Systematic Review and Meta‐Analysis of Medication Reviews Conducted by Pharmacists on Cardiovascular Diseases Risk Factors in Ambulatory Care

Introduction

Cardiovascular diseases (CVDs) are the main cause of morbidity and mortality worldwide, with more than 36% of adults in the United States and 40% in Europe at high risk for developing or with established CVD.^{1, 2} The World Health Organization reported 17.9 million of CVD‐related deaths in 2016, representing 44% of all deaths from noncommunicable diseases, with 85% of these deaths caused by strokes and ischemic heart diseases.^{3, 4} Dyslipidemia, hypertension, and type 2 diabetes mellitus (T2DM) are the most common risk factors in adults, with an estimated 39%, 31%, and 8% affected worldwide, with great impact in mortality, morbidity, and costs of care. However, common strategies to control these diseases appear to be relatively ineffective.^{2, 3, 4}

Pharmacists are increasingly having direct involvement in patient care usually by providing services that have the objective of improving medication management of patients and other healthcare professionals.^{5, 6, 7, 8} There are various types of services, including medication reviews (MRs).^{8, 9} MRs vary from a brief revision of the prescribed medicines to more complex interventions involving patients and physicians, which allow the detection of pharmacological interactions and drug‐related problems such as adverse drug reactions, effectiveness problems, nonadherence, and self‐medication.^{10, 11} Pharmacists‐led interventions have reportedly increased the achievement of therapeutic goals in CVD risk factors such as hypertension and T2DM, decreasing systolic blood pressure (BP) between 6 and 10 mm Hg and glycated hemoglobin (HbA_1c) between 0.46% and 1%.^{6, 7, 8, 9, 10}

Some systematic reviews and meta‐analyses reveal high inconsistencies and heterogeneity on the impact of MR. Possible causes of this problem are the lack of control of confounding factors such as age and other demographic data, months of follow‐up, control groups without usual care or dummy interventions, variability, and fidelity of the intervention including different settings.^{5, 6, 7, 8, 9, 10} These specific setting elements could include access to care teams for proposed action plans, proximity and relationship with prescribers, the physical place of the intervention, and other related factors.^{6, 7, 8, 9, 10} How these differences in ambulatory settings could influence the clinical impact of the pharmacist's provision of MR has not been reported.

The main objective of this meta‐analysis was to analyze the impact of pharmacist‐led MRs on CVD risk factors overall and in different ambulatory settings while exploring the effects of different components of MRs.

Methods

Data Sources and Searches

A systematic review was performed using the PRISMA statement and Cochrane Collaboration recommendations.^{12, 13, 14} Two reviewers (F.M.‐M., A.A.‐C.) performed all of the steps individually, and any discrepancies were decided by a third author (V.G.‐C.). Searches were conducted in PubMed, Web of Science, Embase (through Ovid), the Cumulative Index to Nursing and Allied Health Literature, and the Cochrane Library Central Register of Controlled Trials database, without any time limit (up to May 2019). A manual search in the reference lists of included studies was performed, and grey literature (eg, Google) was also searched. The complete search strategy for each database is available in Table S1.

Eligibility Criteria

Table 1 describes inclusion and exclusion criteria. The Pharmaceutical Care Network Europe (PCNE) categories of MR conducted by pharmacists were used to classify interventions as¹¹: type 1: a basic review of medicines and health problems based on the available medication history in the pharmacy; type 2: an intermediate review with the available medication history in the pharmacy and clinical records or information obtained directly from the patient; and type 3: an advanced review using medication history, clinical records, and information obtained directly from the patient.

During the screening phase (title and abstract reading), articles were excluded if considered irrelevant to the study goals. The full‐text eligibility phase excluded articles that did not fulfill all of the inclusion criteria.

Data Extraction

Standardized data collection forms were used to extract data on the studies’ metadata (eg, author names and year), patients’ characteristics (eg, sample size, mean age, sex, and diseases), type of interventions and its components, setting of intervention, number of visits, PCNE MR category, method of communication with patients and physicians, and clinical outcomes.

Nonpharmacological interventions included education in lifestyle changes, medication use and disease; self‐monitoring of parameters; vitals assessment such as BP, capillary glycemia, or cholesterol measurements; and adherence interventions. Pharmacological interventions consisted of pharmacists suggesting modifications to treatment in detected drug‐related problems or only in CVD risk problems.^{5, 6, 7, 8, 9, 10}

Two ambulatory settings were included. An ambulatory clinic (AC) is defined as a primary care center where health care is mostly provided by general practitioners but could also include specialized outpatient clinics.¹⁵ Community pharmacies (CPs) are legally approved establishments that supply prescription and nonprescription medicines and may provide professional pharmacy services and patient counselling while dispensing.¹⁶

Quality Assessment

The revised Cochrane risk‐of‐bias tool for randomized controlled trials was used to identify the risk of bias. Studies were classified as having low risk, high risk, or some concerns of bias.¹⁷

Statistical Analyses

Pairwise meta‐analyses of the studies were performed for the outcome measures whenever possible. These analyses were conducted using the software Comprehensive Meta‐Analysis version 3 (Biostat).

