We will conduct seven postal surveys to identify which variables (theoretical constructs) predict practice and identify potential areas for intervention. Four surveys will be focus on general medical practice (two in England and two in Scotland), while three will focus on general dental practice. This will allow us to examine variability due to professional group and health care setting. We have chosen two professional groups working in similar but unrelated primary care settings in order to test the possibility that unchangeable organisational or individual factors moderate the relationship between the included variables and clinical practice. To examine the potential effects of variations in the organisation of primary health care, we will compare GPs practicing in Scotland and the north of England.
Tracer activities and outcome measures
We will assess compliance with evidence-based recommendations for five tracer activities. In general medical practice, these will be: norethisterone prescribing for menorrhagia and referrals for lumbar spine x-rays. In general dental practice, these will be: restoration of carious teeth in children, sealants for prevention of caries, and use of x-rays. We have chosen tracer clinical activities based upon the principles of the following criteria derived from Kessner and colleagues [
48] and Irvine [
49] (Table
2). In addition, we sought tracer activities that are attributable to individual clinicians and where routine data collection systems could provide objective measures of professional performance.
Table 2
Criteria for tracer clinical activities
1 | The tracer activity is easy to define |
2 | The tracer activity relates to morbidity that is amenable to improvement by medical care |
3 | There is a sound scientific basis for discriminating between good and less than good performance for the tracer activity |
4 | The effects of non-medical factors on the tracer activity performance should be adequately understood |
5 | Each tracer activity should yield data on enough patients for valid statistical analysis |
6 | Together these activities should span the range of morbidity covered by the health care professional |
7 | Together these activities should span the range of skills required by the health care professionals |
8 | Together these activities should span the range of resources specified by the health care professionals |
Lumbar spine x-ray referrals
Rationale: Lumbar spine x-rays are of limited diagnostic benefit within primary care settings and are associated with significant ionising radiation dosage [
50]. Despite this lumbar spine x-rays are the fourth commonest x-ray request from primary care doctors [
51] with referrals continuing at the rate of 7 patients per 1000 patients mean list size per year [
52]. We have recently completed a trial that found that appropriate indications for referral were not identifiable by case note review for the majority of GP requests [
52]. The trial also observed a reduction in lumbar spine x-rays of 20% without apparent adverse effects following the introduction of educational messages [
52].
Dependent variable: Lumbar spine x-ray requests per 1000 patients per year.
Data sources: We will undertake surveys in geographically distinct areas where general medical practitioners refer to a single radiology department. In addition we will only use departments that have a computerised information system storing details of requesting general practitioner and type of x-ray. After obtaining consent from the study practitioners, we will ask the radiology departments to provide data on lumbar spine x-ray requests for three months following the survey. We have successfully used this method in two different studies across six radiological departments [
51,
52].
Use of norethisterone for management of menorrhagia
Rationale: Norethisterone continues to be used widely in the NHS despite evidence of its limited effectiveness and the availability of effective alternatives [
53]. Almost half a million (497,700) prescriptions were issued in the NHS during 2000.
Dependent variable: Prescriptions for norethisterone per 1000 patients per year.
Data sources: After gaining consent from study practitioners, we will request level 4 PACT data and level 2 SPA data on norethisterone prescriptions. We have successfully used this method in a previous study [
54].
Use of dental radiographs
Rationale: intra-oral radiographs have been shown to still produce a clinically significant diagnostic yield and to benefit individual patients by allowing better informed treatment decisions [
55]. Although modern intra-oral dental films are associated with very low doses of ionising radiation, the total national "volume" of radiographs taken is such that exposures are kept as low as is reasonably achievable. Within this context and that of the recently published Selection Criteria in Dental Radiography Guideline [
55] however, comparison of data from the English and Scottish Dental Practice Boards [
56] has shown that dentists in Scotland are taking fewer than optimal intra-oral radiographs.
Dependent variable: number of intraoral radiographs taken per 100 courses of treatment.
Data sources: Detailed, itemised records of all courses of treatment for NHS patients across Scotland are stored in the MIDAS database at the Dental Practice Board in Edinburgh. This audited information is used for paying dentists and is thus subject to rigorous quality assurance measures.
