A controlled trial of dissemination and implementation of a cardiovascular risk reduction strategy in small primary care practices

Abstract

Objective

To assess the effect of dissemination and implementation of an intervention consisting of practice facilitation and a risk‐stratified, population management dashboard on cardiovascular risk reduction for patients at high risk in small, primary care practices.

Study Setting

A total of 219 small primary care practices (≤10 clinicians per site) across North Carolina with primary data collection from electronic health records (EHRs) from the fourth quarter of 2015 through the second quarter of 2018.

Study Design

We performed a stepped‐wedge, stratified, cluster randomized trial of a one‐year intervention consisting of practice facilitation utilizing quality improvement techniques coupled with a cardiovascular dashboard that included lists of risk‐stratified adults, aged 40‐79 years and their unmet treatment opportunities. The primary outcome was change in 10‐Year ASCVD Risk score among all patients with a baseline score ≥10 percent from baseline to 3 months postintervention.

Data Collection/ Extraction Methods

Data extracts were securely transferred from practices on a nightly basis from their EHR to the research team registry.

Principle Findings

ASCVD risk scores were assessed on 437 556 patients and 146 826 had a calculated 10‐year risk ≥10 percent. The mean baseline risk was 23.4 percent (SD ± 12.6 percent). Postintervention, the absolute risk reduction was 6.3 percent (95% CI 6.3, 6.4). Models considering calendar time and stepped‐wedge controls revealed most of the improvement (4.0 of 6.3 percent) was attributable to the intervention and not secular trends. In multivariate analysis, male gender, age >65 years, low‐income (<$40 000), and Black race (P < .001 for all variables) were each associated with greater risk reductions.

Conclusion

A risk‐stratified, population management dashboard combined with practice facilitation led to substantial reductions of 10‐year ASCVD risk for patients at high risk. Similar approaches could lead to effective dissemination and implementation of other new evidence, especially in rural and other under‐resourced practices.

Registration: ClinicalTrials.Gov 15‐0479.

What is Already Known on this Topic

• Cardiovascular disease (CVD) remains the leading cause of death across the United States and is particularly devastating in rural North Carolina counties several of which are classified as stroke belt “hot spots”.

• Despite evidence‐based guidelines endorsing newer prescribing algorithms designed to attenuate CVD, adoption by primary care clinicians has been slow especially in small, rural, and other under‐resourced practices.

What This Study Adds

• By combining practice facilitation and leveraging EHR data into a useful population management tool, new prescribing guidelines can be more rapidly implemented to achieve significant CVD risk reduction.

• The risk reduction achieved through the intervention is likely to prevent over 5800 cardiovascular events over 10 years resulting in improved cardiovascular health and possible cost savings.

• These methods could be adapted to quickly implement other relatively new evidence into primary care, especially in small, rural, and under‐resourced practices.

1. INTRODUCTION

Cardiovascular disease (CVD) remains the leading cause of death across the United States and is particularly devastating in rural North Carolina counties, ¹ , ² several of which are classified as stroke belt “hot spots” defined by cerebrovascular disease prevalence and mortality. ³ Despite evidence‐based guidelines endorsing new prescribing algorithms designed to attenuate high risk, ⁴ adoption by primary care clinicians, in general, has been slow. Adding to the problem, implementation of CVD risk reduction guidelines has been more challenging in rural areas, contributing to rural‐urban disparities. ⁵

Recently introduced evidence‐based guidelines recommend determination of the 10‐year Atherosclerotic Cardiovascular Disease (ASCVD) risk score to guide decisions on initiation and intensification of medications that reduce CVD risk. Without use of these risk scores, clinicians would simply use isolated cholesterol results, known vascular disease, or a general clinical gestalt to use these therapies resulting in relative under‐prescribing of available treatments. Examples include risk‐based aspirin and statin recommendations published by the United States Preventive Services Task Force (USPSTF). ⁶ , ⁷ However, most primary care practices have not developed a systematic approach to calculating scores or implementing these guidelines. Calculating the risk score can be a barrier for busy clinicians because many electronic health records (EHRs) have limited capacity to do so. In addition, many EHRs do not generate reports on quality measure performance for new measures. ⁵ These limitations are particularly troublesome for small primary care practices that lack EHR programing and financial resources to pay for tailored reports designed to improve care using new evidence with strong clinical impact.

