This review assessed the effect of pharmacist-led self-management interventions for diabetes patients on clinical- and patient reported outcomes in randomized controlled trials. Primary outcomes were HbA1c, self-management skills, and intervention components. Secondary outcomes were blood glucose, blood pressure, BMI, lipids, adherence to medication, quality of life, and diabetes knowledge.

Pubmed, Embase, Cinahl, PsycINFO, Web of Science, and the Cochrane Library were searched from the start date of the database through to September 2017. Keywords used included “pharmacist,” “diabetes,” and “self-management” (Supplementary Table 1). Whenever possible MeSH terms and advanced searched strategies were used (Supplementary Figure 1). The electronic database searches were complemented by manually reviewing the references of relevant reviews and included studies.

The following data were extracted from the included studies: general study characteristics, description of the study population, follow-up time, number and duration of contact moments during intervention, description, and components of the intervention [diabetes education, medication, lifestyle, individual care plan or goal setting, self-management skills, self-monitoring blood glucose (SMBG) and other, group or individual intervention, education for intervention team], clinical outcomes (HbA1c, blood glucose, blood pressure, BMI, lipid profile, and other), and patient-reported outcomes (adherence, diabetes knowledge, quality of life, self-care/self-management, and other) (Supplementary Table 2). Also it was noted whether interventions were tailored according to the patient's needs. A study was categorized as being tailored if the author made this statement in the research paper. The review team did not base the classification of tailoring on literature statements (Kreuter and Wray, 2003; Noar et al., 2007). The study data were extracted by LvE and double checked for eight papers by LvD, KT, and HdG. Any disagreements were discussed until consensus was reached.

The risk of bias in individual studies was assessed with the Cochrane Risk of Bias tool by LvE (Higgins and Green, 2011). This assessment was double checked by LvD, KT, and HdG by assessing the risk of bias in eight studies. Any disagreements were discussed until consensus was reached.

Interventions across the included studies were analyzed and described narratively.

Outcomes were divided into clinical outcomes (HbA1c, glucose levels, blood pressure, BMI, lipids, and other) and patient-reported outcomes (adherence, diabetes knowledge, quality of life, self-care, and other). Results for HbA1c, blood glucose, blood pressure, BMI, lipids, and Summary of Diabetes Self-care Activities Assessment (SDSCA) were pooled in a meta-analysis. Meta-analyses were performed with Review Manager 5.3 by using a random effects model because of clinical heterogeneity across the included studies. Subgroup analyses were performed for the outcome HbA1c, for different intervention elements (follow-up time, baseline HbA1c ≤ 7% and education for intervention team) in order to explain any heterogeneity (I2) across the studies and to explore key intervention components. Sensitivity analyses were performed to test for robustness of the results regarding including studies with a cluster randomization design and studies with a high risk of bias affecting the outcome HbA1c. Results for adherence, diabetes knowledge and quality of life were described narratively.

