All the consultations in this study took considerably longer than the time usually available in primary care, approximately ten minutes [43
]. Thus it is unlikely that, in their current form, the paper-based or computerised decision aids can be easily incorporated in GPs' routine consultations. Centrally, both computerized versions of the DARTS II tool prolonged the risk assessment and decision making phases compared to paper-based guidelines. These computerised decision aids were applied in the context of shared-decision making and it has been noted that actively involving patients in treatment decisions requires additional time [44
]. The computerised decision aids also prolonged the closing phases of the consultation. Reviewing the videos revealed that the GPs were often engaged either in procedural work such as printing off summary reports from the tool or in explaining the 'role' of the computer in the consultation.
In the 10 minutes leading up to the decision point, information-giving and technically-focused conversation dominated proceedings. Although patients' talk was primarily technical, it contained twice the socio-emotional content of the GPs' conversation. Given that the socio-emotional or affective component of clinicians' talk is a key factor in patient's positive evaluation of consultations [46
] and subsequent adherence to treatment [47
], it is essential to establish if the use of technology encourages a shift towards technical language at the expense of inter-personal work. Our results showed no significant difference across the arms of the trial in the proportion of technical to socio-emotional talk shown by GPs. Only the length of time spent working through the tools differed. Thus the technological complexity of the decision aid did not seem to affect the balance of technical to socio-emotional language used by GPs. However, future work should compare the impact of decision aids (in any form) on the technical/affective language balance compared to general consultation discourse.
What was consistent across all consultations in this study was the large amount of information provided to patients which tended to increase as the sophistication of the decision aids increased, albeit with a time cost. Overall, the GPs in this study spent 55% of the consultation giving information and 7% seeking it, which differs from the average proportions of 35% and 23% reported in general medical dialogue [41
]. The patient proportions of 33% and 7% in this study compare with 54% and 6% reported elsewhere [41
]. In this respect the decision aids appeared to be useful tools in presenting patients with information, suggesting that they might encourage an 'informed' model of treatment decision-making [48
] where the clinician provides all the relevant information to patients who then select the treatment they deem most appropriate [5
]. However, the enactment of 'shared' decision-making may remain dependant on the GP/patient social interaction and not directly shaped by technological decision aids.
It was clear that GPs in this study conversationally led the consultation, contributing nearly two-thirds of all utterances and showing more pauses within their speech than patients. More powerful conversants generally talk for longer and are allowed more uninterrupted pauses in their speech [49
]. The verbal dominance seen in this study is similar to the average 60:40 ratio (doctor:patient) reported in general medical dialogue [41
]. Indeed we observed no significant difference in verbal dominance, or clear differences in verbal content, between the 'shared decision making' consultations of the computerised decision aids and the 'paternalistic' consultations involving paper-guidelines. This may appear counter-intuitive, in that one might expect shared decision making consultations to have less verbal dominance than paternalistic consultations. However, it may be that it takes considerable verbal work from doctors to introduce and sustain any model. In addition, such models do not just occur and an initial intention to share may get lost in the unfolding dynamic of the consultation [49
], particularly where clinicians are tasked to work with and through a 'third party' in the consultation, in the form of a decision aid.
GPs and patients both seemed behaviourally engaged in these consultations, showing a great deal of paralinguistic registering of the conversation [34
]. There were a number of significant differences in GP's nonverbal activity across the arms of the trial but no consistent trends. Nevertheless, nonverbal activity appeared to be a key aspect of communication and we found a great deal of mirroring in eye-gaze and illustrative gesturing [49
]. For example, on one occasion a GP was attempting to explain atrial fibrillation to a patient and he used his hands to demonstrate the pumping and fluttering action of the heart. Thus it is clear that unique information can be conveyed visually [50
]. Indeed it has been reported that just 7% of emotional communication is conveyed verbally compared to 22% via voice tone and 55% through visual cues [32
]. However, although non-verbal behaviour has long been regarded to be an important feature of good communication [51
] and an influence on medical outcomes such as patient understanding, compliance and satisfaction with care [52
] it is much less well understood than verbal behaviour [53
]. Thus more attention should be focused on identifying and understanding the contribution of non-verbal behaviour to communication in clinical contexts.
