In the present study, a model of shared decision-making was tested via path analysis. The a priori postulated associations between different SDM-related constructs were examined in a sample of primary care patients. Furthermore, potential confounders (e.g., age, sex, education) of these SDM-related constructs were considered.
The examination of global goodness-of-fit indexes indicated an acceptable congruence between the model and the observed data. Although the index Χ2 turned out to be significant and thus suggested a difference between the model and the observed data, the sample sizes were so large that the significance of the Χ2-test is of little information [33, 34]. Furthermore, the Tucker-Lewis Index and Comparative Fit Index were below the recommended value of .90 and thus showed a non-sufficient match between the data and model. However, normed Χ2 values and the Root Mean Square Error of Approximation indicated an acceptable fit. In summary, it can be concluded that the global goodness-of-fit indexes, though not optimal, support the plausibility of the proposed model. It is possible that a data-driven modification of the measurement model and a post hoc adaptation of the instruments (e.g., through allowing for correlated error terms between items within the scales) would have led to a better global fit. However, refining the measurements of the examined constructs was not the focus of this investigation. As the concept of SDM is still relatively new, both the refinement of existing instruments and the development of new instruments measuring SDM-related constructs are necessary [12, 35–37].
Altogether, our theoretical assumptions could be largely corroborated by the collected empirical data.
Consistent with our hypothesis, higher patient involvement clearly lowered decisional conflict. The effect was quite high, and the finding is consistent with the literature [38, 39]. Therefore, the postulated effect between the process and an intermediate endpoint of SDM could be confirmed. A result that was inconsistent with our expectations was the negative association between the patients’ preference for involvement and their current involvement. In the model, it was assumed that patients who wish to be involved in decision-making would actually be more strongly involved. In the empirical data, the opposite was found. A possible explanation could be that those patients who had a strong involvement preference had high expectations concerning their involvement and thus experienced the actual involvement as unsatisfactory. This explanation could also account for the negative association between preference for involvement and satisfaction with the physician. It is possible that those patients who had high expectations concerning their involvement evaluated their consultation and their physician more critically. Another explanation could be that preference for involvement is highly subjective depending on the context and circumstances . Thus, involvement preference in the specific medical encounter might be very different from the generic preference for involvement in decision-making. These results highlight the importance of collecting data on patient preferences for involvement in addition to the assessment of the experienced involvement to be able to assess the concordance between these measures.
Satisfaction with the physician was clearly affected by patient involvement and decisional conflict. An increased patient involvement affected satisfaction with the physician directly and indirectly through decreased decisional conflict. Thus, decisional conflict may be considered as a mediator between involvement and satisfaction with the physician. The direct and indirect effects of involvement on satisfaction with the physician summed up to a large effect (.54) in the path model. This result is highly consistent with the results of Quaschning et al.  who used a very similar approach. In line with our results, they could explain a high proportion of variance in patient satisfaction by patient involvement. Satisfaction with decision proved to mediate the effect of patient involvement on patient satisfaction. As satisfaction with decision is a construct that can - just like decisional conflict - be categorized as decision outcome  this result is very close to our findings.
The findings were relatively independent of covariates. The most affected of all the constructs was the preference for involvement. Both higher age and lower educational status are known to decrease preference for involvement . In accordance with these findings, a medium influence of age and a moderate influence of education on involvement preference were found in this study. Both physical and mental health-related quality of life covariates were moderately associated with the SDM-related constructs experienced involvement and decisional conflict in the development sample. In the confirmatory sample, only the interrelation between mental quality of life and decisional conflict proved to be relevant. Thus, a low mental quality of life is possibly associated with increased decisional conflict. For the covariates that dropped out in the iterative process of model development (sex, clinical characteristics, and type of decision), it can be assumed that they may not have a substantial effect on the constructs in the examined population.
When interpreting the present results, some limitations should be taken into account. First, on average, the examined sample was of older age and of low-educational level and rural origin. Second, due to the cross-sectional design of the study, the examined associations are correlative and not necessarily causal. The temporal relationships in the model had to be assumed and cannot be confirmed using cross-sectional data. However, these assumptions may be supported by the instructions applied in the study. For example, for data collection on SDM, participants were instructed to rate physician and conjoint behaviour during the decision-making process, while for decisional conflict, they were asked for their experiences after decision-making. Third, the results of our study may also be influenced by the limitations of memory, as patients had to remember their last clinical encounter. Furthermore, all sources of data are derived from the same questionnaire, what might lead to common methods bias. Although the intervention and the two control groups did not differ substantially either regarding demographic and clinical characteristics in baseline comparisons or regarding associations of the investigated constructs in sensitivity analyses, the use of the pooled baseline data might still have introduced some unobserved bias.
As data on the treating physicians were not available, we were unable to account for the hierarchical structure of our data due to patients clustered by physician. This is a serious limitation and may have led to an underestimation of the standard error of our parameters. Additionally, our model only considers one of many possible intermediate and long-term endpoints. With respect to our instruments, it should be taken into account that decisional conflict considers many different facets and thus is quite heterogeneous. Therefore, results on this scale should be interpreted with caution.
Future research on the associations among different SDM-related constructs should incorporate longitudinal data from intervention studies. By using this design, causal effects of changes in patient involvement on decisional conflict and satisfaction with the physician can be clarified. The role of patient preferences for involvement in the process remains poorly understood. Further investigation on this topic is needed, as it is a central construct in recent debates on indications for applying shared decision-making. More empirical evidence on the associations among the SDM-related constructs could lead to a better understanding of the decision-making process. Other constructs of SDM could be incorporated in future investigations. Based on the model, nomological networks for the validation of psychometric instruments could be developed and tested. A better theoretical foundation of SDM could be helpful for increasing the implementation of this promising concept into health care. The model could also help to choose adequate outcome parameter for studies of shared decision making (e.g. a study focusing on patient preferences would need other instruments than a study with focus on the process).