The HHIP uses a multipronged approach to QI, including practice facilitation and technical assistance, on-site and e-learning, and improvement through data transparency. These strategies, described in further detail in this paper, were selected by the study investigators based on a review of patient-centered outcomes research. The HHIP is designed to build internal QI capacity and increase the implementation of evidence-based guideline recommendations for screening and treatment of cardiovascular risk factors. This project builds on the strengths and the previous and ongoing efforts of all Cooperative partners, such as (1) the Southeastern Collaboration to Improve Blood Pressure, a recent trial assessing the use of practice facilitation to improve BP control among adults with hypertension in primary care (UAB) [15]; (2) the Blood Pressure Self-Management Initiative (Alabama Department of Public Health); (3) expertise in promotion of best practices for pharmacotherapy (Auburn); as well as (4) ongoing QI initiatives in primary care (QSource and APHCA).

The HHIP was initially designed as a “stepped wedge” in stages with groups of 10 practices randomly selected to begin the intervention each quarter. However, delays in practice recruitment did not allow for the randomization of practices by quarter, so we instead used a nonrandomized, pre-post design with primary care clinics as the unit of observation. Despite the design change, quantitative analysis of the effect of HHIP on measures of hypertension and smoking cessation will follow the general framework for the analysis of a stepped-wedge study. This includes using preintervention data from all practices to estimate secular trends—changes expected to occur without intervention, as well as estimating effects from the HHIP, and whether there is an enduring change in outcomes after the year of HHIP activities beyond what would be expected due to secular trends. Models will include random effects for sites and for patients to account for multiple patients nested within sites and the potential for multiple observations per patient. Secondary clinic-level analyses will use data aggregated each quarter by clinic.

The Cooperative used several methods of recruitment including existing organization communication channels to recruit primary care practices for inclusion in the HHIP. First, we reached out to clinics in existing networks through the Alabama Practice-Based Research Network, the APHCA, and QSource, leveraging existing relationships for introductions. In addition, we created flyers and promotional materials to send to clinics through email and fax. We also conducted Google searches for clinics in designated areas to find additional practices to recruit while we were visiting enrolled sites. To ensure we reached a broad range of clinics, we aimed for at least 50% of our clinics to be Federally Qualified Health Centers (FQHCs) or look-alikes, which have an important role in increasing access to primary care in medically underserved areas, including rural communities [16,17].

For each practice that expressed interest in participation, we conducted an informational recruitment meeting to further explain the HHIP objectives and procedures. During this initial meeting, interested clinics completed readiness assessments using standardized questionnaires [18,19] and signed an informal letter of agreement to participate in the project. Following the initial meeting, the questionnaire scores and clinic information (eg, anecdotal patient demographics and clinic estimation of BP control rate) were shared with the Cooperative, and practices were voted on to be enrolled in the project. Once a practice was enrolled, a lead practice facilitator was assigned.

Practice facilitators scheduled and conducted a kick-off meeting for the practices that were enrolled in the HHIP. This kick-off meeting served as the start of the intervention where practices received additional information about practice facilitation, completed an “About Your Practice” survey, and initiated processes required to extract baseline data. Enrolled practices were also asked to identify a practice champion as part of the enrollment process. Practice champions are integral to the successful adoption of innovations within a practice, defined as “individuals who dedicate themselves to supporting, marketing, and driving through an implementation, overcoming indifference or resistance that the intervention may provoke in an organization” [20]. The practice champion worked with the practice facilitator to schedule monthly touchpoints throughout the intervention period to implement and track the progress of quarterly QI activities. Clinics were compensated US $1500 for their time and participation in the intervention.