Due to the exploratory nature of the process study, particularly the fact that each video-taped consultation was linked to two subsequent in-depth interviews which was highly labour intensive, the number of patients in this observational analysis was relatively small. Moreover, since the data were not normally distributed, we used conservative non-parametric statistics in our analysis, which was intended to be primarily descriptive. Thus it is possible that real but small differences between the three types of consultation were not detected which may have emerged with a larger sample. Also the efficacy design of the trial meant that just one GP delivered each decision aid and so our findings may be confounded by GPs' style in consultations rather than the decision aid per se. Nevertheless, the GPs' verbal dominance and technical focus was apparent across all the consultations on this study.
Although patients were randomised into the parent trial, we used a convenience sample of the first 29 patients who agreed to be videotaped (31 were asked). The trial eventually recruited 109 patients . In addition, the earlier recruits in our study are more likely to be prevalent cases in the population with a longer experience of atrial fibrillation compared to later incident cases. Moreover, the third arm of the trial (explicit version of DARTS II) was discontinued after eight consultations as a result of data produced by process study. We observed that a number of patients found it difficult to understand the preference elicitation exercise in these consultations and were uncomfortable with this activity [24
]. Thus this divergence between the early process study and the final trial clearly limits our ability to generalise findings from the former to the latter. However, the relative lack of differences in patients' behaviour across the differing types of consultations may help explain why the trial found only transient effects of a computerised decision aid on patients' decision conflict and no difference in actual treatment decisions [54
Due to the complexity of video-based data, a 10 minute slice was sampled for detailed content analysis. The fact that GPs in this study were so focused on information-giving may reflect the fact that conversational content varies in different phases of a consultation [37
] and this work focused on the time just before the decision point. However, within the consultations promoting shared decision-making this should have been the time when patients were most actively involved in the conversation. We found that they were much less vocal than the GPs at this crucial time point.
In addition, consultation timing varied depending on which decision aid was used and so the sampled 'slice' of the consultation may have crossed differing sub-phases. For instance, some physical examination work was included in most time-samples from guidelines consultations, some from implicit tool consultations and none from the explicit tool consultations. However, the lack of clear differences in verbal behaviour between the trial arms suggests that the latter did not strongly affect our analysis.
Perhaps most importantly, the patients in this study were also participants in a trial who had experienced extensive consent procedures, who were not seeing their own GP and who were attending an unfamiliar clinic. This context may further explain the lack of differences in treatment decisions reported by the trial [54
]. Subsequent interview work indicated that some participants regarded themselves as 'subjects' in a research study rather than 'real' patients [26
]. Moreover, conversation analysis of these consultations revealed how the experimental context of these consultations, at moments, overwhelmed the clinical context of the interaction [25
]. In this way, the impact of the trial context on both patients and clinicians may have outweighed any impact of the decision aids themselves. For patients, the eventual treatment decision may not have been perceived as 'real' and for clinicians the unfamiliar patients may have prompted less socio-emotional language than usual and more conversational work to maintain the dialogue.
Although the efficacy trial design (with high internal validity) may limit the applicability of our process data to the 'real world' of primary care, it is important to evaluate the impact and acceptability of complex interventions, such a decision aids, on a limited sample of patients (and indeed clinicians) before wider scale roll out. The work in this study revealed significant difficulties for patients with the extended version of the DARTS II tool, despite earlier pilot work. We also found that working with this decision aid took clinicians and patients much longer than the time usually available for consultations in primary care. Thus further development of this decision aid may be required before the next evaluative stage which should incorporate a more pragmatic (clinically representative) study design.