Use of dental sealants for the prevention of caries in children aged 6–16
Rationale: Dental pit and fissure sealants have been shown to be highly effective in preventing dental caries (decay) in the occlusal (biting) surface of posterior teeth [
57]. Increases in the number of sealants in children aged 6–16 with evidence of pre-existing dental caries would be in accordance with the recommendations of a recent SIGN evidence based clinical Guideline [
57].
Dependent variable: mean number of sealants per 6–16 year old child.
Data sources: MIDAS database at the Dental Practice Board in Edinburgh – see Descriptive data below.
Restoration of carious teeth in children
Rationale: In recent years it has become evident that the proportion of teeth with significant dental caries involving the dentine which have received restorations is unacceptably low in both primary and permanent teeth [
58,
59]. Following the recommendations of a recent SIGN evidence based clinical Guideline [
57] would result in an increase in the number of carious teeth restored..
Dependent variable: Mean number of restored teeth per 6–16 year old child.
Data sources: MIDAS database at the Dental Practice Board in Edinburgh – see Descriptive data below.
Descriptive data
In the surveys, we will ask respondents to provide personal and professional details that might influence observed levels of behaviour. From general medical practitioners we will seek details of their practice list size, total number of FTE practitioners and whether they work full time or part time. This will enable us to calculate rates corrected for 1000 patients mean list size. We will also ask about respondent's gender in the norethisterone survey as we anticipate that female practitioners are likely to see more female patients than are male practitioners. We will allow for the effects of gender before fitting the predictor variables derived from the psychological theories.
For the General Dental Practitioners we will follow previously successful strategies in securing agreement from the dentists to let us have access to their "dental practice profiles". These are routinely derived and maintained centrally by the Dental Practice Division in Edinburgh (with whom the Dental Health Services Research Unit has secure data links). Profiles show mean values of treatments per month for each dentist against averages for the local Health Board and Scotland as a whole. We will also collect appropriate personal data about the dentists and compare this with data from the recent national "Toothousand" census of General Dentists in Scotland.
The range of included activities will also allow us to explore contrasts between types of behaviour and professional groups. For example, the use of radiography is common in both general medical and dental practice. In addition we have chosen conditions that are clinically important in their own right, and for which evidence-based advice on best practice is available. In order to identify which variables (theoretical constructs) predict practice across these five tracer activities and to incorporate a geographical comparison in general medical practice seven surveys will be required. These will be of:
-
General medical practitioners in Scotland to predict prescribing of norethisterone for menorrhagia
-
General medical practitioners in England to predict prescribing of norethisterone for menorrhagia
-
General medical practitioners in Scotland to predict referrals for lumbar spine x-rays
-
General medical practitioners in England to predict referrals for lumbar spine x-rays
-
General dental practitioners in Scotland to predict restoration of carious teeth in children
-
General dental practitioners in Scotland to predict the use of sealants for the prevention of dental caries
-
General dental practitioners in Scotland to predict the use of dental x-rays
Predictor variables (theoretical constructs)
In each survey we will test the predictive power of 14 theoretical constructs (predictor variables) drawn from the six theories (Table
1).
Motivational theories: social cognitive theory
Theoretical construct: self-efficacy.
Measures: Two forms of self efficacy have been identified in relation to health behaviours [
22,
60]. One is the extent to which people feel competent in general to control the behaviour in question). The other is the extent to which people feel competent to control their behaviour in specific situations. We will use the English translation of the Generalized Self-Efficacy Scale (GSES)[
61] to measure generalised self-efficacy in relation to the five tracer activities. The range of situations in which the tracer activity occurs, and characteristics of the situation that make the practitioner feel more or less able to control their behaviour will be elicited in the interviews. These interview responses will be used to generate situation specific self efficacy items using item types [
61].
Number of items: GSES – 10 items; situation specific self-efficacy – 10–15 items.
Response formats: GSES – four choice response (not at all true – exactly true); situation specific self-efficacy – five point Likert type scales.
Scoring: Total score on GSES (range 10–40); total score for situation-specific self efficacy.
Example questions: 'I am certain that I can restore carious teeth in children'; 'I am confident that I can restore a child's carious tooth, even if the child is very anxious.'
Motivational theories: theory of planned behaviour
Theoretical constructs: behavioural intentions, perceived behavioural control, attitude and subjective norm.
