Our findings suggest that the multifaceted comprehensive implementation of a hypertension guideline did not exert an impact on GPs' prescribing of antihypertensive drugs for drug-treated patients with hypertension, even though the participating GPs rated themselves as highly motivated to act according to the guidelines. Increased prescribing of beta-blocking agents for patients with CHD among intervention GPs and the increased prescribing of RAAS for diabetic patients among control GPs are more likely to reflect poor baseline adherence to the guideline than the effects of intervention. The observed ineffectiveness may be due to factors related to 1) intervention, 2) prescribing behaviour, or 3) study design and the measurements.
Intervention
Planning, conducting and evaluating an extensive intervention is challenging, especially when a project is a quality programme aiming at changing healthcare processes rather than a pure clinical study. In extensive interventions, some aims may override others, especially if they are perceived to be more important by the participants [
30]. In order to be adopted effectively, the aims must be explicit and repeatedly discussed with the participants. Our intervention focused on tailoring and implementing a local hypertension care path based on the national clinical guideline. The ultimate aim was to improve the treatment of hypertension patients by following the recommendations of the guideline, including drug therapy. For each session, the specific aims were highlighted and the topics were covered several times (Figure
1). The intervention was multifaceted, consisting of both lecture and small group education and the development of a local care path as well as audit and feedback on the division of tasks and treatment levels. Peer discussions and benchmarking were fundamental parts of the intervention. These methods should have had a small to modest impact on clinical practices, including prescribing [
15].
Prescribing behaviour
Prescribing is a complex behaviour simultaneously affected by several factors of varying intensity; the new information in the guidelines constitutes only one of these [
31]. Among other things, both patients' and doctors' expectations and experiences may affect prescribing, as do the marketing efforts of the pharmaceutical industry. In our study, a decrease in use of diuretics for patients with hypertension only and an increase in the use of new ARBs were observed regardless of patients' co-morbidities and the fact that the guideline did not recommend ARBs as a first-line medication. Similarly, in the Netherlands in late 1990's, Grevings et al. observed that ARB prescriptions are not related to relevant co-morbidities [
32].
Previous observations suggest that in order to achieve change in prescribing practices, personal feedback should be given to physicians [
18,
19]. In our intervention, GPs were informed of the data collection related to their prescribing and their consent to this was requested. However, since the intervention did not include personal feedback on prescribing, it may have been too imprecise to induce prescription-related change.
In other studies aiming at changing prescribing practices of antihypertensive drugs towards the guidelines, the improvement in proportions of target prescribing has been 5-13% units [
21,
33‐
35] and the controls have improved as well (2-6% units). In these studies the interventions have been more individualised. In our study, the prescribing of both the intervention and control physicians changed towards the guidelines. Firstly, we did not offer active interventions to the control GPs but they were informed about the measurements and permission was obtained from them. Secondly, the guideline was published and therefore available on the internet, and thirdly, the control GPs may have participated in local educational activities around the newly published guideline. Organisational traditions seem to modify physicians' prescription practices. Ohlsson and colleagues state that greater similarities seem to be found amongst prescription-related behaviour between physicians within the same health care unit than among those from different units [
36]. In our intervention, peer discussions provided a good opportunity to reflect on prescription practices within the organisation, as well as its own practices against those of peers. Multidisciplinarity may have hindered the focusing of the discussion on prescribing although such a discussion is essential when improving organisational practices and task division.
Study design and the measurements
As the intervention was a pragmatic quality project of one organisation, power analysis was not performed a priori. Type II error, under-powering the study, can not be ruled out. Therefore, it is possible that we failed to reach statistically significant changes even though the intervention actually was effective. For example, two changes in the intervention group - the percentage of patients with diabetes receiving two or more concurrent antihypertensive agents increased by 7.1% units (OR 1.33; 95% CI 0.99, 1.79, p = 0.06) and the percentage of patients with CHD using beta-blocking agents increased by 6.1% units (OR 1.39; 0.99, 1.96, p = 0.06) - nearly reached statistical significance.
