One very effective strategy to achieve good blood pressure (BP) control in primary care is the use of physician/pharmacist collaborative management. Interventions by pharmacists in both community pharmacies and primary care clinics have been shown to significantly reduce BP by both improving medication adherence and intensifying medications. This review will evaluate the strengths and weaknesses of various health services' research study designs that assess various pharmacy interventions to improve BP control. We will also evaluate strategies to measure medication adherence used in research studies, and in some cases, clinical practice. Although poor medication adherence is a major cause of inadequate BP control, suboptimal medication regimens are often more common reasons for poor BP control in typical primary care practice. This review proposes strategies to implement stronger interventions and more robust study designs in comparative effectiveness trials that evaluate team-based care for improving BP control.
It is commonly thought that poor BP control is caused by poor patient adherence, poor guideline adherence by physicians, and limited access to care. However, one study found the most common reasons for “resistant” hypertension referred to a hypertension specialty center to be drug-related causes (61% including suboptimal regimens), patient nonadherence (13%), secondary hypertension (7%), or other (18%).1 Poor medication adherence was only an issue for 15–20% of patients in several studies involving patients who continue to seek care from their medical providers.1,2,3 Several studies have found that the most common reason for poor BP control in the United States appears to be suboptimal treatment regimens and failure to intensify medications (clinical inertia).1,4,5
There are many reasons why physicians might fail to achieve BP control including patient's resistance to adding or adjusting medications, concern about the predictive value or accuracy of office BP measurements, patient-reported home BP measurements, suspected white-coat hypertension, more urgent competing medical problems, BP already close to goal, or the patient experiencing stress on the day of the clinic visit. However, exclusively blaming physicians or patients for this public health problem is somewhat misguided. Hidden in these statistics is the fact that in the United States, and in most countries, the organizational structure does not support physicians and patients adequately.5,6,7 The current health-care reform debate in the United States includes adopting the Patient-Centered Medical Home as a key component of care delivery. In reality, concepts of the Medical Home have been articulated for at least 40 years and involve an ongoing relationship with a personal physician who leads the medical team to improve the quality of care.6,8,9 The Patient-Centered Medical Home is structured such that the point of access to care is organized around the needs of the patient, built upon the relationship with the patient, and their personal physician where a team may form and reform according to the needs of the patient.10,11 The physician may delegate responsibility to the pharmacist to perform a medication history, identify problems and barriers to achieving disease control, and to adjust medications following guidelines. Frequent communication of goal-directed therapy allows the physician to address more acute problems and complications.
Management of chronic diseases within the Medical Home places an emphasis on system changes in health-care delivery, self-management support, clinical information systems, delivery system redesign, decision support, health-care organization, and community resources.6,7,12,13,14,15 One of the most studied areas of system redesign, or organizational change, is the inclusion of pharmacists as members of the health-care team.16,17 A recent meta-analysis identified 298 clinical trials in the United States that evaluated pharmacist-provided direct patient care for various chronic conditions.17 These investigators found significant improvements in hemoglobin A1c, low-density lipoprotein cholesterol, BP, adverse drug events, medication adherence, quality of life, and patient knowledge (P < 0.05).
We will discuss the article by van de Steeg-van Gompel et al. that evaluated an intervention in Dutch pharmacies and is published in this issue of the Journal.18 This review will evaluate the strengths and weaknesses of their study design. We will also review optimal study designs in health services' research involving pharmacy interventions and medication adherence. In addition, one of the authors of this editorial was an owner of a community pharmacy in the Netherlands for 20 years and currently practices as a community pharmacist in the Netherlands lending a unique perspective on Dutch pharmacy services for this review.
van de Steeg-van Gompel et al. conducted a cluster, randomized, controlled study to evaluate a multifaceted training program for community pharmacists in the Netherlands.18 The goal of the intervention was to determine whether pharmacists reacted to the training program and implemented the Medication Event Monitoring System (MEMS). The comparison determined whether the intervention led to more pharmacists using the electronic devices when compared to pharmacists who received only an educational manual. The authors are to be commended for conducting one of the few cluster, randomized trials of a pharmacy intervention. The fact that they also examined barriers and facilitators to implementation of their intervention is also laudable.
