Key Points

Question Is use of interprofessional collaborative practice (ICP) associated with diabetes and hypertension outcomes in primary care patients?

Findings In this systematic review and meta-analysis of 39 comparative studies that evaluated interprofessional team care involving 3 or more professions in primary care for adult patients with diabetes and/or hypertension, ICP was associated with improved hemoglobin A1C (HbA1c), systolic blood pressure, and diastolic blood pressure levels. Primary care ICP was associated with reductions in HbA1c regardless of baseline levels, but the greatest reductions were found with HbA1c levels of 9 or higher.

Meaning The findings suggest that implementation of ICP in primary care may be associated with improved clinical outcomes for diabetes and hypertension in adult patients.

Abstract

Importance Interprofessional collaborative practice (ICP), the collaboration of health workers from different professional backgrounds with patients, families, caregivers, and communities, is central to optimal primary care. However, limited evidence exists regarding its association with patient outcomes.

Objective To examine the association of ICP with hemoglobin A1C (HbA1c), systolic blood pressure (SBP), and diastolic blood pressure (DBP) levels among adults receiving primary care.

Data Sources A literature search of English language journals (January 2013-2018; updated through March 2020) was conducted using MEDLINE; Embase; Ovid IPA; Cochrane Central Register of Controlled Trials: Issue 2 of 12, February 2018; NHS Economic Evaluation Database: Issue 2 of 4, April 2015; Clarivate Analytics WOS Science Citation Index Expanded (1990-2018); EBSCOhost CINAHL Plus With Full Text (1937-2018); Elsevier Scopus; FirstSearch OAIster; AHRQ PCMH Citations Collection; ClinicalTrials.gov; and HSRProj.

Study Selection Studies needed to evaluate the association of ICP (≥3 professions) with HbA1c, SBP, or DBP levels in adults with diabetes and/or hypertension receiving primary care. A dual review was performed for screening and selection.

Data Extraction and Synthesis This systematic review and meta-analysis followed the PRISMA guideline for data abstractions and Cochrane Collaboration recommendations for bias assessment. Two dual review teams conducted independent data extraction with consensus. Data were pooled using a random-effects model for meta-analyses and forest plots constructed to report standardized mean differences (SMDs). For high heterogeneity (I2), data were stratified by baseline level and by study design.

Main Outcomes and Measures The primary outcomes included HbA1c, SBP, and DBP levels as determined before data collection.

Results A total of 3543 titles or abstracts were screened; 170 abstracts or full texts were reviewed. Of 50 articles in the systematic review, 39 (15 randomized clinical trials [RCTs], 24 non-RCTs) were included in the meta-analyses of HbA1c (n = 34), SBP (n = 25), and DBP (n = 24). The sample size ranged from 40 to 20 524, and mean age ranged from 51 to 70 years, with 0% to 100% participants being male. Varied ICP features were reported. The SMD varied by baseline HbA1c, although all SMDs significantly favored ICP (HbA1c <8, SMD = −0.13; P < .001; HbA1c ≥8  to < 9, SMD = −0.24; P = .007; and HbA1c ≥9, SMD = −0.60; P < .001). The SMD for SBP and DBP were −0.31 (95% CI, −0.46 to −0.17); P < .001 and −0.28 (95% CI, −0.42 to −0.14); P < .001, respectively, with effect sizes not associated with baseline levels. Overall I2 was greater than 80% for all outcomes.

Conclusions and Relevance This systematic review and meta-analysis found that ICP was associated with reductions in HbA1c regardless of baseline levels as well as with reduced SBP and DBP. However, the greatest reductions were found with HbA1c levels of 9 or higher. The implementation of ICP in primary care may be associated with improvements in patient outcomes in diabetes and hypertension.