Three of the included studies had a cluster randomized design (Armour et al., 2004; Krass et al., 2007; Mehuys et al., 2011) and 21 were randomized controlled trials (Sarkadi and Rosenqvist, 2004; Choe et al., 2005; Odegard et al., 2005; Doucette et al., 2009; Jameson and Baty, 2010; Taveira et al., 2010, 2011; Cohen et al., 2011; Farsaei et al., 2011; Jacobs et al., 2012; Jarab et al., 2012; Kraemer et al., 2012; Samtia et al., 2013; Butt et al., 2015; Cani et al., 2015; Jahangard-Rafsanjani et al., 2015; Kjeldsen et al., 2015; Nascimento et al., 2015; Wishah et al., 2015; Korcegez et al., 2017; Shao et al., 2017) (Table 1). All studies were published from 2004 onwards. Most of the studies were conducted in North America (Choe et al., 2005; Odegard et al., 2005; Doucette et al., 2009; Jameson and Baty, 2010; Taveira et al., 2010, 2011; Cohen et al., 2011; Jacobs et al., 2012; Kraemer et al., 2012) (9), followed by Asia (Farsaei et al., 2011; Jarab et al., 2012; Samtia et al., 2013; Butt et al., 2015; Jahangard-Rafsanjani et al., 2015; Wishah et al., 2015; Shao et al., 2017) (7), Europe (Sarkadi and Rosenqvist, 2004; Mehuys et al., 2011; Kjeldsen et al., 2015; Nascimento et al., 2015; Korcegez et al., 2017) (5), Australia (Armour et al., 2004; Krass et al., 2007) (2), and South America (Cani et al., 2015) (1). The majority of the studies focused primarily on diabetes mellitus type 2 patients (Armour et al., 2004; Sarkadi and Rosenqvist, 2004; Choe et al., 2005; Odegard et al., 2005; Krass et al., 2007; Doucette et al., 2009; Taveira et al., 2010, 2011; Cohen et al., 2011; Farsaei et al., 2011; Mehuys et al., 2011; Jacobs et al., 2012; Jarab et al., 2012; Samtia et al., 2013; Butt et al., 2015; Cani et al., 2015; Jahangard-Rafsanjani et al., 2015; Kjeldsen et al., 2015; Nascimento et al., 2015; Wishah et al., 2015; Korcegez et al., 2017; Shao et al., 2017) (22), one study included both type 1 and type 2 patients (Kraemer et al., 2012) and one study did not specify the type of diabetes (Jameson and Baty, 2010). In total the included studies represented 3,610 participants with a mean age ranging from 44 to 73 years of age. The median follow-up time was 6 months (Krass et al., 2007; Cohen et al., 2011; Mehuys et al., 2011; Taveira et al., 2011; Jarab et al., 2012; Butt et al., 2015; Cani et al., 2015; Kjeldsen et al., 2015; Nascimento et al., 2015; Wishah et al., 2015; Korcegez et al., 2017), four studies had a follow-up time of less than 6 months (Taveira et al., 2010; Farsaei et al., 2011; Samtia et al., 2013; Jahangard-Rafsanjani et al., 2015) and 10 of more than 6 months (Armour et al., 2004; Sarkadi and Rosenqvist, 2004; Choe et al., 2005; Odegard et al., 2005; Doucette et al., 2009; Jameson and Baty, 2010; Jacobs et al., 2012; Kraemer et al., 2012; Shao et al., 2017).

The interventions in the included studies were all provided by a trained pharmacist, either by the pharmacist alone (Armour et al., 2004; Choe et al., 2005; Odegard et al., 2005; Krass et al., 2007; Doucette et al., 2009; Jameson and Baty, 2010; Farsaei et al., 2011; Mehuys et al., 2011; Jacobs et al., 2012; Jarab et al., 2012; Kraemer et al., 2012; Samtia et al., 2013; Butt et al., 2015; Cani et al., 2015; Jahangard-Rafsanjani et al., 2015; Kjeldsen et al., 2015; Wishah et al., 2015; Korcegez et al., 2017; Shao et al., 2017) or within a multi-disciplinary team (Sarkadi and Rosenqvist, 2004; Taveira et al., 2010, 2011; Cohen et al., 2011). One study did not specify the intervention team, besides including a pharmacist (Nascimento et al., 2015). Most interventions targeted the individual patient (Armour et al., 2004; Choe et al., 2005; Odegard et al., 2005; Krass et al., 2007; Doucette et al., 2009; Jameson and Baty, 2010; Mehuys et al., 2011; Jacobs et al., 2012; Jarab et al., 2012; Kraemer et al., 2012; Samtia et al., 2013; Butt et al., 2015; Cani et al., 2015; Jahangard-Rafsanjani et al., 2015; Kjeldsen et al., 2015; Wishah et al., 2015; Korcegez et al., 2017; Shao et al., 2017) whereas some interventions used group sessions (Sarkadi and Rosenqvist, 2004; Taveira et al., 2010, 2011; Cohen et al., 2011). One study did not specify whether the intervention was offered in an individual or group setting (Farsaei et al., 2011). Fifteen studies reported offering a tailored intervention based on a patient's specific needs (Armour et al., 2004; Odegard et al., 2005; Krass et al., 2007; Jameson and Baty, 2010; Taveira et al., 2010, 2011; Farsaei et al., 2011; Jarab et al., 2012; Kraemer et al., 2012; Cani et al., 2015; Jahangard-Rafsanjani et al., 2015; Kjeldsen et al., 2015; Nascimento et al., 2015; Wishah et al., 2015; Korcegez et al., 2017).

