Can the collective intentions of individual professionals within healthcare teams predict the team's performance: developing methods and theory

Theoretical context

Explanations for clinical behaviour can be investigated using psychological theories which have been successful in predicting behaviour and behaviour change in other settings. Using such a theory-based approach offers the potential of a generalisable framework within which to consider factors influencing behaviour and the development of interventions to modify them. A study by Eccles et al. [3] used six theories to investigate factors associated with prescribing antibiotics for patients with a sore throat among primary care doctors. This showed that the impact of individual beliefs and perceptions on intention to prescribe was high, including both evidence-based and non-evidence based factors, while the impact on behaviour was considerably smaller. Two systematic reviews of the relationship between intention and behaviour in individual HCPs [4, 5] found only 16 eligible studies but suggested that the nature of the relationship was similar to that shown by reviews of much larger numbers of studies in non-healthcare professionals [6]. Data such as these allow clear predictions to be made about the factors likely to change psychological constructs and to change behaviour.

One of the more widely used theories is the theory of planned behaviour (TPB) [7]. The TPB proposes a model about how human action is guided. It predicts the occurrence of a specific behaviour provided that the behaviour is intentional (i.e. the model does not claim to predict behaviours that are habitual or automatic). The TPB model is shown in Figure 1 and depicts the three cognitive variables that the theory suggests will predict the intention to perform a behaviour. While intention is the main precursor of behaviour, perceived behavioural control (PBC) also directly predicts behaviour. For example, a positive intention may be prevented from being translated into action because of an internal or external barrier that the individual perceives as insurmountable.

Because data may reflect the behaviour of more than one HCP, it is thus appropriate to analyse these proxy behaviour data at the aggregated level of a primary care practice. Thus, recorded data can indicate reliably that a patient has been prescribed a statin by one of the HCPs in the practice. However, in order to use a theory-based approach it is then necessary to also consider aggregating individuals' measurement of cognitions (about the prescribing of statins). It would be possible to aggregate measures of individuals' cognitions about clinical behaviours and conditions as a simple mean (as is the practice in the literature on measurement of team-level variables such as team climate[8]). However, the mean may not reflect the organisational, professional, and social processes involved in the team. It may be possible to improve the predictive performance of measures that represent team cognitions by taking account of factors such as individuals' roles, responsibilities, or positions. For example, when identified individuals predominantly perform, or have responsibility for, a behaviour (foot examination of patients with diabetes by a practice nurse), then that individual's intention score could be used as a sole representative measure or used to weight a mean value.

Clinical context

Type 2 diabetes mellitus (DM) is an increasingly prevalent chronic illness and is an important cause of avoidable mortality. Studies of the quality of care for patients with diabetes suggest less than optimum care in a number of areas [9]. In primary care, the management of DM includes glycaemic control, blood pressure control, foot examination for peripheral pulses and neuropathy, lipid control, and weight reduction (retinopathy screening is often organised separately from the practice). Patients are managed by the integrated activities of medical and non-medical members of the primary care team.

Aim

The aim of this study was to explore the relationship between the TPB's direct predictors of behaviour (individuals' intention and PBC) aggregated over HCPs in a number of ways, and practice level behaviour in the context of care for patients with DM in primary care.

The method of aggregation is not simply a statistical device but may reflect different team processes and different theoretical approaches to team-functioning. For example, aggregating intentions by averaging suggests equal weighting of members' views and would suggest team decision-making based on equal and shared communications. Whereas, choosing the highest intention score in the team to represent the relevant 'team cognition score', suggests that the team has allocated roles, with one member specialising in, or having responsibility for, the targeted clinical behaviour; here the underlying model suggests a more complex team structure with more streamlined decision-making. Other methods of aggregating would also test specific role structures, e.g., the team process may be best assessed by selecting the highest intention indicating responsibility for decision-making, along with highest PBC indicating responsibility and capability for the actual behaviour.

Therefore, we investigated the following methods of aggregating respondents' scores for each primary care team: simple mean of all PC team members' intention scores; highest intention score from responding HCPs combined with PBC for either that individual, or, for the highest PBC, the scores of the HCP identified as having primary responsibility for the clinical behaviour, ignoring the scores of other team members.

Design and participants

This was a predictive study of the theory-based cognitions and clinical behaviours concerning the management of patients with diabetes of a sample of primary care doctors and nurses from northeast England, and primary care doctors, nurses, and practice assistants in the Netherlands. We regarded all the healthcare workers within a practice as a team. Data on roles and cognitions were collected by postal questionnaire survey; behavioural data were patient-reported and collected by postal questionnaire survey. Planned analyses explored the predictive value of various aggregations of intention and PBC in explaining variance in the behavioural data.

