Using psychological theory to understand the clinical management of type 2 diabetes in Primary Care: a comparison across two European countries

Background Long term management of patients with Type 2 diabetes is well established within Primary Care. However, despite extensive efforts to implement high quality care both service provision and patient health outcomes remain sub-optimal. Several recent studies suggest that psychological theories about individuals' behaviour can provide a valuable framework for understanding generalisable factors underlying health professionals' clinical behaviour. In the context of the team management of chronic disease such as diabetes, however, the application of such models is less well established. The aim of this study was to identify motivational factors underlying health professional teams' clinical management of diabetes using a psychological model of human behaviour. Methods A predictive questionnaire based on the Theory of Planned Behaviour (TPB) investigated health professionals' (HPs') cognitions (e.g., beliefs, attitudes and intentions) about the provision of two aspects of care for patients with diabetes: prescribing statins and inspecting feet. General practitioners and practice nurses in England and the Netherlands completed parallel questionnaires, cross-validated for equivalence in English and Dutch. Behavioural data were practice-level patient-reported rates of foot examination and use of statin medication. Relationships between the cognitive antecedents of behaviour proposed by the TPB and healthcare teams' clinical behaviour were explored using multiple regression. Results In both countries, attitude and subjective norm were important predictors of health professionals' intention to inspect feet (Attitude: beta = .40; Subjective Norm: beta = .28; Adjusted R2 = .34, p < 0.01), and their intention to prescribe statins (Attitude: beta = .44; Adjusted R2 = .40, p < 0.01). Individuals' self-reported intention did not predict practice-level performance of either clinical behaviour. Conclusion Using the TPB, we identified modifiable factors underlying health professionals' intentions to perform two clinical behaviours, providing a rationale for the development of targeted interventions. However, we did not observe a relationship between health professionals' intentions and our proxy measure of team behaviour. Significant methodological issues were highlighted concerning the use of models of individual behaviour to explain behaviours performed by teams. In order to investigate clinical behaviours performed by teams it may be necessary to develop measures that reflect the collective cognitions of the members of the team to facilitate the application of these theoretical models to team behaviours.