The random effect model was used with the inverse of the variance to obtain pooled effect sizes, and results were reported with a 95% CI and P<0.05. The calculation of 95% prediction intervals (PIs) was performed in preformatted sheets in Excel with the method described by Borenstein and Higgins using mean effect size and its variance (random effect weights), degrees of freedom, and Tau² (estimation measure of the true effect size distribution) in log units (normal approximation).^{18, 19} PIs allow more informative inferences in meta‐analyses (eg, true treatment effects that can be expected in future settings), especially when there is large variation in the strength of the effect (high heterogeneity between studies).^{14, 18, 19}

For the meta‐analyses of dichotomous data (therapeutic goals), the odds ratio (OR) was calculated. For the meta‐analyses of continuous outcomes, the differences between baseline and final values with the corresponding SDs reported by the individual studies (pre‐post correlation of 0.999) were used.¹⁴

For articles that reported 95% CI as a measure of uncertainty, SD was calculated using the size of the samples, the length of the CI, and the value from Student t distribution. When numeric data were insufficient to conduct the pooled analysis, a request was sent to the author by email. If the authors responded negatively or not at all, we excluded the article from the analyses.^{18, 19}

The between‐trial heterogeneity was assessed using the inconsistency index value (I2 statistic) with ranges of <25% (low), 25–50% (moderate), 50–75% (high) and >75% (very high) heterogeneity. Sensitivity analyses were conducted together with analyses for publication and other bias (funnel and scatter plots, Failsafe N) to test the robustness of the results. Subgroup analyses considering setting and components of interventions were performed when possible.¹⁴

Results

Sixty‐nine studies reported data that could be included in the meta‐analyses (Figure 1). One study was excluded from these analyses because it lacked uncertainty data (and the author responded negatively). Forty‐five of these studies were undertaken in ACs and 24 in CPs. The total number of patients was 11 743, with 11 644 included in the meta‐analyses. Of these, 8014 patients were in ACs and 3630 in CPs, with a mean age of 60±7.2 years, and the percentage of men in the included studies was 43±8.8%, without differences between subgroups (Table 2). The mean follow‐up time was 8.35±4.44 months, and there were 5.21±2.52 contacts with patients in average. Most studies provided lifestyle and disease education, and 23 studies considered the opinion of each patient before changing pharmacotherapy. In 39 studies, pharmacists only implemented changes in medications for CVD risk (ignoring other medical conditions). In 48 studies, pharmacists assessed vitals during the interviews and provided self‐monitoring education in 38 studies.

• Download figure
• Download PowerPoint

Risk of Bias

Sixty‐one of the 69 studies included in the meta‐analysis presented low risk or some concerns about bias. The main issues were the impossibility to blind patients to the intervention and the lack of details in the randomization process. Eight studies had a high risk of bias, mostly because of indefinite randomization and the lack of blinded process in the assessment of clinical outcomes. The effect of excluding high‐risk articles from the analyses was explored for each outcome.^{20, 21, 22, 23, 24, 25, 26, 27} Table S2 presents individual risk‐of‐bias analysis.

Clinical Outcomes and Components

Figures 2, 3 through 4 and Tables 3, 4 through 5 present clinical outcomes overall and by individual setting. Figures S1 through S8 contain additional forest plots for each clinical outcome and Table S3 shows the effect of each individual component and type of MR.

• Download figure
• Download PowerPoint

• Download figure
• Download PowerPoint

• Download figure
• Download PowerPoint

Hypertension

Table 3 presents pooled size effects for hypertension outcomes. Mean follow‐up time of the MR service was 8.49±4.99 months, with 5.28±2.59 patient visits. The meta‐analysis for overall BP control (31 studies; n=7031 patients) showed a statistically significant pooled OR of 2.73 (95% PI, 1.05–7.08) (Figure 2). Heterogeneity was high (I²=71%) and the AC subgroup also had a significant PI.

Fifty‐two studies (n=9935 patients) were included in the analysis of systolic BP (SBP) (Figure S1). Heterogeneity was very high (I²=99%) and resulted in significant PIs for the CP subgroup but not for the AC subgroup or overall.

For diastolic BP (DBP) (49 studies; n=9526 patients) heterogeneity was very high (I²=99%), and PIs were not significant overall or in subgroups (Figure S2).

Excluding studies with a high risk of bias, small studies or outliers resulted in similar results for hypertension outcomes (Table 3).