The study by van de Steeg-van Gompel et al. is unique in that the intervention at the patient level was merely the electronic device. Compared to some other studies, this approach would be a rather weak intervention as, for over 80% of patients, poor BP control is not related to poor adherence as noted previously.1,3 Many studies have evaluated the ability of pharmacists to improve medication adherence and/or BP.17 The aforementioned meta-analysis found 54 clinical trials involving adherence, 48% of which had a favorable effect, 24% had mixed results or unclear results, 28% had no effect, and none had unfavorable effects (P = 0.001). Likewise, these authors found 59 studies in which pharmacists assisted with management of hypertension and 85% had favorable results, although 15% had either mixed or no effect (P < 0.001). The mean difference between intervention groups compared to control groups was a reduction of 7.8 mmHg in systolic BP. Another meta-analysis by one of the present authors found significant improvement in BP control in both studies conducted in community pharmacies and in studies involving pharmacists practicing within primary care medical offices.16 The latter meta-analysis found a mean reduction in systolic BP of 9.3 mmHg when the pharmacists managed therapy or recommended medications to the physician compared to a reduction of 4.8 mmHg when nurses conducted the intervention. Likewise, there was approximately an 8.0 mmHg reduction in systolic BP when medication education or adherence were addressed. Therefore, when pharmacist-managed care was employed, BP reductions were considerably greater than other strategies, illustrating the superiority of direct patient care over simple adherence strategies alone. Many of these studies have found that the more potent interventions to achieve BP control involve strategies by the pharmacist to intensify medications.2,3,16
The Dutch authors suggest that MEMS devices should be used clinically to improve adherence. We disagree with that assessment, and the methods to determine and improve adherence are discussed below. The fact that the two study groups were combined makes it impossible to know if any effect was due to regression to the mean, the Hawthorne effect or maturation effects where physicians adhere to guidelines more closely over time. Also, previous controlled trials have found that systolic BP may drop by 10–17 mmHg in the control group, which further highlights the need for strong controlled designs.3,19
We also disagree with the authors that electronic monitoring is ready for clinical use. These devices are not only expensive but few, if any, of the studies that found a benefit of pharmacy services on adherence used electronic devices.17 In one very important study from a military facility, pharmacists used both counseling and unit dose packaging and found marked and sustained improvements in both adherence and BP control.20
Current Dutch Pharmacy Practice: Comparisons with the United States
Dutch pharmacy practice is different from that in the United States. Dutch pharmacies are relatively large, serve 7,000–12,000 patients, and cover a wide geographic area. In contrast, in the US, pharmacies can be found on almost every major intersection in medium and large cities.
Dutch patients usually visit the same general practitioner and same pharmacies (95% of patients) and there is already some interaction and communication between the pharmacist, physician, and the patient. Patients typically obtain refills for chronic medication every 3 months, which is the requirement of Dutch insurance companies which is probably too infrequent to change patient behavior and adherence. On the other hand, apart from performing automated medication review, almost all pharmacies monitor refill adherence and thus discuss drug use with their patients when the refill is too early or overdue. This may explain the high medication adherence in the paper by van de Steeg-van Gompel et al. Implementation of MEMS or other similar technology does not seem necessary for many Dutch patients. Such routine discussions of refill frequency, however, is not a common practice in busy chain community pharmacies in the United States.
Additionally, a number of pharmaceutical care programs are already in place in almost all Dutch pharmacies, including special counseling for all new medications and coaching of patients with a number of chronic conditions by the pharmacy staff. Protocols for counseling for asthma, diabetes, eczema, headache, hypertension, cancer, and dyspepsia were compiled in 2002–2004 and supported by the Dutch pharmacists' association, The Royal Society for the Advancement of Pharmacy.21
Enrollment in such programs is free and medication adherence is regularly discussed. Discussions also occur with the local general practitioners in regular pharmacotherapeutic consultation sessions, which are held every 2 months. During these consultation sessions, optimal pharmacotherapy is discussed on a regional scale. Pharmacists of both independent as well as chain pharmacies (~30%) are involved in this form of discussions. Participation in the sessions is part of the quality criteria for community pharmacy. Increasingly, counseling in Dutch pharmacies is outcome oriented such as strategies to improve BP control rather than simply drug-oriented approaches.21 The article by van de Steeg-van Gompel should be interpreted from that perspective.