. There is a broad international consensus about w hat constitutes high quality care for people with diabetes [2][3][4]. However, despite extensive efforts to im plem ent high quality care [5] both service provision and patient health outcomes remain sub-optimal [6].
Systematic reviews have dem onstrated that a range o f dif ferent intervention strategies to enhance diabetes care pro duce small to m odest improvements in glycaemic control and changes in provider behaviour [5,6]. This is also true for interventions across a range of other medical condi tions [7,8]. W hilst these findings are encouraging it is less clear how to achieve such change reliably as heterogeneity in study design and setting, and the multi-faceted nature of m any interventions makes it difficult to generalise intervention strategies across clinical settings and/or types of health professional. The findings of several recent empirical studies suggest that psychological theories of behaviour can provide a valuable framework for under- standing generalisable factors underlying the clinical behaviour of individual health professionals [9][10][11][12][13][14][15]. This has paved the way for the developm ent o f interventions that target key behavioural processes that are supported by a grounded, empirically tested, scientific rationale [16 18].
One of the m ore widely used and well tested psychologi cal m odels is the Theory of Planned Behaviour (TPB) [19]. Like m any social cognitive models, the TPB is based on the premise that the way people think influences what they do (i.e. that cognitions, such as beliefs and expecta tions, influence behaviour). It proposes a m odel about how hum an action is guided (Figure 1) which predicts the occurrence of a specific behaviour where a person has an intention to perform that behaviour. According to the TPB, specific behaviours can be predicted by the strength o f an individual's intention to enact that behaviour. Intentions are thus the precursors of behaviour and the stronger the intention, the more likely it is that the behav iour will occur. Intention is, in turn, influenced by the individual's attitude towards the behaviour; their percep tions of social pressure to perform the behaviour ("subjec tive norm "); and the extent to which they feel able to perform the behaviour ("perceived behavioural control"). These latter global constructs are m ediated through inten tion, with only perceived behavioural control (PBC) hav ing a possible direct effect on behaviour.
Previous studies conducted in the primary care setting that have used this approach have usually focused on relatively simple behaviours in the context of the m anagem ent of a single acute condition (e.g. [11,13,14]). In such contexts, it is the actions o f one individual that contribute to the subsequent m anagem ent of the presenting acute condi tion (e.g. the prescribing o f an antibiotic for sore throat). In the context of chronic disease m anagement, however, the application of m odels o f individual behaviour, such as the TPB, are m ore challenging. This is because there are several different clinical aspects to the m anagem ent of diabetes, and the behaviours involved in delivering care are usually shared and delivered by a team rather than by one individual. Different groups o f healthcare profession als within a team may also have different, b u t shared roles and responsibilities (e.g. prescribing may be the sole dom ain of GPs; foot inspection may be the sole dom ain of nurses). Alternatively, there may be a specific individ ual within a team or professional group whose role it is to manage a specific aspect o f a patient's care. Thus each aspect o f diabetes m anagem ent may frequently involve n o t only the actions of m ore than one healthcare profes sional b u t th at of different types of healthcare profes sional. So whilst routinely available data on the quality of care that patients receive within a primary care practice will indicate that a clinical action has been performed, it may n o t be possible to identify which individual team m em ber performed it, or the data may be a reflection of the collective actions o f several team members.
This presents a significant methodological challenge to the use o f models of individual behaviour as explanatory frameworks o f clinical behaviours performed by teams as they are n o t norm ally used in this context. Thus the appli cation o f models like the TPB to team behaviours may require an extension of the m odel and possible elabora tion o f the m ethods used to investigate its predictive value.
The current study used the Theory o f Planned Behaviour to identify the cognitions of health care professionals', working within primary care clinical teams, about the m anagem ent of patients with diabetes. In addition to being one of the more widely tested theories in non-clinical populations, this m odel was chosen because it has been shown to be able to predict healthcare professionals' clinical behaviour [9,20]. Furthermore, clinical behaviour is performed within the current ethos of patient-centred care and in the context of situational constraints such as time pressures. The theoretical constructs in the m odel appear well placed to take these issues into account. Spe cifically, Subjective norm (e.g., pressures associated with patient preference) and PBC (whether the clinician has full control over performing all the appropriate behav iours) are proposed to work with Attitude (i.e., the indi vidual's overall evaluation o f the behaviour, arising from perceptions of its advantages and disadvantages) to pre dict intention. Intention predicts behaviour but, within the TPB, the relationship between these two is proposed to be imperfect, with PBC as an effect modifier. The cog nitions o f interest were those that underlie the m anage m ent of two key aspects o f diabetes care; foot examination (predom inantly a nurse or health care assistant behav iour) and the prescribing of statins (a GP behaviour).
To address the m ethodological issue o f relating quality of care data th at represent collective behaviours to individ ual cognitions, the study further explored the relationship between individual cognitions and an independent, prac tice-level measure o f the health care teams' performance in relation to these two clinical behaviours.

R esearch questions
Can the TPB predict: a) the intention of health care professionals to provide two aspects of diabetes care? b) the teams' collective clinical behaviour in relation to two aspects of diabetes care?

M e th o d s D e sig n a n d participants
This was a cross-sectional postal survey o f primary care health professionals in two European countries. Using a theory-based questionnaire, the study formed part o f a process evaluation and was conducted alongside two ran dom ised controlled trials o f different interventions to improve the m anagem ent of patients with diabetes [21,22]. Participants in the study were general practition ers (GPs), practice nurses and assistants, from general practices that were participating in each of the two ran dom ised controlled trials. In the English trial, practices were those recruited to a trial of an "extended" com puter ised diabetes register that incorporated a structured recall and m anagem ent system [21]. In the Netherlands prac tices were those recruited into the PAS trial (The diabetes Passport as an Aid to Structure diabetes m anagem ent in Primary Care) [22]. Adult patients with type 2 diabetes and receiving care from participating trial practices were also invited to take part in postal questionnaire survey asking about the treatm ent they had received at their gen eral practice during previous m onths. In English practices, only patients over the age of 35 years were included and approximately 20% received both GP and specialist care. In Dutch practices patients over the age of 80 years were excluded from participation in the survey, as were patients who received their diabetes treatm ent in secondary care. English practices were situated in three Primary Care Trusts (PCTs) in the north east of England, served by two district hospital-based diabetes registers. Dutch practices were situated in the m iddle and south regions o f the N eth erlands. Both trials reported positive effects o f their respective interventions.