Outcomes of the study were selected based on their associations with patient-level benefit (ie, cardiovascular risk reduction) and avoiding undue burden for practices or providers to measure (Table 2). BP-related outcomes included documentation of hypertension in the EHR when a patient has 2 or more readings with a systolic BP≥130 mm Hg or a diastolic BP≥80 mm Hg, which aligns with the 2017 American College of Cardiology and American Heart Association BP guideline recommendation for hypertension diagnosis [27]. Given the high cardiovascular risk associated with stage II hypertension (systolic BP≥140 mm Hg or diastolic BP≥90 mm Hg) and the prevalence of stage II hypertension in Alabama, the main outcome for intervention efficacy is BP control among adults with hypertension to <140/90 mm Hg. This is also the BP target used in the Centers for Medicare & Medicaid Services (CMS) 165 and National Quality Forum (NQF) 0018 clinical quality measure. Tobacco use outcomes were available through the EHR and included tobacco use screening and cessation intervention. The combined measure of tobacco use screening and cessation intervention aligns with the CMS 138 and NQF 0028 clinical quality measures.

Data on BP and tobacco use are available quarterly from the practice EHRs and retrospective EHR extraction provides preintervention data for all practices. In addition to clinical parameters, we collect data on age, race, and gender, as well as insurance status as available, to assess the impact of our program for these specific subgroups.

The primary model for fixed effects without adjustment for any covariates will have the following structure:

For the primary dichotomous outcomes of hypertension control and documentation of smoking status, Y will represent the log-odds of the binary outcome. In the model “time” will be considered as “calendar time” measured in quarters from the earliest quarter of baseline data and β₁ will be the estimate of the secular trend. “HHIP exposure” will similarly measure the cumulative number of quarters the practice was exposed to HHIP, ranging from 0 in the baseline period, to 4 after a year of HHIP and β₂ will be the estimate of the quarterly change due to HHIP. “Time after HHIP” will similarly quantify the period of follow-up after the year of HHIP and β₃ will be the estimate for any enduring effect of HHIP on rates after the intervention beyond what could be attributed to secular trends. A positive β₃ would indicate that outcomes continued to increase faster among practices after the HHIP intervention than would have been expected and would suggest a durable effect, while an estimate of β₂ that is not significantly different from 0 would be interpreted as practices returning to the change expected by secular trends. In the general framework, the effect due to HHIP will be considered statistically significant if the P value of the model coefficient is <.05.

Before analysis, we will consider smoothed scatter plots of outcomes over time and HHIP exposure; if any trends appear to be importantly nonlinear, we will consider these as categorical to most appropriately model the data. If this is the case, the overall test of the main hypothesis will be the significance of the coefficient for the full-year exposure to HHIP. The general model can be further adjusted for patient-level characteristics (eg, age, race, and gender) and practice-level characteristics (eg, urban vs rural, FQHC vs non-FQHC, Consolidated Framework for Intervention Research [CFIR] scores, and other implementation-related measures) and stratified analyses can be run using subsets of the data. Differential effects by subgroups (eg, race, gender, or practice types) will be tested using interaction terms.

Statistical power was assessed using simulation methods based on the outcome of hypertension control. Briefly, data was simulated for 50 practices, with randomly generated practice-level baseline hypertension control rates of 50% with an SD of 4%. Secular trends were similarly generated for each practice with a mean yearly decrease of 1% with an SD of 0.05%. We assumed a mean effect due to HHIP of 2% per year (0.5% per quarter) assigned randomly with an SD of 1%. Simulations included a range of residual HHIP effects ranging from 0.5% to 1.5%. Simulated datasets began with creating data for 10,000 patients per practice using generating baseline status at random, then iterated forward with each practice’s assigned secular trends and intervention effects, followed by the addition of random perturbations to model the high variability of BP measurements leading to changes in control status at each visit. For multiple iterations, 50 practices were randomly selected, along with a reduced sample of patients equal to 5 times the target sample size to reflect the fact that most patients will not have a visit in a given quarter (eg, if we expect to have data for 200 patients per practice per quarter, the data had an average of 1000 patients per practice). Finally, for each iteration, a random sample was drawn to reflect the mean of 200 patients per quarter. The mixed models were run on each iteration and power was estimated as the proportion of iterations in which coefficients were significant at the P<.05 level. We expect there to be >500 per patient per quarter, which would reflect <50 patients per week at a practice with a single practitioner so these power estimates should be highly conservative. Data from the Physicians Foundation in 2018 found that the average doctor sees 20 patients per day, so we could indeed expect >400 patients with hypertension per quarter for each physician [28]. Under these assumptions, there was >95% power for the HHIP effect with as few as 100 patients per site, which further suggests that we will have adequate power for analyses of patient subgroups analyses. Allowing for 400 patients per site provided 100% power for the secular trends and HHIP effect as well as 90% power for the 1% yearly residual increase from HHIP during follow-up. For analyses restricted to subgroups of practices, including only 25 practices with 400 patients per practice left >90% power for the secular trend and HHIP effect, but reduced power to 64% for the residual effect of HHIP during follow-up. Finally, in simulations of comparisons between subgroups (eg, race or gender), 50 practices with 400 patients per quarter per practice provide >80% power to detect a difference in yearly rates of 1.5% between groups (2% vs 3.5%) using a statistical interaction indicating that we will have adequate power for subgroup analyses. The outcomes related to smoking (principally documentation of smoking status) can be expected to have higher power based on higher baseline rates and greater control of the physicians (leading to lower variability in patient-level outcomes).