Type 2 Diabetes Mellitus

For diabetes mellitus studies, mean follow‐up time was 9.96±6.22 months, with 4.88±2.57 patient visits. The overall OR for achievement of T2DM control (12 studies; n=1805 patients) was 3.11 (95% PI, 1.48–6.52) (Figure 3). Only 1 CP article reported this outcome, and the AC subgroup showed a significant PI. Heterogeneity was moderate (I²=30%). No article had a high risk of bias. Table 4 presents effect sizes for T2DM outcomes.

A total of 3452 patients with T2DM from 25 studies were included in the analysis of the differences in HbA_1c levels (Figure S3). There was very high heterogeneity (I²=99%), which resulted in a nonsignificant PI. Subgroup analysis also showed no significant PI. No study had a high risk of bias.

In the fasting glucose analysis (17 studies; n=2505 patients) there was very high heterogeneity (I²=99%) with nonsignificant PI (Figure S4).

Sensitivity analyses showed no differences except for the exclusion of 3 outliers in diabetes mellitus control, which reduced heterogeneity to 0%, and 1 outlier in HbA_1c, which resulted in a significant PI overall (Table 4).

Dyslipidemias

Table 5 presents dyslipidemia outcomes. Mean follow‐up time was 9.01±6.31 months, with 5.58±2.87 patient visits. Eleven studies (n=2012 patients) reported cholesterol goals (Figure 4), finding a significant OR of 1.91 (95% PI, 1.05–3.46), with moderate heterogeneity (I²=31%). AC had a significant PI. There were no studies with a high risk of bias.

The analysis of total cholesterol had very high heterogeneity (I²=99%) resulting in a nonsignificant PI (Figure S5). There was a significant difference between subgroups (Q=7.91, P=0.005), with ACs having a larger reduction in TC levels than CPs.

Very high heterogeneity (I²=99%) was found in the low‐density lipoprotein cholesterol analysis, with a significant PI in the CP subgroup only (Figure S6). A statistical difference was observed between subgroups (Q=9.62, P=0.002) with a larger effect in ACs. CP analysis included 5 studies versus 15 in the AC subgroup.

For high‐density lipoprotein cholesterol (20 studies; n=2804 patients), there was very high heterogeneity (I²=99%), which led to a nonsignificant PI (Figure S7). There was a significant difference between subgroups (Q=5.25, P=0.022), with a larger effect in ACs versus CPs, but none had statistical significance.

For triglyceride levels (23 studies; n=3185), a nonsignificant PI was observed with very high heterogeneity (I²=99%) (Figure S8).

Excluding small studies or an outlier reduced heterogeneity to 0% and produced significant PI in the control of total cholesterol (Table 5).

Discussion

To our knowledge, this is the first meta‐analysis for MRs that includes a high number of CVD outcomes and uses PIs to account for high heterogeneity. The inclusion of control of hypertension, T2DM, and dyslipidemia and continuous clinical outcomes allowed exploration of a multidimensional effect of the provision of MR by pharmacists. We included a high number of studies and accounted for multiple components of the intervention, exploring the effect of possible bias.

Settings presented significant differences in some outcomes, with the AC subgroup having larger effect sizes in cholesterol values and DBP. This subgroup had significant increases in the achievement of T2DM and TC goals with moderate heterogeneity. Continuous outcomes had high heterogeneity and nonsignificant PIs.

In contrast with community pharmacists, AC pharmacists could directly be part of clinical teams, which may help to increase the acceptance of interventions from physicians, thus increasing the impact of MR.⁹ This assumption could not be tested since only a small number of studies included acceptance rate. The AC group included more patients (almost twice), longer follow‐up times (3±7.3 months difference), and more studies in all outcomes than the CP subgroup. All of these elements could increase effects sizes and heterogeneity at the same time.¹⁴ More studies were undertaken in the AC setting, some with high effects (outliers); however, we found no differences in magnitude and significance of effects when removed from the analyses.

In CPs there were significant decreases in low‐density lipoprotein and SBP values. CP studies tended to be shorter and smaller than AC studies. All CP effects were more affected than ACs when accounting for a high risk of bias and publication bias, with fewer reporting the number of outcomes per study. The lower number of patients within each study could have lowered heterogeneity (increasing statistical significance) and effect sizes in almost all outcomes.^{10, 14}

MR classification had significant differences between types 2 and 3, with advanced MRs providing larger effects in DBP, HbA_1c, and lipids, but this significance is limited because of a small number of pairwise comparisons, and no study with type 1 MR classification resulted from the inclusion criteria (Table S3).^{11, 14} Most of the individual components of the MR service did not have significant effects on outcomes (Table S3). Assessment of BP during visits increased the effect in control of BP and SBP, as patients tend to improve compliance when they are tightly monitored.^{1, 2, 3}