Q uestionnaires
This study used the TPB in the design of a postal question naire survey of healthcare professionals. Four theoreti cally-derived measures were developed, using the standard procedures recom m ended for TPB studies [23], to explore: health professionals' intentions to perform each behaviour (e.g. I intend to inspect the feet o f patients with diabetes who I see during the next m onth), their atti tude towards it (e.g. Overall I think prescribing statins to patients with diabetes is beneficial to them ), their beliefs about perceived social pressure to perform them ("subjec tive norm ", e.g. People who are im portant to me think that I should inspect the feet o f patients with diabetes) and their perceived control over the behaviours (e.g. Pre scribing statins to patients with diabetes is easy). As nurses and health care assistants do no t routinely prescribe stat ins they were only asked about foot examination in the final questionnaire. The response form at for all items was a seven point Likert-type scale, from 1 (strongly agree) to 7 (strongly disagree). Scores on individual items were averaged to produce a composite measure for each con struct, with scores reversed so that a high summary score always indicated stronger or m ore positive beliefs. The questionnaire was pre-tested with six English GPs and the final version cross-validated to ensure theoretical fidelity. Cross-validation was done by both English and Dutch experts for equivalence in English and Dutch languages using translation (from English to Dutch, by a bilingual researcher w ho understood the theoretical constructs) and back-translation (from Dutch to English) by a second bilingual researcher (MB). Discrepancies between the original questionnaire items and the back-translation were identified (by JF) and resolved by discussion with a third bilingual researcher. Copies of the English and Dutch versions of the questionnaire are provided in Addi tional files 1 and 2 respectively.

D a ta collection
In both countries the TPB questionnaire was m ailed to a total of 220 GPs (161 in England and 59 in the Nether lands) and 141 practice nurses and assistants (119 in Eng land and 22 in the Netherlands) at participating trial practices. Participants were also provided with inform a tion about the study and what taking part involved. In accordance with ethical approvals for both trials, consent to participate was given by the return of a completed ques tionnaire. English non-responders received two reminder letters at fortnightly intervals. Dutch non-responders received one reminder letter after 3 weeks.
Theory-based questionnaire data were collected at the end of the intervention period for both studies (Table 1). Patient questionnaires were also m ailed to 4247 patients in both countries at the end of the intervention period (2815 in England and 1432 in the Netherlands). Patients were asked to report w hat m edication they were currently taking and whether or n o t they had had a foot examina tion in the past 12 (England) or 15 (Netherlands) m onths. These patient-reported data were used as a proxy measure o f healthcare teams' performance of two clinical behaviours.