Study setting

The study was based within two randomised controlled trials of interventions to improve the management of patients with diabetes cared for in primary care.

Study practices

In the UK, the study practices were those in three primary care trusts (PCTs) served by two district hospital-based diabetes registers both using the same register software [10]. In the Netherlands, the practices were those in three regions of the middle and south of the Netherlands [11].

Study patients

In the UK, the study patients were those people with type 2 diabetes appearing on the area-wide diabetes registers, aged over 35 and receiving diabetes care exclusively from the DREAM trial (The D iabetes RE call A nd M anagement system trial) [10] practices, or shared between study practices and hospital. At the time of the study, approximately 20% of patients received both general practitioner (GP) and specialist care, though there was no formal shared-care scheme in operation in the practices studied. In the Netherlands, patient reported outcomes were gathered from patients with type 2 diabetes, who were younger than 80 years and registered with practices participating in the PAS trial (The diabetes P assport as an A id to S tructure diabetes management in primary care trial) [11]. Patients managed in secondary care were excluded from the PAS trial.

Predictive measures

Theoretically-derived measures were developed following the operationalisation protocols of Ajzen [7, 12]. Twelve UK primary care doctors and practice nurses were interviewed about three behaviours (measuring blood pressure, foot examination, prescribing statins). The schedule for these semi-structured interviews was designed to elicit responders' beliefs relating to the constructs of the TPB. Primary care doctors and practice nurses were encouraged to talk freely about these beliefs, and any ambiguities were clarified using appropriate prompts. Interviews were tape recorded, transcribed, and content analysed. Beliefs frequently mentioned in the interviews were used to design items in a questionnaire that was developed for each of the three behaviours. The response format for all items was a seven point Likert-type scale, from one (strongly agree) to seven (strongly disagree). This initial draft of the questionnaire was pre-tested with a further six UK primary care doctors for style and clarity of content and to determine completion time. Minor revisions of wording were made to the questionnaire based on their comments. The final questionnaire used in the UK covered three behaviours, both 'indirect' and 'direct' measures of the theoretical constructs [7, 12] and consisted of 154 items, including questions about the size of practices and demographic details. For the Netherlands survey, because of concerns about respondent burden, a shortened set of the questions from the UK questionnaire was used covering only two of the three behaviours and using only direct measures. The relevant questions from the UK set were translated into Dutch and then back translated into English (and adjusted where necessary) to ensure that the meaning was the same for the UK and Dutch studies.

The questions measuring intention and PBC for the two behaviours of prescribing statins and examining patients' feet are shown in the Appendix. Scoring was adjusted so that a high score indicates a strong intention and a high degree of perceived control.

Outcome measures

In the UK, as part of a larger patient reported outcomes survey [10], patients with DM were asked the following two questions. First, 'please provide as much information as you can in the box below about ALL the medication you have taken over the last four weeks '; any report of a statin was identified. Second, they were asked, 'over the last 12 months did you have any of the tests or investigations listed'; the list included: 'test of feeling on your feet'; a positive response was taken as an indication of having a foot examination.

In the Netherlands, patients were asked to report on the medication they were currently taking and whether or not they had had their feet examined in the past 15 months.

For both countries, responses were used to calculate the percentage of patients per practice who reported taking a statin, and the percentage of patients per practice who reported having their feet examined.

Procedure

In both the UK and the Netherlands, the questionnaire was mailed to all primary care doctors, nurses, and (in the Dutch practices) practice assistants at participating trial practices at the end of the intervention period. In the UK, two reminder letters were sent to non-responders at fortnightly intervals. Dutch non-responders received one reminder letter after three weeks. Patient reported outcomes were also collected by postal questionnaire at the end of the intervention period of both trials.

Analytical approach

Internal consistency of multi-item measures [of intention and PBC] was assessed using Cronbach's alpha (for measures with more than two items) using an acceptability criterion of α >0.6, and Pearson's correlation coefficient (for two-item measures) using an acceptability criterion of r >0.25.

