Healthcare professionals are continually exposed to new research findings that could contribute to more effective and efficient patient care. Unfortunately, the transfer of research findings into practice does not happen as readily as desired [1], and many authors have documented gaps between evidence-based practices and the routine clinical practice of healthcare professionals [2, 3].

A wide range of factors can influence the clinical practice of healthcare professionals [4], including individual motivational predispositions to change as well as economic, political, and organizational contexts. However, our understanding of these factors and optimal approaches to change healthcare professional behaviour is incomplete. This has led to calls for more theory-based research to better inform the design of interventions to change healthcare professionals' behaviour [1, 5, 6]. Although several theoretical perspectives could be used to explore the determinants of the healthcare professionals' behaviours, most or many clinical practice adoption decisions are individual professional decisions [7]. Consequently, it would be useful to obtain a better understanding of the individual mechanisms of the adoption of new behaviours from social psychology theories [8]. For the purpose of this review, social cognitive theories refer to theories where individual cognitions/thoughts are viewed as processes intervening between observable stimuli and responses in real world situations.

The problem of understanding why healthcare professionals do or do not implement research findings can be viewed as similar to finding out why people in general do or do not adopt a given behaviour such as health-related habits. This has been extensively investigated, and social psychological theories have already demonstrated their value. For the prediction of health-related behaviours, there are several social cognitive theories that predict moderate to large amount of the variance of intention and behaviour [9].

It is surprising that relatively little attention has been given to reviewing published studies applying social cognitive theories investigating healthcare professional behaviours. It is only recently that two publications have reviewed specific aspects of theory-based studies of healthcare professional behaviour and practice. Eccles and colleagues [10] concluded that intention was a valid proxy measure for behaviour among clinicians (i.e., physicians, nurses, pharmacists, other health workers). They did not quantify the strength of association between intention and behaviour among healthcare professionals, but based on the review of ten prospective studies, they concluded that this association was similar in magnitude to that reported for non-professional populations. For example, in a quantitative summary of meta-analyses, Sheeran estimated that, on average, 28% of the variance in behaviour (R²) is accounted for by intentions [11].

A review by Perkins and colleagues [12] was limited to applications of the theories of reasoned action (TRA) [13] and planned behaviour (TPB) [14] to understand clinicians' behaviour (i.e. physicians, nurses, pharmacists, other health workers). They found very few studies (N = 19), and only half of them (N = 9) included a measure of behaviour (eight self-reported; one objective from medical record). As in the review by Eccles and colleagues [10], they also did not quantify the strength of association between TRA/TPB constructs and actual behaviour, but nonetheless concluded that different constructs of these two theories predict intention and behaviour among different groups of clinicians.

Obviously, more information is needed regarding the usefulness of social cognitive theories to understand and predict healthcare professionals' intentions and behaviours. The aim of this study was to review systematically the literature to quantify to what extent studies based on social cognitive theories explain intention of healthcare professionals to adopt clinical behaviours and predict health professionals' clinical behaviour. Given that any of several social cognitive theories could have been used to investigate healthcare professional behaviours, this review was not limited to applications of the TRA and TPB. Other social theories such as Bandura's social cognitive theory [15], Triandis' theory of interpersonal behaviour [16] and others theories of behaviour were included as well.

The present study examined the efficacy of studies based on social cognitive theories in explaining intention and predicting the clinical behaviour of healthcare professionals. By means of a systematic review, the overall efficacy was evaluated and the effect of factors that could affect the efficacy of prediction was also examined. Overall, the efficacy of prediction of behaviour was equivalent to values reported in several meta-analyses of the TPB, the most widely used social cognition model of health behaviour. For instance, between 25.6% and 34% of explained variance in behaviour was reported for applications of the TPB [9, 99]. The current frequency-weighted mean R² of 0.31 for the prediction of healthcare professional' behaviours compares very favourably to these figures. Regarding the prediction of intention, however, the value observed in the present study (59% explained variance) was higher that the values reported for applications of the TPB (33.7% in Conner and Sparks [9], and 40% in Godin and Kok [99]). A possible explanation for this is that the present review was not limited to the TPB. Other theories were investigated and consequently variables other than those identified in the TPB were considered in the prediction. For instance, role beliefs and moral norm are important variables in Triandis' theory that emerged as substantial determinants of intention.