Statistical analyses
The internal consistency of multi-items measures was assessed using Cronbach's alpha (for measures with three items) and Pearson's correlation coefficient (for measures with two items), using an acceptability criterion of a > 0.6, and r > 0.25 respectively.
Though we have previously shown that predictors of intention differed by trial group within the English study [24] we found no evidence of a trial group effect on inten tion or behaviour, Data were therefore analysed as two cross sectional studies by pooling the data from trial inter vention and control arms within each country. Each study was individually powered to answer a specific set of research questions. One of the aims of the pooling the data in this analysis was so that we could formally com pare the results from the two countries. This involved comparing of group of 46 with a group of 69 practices for the prescription o f statins and comparing a group of 65 with a group of 110 practices for the recording of feet inspections. These sample sizes gave us 80% power to detect a strength of correlation between two variables (Pearson product m om ent correlation coefficient) of 0.27 (UK sample), 0.34 (NL sample) 0.21 (com bined sample) respectively for the recording of foot inspections and 0.33 (UK sample), 0.40 (NL sample) 0.27 (com bined sample) respectively for the prescription of statins, assuming a type 1 error rate of 5%. It was n o t possible to attribute patient-reported outcomes to individual health care professionals so these behav ioural data were aggregated to the team level. This aggre gated variable was the percentage of patients reporting foot examinations or statin use for each general practice. W ithin each practice, individual health professionals were assigned the aggregated variable for each of the two behaviours. Planned analyses explored the predictive value o f the TPB m odel in explaining variance in health professionals' intention and their assigned behaviour scores. Relationships between the antecedents of inten tion (attitude, subjective norm and perceived behavioural control) and intention and between intention and clinical behaviour for b oth foot examination and the prescribing o f statins were examined using correlation and m ultiple regression analysis. As the TPB allows for a direct effect of perceived behavioural control (PBC) on behaviour, PBC was included in the m odels predicting behaviour. An interaction term was fitted to test for a country effect in all the regression analyses. As both host studies were ran dom ised controlled trials interaction terms were fit into a regression m odel to test for any respective trial effects on the outcome variables. The appropriateness of regression m odels was assessed by examining plots of residuals.
Non-response comparisons of practice size (the num ber o f GPs and nurses per practice) were m ade using Pearson's Chi-square.   Mean scores on the TPB cognitive variables, correlations and rates of patient-reported foot examination and patient-reported statin use are shown in Table 2, for both countries.

Predicting intention (individual-level outcome variable)
Attitude, subjective norm and PBC were regressed on intention to inspect feet (Table 3, Model 1). Attitude sig nificantly predicted intention to inspect patients' feet for both English and Dutch health professionals. Subjective norm significantly predicted intention for Dutch health professionals; no significant interaction was found between country and subjective norm (ß = -.286, p = 0.117), indicating that there is no difference in the im por tance of this variable between the two countries. There was no m ain effect for Country in this model. Together attitude and subjective norm explained approximately 34% o f the variance observed in health professionals' reported intention to inspect feet.

Predicting behaviour (team-level outcome variable)
Intention and PBC were regressed on behaviour (Table 3, Model 2). Neither intention nor PBC predicted foot inspection behaviour. As there was a significant difference in m ean rates o f patient reported foot inspection between the two countries, a "country" variable was allowed into the overall model. An interaction term was also fitted to formally test the relationship between PBC and country. The interaction was non-significant (ß = .022, p = 0.343).
There was a significant m ain effect o f Country.

Prescribing Statins
Scores for subjective norm and perceived behavioural control over prescribing P r e s c r i b in g S t a t in s (1 = strong disagreement; 7 = strong agreement).

Predicting intention (individual-level outcome variable)
Attitude, subjective norm and PBC were regressed on intention to prescribe statins (Table 3, Model 1). Attitude significantly predicted intention for Dutch GPs. However, no significant interaction was found between country and attitude (ß = .280, p = 0.347). There was no other appar ent country effect. GPs' attitudes towards prescribing stat ins explained approximately 40% of the variance observed in their reported intention to perform this behaviour.

Predicting behaviour (team-level outcome variable)
Intention and PBC were regressed on behaviour (Table 3, Model 2). Neither intention nor PBC predicted statin pre scribing behaviour. As there was a significant difference in m ean rates o f patient-reported statin use between the two countries, a "country" variable was also allowed into the overall model. An interaction term was also fitted to for mally test the relationship between PBC and country. The interaction was non-significant (ß = -2.259, p = 0.402) indicating that the relationship between PBC and pre scribing behaviour did not differ between countries. There remained a significant country effect not explained by the TPB constructs.