In response to the slow diffusion of new evidence in primary care, the Agency for Healthcare Research and Quality (AHRQ) launched the EvidenceNOW (EN) initiative. EN sponsored seven cooperatives across the United States to implement and test strategies designed to reduce CVD risk for patients in small primary care practices. Heart Health Now (HHN) is the North Carolina Cooperative for EN. While all EN cooperatives used 3‐12 months of practice facilitation (PF) to implement cardiovascular disease prevention strategies, the NC Cooperative was the only one that used risk stratification and targeting of patients at higher risk as a key component of the intervention. The conceptual model for HHN coupled the Institute for Healthcare Improvement's Model for Improvement (http://www.ihi.org/resources/Pages/ HowtoImprove) with novel informatics support. Onsite PF was the vehicle for disseminating this model and helped practices implement through associated workflow design. The practices’ own EHR was used to build the informatics tool—a dashboard of patients stratified by 10‐year ASCVD risk to pose a sense of urgency for CVD risk reduction by not only documenting each patient's risk but by highlighting those at highest risk who would most benefit from immediate risk reduction treatments. Targets for improvement included newer evidence described by the recent USPSTF recommendations for aspirin use for primary prevention ⁶ and the 2013 ACC‐AHA recommendations for statin treatment. ⁴ We also applied underused older recommendations (smoking cessation counseling and better hypertension control) in our dashboards as facilitation targets. In this report, we assess the effect of the HHN intervention on 10‐year ASCVD risk for the population of patients defined at high risk (≥10 percent) because, the above recommendations, based on expert consensus, target this group for medical treatment in addition to lifestyle modifications.

2. METHODS

2.1. Study design

We performed a stepped‐wedge cluster, randomized trial to assess the effect of the HHN intervention on 10‐year ASCVD risk for individuals with ≥10 percent risk at baseline. Practice data were collected during an initial period in which no practice cohorts were yet exposed to intervention. At regular succeeding intervals (the step), a cohort was randomized to cross over from control to intervention. This process continued until all cohorts crossed over to intervention. Data collection continued throughout the study; therefore, each cohort contributed to observations under both control and intervention periods. The stepped‐wedge design provided a rigorous control group that accounted for secular trends while allaying ethical concerns around withholding an intervention with potential benefit for a large, at‐risk, population.

2.2. Participants

NC primary care practice sites staffed by 10 or fewer clinicians (physicians and/ or advanced practitioners) were eligible. We identified practices through their participation in NC Medicaid's medical home program, Community Care of North Carolina (CCNC). Note that the medical home aspect connotes that patients were assigned a practice as their primary care practice “home”. Functionality as a recognized patient‐centered medical home or in QI‐related work was highly variable and was measured with surveys as described in Section 2.4. Our focus was on independent practices, Rural Health Centers, and Federally Qualified Health Centers (FQHC) that lacked organizational support for workflow redesign and development of tailored quality reports derived from their EHRs. CCNC representatives facilitated recruitment over the first 7 months of the project (May 15, 2015, through December 15, 2015). See the CONSORT diagram in Figure 1. Practices were assigned to high‐readiness or low‐readiness groups based on our ability to extract data needed to build the population dashboards for participating sites. Within readiness groups, practices were randomly assigned to one of three cohorts with sequential start dates for the intervention. Of 292 enrolled, 47 practices withdrew before their actual start and 26 never provided data feeds needed for analysis. Therefore, 219 met our criteria for analysis. Practices were randomly assigned to start dates creating 6 cohorts. Prior to the assigned start dates, all practices served as controls and then remained as controls until beginning their particular intervention phase. The three cohorts of high‐readiness practices received the intervention first followed by the three low‐readiness cohorts. The first cohort started the intervention during January 2016 and then each subsequent cohort joined every other month thereafter. All practices began the intervention by the end of 2016 and finished by December 31, 2017.

FIGURE 1.

CONSORT diagram for practice participation in heart health now[Color figure can be viewed at wileyonlinelibrary.com]

2.2.1. Human subjects

Given that our intervention was centered on the delivery of already recommended practices and was practice rather than individual level, the University of North Carolina at Chapel Hill IRB judged HHN as human subjects exempt.