This systematic review also showed that a number of factors affect the efficacy of prediction of intention/behaviour. On this regard, type of health professionals and behaviour categories, sample size, psychometric qualities, method for assessing behaviour, level of correspondence between the operational definitions of intention and behaviour required special attention.

Variations in the efficacy of prediction of intention and behaviour were observed between types of healthcare professionals. In the prediction of behaviour, the best predictive models were observed for healthcare professionals other than physicians and nurses, whereas the best prediction of intention was observed among the nurse samples. Similarly, important variations in explained variance of professionals' behaviours and intentions were observed between behavioural categories. It is not clear what underlies these variations in efficacy of prediction, but one possible explanation could be the nature of the behaviour to be performed and the context of practice. This was particularly evident in prospective studies among physician samples, in which these two elements were defined more vaguely probably because the clinical practice of physician is more difficult to define accurately. This interpretation is further supported by our observation that the operational definitions of intention in terms of action and context for the prediction of behaviour were generally more precise in other healthcare professional samples compared with the studies of physician samples. Given the complexity of clinical-related behaviours, and particularly for diagnostics and treatment decisions, behaviour adoption could be modulated by several aspects of the context, such as patients' acceptability or preference for a given treatment, characteristics of the health problems, new versus usual patients, patients with multiple symptoms, antecedents or counter indications for a given type of medication, etc. Consequently, the accuracy of intention to predict future behaviour is reduced. Obviously, further research should pay more attention not only to the definition of the targeted behaviour, but also to its context of realization. As such, the use of vignettes could be a useful avenue to define more specifically the context of behavioural performance. For instance, Harrell and Bennett [22] successfully used a vignette to predict prescribing behaviour among a physician sample. They were able to explain 26.8% of variance in a behaviour assessed objectively. Thus, the use of vignettes could help healthcare professionals to better define the context of behavioural performance and formulate their intention more accurately. Consequently, the efficacy of social cognitive theories to understand healthcare professionals' behaviour could be improved and the findings could be more appropriate to inform future interventions.

Other methodological aspects were also scrutinized in the present review, and obviously they require special attention given their significant impact on the efficacy to explain intention and predict behaviour. For instance, when an objective measure of behaviour was obtained, the efficacy of prediction was much lower than when self-report measures were used. This observation is congruent with the results reported by Armitage and Conner [100] for the prediction of behaviour. They observed a significant difference between the proportion of variance explained when behaviours were observed (R² = 0.20) compared to self-reported (R² = 0.31). It can be argued that the objective assessment of behaviour is less subject to several biases (including reporting bias) than self-reports and consequently is more accurate in measurement. However, the majority of the studies using an objective measure of behaviour did not comply with the principle of correspondence between intention and behaviour, as recommended by Fishbein and Ajzen [13] (and acknowledged by most theorists in social psychology). Again, the main discrepancies were noted for the action and context dimensions; that is, the action and context mentioned in the statement of intention did not fully correspond to the behavioural measured obtained. For example, in the study by Sauls [30], the intention of intra-partum nurses was formulated with respect to several specific actions related to professional labour support during childbirth. However, the measure obtained as the behavioural outcome was the patients' length of labour. This resulted in a lack of correspondence between what was measured and what was intended. In summary, one cannot eliminate flaws in methods as an explanation for the poor efficacy in prediction when objective measures were taken. This appears to be an important point that will require further investigation.

Another methodological aspect affecting the efficacy in prediction is sample size. A lower prediction was observed among studies with smaller sample sizes. This observation supports the thorough analysis by Rashidian and colleagues [18] who estimated the sample size that should be used for a random survey of prescribing intention and actual prescribing for a study based on the TPB. Based on the variance inflation factor method, they suggested that a sample size of 148 should be recruited. This suggests that studies of healthcare professionals' behaviours should be planned in order to recruit the appropriate number of participants. If this condition is not met, the potential to obtain an efficient predictive model is reduced.