The interventions in the included studies varied in the intensity as well as the number and type of components. The intensity, measured as the frequency of contact moments, differed across the studies from once a week to once every 3 months. Face-to-face contact with the pharmacists (Armour et al., 2004; Sarkadi and Rosenqvist, 2004; Krass et al., 2007; Doucette et al., 2009; Taveira et al., 2010, 2011; Cohen et al., 2011; Mehuys et al., 2011; Jacobs et al., 2012; Kraemer et al., 2012; Samtia et al., 2013; Butt et al., 2015; Cani et al., 2015; Jahangard-Rafsanjani et al., 2015; Kjeldsen et al., 2015; Nascimento et al., 2015; Wishah et al., 2015; Korcegez et al., 2017) (18) as well as a combination of face-to-face contacts and telephone contact with the pharmacists (Choe et al., 2005; Odegard et al., 2005; Jameson and Baty, 2010; Farsaei et al., 2011; Jarab et al., 2012; Shao et al., 2017) (6) were reported in the studies. The total contact time varied across the studies, though not all studies reported this information (Armour et al., 2004; Sarkadi and Rosenqvist, 2004; Doucette et al., 2009; Farsaei et al., 2011; Mehuys et al., 2011; Taveira et al., 2011; Jacobs et al., 2012; Samtia et al., 2013; Cani et al., 2015; Nascimento et al., 2015; Korcegez et al., 2017; Shao et al., 2017). (Table 1). Fifteen studies included diabetes education (Sarkadi and Rosenqvist, 2004; Krass et al., 2007; Taveira et al., 2010, 2011; Cohen et al., 2011; Mehuys et al., 2011; Jacobs et al., 2012; Jarab et al., 2012; Samtia et al., 2013; Butt et al., 2015; Cani et al., 2015; Jahangard-Rafsanjani et al., 2015; Wishah et al., 2015; Korcegez et al., 2017; Shao et al., 2017) either about diabetes in general or about acute and chronic complications. Education on medication was provided in 21 studies (Armour et al., 2004; Sarkadi and Rosenqvist, 2004; Choe et al., 2005; Odegard et al., 2005; Krass et al., 2007; Doucette et al., 2009; Jameson and Baty, 2010; Cohen et al., 2011; Farsaei et al., 2011; Mehuys et al., 2011; Taveira et al., 2011; Jacobs et al., 2012; Jarab et al., 2012; Samtia et al., 2013; Butt et al., 2015; Cani et al., 2015; Kjeldsen et al., 2015; Nascimento et al., 2015; Wishah et al., 2015; Korcegez et al., 2017; Shao et al., 2017) and included education about adherence, dosage, drug-related problems, indication, storage, and use. In 19 studies education on lifestyle, including diet, exercise, foot care, and/or smoking cessation were part of the intervention (Armour et al., 2004; Sarkadi and Rosenqvist, 2004; Odegard et al., 2005; Krass et al., 2007; Jameson and Baty, 2010; Taveira et al., 2010, 2011; Cohen et al., 2011; Farsaei et al., 2011; Mehuys et al., 2011; Jacobs et al., 2012; Jarab et al., 2012; Samtia et al., 2013; Butt et al., 2015; Cani et al., 2015; Jahangard-Rafsanjani et al., 2015; Wishah et al., 2015; Korcegez et al., 2017; Shao et al., 2017). In 19 studies the intervention included self-management skills support (Armour et al., 2004; Sarkadi and Rosenqvist, 2004; Choe et al., 2005; Odegard et al., 2005; Krass et al., 2007; Doucette et al., 2009; Jameson and Baty, 2010; Taveira et al., 2010, 2011; Cohen et al., 2011; Farsaei et al., 2011; Jacobs et al., 2012; Kraemer et al., 2012; Samtia et al., 2013; Jahangard-Rafsanjani et al., 2015; Nascimento et al., 2015; Wishah et al., 2015; Korcegez et al., 2017; Shao et al., 2017) and in 15 studies participants were trained in self-monitoring blood glucose (Armour et al., 2004; Sarkadi and Rosenqvist, 2004; Odegard et al., 2005; Krass et al., 2007; Jameson and Baty, 2010; Cohen et al., 2011; Taveira et al., 2011; Jarab et al., 2012; Kraemer et al., 2012; Samtia et al., 2013; Butt et al., 2015; Cani et al., 2015; Jahangard-Rafsanjani et al., 2015; Korcegez et al., 2017; Shao et al., 2017). A total of 14 studies used either an individual care plan or goal setting to improve diabetes outcomes (Armour et al., 2004; Odegard et al., 2005; Krass et al., 2007; Doucette et al., 2009; Taveira et al., 2010, 2011; Cohen et al., 2011; Farsaei et al., 2011; Jacobs et al., 2012; Kraemer et al., 2012; Jahangard-Rafsanjani et al., 2015; Kjeldsen et al., 2015; Wishah et al., 2015; Korcegez et al., 2017). Other less common interventions were the use of a diabetes diary (Farsaei et al., 2011; Butt et al., 2015; Jahangard-Rafsanjani et al., 2015), medication reviews by a pharmacist (Armour et al., 2004; Choe et al., 2005; Krass et al., 2007; Doucette et al., 2009; Jacobs et al., 2012; Korcegez et al., 2017), and providing participants with written information (Sarkadi and Rosenqvist, 2004; Jarab et al., 2012; Cani et al., 2015; Jahangard-Rafsanjani et al., 2015; Nascimento et al., 2015; Wishah et al., 2015; Korcegez et al., 2017).