Although pharmacist-assisted management of hypertension is rare in community pharmacies in the United States, these services are becoming increasingly more common in community pharmacies in many states. In contrast, physician/pharmacist collaborative models are quite common in family medicine and internal medicine residency training programs, academic health centers, Veterans Affairs clinics, and staff model managed care organizations such as Kaiser Permanente in Colorado and Group Health in Seattle.22,23,24,25 Because one of the biggest issues for physicians is inadequate time to focus on hypertension, allowing pharmacists to provide direct patient care and pharmacotherapy management for chronic diseases would improve hypertension control and is supported by numerous publications.16,17 We believe strategies to increase community pharmacist knowledge of guidelines and improve their skills for proper medication management, would be a much better utilization of their expertise than simply informing patients about how to use MEMS. In fact, several examples in Canada, Australia, Portugal, and the United States of just such training programs for community pharmacists have been successful for single-disease states.16,26,27,28,29,30,31,32 After focusing initially on different disease states in the Netherlands, a broader approach is now being taken, and most pharmaceutical care training programs concentrate on polypharmacy.33
Assessing Medication Adherence
Methods to evaluate medication adherence primarily in research studies include patient self-report, prescription database assessment, pill counts, and electronic monitoring.34,35,36,37 Each of these approaches has advantages and disadvantages for measuring adherence.
In clinical settings, patients are commonly asked “Are you taking your medications?,” which is clearly inadequate to assess adherence. A better approach to patient self-report would be to ask open-ended questions such as: “How many times in the last week (month) have you missed your medication?” We have used this question for nearly 30 years and prefer this question because, if it is asked in an empathetic manner, will often provide valuable information about missed doses. This question should be asked for each medication and followed by specific questioning if problems are identified. One of the most common research tools to assess adherence is the validated self-report by Morisky.36,37 This tool has four questions: “Do you ever forget to take your medicine?,” “Are you careless at times about taking your medication?,” “When you feel better do you sometimes stop taking your medicine?,” and “Sometimes if you feel worse when you take the medicine, do you stop taking it?” These investigators and one other study38 noted a strong relationship between self-reported adherence with BP control. These investigators have more recently validated an eight-item self-report tool in 1,367 patients with hypertension.39 Either tool could be used in clinical practice.
Kressin et al. and Rose et al. assessed the validity of self-report and its relationship to BP control.40,41 They combined questions from Choo42 and Morisky et al.36,37 to create a six-item questionnaire. In a study of 793 patients, Rose et al. found that if patients affirmed 2 out of 6 of these questions they were significantly more likely to have uncontrolled BP (odds ratio, 1.86, P < 0.001).40
Other validated self-report instruments include the Brief Medication Questionnaire (BMQ),43 the Medication Adherence Self-Efficacy Scale,44 and the Hill-Bone Compliance to High Blood Pressure Therapy Scale.45 Although self-report may overestimate adherence, it is convenient, cost-effective, and an admission of poor adherence has good predictive value for BP control.35,42
Another important research approach to assess adherence are pharmacy refill records where large databases can be easily screened. However, these too may overestimate adherence if patients hoard their medications or if they automatically request mail refills in spite of poor adherence.35,42,46,47 Another limitation with refill data occurs when physicians verbally change dosing directions, thus making refill information inaccurate and difficult to assess. It is impossible to tell from refill records whether a medication was discontinued due to patient nonadherence or a purposeful decision by the physician. Even with these limitations, refill records are extremely valuable in research studies.
Pill counts to measure adherence are primarily used in research studies.3,34,35,42,48 Pill counts have only moderate correlations with other adherence measures and overestimate adherence if patients remove pills and place them in other containers.35,42 Pill counts have been used commonly in large, multicenter clinical trials in hypertension where patients are dispensed their BP medications from study investigators. The African American Study of Kidney Disease and Hypertension (AASK) found that good adherence was achieved in 68% of patients with pill counts and 47% with MEMS.34 Interestingly, only 16–18% of patients who were found to be nonadherent using either or both methods achieved BP control, whereas 50% of those who were adherent to both methods achieved BP control. If pill counts are used in clinical trials, it would be ideal if they are conducted out of the sight of the patient to minimize the Hawthorne effect but it is difficult to blind pill counts from patients. Although pill counts are useful when the medications are dispensed by the study investigators, they are less reliable when patients fill their medications from multiple community pharmacy sources.
Some consider electronic monitoring to be the gold standard for assessing adherence because it measures overall adherence and the actual time when a dose is taken.42 However, electronic monitoring is only accurate if patients consume the proper dose each time the cap is opened. Sources of error include patient opening the cap without consuming the medication, incomplete closure of the cap (some patients leave the cap off for convenience), or removing several doses of pills at one time. Because the cost for an electronic monitor is high, this method has been used primarily for research purposes. Other electronic devices are currently being tested that may be more cost-effective. However, more thorough assessment of MEMS compared to other memory enhancing approaches such as unit dose packaging or pharmacist counseling is necessary.20
Poor medication adherence may be intentional (patient simply stops taking medications or stretches doses for cost reasons) or unintentional (patient forgets).