2.3. Variables

The intervention consisted of educational tools (web‐based modules and live webinars), onsite PF, and utilization of a practice‐specific, cardiovascular population management dashboard that included risk stratification and monthly audit and feedback with measure‐specific run charts to guide quality improvement (QI). Run charts were provided at both the practice and individual clinician level and were used for comparisons to baseline without specific benchmarking. Practice facilitators (also known as a practice coaches, QI coaches, and practice enhancement assistants) are specially trained individuals who work with primary care practices “to make meaningful changes designed to improve patients’ outcomes. They help physicians and quality improvement teams develop the skills they need to adapt clinical evidence to the specific circumstance of their practice environment”. ⁸ The PF for HHN were associated with the North Carolina Area Health Education Centers Practice Support Program. Practice facilitators have past histories as practice managers, nurses, QI managers, clinical educators, and social workers with experience working in outpatient settings. They are trained using online modules then a one‐ to two‐month apprenticeship visiting primary care practices with an experienced practice facilitator. All facilitators then participate on monthly calls, a real‐time list serve, and biennial face‐to‐face professional development meetings. For HHN, they worked with practice teams to establish QI approaches and workflows to implement cardiovascular risk reduction strategies using the dashboard and reports from EHRs. Eighteen facilitators were committed to HHN at any one time. Previous engagement with participating practices was minimal to none. On average, practice facilitators spent an hour per month at each practice site and had a total of one hour of email and web‐meeting interactions in‐between.

The dashboards were created using daily uploads from EHRs to a centralized registry. The registry was used to calculate ASCVD 10‐year risk scores per published methodology ⁴ and produce practice‐specific information that included ASCVD risk stratification of adult patients aged 40‐79 years. The formula used to calculate 10‐year ASCD risk include age, gender, race, systolic blood pressure, diastolic blood pressure, total cholesterol, HDL cholesterol, diabetes history, smoking history, and hypertension treatment status. The patient ranking and risk scores were linked to checklists of unmet risk reduction treatment opportunities. The dashboards were then used by practice leaders as a “reminder” function so providers or designated staff could flag high‐risk patients who were missing important treatments. During visits, facilitators incorporated high‐leverage principles in their work with practice teams that included change management strategies, implementation of the dashboards, workflows to incorporate dashboards, rapid cycle QI, and the use of protocols, templates, and decision support prompts to promote risk reduction. They also emphasized options with each practice team that systematically engaged high‐risk patients through outreach on a practice population basis as opposed to simply waiting for next scheduled appointments. For example, a practice might establish Wednesday afternoons for ASCVD risk reduction clinic or create schedule allotments in a weekly series to accommodate patients at high risk in a systematic, prioritized manner. After initial patient engagement discussing high risk, practices with licensed nurses could use standing orders for treatment intensification without always requiring face‐to‐face visits. Practices limited to nonlicensed staff could only make adjustments using serial visits. In addition, PFs ensured that practices had access to local resources to support heart healthy diets and physical activity.

Key measures targeted by the intervention included standard measures—hypertension control, aspirin for established CVD, and counseling for tobacco cessation. We added the new measures for statin prescribing defined by the AHA‐ACC ⁴ using the 10 percent ASCVD 10‐year risk threshold recommended by the USPSTF ⁷ and risk‐based aspirin for primary prevention per the USPSTF. ⁶ Dates of treatment initiation and patient achievement of specific CVD risk reduction measures were also derived from daily EHR uploads.

2.4. Data collection

Our analysis focused on 219 practices. Data extracts were obtained from practices on a nightly basis directly from their EHR to the research team registry at the Cecil G. Sheps Center for Health Services Research at UNC (N = 26) or CCNC (N = 193). Data were submitted from the fourth quarter of 2015 through the second quarter of 2018. For each practice, we updated risk calculations, individual measures, run charts for practices and individual clinicians (graphic representation of the proportion of patients achieving specific measures over time), and treatment opportunities monthly. Each report represented a rolling 12‐month measurement period (eg, the measurement period for the fourth quarter of 2015 covers calendar year 2015) and included data from all patients who visited the practice during the entire 12 months.

At baseline, defined as between enrollment and initial data collection, two surveys were administered to all practices. These surveys were used to assess factors internal and external to the practice that might affect the impact of the intervention. The Practice Characteristics Survey requested information about practice size, ownership, QI capacity, EHR capabilities, participation in payment or quality demonstration programs, recent organizational changes (eg, loss of one or more clinicians), and other practice characteristics. A mailed and emailed invitation including a web‐based survey link was sent to an identified contact. Starting one week following the invitation, nonrespondents received up to three email reminders and two telephone reminders. Respondent practices received $100 for completed surveys.