The intervention GPs, the participants in the quality project, were voluntary. Similarly the controls were invited on a voluntary basis, in order to reduce selection bias. This self-selection of active GPs may reduce the generalizability of the results to the general GP population. Furthermore voluntary participants in studies may be active in the adoption of new information and therefore correspondingly active in bringing about change.
We found no differences in measured patient characteristics between intervention and control patients or within either group before and after the intervention. The case-mix in terms of age, gender and measured morbidities probably did not affect the results as we stratified the patients by co-morbidities. However, we did not control for other patient-related factors such as socioeconomic status or use of other health care services.
In Finland, the Special Refund for diabetes is often delayed [
37]: we therefore chose purchases of antidiabetic drugs as the definition of the disease. Diagnosis of CHD based on a Special Refund is specific, but may not be sensitive. Consequently, patients with CHD may have been misclassified into the hypertension only group. This, in turn, may have led to an overestimation of the use of beta-blockers as hypertension medication. Previous studies, however, have shown that in Finland, beta-blocking agents are preferred to other antihypertensives [
9,
13,
38].
In the following, we will focus on measuring the impact of an intervention accurately in the context of pharmacy claims databases and give specific consideration to our outcome measures in the light of the issues raised by Maclure et al. (2006) in their review of measuring prescription-related improvements in the primary care setting [
39].
We measured change in prevalence of use of various antihypertensive medications. When measuring prescribing, the denominator, i.e. the definition of the patient group in which the outcome is intended to be measured, is critical. As Maclure and co-workers stated, every patient in the denominator should be at risk of entering the numerator [
39]. Therefore, new patients visiting a physician - i.e. incident users of any antihypertensives, those switching from one therapeutic class to another and those adding a drug to their previous medication - would represent a more valid patient group for measuring changes in prescribing. In our study the denominator consisted of both new and previous users of any antihypertensive drugs - potentially leading to the dilution of the intervention's observed effect. One explanation is that, from the clinical point of view, there is no need for change if the drug is effective and well tolerated, even if the treatment was not recommended in the guideline as a first line medication. On the other hand, because switching was not analysed we may have overestimated the concurrent use of various drugs. Our analysis of ACEI and ARBs as a single group (RAAS), where switching between the drugs is usual, diminishes this bias.
We only had data on purchased prescriptions reflecting both doctors' and patients' actions. Combining patient records and claims data would provide more valid information on doctors' practices. Furthermore, we retrieved claims for the prescriptions written by the study GPs only and may have overestimated the number of patients on monotherapy in case other physicians prescribed hypertension medication to the same patients. Finally, the duration of the follow-up and the timing of the measurements have an influence on whether an intervention's outcomes and impact can be detected [
39]. In Finland, a prescription is valid for one year from the date it is issued; thus, patients with chronic conditions and a good treatment balance need to visit a doctor or renew their prescriptions at least once a year. For this reason, our three-month time window may reflect prescribing during the preceding year. A longer follow-up would be needed to detect changes, especially if only new patients were included in the analysis.
Implications
The latest update of the Current Care Hypertension Guideline published in 2009 included several new recommendations [
40]. A combination of RAAS and CCB was recommended over one of RAAS and diuretics, especially for high-risk patients, and combinations with beta-blockers were discouraged. Concern has been expressed on how the new recommendations will be translated into clinical practice, especially in a country characterised by the frequent use of beta-blocking agents.
Practical tools, such as quality indicators, could facilitate guideline implementation. Various indicators have been used to measure the quality of prescribing and monitor change [
41‐
45]. However, the development processes of various indicators may not be explicit [
46]. Measures should be focused, valid in every sense (content, face, concurrent, construct) and feasible. The evidence base underlying prescribing indicators should be clearly described; for example, in order to have strong content validity, indicators should be based on up-to-date guidelines [
47,
48]. They should be developed with three aims in mind: the evaluation of clinical practices at the personal and organisational level as well as providing national benchmarking data. Comprehensive guideline implementation and health care management could be strengthened if, when using indicators, areas with the potential for improvement were recognised prior to conducting implementation and quality improvement activities. Researchers would also benefit; evidence on changing prescribing and implementation would be easier to collect. Furthermore, claims databases and electronic patient records should support data collection on clinical practices and patient outcomes.