Inclusion and exclusion study criteria were based on specific (1) study designs, (2) settings, (3) participants, (4) interventions, (5) comparators, and (6) outcomes. Randomized controlled trials (RCTs), cluster RCTs, and cross-over RCTs were eligible. Case reports, case series, non-randomized evaluations, reviews, meta-analyses, conference proceedings, policy papers, study protocols, and expert opinions were excluded. Studies based in a community/ambulatory care setting were included. Studies were considered if they included adult outpatients (18 years or over) with a diagnosis of hypertension, treated or not treated, and if they evaluated the effect of pharmacist interventions—directed or in collaborative care—in outpatients with hypertension compared to usual care. We distinguished between outpatient clinics (hospital or medical facility without overnight stay) and community care (e.g., at a community pharmacy or general practitioner). We included interventions that were delivered by a pharmacist directed or in collaborative care. Outcomes of interest were the mean difference in BP change, or BP control (BP below a predefined target level) at follow-up. We considered all publications in English, French, and German and searched all databases from inception to the date of search.

We searched the following electronic databases: MEDLINE (Ovid) (from 1946 on), Excerpta Medica database (Embase) (from 1947 on), Cochrane Central Register of Controlled Trials (CENTRAL) (from 1947 on), Cochrane Database of Systematic Reviews (CDSR) (from 1995 on), CINAHL (EBSCO) (from 1937 on), Web of Science (from 1900 on), JBI EBP Database (Ovid) (from 1998 on), and Tripdatabase (from 1997 on). The search for unpublished studies included the Grey Literature Report (New York Academy of Medicine, http://www.greylit.org). The search was conducted on 26.03.2024.

The specific search strategies (Supplementary Table S1) were developed by an experienced medical librarian in systematic review searching (BK) in consultation with the project team. They were constructed to include the two main concepts of this systematic review: “hypertension” and “pharmacist intervention”. A three-step search strategy was used in this review. First, an initial limited search of MEDLINE (Ovid) was undertaken using the search terms “Pharmacist intervention”, “Pharmacists”, “Pharmaceutical Services”, “Pharmacy Service, Hospital”, “Pharmacies”, “Pharmacy”, “Hypertension”, “Blood pressure”. Second, an analysis of the text words contained in each article's title, abstract, and index terms was undertaken to expand the list of search terms. Based on the results of this analysis, a more thorough search was conducted in the chosen databases. The search strategy for MEDLINE was created first and was then adapted for each database, including all identified keywords and index terms. Third, the reference lists of all included studies selected for critical appraisal were searched by hand and cited reference searches for all included studies were conducted in Web of Science to find any additional studies not identified during the initial search processes. The methodology search filter to limit retrieval to appropriate study designs, a modified version of the Cochrane Highly Sensitive Search Strategy, was used to identify randomized trials (10).

Study records retrieved by electronic searching were uploaded to the systematic review management software Covidence (14). After the removal of duplicates, titles, and abstracts were independently screened by two reviewers (VG and ST) for inclusion. The reviewers indicated whether a citation is potentially relevant (met inclusion criteria), is clearly not relevant (met exclusion criteria), or if the information is insufficient to make a judgment. We obtained full-text publications for all titles/abstracts that appear to meet inclusion criteria or where there is any uncertainty. Full-text publications were independently examined by two reviewers (VG and ST) to select studies for inclusion. Reasons for exclusion of ineligible studies were recorded. Any disagreement was resolved through discussion and, if required, by consulting a third review author (AC or VS). The selection process was recorded in detail in a Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram (12).

Data extraction was conducted using a prespecified data extraction template in the systematic review management software Covidence (14). Data was independently extracted by two reviewers (VG and ST) from each eligible study. Using a structured data collection form, these two reviewers independently extracted the data listed in Supplementary Table S2.

Primary outcome data to extract was the mean difference between intervention and control group in systolic and diastolic BP change from baseline to follow-up and the corresponding standard error (continuous outcome). If not reported in studies, the mean difference in BP change and standard error was calculated from reported information (BP change per study group, BP at baseline, BP at follow-up, and standard deviation, corresponding confidence intervals or p values of subgroup analyses). If no BP change or BP baseline information was available, the mean difference between intervention and control group at follow-up was used (15). For BP control (dichotomous, secondary outcome), we extracted the proportion of participants reaching a pre-defined BP target level. The target BP could differ from one study to the other.

We classified pharmacist interventions using the following pre-defined categories: (1) pharmacist-directed care (pharmacist initiating and managing care) and (2) pharmacist collaborative care (pharmacist collaborating in interventions conducted by a multidisciplinary healthcare team) (16). Based on the Cochrane Effective Practice and Organization of Care (EPOC) taxonomy interventions, we categorized pharmacist interventions by different target levels (patient, healthcare provider) and types (educational approach, e.g., targeted toward patients to improve their lifestyle; feedback, e.g., to healthcare providers to adapt medication; use of reminder tools, e.g., drug adherence aids) (17).