The Practice Member Survey, administered in a similar time frame to the Practice Characteristic Survey, was sent to five clinical and nonclinical staff members unless the practice had fewer personnel. Items on this survey focused on practice members’ perceptions of practice capacity for QI, practice readiness for change, personal burnout, and clinical staffs’ perceptions about cardiovascular guidelines. Practice Member Survey data collection procedures mirrored Practice Characteristics Survey procedures. Respondents received $30 for completed surveys.

2.5. Statistical analysis

The group targeted for proactive intervention were patients aged 40‐79 years whose calculated ASCVD 10‐year risk score was ≥10 percent at baseline. As the primary outcome, we report change in the average 10‐year ASCVD risk score for all patients in participating practices five quarters after the intervention began compared to baseline. To determine postintervention risk scores, we applied formulas published by Lloyd‐Jones et al regarding the impact of blood pressure change, aspirin use, and statin use on patients’ baseline ASCVD scores. ⁹ We used smoking status to calculate baseline scores but did not include smoking cessation in postintervention calculations because we only recorded counseling and not confirmed cessation.

We calculated descriptive statistics on all patient‐ and practice‐level variables and assessed change in 10‐year ASCVD risk for all participants and by demographic subgroups. Change in 10‐year ASCVD risk was calculated as the difference between first available score prior to start of intervention, to postintervention as determined 15 months after the initiation of the one‐year intervention. One‐sample t tests were used to assess differences.

Linear regressions were performed to identify single and multiple associations between patient‐ and practice‐level characteristics and reductions in ASCVD score. The general equation employed was:

ASCVDi,qq = 5 − ASCVDi,qq = 0 = yi = β’Xχiβ + εi, εi ~ N(0, σ2), where yi is the response value for subject i, q is quarter, q = 0 is baseline prior to the start of the intervention and q = 5 is the quarter after the intervention, β is the vector of regression coefficients, and εi is the error term. In adjusted results, we examined the potential association of patient‐level characteristics including sex, age, race, ethnicity, and median household income by zip code. In a separate model, we also examined baseline practice‐level factors including rural location, type of practice organization (clinician owned, hospital owned, academic, Rural Health Center, FQHC), number of clinicians, staff to clinician ratio, patient visits per clinician, practice disruptions, payer mix, patient‐centered medical home recognition status (PCMH), practice readiness, and practice adaptive reserve scores (change capacity). ¹⁰ For models 1 and 2, univariate models were first fit after which significant variables were incorporated into a single model. We then used significant associations from these models to build a combined model to examine which patient and practice associations remained significant. Additionally, we used the stepped‐wedge design using linear mixed models (LMMs) with random intercepts for subject to account for the preintervention controls and calendar time. ¹¹

Because of reports published after HHN began suggesting a less robust cardiovascular risk reduction attributable to aspirin than incorporated by Lloyd‐Jones, ¹² , ¹³ we performed post hoc sensitivity analyses for both unadjusted and adjusted models that excluded aspirin benefit.

The R Statistical software ¹⁴ was used for all analyses.

3. RESULTS

3.1. Demographic data

A total of 437 556 patients in 219 practices had ASCVD risk calculated with 146 826 scoring at a 10‐year risk ≥10 percent at baseline. See Table 1 for patient characteristics. Note that 65 percent of patients were non‐Hispanic White, 24 percent non‐Hispanic Black, and 3 percent of Hispanic ethnicity. Fifty‐two percent resided in rural or micropolitan (a nonmetropolitan labor area with a population of 10 000 to 49 999) areas. Practices averaged 6.4 providers per site. Fifty‐four percent were clinician‐owned and 26 percent were FQHCs. The average payer mix per practice consisted of Medicare 32 percent, Medicaid 14 percent, dual 9 percent, uninsured 10 percent, and commercial insurance 34 percent. Practice characteristics are summarized in Table 2.

TABLE 1.