Many different outcome measures were reported by the included studies (Table 1). They were divided in clinical and patient-reported outcomes.

All studies reported HbA1c as an outcome measurement for their intervention. A meta-analysis was performed, with one study excluded because of an insufficient number of participants reporting HbA1c at the final follow-up (Kjeldsen et al., 2015).

The meta-analysis (Figure 2) shows an overall significant effect in favor of the intervention on HbA1c, with HbA1c levels improving by a mean of 0.71% (CI −0.91, −0.51; overall effect P < 0.0001). Several subgroup analyses were performed based on different study characteristics (Table 3, Supplementary Figures 2A–I). None of these subgroup analyses showed a significant difference between groups.

Other clinical outcomes reported were blood glucose levels, blood pressure, BMI, and lipid profile (Table 4, Supplementary Figures 3–6). Meta-analyses showed no significant reduction for blood glucose levels, but a significant improvement in systolic- and diastolic blood pressure (−5.20 mm Hg [−7.58; −2.92] and −3.51 mm Hg [−6.00; −1.01], respectively), BMI scores (−0.49 kg/m2 [−0.79; −0.19]), total cholesterol levels (−0.19 mmol/l [−0.33; −0.05]), LDL-C levels (−0.16 mmol/l [−0.26; −0.06]), and HDL-C levels (0.32 mmol/l [0.02; 0.61]).

The risk of bias within studies was assessed with the Cochrane Risk of Bias tool. All but two (Taveira et al., 2011; Jarab et al., 2012) studies were subjected to some form of bias either at high risk or at an unclear risk due to lack of information (Supplementary Figure 7). In total, eight studies were considered to have a low risk of bias (Sarkadi and Rosenqvist, 2004; Jameson and Baty, 2010; Taveira et al., 2011; Jacobs et al., 2012; Jarab et al., 2012; Butt et al., 2015; Cani et al., 2015; Wishah et al., 2015).

The funnel plot for the pooled results of HbA1c can be considered symmetric and indicates that it is unlikely publication bias has been introduced in the analysis (Supplementary Figure 8).

Two sensitivity analyses were performed. In the first sensitivity analysis the studies with a cluster randomization design were excluded, because none of these studies corrected for the clustering effect. The clustering effect is known for potential overestimation of the effect of the intervention (Killip et al., 2004). After excluding these studies the weighted mean difference of HbA1c for the patient-level randomized studies was −0.76% [−1.00; −0.52]. This difference is of the same magnitude as the difference observed when including all studies.

The second sensitivity analysis was performed using only the eight studies with a relatively low risk of bias from influences on HbA1c. The weighted mean difference of HbA1c for studies with a low risk of bias was −0.84% [−1.11; −0.57]. This difference is also of the same magnitude as the difference observed when including all studies.

This review found evidence that pharmacist-led self-management interventions are beneficial for diabetes patients. All of the included studies used proxies to measure the effect of self-management interventions; only a minority directly measured the effect of self-management interventions on self-management skills. Overall, pharmacist-led interventions had a positive effect on HbA1c values, blood pressure, BMI, and self-management skills as shown by the results of the meta-analyses. Also the results suggest pharmacist-led self-management interventions improve adherence to medication, diabetes knowledge, and quality of life.

The results on HbA1c values in the meta-analysis showed a significant effect of pharmacist-led interventions. The magnitude of this reduction (−0.71% [−0.91; −0.51]) can be considered as clinically relevant and can be associated with risk reduction in microvascular complications (Stratton et al., 2000). These findings are in agreement with the findings of Machado et al. (2007), who reported a pooled effect of −1.00 ± 0.28% on HbA1c values. However, in their review all kinds of pharmacist interventions for diabetes patients were included. Compared to systematic reviews on the effect of self-management interventions by either a physician, nurse or diabetes educator, the effect of pharmacist-led self-management interventions was over three times larger (Sherifali et al., 2015). The added value of pharmacist-led interventions for diabetes goal attainment is supported by the findings of Greer et al. (2016), who reported a relative risk (1.83 [1.44; 2.33]) in favor of diabetes patients receiving pharmacist-led disease management. The diversity of intervention contents in the included studies is also highlighted in previous reviews (Machado et al., 2007; Greer et al., 2016; Pousinho et al., 2016).