Measures: The strength of behavioural intentions, perceived behavioural control, attitudes towards the behaviour and subjective norms will be assessed using standard item wordings and response formats [
47].
Number of items: 3–5 items per construct.
Response formats: 7 point Likert type scales indicating degree of agreement with the item.
Scoring: Mean score for each construct.
Example questions: 'I feel under social pressure to use dental sealants in the next month' (subjective norm), 'I would like to avoid prescribing norethisterone for patients, but I don't really know if I can' (perceived behavioural control). In addition to these four theoretical constructs, we will measure the strength of beliefs underlying attitude, subjective norm and perceived behavioural control, as specified in the theory of planned behaviour.
The three belief based measures will incorporate the beliefs identified in the interviews but will be assessed using standard item formats. For each of these measures there will be 5–6 belief items matched with 5–6 evaluation items (e.g., 'If I refer a patient for a lumbar x-ray they will be reassured' – a behavioural outcome belief, matched with 'Reassuring patients with low back pain is important' – an evaluation of outcome). Each of these items will be measured on 7 point Likert type scales indicating degree of agreement with the item. Belief item scores will be multiplied with evaluation scores and then combined to form the overall scale.
Action theories: operant conditioning
Theoretical constructs: anticipated consequences of the behaviour; frequency of performing the behaviour in the past.
Measures: The anticipated consequences for the doctor or dentist themselves of performing each tracer activity will be assessed through responses to a series of 'if-then' statements. The range of potential consequences (positive and negative) for each tracer activity will be elicited in the interviews and used to generate items for the questionnaires. The frequency of past behaviour will be assessed by asking the doctor or dentist to estimate how many patients they have seen with tracer condition over a period of time and what proportion they have treated. Number of items: anticipated consequences – 10–15; frequency of past behaviour – 2. Response formats: anticipated consequences – five or seven point Likert type scales indicating degree of agreement with the item; frequency of past behaviour – open questions.
Scoring: anticipated consequences – total score for positive consequences, total score for negative consequences; frequency of past behaviour – proportion of patients treated per number of patients seen in a specified period.
Example questions: Anticipated consequences: 'If I refer a patient for a lumbar spine x-ray, then I will maintain a good relationship with them'; 'If I x-ray a patient's teeth, then the practice will receive a fee'. Frequency of past behaviour: 'Approximately how many patients have you seen with menorrhagia in the last six months?' followed by 'for what proportion of these patients did you prescribe norethisterone?'.
Action theories: implementation intentions
Theoretical constructs: extent of prior planning.
Measures: To date, most studies of implementation intentions have manipulated them in experimental situations, rather than measured them [
32,
33]. Hence standard measures of implementation intention have not been developed. We will use one of open questions developed to assess stage of change (see below), but coded to describe the extent of prior planning. Number of items: 1–2 Response formats: open responses.
Scoring: Responses will be coded by two independent coders into three categories: no evidence of a plan, some evidence of planning, and clearly specified plan.
Example questions: 'if you have thought about changing your practice, what have you decided to do?'; 'how will you go about this?' (please describe)
Action theories: self-regulatory theory
Theoretical constructs: Perceived identity, cause, controllability, duration and consequences of the condition; emotional response to the condition (state anxiety).
Measures: Perceived identity, cause, controllability, duration and consequences of the condition will be assessed using questions derived from the Illness Perception Questionnaire [
62]. To measure emotional response we will use the state anxiety items from the Spielberger State-Trait Anxiety Inventory [
63]. We will preface the items by asking the respondents to think about patients with this condition and to answer the questions according to how they feel about them.
Number of items: cognitive representations – 38 items; emotional response – 20 items.
Response formats: identity – the number of symptoms endorsed as being part of the illness from a list of 15 symptoms (score range 0–15); cause, cure/control, duration, consequences – items rated on five point scales from strongly agree to strongly disagree; state anxiety – four choice response (almost never – almost always).
Scoring: Total score for each of the five dimensions of cognitive representation; total score for state anxiety.
Example questions: 'Menorrhagia is a serious condition' (consequences); 'Norethisterone will be effective in curing menorrhagia' (cure/control); 'Low back pain will last a long time' (duration); 'I feel like a failure' (state anxiety).