Characteristics of patients at high cardiovascular risk (10‐year Atherosclerotic Cardiovascular Disease Risk Score ≥ 10%) receiving care from Heart Health Now participating practices

Variable	Percentage of patients (N = 146 826)
Female gender	46
Race
Black race	24
White race	65
Race missing	9
Hispanic ethnicity	3
Rural residence	52
Hypertension (ICD‐9 or ICD‐10 code a in the electronic health record)	76.8
Diabetes (ICD‐9 or ICD‐10 code a in the electronic health record)	32.2
Hypercholesterolemia (Total cholesterol > 200 or LDL > 130)	24.5
Current smoker	20.0
History of Ischemic Vascular Disease (ICD‐9 or ICD‐10 code a in the electronic health record for any arteriovascular condition)	13.6
Mean age in years (standard deviation)	65 (8)
Median household income by zip code in dollars (standard deviation)	46 622 (12 480)

^aInternational Classification of Diseases 9th or 10th revision codes.

TABLE 2.

Practice characteristics for sites participating in Heart Health Now that received practice facilitation and submitted data

Variable	Percentage of practices (standard deviation) N = 219
Location rural or micropolitan	52
Payer mix (mean % insurance covered among practices)
Medicare	32 (16)
Medicaid	14 (10)
Dual Medicare and Medicaid	9 (9)
Commercial	34 (17)
Other insurance	2 (6)
No insurance	10 (12)
Practice ownership type
Clinician owned	54
Federally qualified or rural health center	26
Hospital owned	10
Academic center	9
Patient‐centered Medical Home Recognized	59
Mean number of providers per practice	6.4 (4.6)
Mean provider visits per day	22 (6)
Mean adaptive reserve scale score	0.71 (0.12)

3.2. Main findings

At baseline, the mean 10‐year ASCVD risk score among patients at high risk was 23.4 percent (SD ± 12.6 percent). After intervention, the mean score was 17.1 percent (SD ± 11.5 percent), an average absolute risk reduction of 6.3 percent (95% CI 6.3, 6.4). An absolute risk reduction of 4.0 percent was attributable to the intervention when incorporating the stepped‐wedge controls and calendar time in the analysis. In the sensitivity analysis excluding any estimate of beneficial effect for aspirin, the postintervention risk score was 17.5 percent (SD ± 11.7 percent) while the absolute benefit attributable to the intervention remained at 4.0 percent using the stepped‐wedge. Seventy‐seven percent of high‐risk patients experienced risk reduction. Bivariate analyses indicate all subgroups showed risk improvements postintervention (Table 3). In the patient model, male gender, age >65 years, low‐income (<$40 000), metropolitan residence, and Black race (P < .001 for all variables) were each associated with greater risk reductions. When practice variables were added, higher staff to clinician ratio, number of clinicians and patient visits were associated with improved patient‐level risk scores after the intervention while payer mix and hospital ownership had negative effects. Table 4 shows the significant results for the combined model.

TABLE 3.

Results of bivariate analysis showing means and overall risk reduction of 10‐year Atherosclerotic Cardiovascular Disease Risk Score of patients at high risk (≥10%) at baseline vs 3 months postintervention

Variable	Mean baseline risk score (standard deviation) N = 146 826	Postintervention risk score (95% confidence interval)	Absolute risk reduction (95% confidence interval)
Total population	23.4 (12.6)	17.1 (11.5)	6.35 (6.29, 6.41)
Age ≥ 65 y	27.4 (13.5)	21.0 (11.9)	6.48 (6.39, 6.56)
Age < 65 y	17.9 (8.5)	11.7 (8.1)	6.17 (6.09, 6.25)
Median income < $40 000	23.9 (12.8)	17.5 (12.0)	6.45 (6.32, 6.57)
Median income ≥ $40 000	23.0 (12.3)	16.8 (11.0)	6.33 (6.26, 6.40)
White race	23.4 (12.6)	17.3 (11.3)	6.19 (6.12, 6.26)
Black race	23.5 (12.3)	16.7 (11.7)	6.76 (6.62, 6.90)
Male	24.5 (12.9)	18.1 (11.4)	6.48 (6.40, 6.57)
Female	22.3 (12.1)	16.1 (11.4)	6.19 (6.10, 6.29)
Hispanic ethnicity	21.2 (11.7)	15.1 (10.6)	6.01 (5.55, 6.47)
Non‐Hispanic	23.5 (12.6)	17.1 (11.45)	6.33 (6.26, 6.40)
Metropolitan dwelling	23.3 (12.4)	16.6 (11.2)	6.69 (6.60, 6.78)
Rural Dwelling	23.5 (12.7)	17.5 (11.8)	6.03 (5.95, 6.11)

^*All risk score reductions within each subgroup are statistically significant (P < .05).