Decreasing cholesterol or BP values would be expected to happen faster and to require fewer visits than improving diabetes mellitus outcomes, but we found no differences in follow‐up times or the number of visits for the included studies or in regard to the observed effects.^{1, 2, 3} In hypertension, a significant increase overall in achievement of BP goals was found. Analyses show nonsignificant decreases in SBP and DBP (only CP achieved PI significance in SBP) but with high heterogeneity. Excluding small studies in the SBP analysis had no effect in the AC subgroup but decreased the effect in CPs and prevented significant PIs, which, together with an asymmetric funnel plot, suggested that there was a risk of publication bias in the CP subgroup (excluding outliers produced the same effect as they were mostly in CPs).¹⁴ The small effect in DBP could be explained by the fact that most included patients were older adults, who often have isolated systolic hypertension.^{2, 3}

In T2DM, an overall significant increase in achieving HbA_1c goals was observed. Only 1 CP study reported this outcome despite most studies reporting HbA_1c percentages and being a key outcome, which prevented subgroup analysis.⁹ In dyslipidemia outcomes, the control of total cholesterol increased significantly overall and in ACs even while removing outliers or small studies, but not in continuous variables (except for low‐density lipoprotein cholesterol in CP with a low number of studies) as heterogeneity was high.

Previous reviews reported significant reductions in SBP, DBP, HbA_1c, and cholesterol values, and our analysis reported a similar magnitude in clinical changes.^{6, 7, 8, 9, 10} However, we found that at a larger number of studies and when accounting for heterogeneity, statistical significance was lost in most continuous outcomes (as shown by nonsignificant PI). Nevertheless, our results support a significant effect in the control of these cardiovascular risk diseases by pharmacist‐led MRs, even when accounting for high heterogeneity.

There are only a limited number of studies that measure the impact of MR services by other health professionals. Nurses generally show lower effects than pharmacist‐led MRs in similar outcomes, but interventions that included both pharmacists and nurses seemed to provide better outcomes.^{6, 7, 8, 9, 10, 90, 91, 92, 93}

Previous evidence has suggested that heterogeneity in pharmaceutical care studies could be accounted for by some major causes such as differences in sampling, patient demographic and clinical characteristics, differences in intervention components, and fidelity of the intervention.^{9, 10, 96, 97} We found that most studies had similar patient characteristics such as age, sex percentage, and baseline health conditions of patients. Interestingly, excluding outliers had no effect on the magnitude of point estimate or heterogeneity.

The effects of individual components of the MR service were examined, but the description of interventions was both vague and varied greatly. Most studies did not include key points such as acceptance rate for interventions and fidelity of the pharmacists to provide MR, which could have effects on outcomes.^{96, 97} The interaction between physicians and pharmacists was poorly described in many studies, therefore sensitivity analysis could not be performed. It would be optimal when generating evidence to have and use a standardized intervention that clearly defines the components and characteristics of the intervention, ie, dose and fidelity so that this source of heterogeneity could be ameliorated.^{96, 97}

Intermediate and advanced MR services seem to provide benefits in controlling cardiovascular risk diseases as a result of many factors such as resolution of drug‐related problems, increase in medication adherence, simplification of therapies, and reduction of clinical inertia (common in cardiovascular conditions).^{5, 6, 7, 8, 9, 10, 11} We believe that the increase in control of hypertension, T2DM, and dyslipidemias of pharmacist‐led MR and its positive effects in most clinical outcomes support the implementation of this service, but more evidence is necessary regarding the in‐depth description of components to optimize its effects.

Study Limitations

This study has several limitations. Moderate to high heterogeneity was observed, which represented a difficulty in establishing the true impact of MR. Individual effects of components of the MR interventions could not be adequately compared. Because of large variability in the number of reported components in many studies, combined effects meta‐regressions could not be performed, therefore paired analyses using means and P values for significance had to be used. These results could be biased by the combined accumulation of type I errors for the large number of studies, thus its results should be interpreted with caution.¹⁴ There could be some risk of bias as a result of the exclusion of languages other than English and Spanish, with differences in cultural and healthcare system organization.

Conclusions

There is evidence to conclude that MRs provided by pharmacists may improve control of BP, cholesterol, and T2DM as significant effects sizes and PIs were found overall. We could not conclude that MR was better than usual care in most continuous clinical outcomes. Although effect sizes were positive with significant CIs for all analyses and settings, PI lacked significance in these outcomes. ACs had significant effects in the achievement of control of diabetes mellitus and high cholesterol, while CPs had significant decreases in SBP and low‐density lipoprotein cholesterol values, but larger studies are needed to further explore these differences. Advanced MRs in ACs could have larger effects in diabetes mellitus and cholesterol outcomes, but more evidence is needed. To ensure that there is optimization of research resources and for healthcare systems to adopt MR as usual practice, international standards should be set for the evaluation of MR services including defining in detail the target population and the MR intervention.

Disclosures

None.

Footnotes