Discussion
This study has shown that the variables specified by the Theory of Planned Behaviour were im portant predictors of health professionals' intention to inspect feet and to prescribe statins. Primary Care health professionals' atti tudes towards both the clinical behaviours investigated and their perceived social pressure to perform them accounted for a significant am ount o f the variance in their intention to provide these elements of diabetes care. This was found to be true in general for health professionals from two European countries in relation to inspecting the feet of diabetic patients. However, we did n o t find a rela tionship between health professionals' intention, or their perceived behavioural control m easured at the individual level and our patient-reported measure o f behaviour (which reflected team-level behaviour). This is despite the findings of two recent systematic reviews suggesting that social cognition m odels of behaviour, which have been successfully used to predict behaviour and behavioural change in non-clinical populations, can be usefully applied to clinical behaviour at the individual level [9,20].
This difference between the results of individual level studies and the present study predicting team behaviours may result from lack of correspondence between the measures of cognitions and behaviours. Fundamental to the Theory of Planned Behaviour is Fishbein's "TACT" principle o f correspondence [25]; which is that measures o f intention and behaviour m ust be specified at the same level of generality. Measures correspond if they relate to the same operational definitions of the: Target of the action (in the present study this is any patient with type 2 diabetes); Action to be performed (e.g. foot examination); Context in which the action is performed (e.g. during a consultation) and the specified Time period (e.g. over the next/last m onth). M o d e l S t a n d a r d is e d S t a n d a r d is e d S t a n d a r d is e d S t a n d a r d is e d S t a n d a r d is e d S t a n d a r d is e d For foot examination, the measures used in the present study to assess this behaviour adhered closely to this "TACT" principle in that the wording of the questionnaire items in our patient-reported measure corresponded closely to those in the health professional measure. Thus it is unlikely th at poor correspondence between the word ing of these measures used to quantify intention and behaviour for foot examination contributed to error [9 ].
For the prescribing o f statins, however, one question in the patient report measure may have been too general; rather than ask patients if they had been prescribed statins we asked them to list all the m edication they had taken in the past 4 weeks. W ording the question this way changed the focus o f whose behaviour we were asking about (and The ability o f social cognition theories like TPB to predict clinicians' behaviour has been dem onstrated in studies using both self-reported and objective (observed) meas ures (varying between 13% [20] and 20% [26] for objec tive measures), though the am ount of variance in behaviour that is explained by such m odels is consistently lower when an objective measure of clinician behaviour (like patient report) is obtained [9,26]. There are several factors th at could account for the finding that social cog nitive m odels predict intention m ore strongly than they predict behaviour. Among them is the "intention-behaviour" gap. There is a considerable literature that addresses this gap (e.g. [27]) which highlights the im portance of "post-intentional" factors that intervene to mediate an individual's behaviour, given the existence of a strong intention. However, while it is highly possible that such factors contributed to the findings presented here, an alternative explanation for the lack of an observed associ ation between intention and behaviour in the present study could be a lack of "correspondence" between indi viduals' cognitions and the aggregated measure of behav iour that we used. i.e., the predictors (including intention) were m easured at the level o f the individual clinician and behaviour was m easured at the practice level.
This latter explanation presents a methodological chal lenge to the use of social cognitive m odels to investigate clinical behaviours as it is n o t always possible to achieve such a precise link between the measures of cognition and behaviour. This is a problem which is amplified in the investigation of behaviours that are performed within the context of a team; some behaviours may be shared (e.g. foot inspection may be the role of more than one nurse or health care assistant and the prescribing of statins the role o f more than one GP) and others may contribute cum ula tively to a single aspect of care (e.g. in the weight manage m ent o f people with diabetes a nurse may provide lifestyle counselling, a dietician give dietary advice and a GP pre scribe a weight loss medication).
Hence for the behaviours investigated in the present study it was n o t possible to link the measures of intention and behaviour so precisely. Instead, patient-reported rates of statin use and foot inspection were aggregated to practice level and the m ean value assigned to individual health professionals within each practice. This strategy assumes that each health professional has an equal role in the per formance of the behaviour o f interest -i.e. that the behav iour is a shared role. Where this is no t the case -w hen for example a single GP takes the lead in providing care for patients with diabetes in one practice, or it is the role o f a single nurse to examine patients' feet -this strategy reduces the specificity of this measure of behaviour. Fur ther more, the latter scenario w ould no t necessarily result in other team m em bers having less favourable attitudes etc towards the clinical behaviours investigated here. They may, however, have little or no intention to perform those behaviours because they are confident that these actions will be covered by other members of the clinical team, reducing the ability of this measure to predict behaviour. Thus some alternative m ethods of aggregating the collec tive cognitions of the team m ight lead to stronger predic tion of the collective behaviour.
There are additional problem s in the m easurem ent of the clinical behaviours. We used patient reported measures as these were the only measures in com m on for these behav iours across the two host trials. While patient-and self report measures are com m only used as proxies for actual behaviour in im plem entation research, these, along with other frequently used proxy m easurem ent m ethods, do have limitations which can threaten their validity. The patient data used in the present study may have been biased by the low response rates to the patient survey; while 69% o f useable responses were obtained for the Dutch patient questionnaire only 51% were obtained for the English patient sample. In addition, we did n o t have sufficient inform ation about the approached samples that w ould allow further evaluation of how representative those responding were o f the respective patient popula tions. Encouragingly, the rates of statin use and foot inspection reported by the English patients in this study are supported by additional data from medical records reported elsewhere [28]. Data from this addition source suggest that there was no difference in levels of clinician performance as reported in the adjusted record based data and the unadjusted patient-report based data. This pro vides some evidence that these proxy measures may pro vide an adequate measure of actual rates o f statin prescription and foot inspection.