TABLE 4.

Multiple linear regression results for variables significantly associated with individual risk reduction for patients at high risk (10‐year Atherosclerotic Cardiovascular Disease Risk Score ≥ 10%) after the Heart Health Now intervention

Variable	Estimate	P‐value
Male gender	0.54	<.001
Age ≥ 65 y	0.70	<.001
Black race (compared to White)	0.59	<.001
Median household income < $40 000	0.29	<.001
Hospital Owned (vs all other ownership models)	−1.50	<.001
Number of clinicians	0.11	<.001
Staff to clinician ratio	0.25	.003
Visits per day	0.02	.04
Lower percentage of Medicare patients	0.05	<.001
Lower Percentage of Uninsured	0.03	<.001
Lower Percentage of Commercial Patients	0.04	<.001

Using the results of the LMMs, we decomposed the variability in final adjusted risk scores into variability over time within subject, between subjects from the same facility, and between practices. To examine the amount of variation at each level, that is, to compare variability over time within subject, between subjects from the same facility, and between practices, we ran the unconditional means model and estimated variance components. From the estimates, 31 percent of total variability in final adjusted risk score (outcome) was due to differences over time for each subject, 66 percent of total variability is due to differences between subjects in the same practice, and only 3 percent of total variability is due to differences between practices. Also, we examined quadratic and cubic time predictors but they were not significant. Therefore, we only considered time as a linear predictor at the subject‐level.

3.3. Validation

To help validate that risk reduction results could reasonably be explained by intervention‐related therapeutic changes, we established that over 50 500 postintervention patients experienced a reduction of systolic blood pressure of at least 5 mm Hg and over 39 200 had a systolic reduction of at least 10 mm Hg. Average systolic blood pressure dropped from 135 to 128 mm Hg. Regarding aspirin use, we observed 11 384 new prescriptions. We also observed 20 376 new statin prescriptions. The HHN protocol did not require repeat LDL cholesterols over time. In the subgroup (N = 36 322) that happened to have both baseline and intervention LDL‐C values, the average LDL‐C dropped from 105.3 to 100.6 mg/dL among those eligible for new statin prescriptions. Among 8746 patients who had LDL‐C values at both time points and indeed received new statin prescriptions, these values were 111.2 mg/dL and 103.4 mg/dL, respectively.

4. DISCUSSION

The adoption of important evidence into clinical practice has often been slow, even when interventions demonstrate improved survival in multiple trials. ¹⁵ , ¹⁶ We selected small and rural primary care practices because CVD rates are higher in rural regions ² , ³ and small practices typically have fewer resources to implement new evidence, apply population management techniques, and redesign workflows. ¹⁷ , ¹⁸ , ¹⁹ Our study engaged these practices with a practice support intervention combining PF with population‐based, informatics tools. This strategy was designed not only with specific intent to drive CVD risk improvement but also generally, to accelerate implementation of new evidence in primary care.

In comparison with past efforts using single interventions to achieve limited improvements, our intervention combined both onsite practice facilitation and informatics support. With our one‐year intervention, we saw a substantial reduction in ASCVD risk score for patients identified as high risk (ASCVD risk ≥ 10 percent) through the implementation of two newer evidence‐based measures (risk‐based statin and aspirin prescribing) and by attaining better adherence to standard blood pressure management parameters. In studies examining informatics support alone, the effect on CVD risk reduction through primary prevention has been limited. Karmali et al recently published a Cochrane Review on the use of risk scoring to improve primary prevention of CVD. ²⁰ Incorporating 41 randomized controlled studies, they demonstrated no statistically significant effect of clinically integrating risk stratification on prevention of CVD events though there were slight increases in prescription of lipid lowering and blood pressure medications. In a similar review published in 2019, Groenhof et al assessed the effects of clinical decision support tools and saw no effect on cardiovascular risk factors or treatment target attainment. ²¹ Peiris et al performed a cluster, randomized trial in 60 Australian primary care practices that utilized a software program integrated into the EHR that provided point of care, CVD risk scores coupled with clinical recommendations. Despite the availability of these point of care tools for primary prevention targeting 38 725 patients at high risk, the process of risk factor measurement improved but prescription of risk‐reducing medications did not compared with the control group. ²²