Introduction

Diabetes and hypertension are substantial causes of heart disease and stroke, which are leading causes of death in the US.1,2 In 2018, 34.1 million people (13% of the US population) had diabetes1 and 108 million (45% of US adults) had hypertension.2 Given the complexity of diabetes and hypertension management, team-based care with physicians, nurses, pharmacists, dietitians, and other health care professionals can be an effective approach.3-6

The World Health Organization defines interprofessional collaborative practice (ICP) as a situation in which “multiple health workers from different professional backgrounds work together with patients, families, carers, and communities to deliver the highest quality of care.”7(p7) According to Wagner et al,8 the use of ICP is the key to achieving the quadruple aim of “improving patient health, enhancing patient experience, reducing health care costs, and improving the work life of providers and staff.”8(p1) Characteristics of ICP teams include shared goals, clarity of roles, effective communication, and shared decision-making.4,9

Although ICP is recognized as a central component of providing optimal primary care, to our knowledge, there is limited evidence of its role in patient-oriented health outcomes. Two systematic reviews reported conflicting results for ICP in patients with diabetes.10,11 One systematic review of 8 studies showed a nonsignificant reduction in hemoglobin A1C (HbA1c) when comparing team-based care with usual care.10 In contrast, another review of 7 trials found that team-based care was associated with improved HbA1c levels compared with controls.11 A 2019 systematic review and meta-analysis of 35 studies reported that, compared with usual care, team-based care was associated with improved HbA1c, systolic blood pressure (SBP), and diastolic blood pressure (DBP) levels.6 This study included randomized clinical trials (RCTs) only up to 2015 and was not focused on assessing ICP by at least 3 professions in primary care settings.

A previous scoping review (2000-2013) examined the breadth of information on ICP in primary care and reported broad consequences associated with patient outcomes.12 This review, without meta-analysis, found 8 studies reporting positive differences in HbA1c and 10 reporting positive differences in BP when ICP was compared with controls. Conversely, 6 additional studies reported no differences in HbA1c, and 3 reported no differences in BP.12 Therefore, results are mixed in assessing ICP in patients with diabetes and hypertension, and an updated systematic review and meta-analysis is warranted to expand applicable knowledge. Our systematic review and meta-analysis was an extension of the scoping review,12 with a literature search updated to 2020 that examined ICP compared with usual care and controls using HbA1c, SBP, and DBP in patients with diabetes and/or hypertension receiving primary care.

Methods

Study Selection
To be eligible for inclusion in the systematic review, studies had to use a comparative design and evaluate ICP in adults with diabetes and/or hypertension receiving primary care. We selected studies that reported evidence of ICP involving 3 or more health professions; primary care practice; adults having diabetes and/or hypertension; assessment of HbA1c, SBP, or DBP levels; and statistical evaluation of ICP. Non-English records, reviews, meta-analyses, drug trials, case studies, editorials, and news articles were excluded. To be included in the meta-analysis, the reported comparative data had to be sufficient to calculate a standardized mean difference (SMD).

Definitions for ICP and Primary Care
For the present study, an ICP team was defined as a collaboration among individuals from at least 3 different health professions. At least 1 member of the team needed to serve as the primary care professional bearing the authority to diagnose and initiate treatments.7,13,14 Consistent with the previous scoping review, the Starfield definition of primary care was used, which defines primary care as being the first point of entry to a health care system, person focused (not disease oriented), and integrating care from outside professionals.12,15,16 The 4 key features of primary care service delivery include access (easy to establish contact with a professional who has gatekeeper roles), longitudinality (timely and complementary patient–health care professional experience), comprehensiveness (meeting a broad range of health needs), and coordination of care (integration of services received from external/specialty health care professionals).12,15,17

Search Strategy
A systematic search was conducted in March 2018 using resources including MEDLINE; Embase; Ovid IPA; Cochrane Central Register of Controlled Trials: Issue 2 of 12, February 2018; NHS Economic Evaluation Database: Issue 2 of 4, April 2015); Clarivate Analytics WOS Science Citation Index Expanded (1990-2018); EBSCOhost CINAHL Plus With Full Text (1937-2018); Elsevier Scopus; FirstSearch OAIster; AHRQ PCMH Citations Collection; ClinicalTrials.gov; and HSRProj. Results were limited to English and initially to publication years from January 2013 to 2018; this start year was selected to build on the previous scoping review (2000-2013).12 A research librarian who participated in the scoping review assisted with our search. The search strategy for MEDLINE is described in eMethods 1 in the Supplement. In addition, an abbreviated search update was performed (2018 to March 2020), using Ovid MEDLINE and Cochrane Library databases.