Lim itations
This study is limited by the low response rate to the Eng lish survey [24]. This was particularly low at the individual level for b oth behaviours (37%), bu t im proved at practice level (statin use 57%, foot inspection 79%). This may have been due to greater respondent burden for the Eng lish HPs as the English survey instrum ent consisted 154 items and covered three behaviours. However, while non response analysis indicated that nurse respondents were over-represented in the English dataset, both the English and the Dutch practices responding to the surveys were largely representative o f practices enrolled on the two tri als.
The psychological m odel we used relates to the intentions and behaviour of individuals b u t the two aspects of diabe tes care that we examined are performed in the context of the team m anagem ent o f this chronic disease. As we did n o t survey all practice staff within each participating prac tice, it is feasible that the cognitions of key health profes sionals whose role involved providing the behaviours of interest were n o t included in the study. It is also possible that either one or both of the behaviours measured were n o t performed by all health professionals who did respond to the survey. Allocating our aggregated measure of behaviour to these respondents assumed that they had. A further lim itation may have been our use of an accepta bility criterion of r > 0.25 for internal consistency for 2item measures. However our 2-item measures were found to be well above this m inim um threshold. These m ethod ological lim itations had the potential to reduce the corre spondence of the measures used and thus the predictive ability o f the TPB m odel to explain clinicians' behaviour in the context of a team setting.

C o nclu sio n
The findings of this study are very exploratory in nature and suggest associations rather than causes. Despite its limitations however, this study has identified m odifiable factors underlying health professionals' intentions to per form two clinical behaviours, providing a rationale for the developm ent of targeted interventions. This study adds to the growing body o f evidence that psychological models of hum an behaviour may be of value in the prediction of health professionals' intentions to perform clinical behav iours. However, we did no t observe a relationship between health professionals' intentions and our proxy measure of team behaviour. Importantly, the study also highlights significant methodological challenges to the use of social cognitive models of individual behaviour to explain behaviours performed as part of the team m an agement o f chronic diseases like diabetes.
The lack o f a direct link between individuals' cognitions and behaviour com prom ised the correspondence between measures (a fundam ental feature o f the TPB) and may explain the lack of association between intention and behaviour. However, in order to use a theory-based approach to behaviours that are performed in the context o f a team -such as diabetes care -it may be necessary to develop the m easurem ent of the theoretical constructs to facilitate their application to team behaviours. It may, for example, be necessary to consider different strategies for aggregating scores that represent individuals' cognitions w hen their collective behaviours contribute to a single outcome. This is the subject o f a separate methodological paper by the authors.