The results also indicated that good psychometric values are essential to explain a greater proportion of the intention variance. It has been documented that the reliability of a scale affects its predictive power; poor prediction results from poor reliability [101]. This effect was not observed for the prediction of behaviour, but the number of studies was relatively small compared to the number of studies available for the analysis of intention.

To guide the analysis of the variables measured to predict intention and behaviour, we used the comprehensive approach suggested by Michie and colleagues [8]. This approach was found to be very useful to capture most of the dimensions that were used to study healthcare professionals' behaviours. Notwithstanding the quality of their classification, we added two categories to their method: moral norm and habit/past behaviour. This decision is supported by the finding that moral norm as a single construct was found to be a significant determinant of intention seven out of ten times when assessed. It is also likely that with the addition of studies on the prediction of behaviour, the importance of past behaviour/habit will progressively emerge. This anticipated result is based on the observations of Verplanken and Woods [102] who demonstrated that habitual behaviour performed in a stable context is more difficult to change. Given that many of the behaviours performed by healthcare professionals could be categorized as habitual because they are typically performed in a stable context, this aspect should be documented in future studies. Unfortunately, at this time, it is not possible to verify this assumption as the number of applications was not sufficient.

One of the key questions addressed by this review is which theory or theoretical construct is the most relevant for the study of healthcare professionals' behaviours. Our results suggest that the TPB is an appropriate theory to predict behaviour, whereas Triandis' theory better captures the dynamic underlying intention. Indeed, the two categories of variables predicting behaviour most often (when assessed) were intention and beliefs about capabilities. This latter category includes the concept of perceived behavioural control, one of the TPB determinants of behaviour alongside intention. Concerning the determinants of intention, the situation is more complex, because five categories of variables significantly contributed to its prediction (i.e., most of the time when assessed). These categories of variables were: beliefs about capabilities, beliefs about consequences; moral norm; social influences; and role and identity. According to Triandis' theory, these variables would correspond to facilitating factors, cognitive attitude, moral norm, social norm, and role beliefs, respectively. Finally, even if habit did not emerge as one of the important determinants predicting behaviour, it has been added because according to Weinstein [103] its effect should be controlled in longitudinal studies. Thus, direct links with both intention and behaviour are anticipated. Interestingly, this variable is also included in Triandis' theory. We have illustrated the interrelationship of these variables in the prediction of intention and behaviour in Figure 2. We do not imply that other factors are not important, but it appears from our analysis, that the integration of the variables presented in Figure 2 summarizes the majority of our observations.

A number of limitations should be noted. First, a limited number of studies predicting behaviour were identified. It appears that most of the effort invested was concerned with understanding intention. Not much attention has been given to prospective studies aimed at predicting behaviour. More studies of behaviour prediction are therefore strongly needed to understand which factors underlie the cognitive process of decision-making in clinical-related behaviours. Second, in our analysis of the efficacy of prediction, we did not control for the number of variables included in the predictive models. We acknowledge that this might have inflated the relative performance of some theories over more parsimonious ones.

In conclusion, this study was the first systematic review aimed at investigating applications of different social cognitive theories for the study of clinical-related behaviours of health professionals. This is an important first step in identifying variables explaining intention and predicting clinical-related behaviours. Nonetheless, a number of methodological factors were identified as potential moderators of the efficacy in prediction of studies based on social cognitive theories. Future studies should take into consideration methodological aspects in order to contribute to the development of a significant corpus of data on the clinical behaviours of healthcare professionals. In particular, special care should be given to better define the context of behaviour performance. In addition, we noted that there is an important lack of prospective studies predicting healthcare professionals' clinical-related behaviours; only 16 studies were identified. Thus, there is an urgent need of additional prospective studies based on sound theoretical frameworks. We hope that the information provided in this review of the scientific literature will be useful to researchers in the planning of studies that may lead to improved strategies to change healthcare professionals' behaviours.