Data Collection
This study followed the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guideline for data abstraction in the systematic review.18,19 A pharmacist who practices in interprofessional primary care (J.K.L.) led the review and data collection. A dual review process, having 2 teams of 2 reviewers, was used for study inclusion and data extraction using previously tested standardized forms to minimize variability. Each reviewer independently screened articles and extracted data, then met to reconcile the differences by consensus. We collected study characteristics; participant characteristics; team makeup, features, and functions; and clinical outcomes of HbA1c, SBP, and DBP.

Outcomes and Data Analysis
The data for primary outcomes (HbA1c, SBP, and DBP) were analyzed separately. The SMD (outcome measure that indicated the difference in effect between ICP and comparison) was calculated for each study. Subsequently, the SMDs were pooled using a random-effects model, and a forest plot was constructed. The SMD provided an overall effect estimate of the ICP. The size of the SMD is considered as small (<0.2), moderate (0.2–0.8), or large (>0.8).20 For each outcome, a fail-safe N was calculated to determine the number of studies with no difference required to change a significant result to no difference. The I2, which measures the percent of variation owing to factors other than random variation, was used to determine whether excessive nonrandom variation was present. Presence of publication bias was evaluated using a funnel plot and Kendall τ rank correlation.

The studies were stratified by design (RCT, prospective cohort, retrospective cohort, and pre-post studies), and the analysis was repeated to determine whether the SMD was associated with study design. For HbA1c, stratification by baseline HbA1c was performed to identify associations of ICP with patient cohorts having varied diabetes control status.21 In addition, the leave-one-out method was conducted to determine whether specific studies had a substantial role in the pooled SMD. Data analysis was conducted using Comprehensive Meta-Analysis (CMA) software (Biostat Inc). The CIs reported in CMA were corrected using the method of Hartung-Knapp-Sidik-Jonkman.22 The a priori P value was .05. The meta-analysis process and data are shown in eMethods 2 in the Supplement.

Risk of Bias Assessment
Because we included diverse study designs, the tools based on the framework of the Cochrane Collaboration recommendations for Effective Practice and Organization of Care were used.23 These tools were developed for bias assessment of RCTs, non-RCT cohorts, and pre-post studies. Each item was ranked low risk of bias, unclear, or high risk of bias. A dual review was performed with consensus generation.

Results

Study Selection and Characteristics
We identified 6316 articles from the 2013-2018 searches and 175 articles from other sources. After removing duplicates, the review teams screened 3543 titles or abstracts then reviewed 170 abstracts or full-texts to assess 63 articles for eligibility, including the 12 relevant articles from the previous scoping review12 and 5 from the abbreviated search update (2018 to March 2020). Of these, 13 records were excluded for having 3 or fewer health professions or no usable outcome measures, leaving 50 articles retained in the systematic review. A final 39 studies were included in the meta-analysis24-62 after 11 studies were excluded because of inadequate data (eTable in the Supplement).63-73 Figure 1 depicts the inclusion process of the systematic review and meta-analysis.

Characteristics of the 50 studies included in the systematic review are listed in Table 1. Of the 39 studies included in the meta-analyses, 15 were RCTs,24-38 7 were prospective cohort trials,39-45 1 was a retrospective cohort,46 and 16 were pre-post studies.47-62 Sample size ranged from 40 to 20 524, and study duration ranged from 3 to 24 months. Among the studies that reported patient age and sex, the mean age ranged from 51 to 70 years, and the percentage of male participants ranged from 0 to 100. Studies were most often conducted in the US (n = 18), followed by Brazil (n = 4) and Canada (n = 4), and in an ambulatory care clinic/center/office (n = 9) and community health centers (n = 8). Table 1 also lists ICP team members, roles, main features/process, name of intervention program/model if specified, and other notable intervention details. The team makeup varied widely from the number of professionals involved to types of professions included (3-10). Most teams involved physicians as primary care professionals (n = 36), and most often included professionals from nutrition (n = 33), nursing (n = 32), and pharmacy (n = 20). Similarly, interprofessional team function and intervention features reported by the included studies varied.