We were interested in the relationship between practice-level behaviour and aggregations of individuals' cognitions (intentions and PBC), and investigated this using multiple regression analysis. We conducted analyses to reflect four possible team patterns. First, we argued that the behaviour was likely to be driven equally by the individual intentions of all the practice members; we therefore calculated a mean value for each practice. It was likely that we would both get responses from single-doctor practices and get single responses (from either a nurse or a doctor) from multi-doctor practices. Under these circumstances the concept of a mean value was less meaningful, and therefore we repeated the analyses including only those practices from which we received more than one response. Second, we considered that behaviour could be most driven by the individual with the highest intention (and their PBC) within the practice, and so used these measures as predictor variables. Third, we considered that the behaviour could be the product of one team member having a strong intention, and another team member having a high level of PBC. An example of this would be the situation where a nurse had a high intention to perform the behaviour and a doctor had a high PBC score as a consequence of knowing that the nurse intended to perform the behaviour. Fourth, we considered that behaviour was most likely to be driven by the individual whose role it was to perform the behaviour. Therefore, for foot examination, we considered that this could be the role of a nurse. The statin analysis was restricted to doctors.

As the TPB predicts a direct effect of both intention and PBC on behaviour, both were included in the regression analyses.

We also explored a country effect (to allow for both 'real' and methodological differences between them) and the number of responses per practice. Although both host studies were randomised controlled trials, we analysed them as two cross-sectional studies on the basis that any effect of the interventions on behaviour would be mirrored by a change in cognitions, and that the relationship between cognitions and behaviour should therefore persist, whether or not the trial changed the levels observed in the intervention group.

Ethical approval

The UK study was approved by the South Tyneside, Southwest Durham, Hartlepool, and North Tees Local Research Ethics Committees (LRECs). The Dutch study was approved by the ethics committee of Radboud University Medical centre, Nijmegen, The Netherlands.

Prescribing statins

The three-item measure of intention had a Cronbach's alpha of 0.95. The two item measure of PBC had a Pearson's Correlation Coefficient of 0.37 (p < 0.001). In UK practices, the overall mean (sd) of the practice mean intention score was 4.8 (1.5), and in Dutch practices this was 5.6 (1.3) (mean difference (95% CI) -0.7300 (-1.4 to -0.04) p = 0.038). Similar values for the strongest intention were, for the UK practices, 5.2 (1.5) and for the Dutch practices 5.7 (1.3); these were not significantly different. The mean intention score (from participating HCPs) within each practice was significantly correlated with the highest intention score within that practice (Pearson Correlation Coefficient 0.93, p < 0.001), but neither was significantly correlated with the practice mean percentage of patients taking a statin.

When using the highest intention score for each practice, none of highest intention, PBC of the highest intender, or highest PBC in the practice predicted the prescription of statins (Table 2). Again, the country effect is apparent and of the same order of magnitude and significance. When PBC was removed from the model, intention still did not predict behaviour, and there was no additional effect of an interaction term between intention and country.

Foot examination

The three-item measure of intention had a Cronbach's alpha of 0.96. The two-item measure of PBC had a Pearson's Correlation Coefficient of 0.44 (p < 0.001). In UK practices, the overall mean (sd) of the practice mean intention score was 4.9 (1.3), and in Dutch practices this was 4.4 (1.4); these were not significantly different. Similar values for the strongest practice intention were, for the UK practices, 5.9 (1.3) and for the Dutch practices 5.1 (1.6) (Mean difference (95%CI) 0.78 (0.14 to 1.43), p = 0.018). The mean intention score for a practice was significantly correlated with the highest intention score within that practice (Pearson Correlation Coefficient 0.78, p < 0.01) and the highest intention score was also significantly correlated with the practice mean percentage of patients reporting a foot examination (Pearson Correlation Coefficient 0.29, p < 0.01).

In a regression model (Table 2) including both mean intention and mean PBC, neither significantly predicted behaviour but there was a significant 'country effect' with UK practices being 14% more likely to inspect feet. When PBC was removed from the model, intention still did not predict behaviour, and there was no additional effect of an interaction term between intention and country. A similar analysis restricted to the smaller number of practices where there was more than one respondent produced a similar pattern of results, though the country effect was not significant.

The highest intention score in a practice belonged to 38 nurses (24 of whom were from practices where intention scores were available for both primary care doctor and nurse respondents) and 39 primary care doctors (eight of whom were from practices where intention scores were available for both primary care doctor and nurse respondents). In the remaining six practices, this score was the same for both nurse and primary care doctor, and the regression used the scores for individuals who have both the highest intention and the highest PBC. The highest practice intention was a significant predictor of foot examination. Again, there was a significant country effect, with reported feet inspections being 11% fewer in ND practices than UK practices (p = 0.011). Removing PBC, including an interaction term for intention/country and including type of healthcare professional (thus exploring professional role) did not significantly change the model. Finally, the analysis was repeated using the highest intention score for the practice and the strongest PBC score for the practice. In this model, the PBC score is predominantly that of the primary care doctor respondents. This analysis produced results similar to the previous one.