the efficacy of interventions designed to bring about a change in clinical practice [86]. However, most of these trials assessed a change in health professionals' behaviour without assessing the underlying decision-making process that lead to such behavioural change. This review provides a list of standardized measures of the physician perspective of the clinical decision-making process, an essential step prior to behavioural change. Moreover, most of the included instruments provided some account of their conceptual or theoretical underpinnings. This is important because more attention needs to be given to the combination of different theories that could help us understand professional behaviours [87–90]. Therefore, this review provides health services researchers and educators with a set of standardized and theory-driven instruments that have the potential to improve the quality of implementation studies and by extension our understanding of health professionals' behaviour changes.

Second, this review provides evidence that health services researchers are beginning to use a dyadic and relationship-centered approach to clinical decision-making [91–93]. In other words, health services researchers are moving from studying groups of patients and health professionals separately to studying both simultaneously. For example, five of the six most recently developed instruments had corresponding patient versions [57, 58, 60–71, 77–79, 81, 82]. Moreover, for the authors of two of these instruments, evidence of validity and reliability data was available only for the combined use of the physician's and patient's questionnaires [68, 77–82]. This observation suggests that, increasingly, the clinical decision-making process is perceived as not being dissociable from the complex aspects of interdependence occurring between the physician and the patient. Indeed, the patient-physician relationship is an important component of physicians' satisfaction with their job [93]. Physicians' judgements about their experience with individual patients both reflect and shape what takes place during office visits and beyond [84]. This symmetry supports empirically what has previously been described on the basis of personal needs, namely, that both the physician and the patient have the same human needs for connection which can be fulfilled in the clinical encounter [84]. Therefore, future research in the field of clinical decision-making should foster the use of patient and physician versions of a similar instrument. In line with the growing interest for shared decision-making, this may allow for a more comprehensive assessment of the complexity of the clinical decision-making process and thus of its dynamic and reciprocal nature [65].

Third, this review highlights the need for further methodological development in studies assessing the perception of physicians of the decision-making process within specific clinical encounters. None of the authors of the included instruments provided data on the responsiveness of their instruments (i.e. the extent to which the instrument measures physician change over time). Also, 'within physician' clustering of multiple data points (i.e. non-independence of data) produced statistical challenges that were dealt with inconsistently by their developers. In one instrument, clustering of multiple data point under each physician was taken into account for the factorial analysis but not for the reliability analyses [84, 85]. Authors who took clustering into consideration used one of three strategies: average score per physician [55, 56, 59, 74], one data entry per physician [47–53, 75] or bootstrapping [57, 64–67, 69–71, 84, 85]. Therefore, methodological development in this area will be needed to ensure that responsive instruments and adequate analytical approaches are used in studies assessing the perception of physicians of the decision-making process within specific clinical encounters.

Lastly, for the included instruments, the mean number of items ranged from 6 to 37 items (mean = 16.7). It remains a challenge for health service researchers to develop sound measurements for conducting implementation studies that will minimize the burden to participating physicians. In our own experience, and in line with what has been reported in the literature, there appears to be an association between instrument length, defined in this systematic review as the number of items included in an instrument, and physician participation in studies [94]. This is perhaps even more apparent for health professionals' self-administered questionnaires after a specific clinical encounter. As such, our results provide some valuable insight or benchmarking about the number of items included in the instruments that are currently available for conducting studies on clinical decision-making with physicians.

This review has a number of limitations. Studies reporting the development of instruments are generally not well-indexed in electronic databases [95]. In this review, the search strategies used may not have been optimal even though we consulted with three experienced information specialists. It is possible that some eligible instruments as well as relevant publication regarding the included instruments were not included in this review. Also, clinical decision-making is moving from a unidisciplinary perspective to an interdisciplinary perspective [20]. Therefore, the included instruments might not be representative of on-going developments in healthcare decision-making. Indeed, recent health services policy documents clearly indicate the need for patient-centered care provided by an interprofessional team [96]. However, in a review on barriers and facilitators to implementing shared decision making in clinical practice as perceived by health professionals, the vast majority of participants (n = 2784) enrolled in the 28 included studies were physicians (89%) [97]. This suggests that more will need to be done to enhance an interprofessional perspective to shared decision making, a process by which a patient and his/her healthcare providers engage in a decision-making process. We firmly believe that the instruments that were identified throughout this review could be further developed using this interprofessional perspective.

Lastly, it is interesting to note that for the eleven included instruments, the mean score of items on the STARD was 12.4 (range: 2 to 18). It is important to emphasize that seven of the included instruments were published before the 2003 STARD criteria. Although, this mean score compared well to the mean scores of items on the STARD that were reported in test accuracy studies in reproductive medicine: 12.1, future research in this field will need to improve the reporting of the development of instruments that would assess healthcare professional's perspective of the decision-making process

Conclusion

This systematic review provides valuable data on instruments that assess the perception of physicians of the decision-making process within specific clinical encounters. It can be used by educators and health services researchers as a repository of standardized measures of the physician perspective of the clinical decision-making process and we hope of other healthcare providers. It was not our intention to identify the "best" instrument but rather to offer options to the target audience. We believe that based on the context of its intended use, a process of weighting its limitations and strengths and other factors faced by its potential users, most if not all of the identified instruments might play a valuable role in the future. This systematic review also sent an important signal: in the XXI century, the clinical decision-making process might only be adequately assessed by using a dyadic approach. In this regard, some of the identified instruments might be more attractive than others. However, more research is needed to investigate the validation of these instruments. More specifically, for the production of evidence on the validity and reliability data of the instruments, analytical methods that take into account within physician clustering is required. For all the included instruments, the development of corresponding patient versions should be encouraged. The combined use of the patient version with its respective healthcare professional version will help capture the complexity of the clinical decision-making process and thus of its dynamic and reciprocal nature. Only then will a new and more comprehensive understanding of health-related decision-making in the context of specific clinical encounters be possible.