Study Outcomes
Hemoglobin A1C
In data pooled from 34 studies (N = 12 599) shown in Figure 2, ICP was associated with reduced HbA1c for all groups regardless of baseline HbA1c levels, although the SMD varied between the groups. For group 1 (mean baseline HbA1c, 7.4), the SMD was small at −0.13 (95% CI, −0.20 to −0.06; P < .001); for group 2 (mean baseline HbA1c, 8.6), the SMD was borderline moderate at −0.24 (95% CI, −0.39 to −0.08; P = .007); and for group 3 (mean baseline HbA1c, 9.9), the SMD was large at −0.60 (95% CI, −0.80 to −0.40; P < .001). The SMD was significantly greater for group 3 than for either group 1 (P < .001) or group 2 (P = .002), but the SMDs for group 2 and group 1 did not differ (P = .08). The SMD increased 80% from group 1 to group 2 and 250% from group 2 to group 3. Given the substantial differences among these groups, no overall SMD was calculated. Heterogeneity (I2) also varied in group 1 (I2 = 42.9%), group 2 (I2 = 79.9%), and group 3 (I2 = 81.5%), indicating significant between-study variations. In the leave-one-out analysis, removal of 1 study52 in group 2 reduced the group SMD by 27% from −0.24 to −0.17, which would have contributed to the heterogeneity of group 2. No other study changed group SMDs more than 18%. Heterogeneity was not associated with the number of professions involved in ICP; the correlation between the number of professions and decrease in HbA1c was not significant. The correlation of study duration and HbA1c effects was also nonsignificant.

The association of ICP with HbA1c differed by study design (overall P = .03 for differences between the 3 types of studies) (eFigure 1 in the Supplement). The SMD was greatest for RCTs (SMD = −0.46; 95% CI, −0.65 to −0.27; P < .001), less for pre-post studies (SMD = −0.26; 95% CI, −0.40 to −0.12; P = .002), and least for prospective cohort studies (SMD = −0.14; 95% CI, −0.33 to −0.05; P = .11). Only the RCTs and prospective cohort studies differed significantly (P = .007), with no statistical difference between the RCTs and pre-post studies (P = .12) or pre-post studies and prospective cohort studies (P = .08). However, the research design was confounded by baseline HbA1c levels. The mean baseline HbA1c level for the prospective cohort studies was 7.5%; for pre-post studies, 8.4%; and for RCTs, 9.1%; which is similar to the baseline HbA1c levels and the SMDs for HbA1c reduction. In the funnel plot (eFigure 2 in the Supplement), missing studies in the right lower quadrant were noted, and Kendall τ rank correlation was significant (τ=−.37; P = .002), indicating likely publication bias. The fail-safe N = 2068 suggested that 2068 studies showing no effect are needed to reduce the SMD to 0.

Systolic Blood Pressure
In data pooled from 25 studies (N = 35 618), shown in Figure 3, ICP was associated with a moderate effect on SBP; the overall SMD was −0.31 (95% CI, −0.46 to −0.17; P < .001). However, the SMD varied by study design. The SMD was significant for ICP in RCTs (SMD = −0.37; 95% CI, −0.62 to −0.11; P = .009) and the retrospective cohort study (SMD = −0.08; 95% CI, −0.11 to −0.06; P < .001) but not for prospective cohort studies (SMD = −0.28; 95% CI, −0.66 to −0.09; P = .10) or pre-post studies (SMD = −0.27; 95% CI, −0.58 to −0.04; P = .08). The SMD for the retrospective cohort study was significantly smaller than the SMDs for RCTs (P = .02) and pre-post studies (P = .02) but not statistically different from the SMD for prospective cohort studies (P = .29). Nonetheless, when excluding the retrospective cohort study, there was no difference in the SMD between RCTs, pre-post studies, and prospective cohort studies. Heterogeneity among the studies was high (I2 = 95.4% overall). Heterogeneity was also high among within-design groups: prospective cohort studies (I2 = 98.2%), RCTs (I2 = 86.4%), and pre-post studies (I2 = 84.1%). In the leave-one-out analysis, removal of 1 study43 decreased the overall SMD by 23%, contributing to heterogeneity. The SMD was not associated with baseline SBP levels (for SBP<130 vs SBP≥130; P = .76). The funnel plot (eFigure 3 in the Supplement) showed missing studies to the right of the mean. The Kendall τ rank correlation between SMD and SE was significant (τ=.22; P = .008), indicating likely publication bias. The fail-safe N was 1812 studies.