Using MEMS to improve adherence in the community might be effective for patients who forget to take their medications. Patients who intentionally do not take their medications or cannot afford their medications will probably not be influenced by MEMS. In these cases, reasons for poor adherence can be complex and behavioral counseling approaches would be much more appropriate.
Another disadvantage of MEMS is that patients with hypertension often require 2–4 medications and the cost of MEMS would be prohibitive if used for each medication. In addition, patients may be selectively nonadherent to specific medications making selection of only one agent problematic. A strength of the paper by van de Steeg-van Gompel et al. was that the pharmacists had a protocol to prioritize which BP medication would be selected for MEMS based on the greatest likelihood of poor adherence.
Another issue with electronic devices is that patients are aware that the devices are measuring their adherence, and therefore these devices will change patient behavior in a clinical trial. Of course, this was the reason van de Steeg-van Gompel et al. used MEMS as an intervention strategy. However, the device was used in both groups as the intervention was whether the pharmacists enrolled patients to the program. A stronger design would have been to compare MEMS alone to another strategy such as MEMS plus counseling by the pharmacists.
There are strengths and limitations of all of the strategies to assess medication adherence.35 For this reason, it has been recommended that research studies in hypertension utilize two different and complimentary methods for adherence assessment.35 This suggestion is reasonable for efficacy studies that are being performed under ideal circumstances. However, for effectiveness trials that attempt to replicate clinical practice and minimize the Hawthorne effect, it may only be possible to use one method to evaluate medication adherence.
We believe, however, that validated self-report questionnaires are the most convenient and cost-effective strategy to evaluate adherence in the clinical setting. As for strategies to improve adherence clinically, we believe they must be tailored to the likely reasons for nonadherence such as medication cost, fear of adverse reactions, general concern about taking medications long-term or forgetfulness. MEMS may have a role for those who forget to take their medication, but this has not been adequately proven when compared to other approaches.35
Selecting the Proper Pharmacy Intervention to Improve BP Control
It is interesting that van de Steeg-van Gompel et al. chose the number of patients enrolled to use MEMS as the outcome measure for this study. In this study, the pharmacist did not control referral by the physician, which likely resulted in low enrollments. The pharmacist apparently simply explained MEMS, did little adherence counseling and did not make therapeutic recommendations to physicians that have been shown to markedly improve BP control.2,3 In addition, this intervention was infrequent as the pharmacists did not discuss the MEMS data with the patient until 2 months after enrollment. BP was measured at 5 months with no patient intervention between 2 and 5 months. Thus, the intervention and follow-up were infrequent and short so potency was low and it is difficult to determine the long-term effects. The authors suggest that patient enrollment may have been higher if physicians enrolled patients. However, in our experience, this is a recipe for failure as busy physicians rarely refer patients to studies or enroll patients themselves. Other studies in community pharmacy have found successful enrollment by pharmacists.49
Selecting the Outcome Measures, Covariates, and Analyses
Research designs must test the comparative effectiveness of different approaches but the intervention and study design must be as strong as possible.
The ideal outcomes for pharmacy-related hypertension studies should include mean BP, BP control, medication adherence, and perhaps other measures. Blood pressure is an outcome with several problems with both reliability and validity as many biases and errors in measurement can occur.50,51 Clinic BP values are unreliable because the patients frequently are not rested properly, the wrong cuff size is used, feet are not flat on the floor, the back is not supported in a chair and the arm is not supported at heart level. Reliability and validity can be improved by training observers, using validated automated devices, standardizing technique and patient positioning, and averaging multiple BP readings.48,50
Another strength of the article by van de Steeg-van Gompel et al. is that the BP values were measured with an automated device and three values were averaged. In contrast, some studies have used the pharmacists to measure the BP using aneroid or mercury devices, making the outcome potentially biased. Although blinded data collection is ideal, it is often extremely difficult in health services' studies. Therefore, at a minimum, the data collection should be done by an individual not invested in the results such as a research assistant or study nurse.