Two reviewers (VG and ST) independently assessed the risk of bias for each study using the “Cochrane Risk of Bias Tool” for randomized trials (RoB 2) (18). This tool assesses the risk of bias according to the following domains: 1. Randomization process; 2. Effect of assignment to intervention; 3. Missing outcome data; 4. Measurement of the outcome; 5. Selection of the reported result. The risk of bias assessment for cluster-randomized trials and randomized crossover trials was performed with the specific RoB 2 tools for these study designs (18).

We classified the risk of bias for each domain as either “Low risk”, “Some concerns” or “High risk” and provided information from each study together with the reasons for our evaluation (10, 18). Given the type of RCTs included in our review, blinding the participants and the research teams was usually not feasible; only the outcome assessment could be blinded. The quality of BP measurement was also systematically assessed along three criteria: (1) use of clinically validated BP measurement devices; (2) training of outcome assessor; (3) measurement of BP out of the office. In accordance with the revised Cochrane risk of bias tool for randomized trials, RoB2, we derived an overall study risk of bias as follows: “Low risk” with all domains at low risk of bias, “Some concerns” with at least one domain of some concern, “High risk” with high risk of bias in at least one domain or with some concerns for two or more domains (18). We resolved any disagreement in quality assessment through discussions and involvement of an arbitrator (AC or VS) where necessary.

For the continuous outcome, we used the mean difference in BP change between groups whenever available. When this measure was not reported, we used the mean difference in BP between groups at the end of follow-up (15). Calculations for the between-group standard error were based on calculations provided in the Cochrane Handbook (10). The pooled effect was calculated as weighted mean differences in BP between intervention and usual care groups, with 95% confidence intervals (CI). For the dichotomous outcome of BP control, we estimated the pooled relative risk (RR) comparing intervention vs. usual care groups, with 95% CI.

All analyses were conducted with R 4.2.2 (The R Foundation) using the meta software package (19). For the secondary outcome BP control, we calculated the relative risk as the ratio in the proportion of BP control (%) at follow-up in pharmacist intervention group vs. usual care group. A random effects model was used to estimate the pooled effects and results displayed in forest plots. Between-studies heterogeneity was quantified using the I² statistic. To explore the possible sources of heterogeneity, we conducted subgroup analyses by categories of selected study characteristics: region, setting, type of pharmacist care, healthcare team composition, intervention characteristics. To assess the robustness of our results, we performed sensitivity analyses: among studies at high quality and with larger study size. Publication bias was assessed by visual inspection of funnel plots, and funnel plot asymmetry was examined using the Egger test (10, 20). The Grading of Recommendations Assessment, Development and Evaluation (GRADE) framework was applied to assess the strength of the body of evidence for this systematic review (21). Domains used to assess the certainty of the evidence included risk of bias, inconsistency, indirectness, imprecision, and publication bias. The certainty of evidence was assessed for systolic BP both by including all studies and by excluding studies with a high risk of bias.

A total of 2,330 study records were identified by electronic database searching (Supplementary Table S1) and loaded to the systematic review management software Covidence (14). After removal of duplicates, 2,218 studies were independently screened based on title and abstract by two authors, and 274 full texts were evaluated for eligibility (Figure 1). A total of 95 studies, all published in the English language, were included in our review (23–114), including 31,168 participants (control 16,157, intervention 15,011). The characteristics of each study are presented in Supplementary Table S3. Reasons for exclusion of studies (n = 179) were most commonly not appropriate study design (n = 67), not appropriate patient population (n = 62) and not appropriate intervention (n = 29) (115–117).

A summary of studies and patient characteristics is shown in Table 1. Included studies were published between 1973 and 2023 and conducted in different regions (North America: n = 47, Europe: n = 17, other: n = 31). The characteristics of the pharmacist interventions are summarized in Table 2. In 75% of the studies, the intervention was led by the pharmacist and in 25% in collaboration with other healthcare providers. In the 24 pharmacist collaborative care studies, the team composition was pharmacist/physician (N = 12); pharmacist/nurse/physician (N = 4); pharmacist/nurse/physician/diabetes educator (N = 2), or pharmacist/nurse (N = 6). Interventions targeted patients in 93% and healthcare providers in 61% of the studies. The types of intervention comprised patient education in 88%, healthcare provider education in 14%, feedback to healthcare providers (e.g., medication change) in 49%, and patient reminders (e.g., drug adherence aids) in 24% of the studies, respectively. Feedback to healthcare providers included medication change, either after independent pharmacist prescribing or recommendations to physicians. Further details of study and pharmacist intervention characteristics for each included study are displayed in Supplementary Table S3.