Diastolic Blood Pressure
In data pooled from 24 studies (N = 35 606), shown in Figure 3, ICP was associated with a moderate effect on DBP; the overall SMD was −0.28 (95% CI, −0.42 to −0.14; P < .001). However, the SMD varied by study design. The SMD was significant for ICP in the RCTs (SMD = −0.36, 95% CI, −0.63 to −0.10; P = .01) and pre-post studies (SMD = −0.17; 95% CI, −0.27 to −0.07; P = .005) but not in the prospective cohort studies (SMD = −0.29, 95% CI, −0.79 to 0.21; P = .19) or retrospective cohort study (SMD = 0.00, 95% CI, −0.03 to 0.03; P = .87). The SMD for the retrospective cohort study was significantly smaller than the SMDs for the RCTs (P = .006) and pre-post studies (P < .001) but not statistically different from the SMD for prospective cohort studies (P = .39). Nevertheless, there was no difference between the SMDs for the RCTs, pre-post studies, and prospective cohort studies (P = .31). Heterogeneity was high among the prospective studies (I2 = 98.9%; P < .001) and RCTs (I2 = 86.1%; P < .001) but not among the pre-post studies (I2 = 39.7%; P = .13). In the leave-one-out analysis, the removal of 1 study43 reduced the SMD by 24%, contributing to the heterogeneity. The SMD was not associated with baseline DBP levels (for DBP<80 vs DBP≥80; P = .45). No publication bias was noted; the funnel plot showed no missing studies (eFigure 4 in the Supplement), and the Kendall τ rank correlation was nonsignificant (τ=.22; P = .14). The fail-safe N was 1539 studies.

Bias Assessment
The bias assessment for studies included in the meta-analyses are presented in Table 2. Overall, RCTs scored a low risk for most factors, but there was a mixed unclear and high-risk majority for “knowledge of allocated interventions” (n = 9) and “contamination” (n = 8). The non-RCT studies showed most high-risk scores for “allocation sequence generation” (n = 21) and “concealment of allocation” (n = 18) and mixed unclear and high-risk scores for “dropouts, attrition” (n = 19) and “knowledge of allocated interventions” (n = 14).

Discussion

A notable finding from the current meta-analysis (n = 39) is that ICP was associated with reduced HbA1c levels regardless of the baseline HbA1c level and decreased SBP and DBP in adult primary care patients with diabetes and/or hypertension. The ICP effect estimate was substantial for patients with a baseline HbA1c greater than or equal to 9 (250% larger than the effect estimate for baseline HbA1c≥8 to <9), but no correlation was found between baseline BP levels and ICP. Although ICP teams (≥3 different professions) delivered varied interventions within diverse primary care settings, the association was significantly positive across all SMDs, with the largest effect size for the highest baseline HbA1c group and a moderate effect size for both SBP and DBP. For HbA1c, 2068 negative studies are needed to negate the favorable effects by ICP. For SBP and DBP, important clinical measures of hypertension and cardiovascular status for diabetes, 1812 and 1539 negative studies, respectively, are needed to refute the effects of ICP.

To our knowledge, this is the most up-to-date and inclusive systematic review and meta-analysis on ICP in primary care for patients with diabetes and/or hypertension (50 studies in systematic review and 39 in meta-analysis). While previous research has assessed the association between team care and diabetes and hypertension outcomes, the latest search, to our knowledge, ended in 2015 in an RCT-only meta-analysis.6 Conducted in controlled environments involving specified patient populations and using precise interventions, RCTs have a superior study design with a lower risk of bias. Yet, the findings from RCTs may lack real-life scenarios and patient behaviors in response to clinical interventions that more closely reflect everyday experience. Moreover, previous research included teams of at least 2 professionals in various settings, whereas we included ICPs of at least 3 health professions in primary care. Among the 35 studies in the 2019 meta-analysis,6 only 2 studies overlapped with the 39 studies included in our meta-analysis,27,37 indicating differences in research scope.