This study has several strengths. All of the studies included measured HbA1c values, which made it possible to compare the effect of the described interventions in a meta-analysis. Also the results for blood glucose, blood pressure, BMI, lipids, and self-management skills could be pooled in meta-analyses.

Though most studies used proxies to measure the effect of pharmacist-led self-management interventions, a few studies directly measured self-management. The results of these studies reveal a positive direct relation between the self-management intervention and the development of self-management skills in diabetes patients. This is most likely because the interventions in almost all of the included studies addressed medication and medication-related problems that are rather common among diabetes patients (Haugbolle and Westh Sorensen, 2006; Kempen et al., 2014).

This study also has some limitations. The reporting of the interventions and study results were very limited in some of the studies (Armour et al., 2004; Choe et al., 2005; Odegard et al., 2005; Krass et al., 2007; Doucette et al., 2009; Taveira et al., 2010; Cohen et al., 2011; Farsaei et al., 2011; Mehuys et al., 2011; Kraemer et al., 2012; Samtia et al., 2013; Jahangard-Rafsanjani et al., 2015; Kjeldsen et al., 2015; Nascimento et al., 2015). This made the risk of bias assessment difficult. However, the sensitivity analysis showed that excluding studies with a high risk of bias did not materially change the results of the meta-analysis of the HbA1c values.

The most frequently used instrument to measure self-management in diabetes patients was the SDSCA questionnaire. However, the SDSCA questionnaire pays limited attention to medication related issues (Toobert et al., 2000). Therefore, this questionnaire may not be the best instrument to measure the effects of pharmacist-led and medication-related self-management support. A more suitable instruments for instance might be the MUSE questionnaire (Medication Understanding and Use Self-Efficacy Scale), which focuses on medication use and knowledge (Cameron et al., 2010). This scale can be used among patients with any level of health literacy.

The interventions reported in all of the included studies can be considered as complex interventions, because all of them consisted of multiple components. Also the mechanisms of action for implicating practice were complex as this depended on both the pharmacists delivering and implementing the intervention and the patient implementing it into daily life (Kelly et al., 2017). In this review we have shown that these complex interventions have a positive influence on various diabetes related outcomes. Subgroup analyses did not provide evidence which of the components were essential for the effect. More sophisticated analyses, such as meta-regression analyses or modeling, could have given more insight into key components (Viswanathan et al., 2017). However, this was not possible due to the limited number of studies, data available and the different ways in which the data was presented in the included studies. Although, we have described the different components of pharmacist-led self-management interventions, the ideal composition of intervention components is still a black box.

The overall results of our study argue that pharmacists take an active role in improving patient diabetes self-management since the effectiveness of pharmacist-led interventions is at least comparable to that of other healthcare providers (Sherifali et al., 2015). Although we were unable to identify specific factors contributing to the success of pharmacist-led self-management interventions, a tailored approach seems to be preferable for future developments (Linn et al., 2011, 2013; Harrington and Noar, 2012). In line with findings of previous studies; self-management needs depend on personal characteristics and development (Bos-Touwen et al., 2015) and self-management support should focus on how to identify problems and how to take appropriate actions (Bodenheimer et al., 2002). Another important factor for successful interventions might be the intensity of contact moments over time, with the intensity of contact moments appearing more important than the length of the intervention. This is demonstrated by Krass et al. (2011) and Odegard et al. (2005) who found that prolonging the follow-up time without sustaining the contact frequency did not further improve HbA1c values. Moreover, some patient groups are more vulnerable to having low self-management skills than others. For example, patients with a low level of health literacy may benefit much more from self-management support compared to more health literate diabetes patients (Fransen et al., 2012, 2015). Summarizing the evidence, pharmacists should offer self-management support to diabetes patients in order to improve clinical- and patient reported diabetes outcomes.

Future research into self-management support should focus on developing an intervention from a multidisciplinary perspective to combine the knowledge from the different disciplines involved in diabetes care. Most studies only focus on the role of a single healthcare professional. Combining the strengths of different disciplines might increase the effect of the intervention. Particular emphasis should be placed on vulnerable patient groups and using valid measurements of self-management skills in multiple dimensions.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.