The implementation outcomes evaluation will be guided by Proctor’s Framework for Implementation Outcomes [29], and we will assess implementation using the CFIR [21,30]. Our proposed evaluation of the implementation of the HHIP will be guided by the tenets of realist evaluation. The realist evaluation approach recognizes that there are many interwoven factors operating at different levels within a setting, making this evaluation method a better approach to complex interventions than a more traditional, noncontextual cause and effect evaluation approach [31].

Realist evaluation investigates the “context-mechanisms-outcomes pattern configuration.” Realist evaluation does not just consider outcomes to determine if an intervention is successful but analyzes the outcomes to discover if the speculated mechanism or context theories are confirmed. The overarching HHIP evaluation questions are: What works, for whom, in what way, to what extent, in what contexts, and how? To better understand context-specific mechanisms that may influence the successful implementation of HHIP, we will draw on a menu of constructs from the CFIR across 5 domains [21,30]. These domains include the (1) intervention characteristics, or aspects of the intervention that may have a positive or negative impact on implementation success, (2) outer setting, which includes the external forces that may influence successful implementation, (3) inner setting, which includes the organizational characteristics of the implementing organization, (4) characteristics of the individuals, including the knowledge, attitudes, beliefs, and personal attributes of those implementing, and (5) process of implementation, which refers to the stages of implementation and participation of key influencers in each of those stages.

Of the 39 constructs described in the CFIR, our evaluation will focus on the 23 constructs that are most relevant to this project (Table 3).

Before implementation at each clinic, we administer the Organizational Readiness for Implementing Change (ORIC) survey, a psychometrically-validated instrument developed by Shea and colleagues [18], based on the constructs of Weiner’s Organizational Theory for Implementation Effectiveness [32]. At baseline and again at the conclusion of the HHIP, the evaluation team will survey the participating chief medical officer, a senior practice leader, or the practice champion of each clinic using the 14-item Change Process Capability Questionnaire (CPCQ) to assess a practice’s capacity to improve care using QI strategies [19].

Drawing on Proctor’s Conceptual Framework for Implementation Outcomes, we will evaluate the implementation of the HHIP. Proctor et al [29] describe 8 implementation outcomes, including acceptability, appropriateness, feasibility, costs, adoption, fidelity, penetration, and sustainability. We will address the acceptability, appropriateness, feasibility, and sustainability of the HHIP with validated questionnaires [33,34]. Fidelity and penetration will be assessed with administrative data collected as part of delivering the intervention (eg, number and duration of contacts with clinic). In addition, we will use semistructured interviews, approximately 30 minutes in duration and conducted through Zoom (Zoom Communications, Inc), with clinic staff members and providers from 8 to 10 clinics. Guided by the CFIR, the interviews will focus on the context for implementing the HHIP. Table 4 describes the outcomes that we will measure and our method or measurement.