Results of the risk of bias assessment for each included study and the five domains (randomization process; deviations from intended interventions; missing outcome data; measurement of the outcome; selection of the reported result) are shown in Supplementary Figure S1. Overall, the quality of the studies was relatively low. Some 54 studies (57%) were at high risk of bias, 41 studies (43%) raised some concerns, and no study was assessed at low risk. The method of blood pressure measurement was assessed as appropriate or probably appropriate in 66 studies (69%), inappropriate or probably inappropriate in 5 studies (5%), and missing information in 24 studies (25%).

The forest plots of the mean change in systolic and diastolic BP change are displayed in Figure 2 (76 studies, N = 27,057) and Figure 3 (71 studies, N = 22,652). The pooled results showed that pharmacist care services alone or in collaboration with other healthcare professionals reduced systolic/diastolic BP by −5.3 (95% CI: −6.3 to 4.4)/−2.3 (95% CI: −2.9 to −1.8) mmHg. There was a large between-study heterogeneity, with an I² of 86% for systolic BP and of 75% for diastolic BP.

Results of subgroup analyses are displayed in Table 3. Overall, there were no major differences in the effect on BP according to study characteristics or types of intervention. No difference in BP reduction was observed by region of study and pharmacist directed/collaborative care. The effects on systolic BP tended to be slightly larger if the intervention was in an outpatient clinic [−6.0 mmHg (95% CI: −7.2 to −4.8)], collaborative [−6.2 mmHg (95% CI: −7.6 to −4.9)], based on healthcare provider education [−6.1 mmHg (95% CI: −9.2 to −3.0)], or through healthcare provider feedback [−6.1 mmHg (95% CI: −7.2 to −4.9)].

Sensitivity analysis restricted to studies at relatively low risk of bias yielded similar results (mean difference of −4.8 mmHg (95% CI: −5.9 to −3.7 mmHg; I² = 87%) for systolic BP (N = 39, including 21,339 participants) and −2.3 mmHg (95% CI: −3.0 to −1.6 mmHg; I² = 63%) for diastolic BP (N = 35, including 16,780 participants), respectively), but the between-study heterogeneity was much smaller. The forest plots of the mean change in systolic and diastolic BP restricted to studies with a relatively low risk of bias are displayed in Supplementary Figures S2, S3. Sensitivity analysis excluding studies of small size (less than 50 total participants) resulted in a mean difference between pharmacist intervention and usual care of −5.1 mmHg (95% CI: −6.1 to −4.2 mmHg; I² = 87%) for systolic BP (N = 70, 26,837 participants) and −2.2 mmHg (95% CI: −2.7 to −1.7 mmHg; I² = 74%) for diastolic BP (N = 65, 21,766 participants). These differences were of the same magnitude compared with the differences observed when all studies were included.

The secondary outcome BP control at follow-up was reported in 62 studies (N = 22,657). At baseline, the median proportion of participants with controlled BP was 0% (25th to 75th percentile: 0.0%–37.4%) in the usual care group and 0% (25th to 75th percentile: 0.0%–35.4%) in the pharmacist care group. At follow-up, the median proportion of participants with controlled BP was 40.8% (25th to 75th percentile: 29.9%–57.0%) in the usual care group and 61.2% (25th to 75th percentile: 46.1%–69.6%) in the pharmacist care group. The mean difference in the change from baseline to follow-up in the proportion of participants with controlled BP was +17.3% (95% CI: + 13.3% to +21.2%) between pharmacist care and usual care group. The pooled results showed a relative risk for BP control after pharmacist intervention of 1.37 (95% CI: 1.27 to 1.47; I² = 80%; Supplementary Figure S4).

Potential publication bias was observed for both systolic and diastolic BP as suggested by the asymmetry of the funnel plots (Supplementary Figure S5), with Egger test p-value of 0.03 for systolic BP and <0.01 for diastolic BP. When limited to relatively high-quality studies, the funnel plot asymmetry disappeared (Supplementary Figure S6), with Egger test p-values of 0.65 for systolic BP and 0.07 for diastolic BP. The GRADE framework was applied to assess the strength of the body of evidence for this systematic review (Supplementary Table S4) (21). The certainty of evidence for systolic BP was assessed as low when including all studies (the certainty was downgraded due to study limitations and publication bias) and moderate when excluding studies with a high risk of bias (downgrading for study limitation). This means that the true effect is likely to be close to effect estimate, but the possibility that it is substantially different exists.