Stage of change
Theoretical construct: stage of change.
Measures: Questions adopting a similar format to those used in studies using the transtheoretical model of behaviour change or the precaution adoption process model [
43,
64,
65].
Number of items: 4–5 questions.
Response formats: yes/no or open response.
Scoring: Responses will be coded by two independent coders into four categories: unaware, unmotivated, motivated but not acting, and taking action.
Example questions: 'Have you heard about the evidence relating to norethisterone and menorrhagia' (no = unaware), 'if you have heard about this evidence, have you ever thought of changing your practice' (no = unmotivated), 'if you have thought about changing your practice, what have you decided to do?' (responses could include deciding that no action is required, or deciding to change practice in some way); 'if you have decided that you need to change your practice, have you done anything about it? (yes = taking action; no = motivated but not acting).
Measures will be developed for each predictor variable for each survey. We will use existing measures as a starting point in this process (wherever possible), and we will follow the standard procedures that have been described to develop measures of these theoretical constructs (e.g. [
66]). Initially we will conduct semi-structured interviews with a purposive sample of 10–15 practitioners for each survey. The purpose of these interviews is to identify beliefs and attitudes relevant to the specific tracer activity and its associated health problem. The responses to these questions will be used to generate standard questionnaire items to assess the variables. Each questionnaire will then be piloted for clarity and acceptability to practitioners. The reliability of the measures will be assessed prior to analysis, using Cronbach's alpha to assess internal reliability and confirmatory factor analysis to identify and discard redundant items. The construct validity of the measures will be assessed prior to analysis by examining correlations between predictor variables that are expected to be similar (convergent validity) and dissimilar (discriminant validity). Measures which do not meet minimum criteria for reliability and validity (i.e., r < 0.7) will not be entered into the analysis.
Sample size and analysis
The surveys will generate at least ordinal level quantitative data. The relationships between predictor and outcome measures in each survey will be assessed primarily using multiple regression analysis and structural equation modelling – a procedure that utilises the observed covariance matrix. Power calculations for multiple regression analysis depend on the number of cases per predictor variable. A minimum sample size of 50 + 8 m, where m is the number of predictor variables, is recommended for testing the multiple correlation, and 104 + m for testing individual predictors [
67,
68]. We have 14 predictor variables, requiring a minimum sample size of 162 per survey to test the multiple correlation, or 118 to test individual predictors. We will aim to achieve a final sample size of 200 respondents per survey (1400 respondents in total) to ensure that the sample size is sufficient to take the effects of covariates into account. Sample sizes for multiple regression can also be estimated using the change in R
2. Using an R
2 of 48% (as found in our previous work on antibiotic prescribing among GPs [
20]), a study using 200 cases will have 90% power to detect a 3% increase in the R
2 at the 5% significance level for each additional covariate. Our proposed sample size of 1400 should ensure that the observed covariances are good estimates of the population covariance. An observed correlation of 0.5 will have a standard error of approximately 0.02.
Two potential sources of response bias could apply in this study: non response to the questionnaire, and non consent to accessing behaviour data. Cummings et al [
69] found an average response rate of 61% in a random sample of studies using surveys mailed to physicians. Recent studies that we have undertaken have suggested that we should achieve at least this response rate. The survey of GPs in Grampian used a questionnaire similar to those in this study, and achieved a response rate of 70% [
20]. Assuming a conservative response rate of 60% and a total final sample size of 1400 (800 GPs and 600 dentists), we will distribute questionnaires to 334 practitioners per survey (1336 GPs and 1002 dentists). We do not know how many of these practitioners will agree to let us access personal performance data. Sensitivity analyses will be used to assess the possible bias due to non response and non consent.
Logistic and linear multiple regression analysis will be used to assess the overall predictive power of the predictor variables (theoretical constructs) within each survey. Analyses will allow for the impact of two possible confounding variables: the number of patients that could potentially be seen by the GP or dentist (calculated as total practice size divided by the number of partners and adjusted to reflect full-time or part-time work); and the gender of the GP (a potential confound in the survey of norethisterone prescribing). In the final phase of the project, the findings from the seven surveys will be analysed simultaneously adopting a random effects approach [
70] to investigate whether the relationships between predictor variables and outcome measures are modified by (a) professional group and (b) geographical location.