To strengthen the confidence to detect the directly aligned effects of ICP, we strictly adhered to the prespecified inclusion criteria and required the use of explicitly stated data from each study. Therefore, in study selection, we excluded studies that did not clearly report involvement of at least 3 professions in primary care. For example, a study of pharmacists working with physicians and other health care professionals on patients with diabetes that provided no specification for “other providers” was excluded.74 Further, we excluded studies with outcome measures reported in a format that was not suitable for SMD calculation from the meta-analysis. For bias assessment, we used tools specific for rating RCTs and non-RCTs and found RCTs appraised as having a lower risk of bias compared with non-RCTs.

Heterogeneity was substantial for all of the outcomes (HbA1c, SBP, and DBP). For HbA1c, baseline HbA1c likely contributed to the heterogeneity, but significant heterogeneity remained within the HbA1c groups. For SBP and DBP, we found no association between baseline BP levels and BP reduction; however, the heterogeneity was high. Study design may have been a factor in the heterogeneity, but it was difficult to assess for HbA1c given the confounding by baseline HbA1c levels. The BP stratification by study design revealed significant differences in overall SMD for SBP and DBP, with RCTs and prospective studies showing larger effect sizes compared with the other designs. Such differences may stem from studies with more control having the intervention group receive all aspects of the intervention, whereas less controlled studies may have missing intervention aspects or contaminated comparison groups. The number of professions included in the ICP teams did not seem to contribute to the heterogeneity. The study duration also varied (3-24 months), yet the association of study duration and HbA1c was not significant. Hence, heterogeneity may be associated with factors that were not assessed in this meta-analysis, such as intervention dose-effect.

Sources of variation were also likely due to differences in sample size and population, setting, and possible publication bias. Sample size may have similar effects as the study design; for example, smaller studies may be easier to control than very large studies. Simultaneously, studies with a small sample size may have been underpowered to detect the intervention effect, and biased selection may have taken place. There was a varying degree of diabetes control among the participants indicated by baseline HbA1c levels, which may mean that the source populations were varied. Although the mean age ranged from 51 to 70 years, only 2 studies reported a mean age greater than 65 years. While all ICP teams delivered primary care (18 in the US and 21 elsewhere), study settings varied from ambulatory care clinics to community health centers, public health centers, Veterans Affairs health systems. and other settings, with differing resources and infrastructures for ICP provision. Publication bias, which can also be a factor in variation among included studies, was found to be likely for HbA1c and SBP.

Similar to previous findings,21,75,76 we uncovered inconsistencies among the number and types of professionals involved in ICP, how the team functioned, and types of interventions delivered. The number of professions ranged from 3 to 10, which suggests differing interventions delivered by diverse expertise. The focus of our study, however, was to assess ICP and not the addition of specific health care professionals. The secondary analysis showed no association between the number of professions in ICP and HbA1c reduction. The teamwork and communication strategies varied, although colocation was most often reported (n = 30), followed by having shared electronic medical records (n = 10) and weekly or biweekly team meetings (n = 7). Regarding the interventions, 13 teams provided joint/group educational sessions and 11 had shared/group visits. With such diversity, identifying an ideal team feature and function for effectiveness and efficiency, perhaps tailored to patient risk, may be an appropriate future research area.

Limitations
This study has limitations. No determination of differences in the source population was evaluated, such as educational level that may be a factor in medication adherence, lifestyle modifications that can affect outcomes, or insurance information that may reveal socioeconomic status. Neither the degree of integration among team members in primary care nor the intervention intensity was clearly specified in most studies. Study funding sources were also not considered. Despite these limitations, we assessed an ample number of studies that used the equivalent outcome measures. Worldwide, health care is transforming rapidly, with team-based care suggested for diverse patients. Concurrently, aging populations with chronic conditions may overwhelm primary care systems. ICP appears to be a plausible option for areas with limited access to care and in patients with poorer diabetes control. Using our findings, primary care practices may wish to consider providing ICP involving at least 3 professions to improve diabetes and hypertension outcomes.

Conclusions

The results of this systematic review and meta-analysis suggest that there is a positive association of ICP in primary care with HbA1c, SBP, and DBP levels in adult patients with diabetes or hypertension. Adults with diabetes and/or hypertension should receive team-based care to improve outcomes.