The analyses must control for as many covariates as possible, which is particularly important if imbalances between groups are possible. At a minimum, such covariates should include: gender, race, education, insurance status, household income, marital status, smoking status, alcohol intake, body mass index, number of coexisting conditions at baseline, number of baseline medications of interest, baseline medication adherence, and total number of clinic visits. The number of clinic visits is important as studies have found that BP control is more likely when patients see providers more frequently.3
Strong clinical trials and quality health services' research address dropouts in the analyses. Multiple methods and modeling should be employed for missing data to determine whether the results are robust.2,3 Subjects who drop out of a trial may be less adherent to therapy and less likely to achieve control of their condition. One common approach is to perform sensitivity analyses where different scenarios are considered to determine whether the results remain significant under different assumptions. For instance, patients with missing data in the intervention arm are assumed to have uncontrolled BP, while missing data for control subjects are assumed to be controlled. Although this approach will make it more difficult to show differences between groups, two recent studies used this approach and the results remained significant under the most conservative assumptions.2,3 If the results are still significantly different between groups under different assumptions, the results of the intervention are robust and there should be greater confidence in the findings.
Types of Study Designs
Most of the published pharmacy services studies would be classified as efficacy studies where the intervention is delivered under ideal circumstances. In efficacy trials, investigators attempt to achieve high fidelity to the intervention and the proposed timeline.3 Another problem with studies of pharmacy interventions is that there are only a few that involve more than one or two medical offices or more than a few presumably highly motivated intervention pharmacists. As the two meta-analyses show, we believe there is sufficient evidence that pharmacist management improves BP control and poor medication adherence.16,17
There is, however, a need for studies that demonstrate that such pharmacy interventions are effective when scaled up in large numbers of medical offices or pharmacies and if the interventions can be generalized to diverse populations. Such studies include comparative, prospective effectiveness trials that attempt to evaluate the intervention under typical practice situations.52 Effectiveness studies require large numbers of patients, providers, intervention pharmacists, and medical offices to prove that the intervention can be scaled up and implemented more broadly.
Implementation studies are similar to effectiveness studies but they evaluate whether the intervention can be adopted in routine practice.52,53,54 These studies typically measure variables such as whether providers feel empowered to deliver the intervention and/or barriers and facilitators to implementation. In this regard, the article by van de Steeg-van Gompel et al. is testing whether the pharmacists implemented the MEMS intervention, which is another unique feature of their trial.
Cluster, Randomized Designs
Hypertension specialists typically consider randomized, controlled trials to be the gold standard for evidence. These studies randomize patients to one of several blinded treatments. Blinding is straightforward with antihypertensives. However, randomization by patients is a weak design when considering health services research. A physician could have patients in both the control and intervention groups. If the pharmacist makes treatment recommendations to the physician, then contamination would be likely for patients in the control group.
Alternatively, randomization may also occur at the level of the physician or practice (clinic). Many believe that contamination is inevitable if physicians are randomized, as intervention physicians may change the behavior of physicians in the control group. Such contamination is probably unlikely in most busy offices and one of us has found that, based on two published cluster, randomized trials, randomization at the physician level would not have changed the results2,3
Many health services' researchers believe that the cluster, randomized design where clinics are randomized is the strongest methodology.55 One major disadvantage of such designs is the potential for unevenness in clinic operations or patient populations. Although such differences should always be measured and controlled in the analyses, it may still be difficult to make inferences from the results. Methods to minimize these issues include stratifying and matching clinics using as many known variables as possible and including large numbers of clinics.52 Generally, two arm studies should include at least 16 clinics and proper analyses must be performed.55 Uniquely, power is determined by the number of clusters (clinics) and variability within clinics rather than simply the number of subjects or their variability. Thus, the study by van de Steeg-van Gompel et al. is strengthened because it is a cluster, randomized design.
There is a critical need for more comparative effectiveness studies of pharmacy-based interventions, use of team-based care within primary care offices and strategies to improve medication adherence. Hypertension specialists who conduct such studies, or read such articles, must pay particular attention to the type of intervention, the outcome measure of interest, study design, data collection, and analysis. As this field continues to evolve, we can all hope that such progress yields better BP control in the population.
B.L.C. is currently supported by the National Heart, Lung, and Blood Institute, RO1 HL091841 and 1RO1 HL082711, the Agency for Healthcare Research and Quality (AHRQ) Centers for Education and Research on Therapeutics Cooperative Agreement #5U18HSO16094 and the Center for Research in Implementation in Innovative Strategies in Practice (CRIISP), Department of Veterans Affairs, Veterans Health Administration, Health Services Research and Development Service (HFP 04-149). The views expressed in this article are those of the author and do not necessarily reflect the position or policy of the Department of Veterans Affairs. We acknowledge the assistance of Meaghan Rogers, Amy Martin, and Gang Fang, who reviewed this manuscript.
The authors declared no conflict of interest.