Practice facilitation is a multimodal approach incorporating a range of intervention components to address the challenges of implementing evidence‐based guidelines in primary care settings. The degree of informatics support available to practice facilitators is often variable. In a meta‐analysis assessing the impact of PF on adoption of evidence‐based guidelines, Baskerville and colleagues note the effect as “moderately robust” in improving prevention services. ²³ Also, PF has previously been shown to enhance care for diabetes ²⁴ , ²⁵ and asthma. ²⁶ Wang, in a more recent review of PF on chronic disease management, showed generally positive results on cardiovascular measures such as blood pressure control and cholesterol treatment in studies involving fewer than 50 practices. ²⁷ However, others have reported significant barriers to CVD risk reduction. ²⁸ In a recent pragmatic facilitation trial conducted in Canada focusing on improving eight facets of cardiovascular care in 84 primary care practices, the intervention failed to improve adherence to best practice guidelines. ²⁹ This latter trial did not use risk stratification and it was unclear whether the practice facilitators had access to run charts or other data tools to help practices iteratively improve. In our results, utilizing practice facilitation with access to informatics tools, the 4.0 percent absolute 10‐year ASCVD risk reduction would translate to 5800 events over 10 years if the new treatments were maintained over time. This benefit is much more substantial then past CVD risk interventions using practice facilitation or informatics tools alone.

Our findings suggest that PF with informatics support can effectively promote implementation of evidence‐based guidelines to reduce CVD risk among patients managed in small, rural practices and speed up diffusion of newer evidence as incorporated in the 2013 AHA‐ACC cholesterol and 2016 USPSTF aspirin guidelines. ⁴ , ⁶ While facilitation helped practices, leverage key changes including data base implementation, team‐based protocols, EHR templates, decision support, and population management techniques, the dashboard provided risk stratification, run charts, and reports to overcome barriers of their nascent EHRs. ⁵ By providing automated risk‐stratification tools not available via the EHR, providers became aware of risk‐based recommendations, and by meeting with practice facilitators, were able to discuss options for using these tools to engage patients at high risk and rapidly reduce risk within parameters of their practice. The resultant increase in statin use, over 20 000 new prescriptions as described in Section 3.3, is noteworthy as there are compelling data that statins reduce the risk for CVD among those with and without previous vascular disease, ³⁰ , ³¹ , ³² that greater intensity of statin therapy yields greater reduction in risk for vascular events, ³³ , ³⁴ and that low adherence to statins is associated with greater risk of dying. ³⁵ As noted by Baskerville et al, tailored approaches similar to HHN are associated with more improvement than one size fits all QI approaches. ²³

Our sensitivity analysis showed that aspirin use had a minimal effect in the overall risk reduction attributable to our intervention. Although evidence for aspirin use in primary prevention is less compelling and the ASCVD risk reduction effect is less pronounced than that of statins, ⁹ the USPSTF recommended low‐dose aspirin to reduce the CVD risk among patients at increased risk for this disease. ⁶ , ³⁶ The recent completion of three randomized controlled trials addressing the use of low‐dose aspirin for primary prevention in older patients ³⁷ , ³⁸ , ³⁹ has led the AHA and others to recommend a more nuanced and parsimonious approach to low‐dose aspirin use for primary prevention. ⁴⁰ , ⁴¹ Our algorithm in HHN emphasized aspirin use for primary prevention only for patients 40‐59 years of age with an ASCVD risk score ≥10 percent. These parameters are consistent with these latest approaches. Therefore, the risk tools and population strategies used in HHN remain relevant in the aspirin realm though the minimal benefit demonstrated through our sensitivity analysis suggests that aspirin for primary CVD prevention provides only marginal impact relative to the whole HHN intervention.

Similar to statins, there is robust evidence that lowering BP with medication is efficacious ⁴² yet underused. ⁴³ Even though evidence for BP lowering is older and better established, the availability of risk scores coupled with clear demarcation of individual treatment opportunities on the dashboard likely added more urgency to blood pressure treatment intensification for the high‐risk group.

Strengths of our study include the stepped‐wedge design that provided a control group to assure that the intervention was effective apart from secular trends, the pragmatic nature of the trial, the inclusion of over 200 practices, and the ability to discern a measurable effect in a population of over 140 000 patients. In addition, we enrolled practices in rural areas, where risk for CVD is very high and focused on patients at high risk who gain the largest absolute benefit from risk reduction efforts. Importantly, the intervention yielded CVD risk reduction that was consistent among all subgroups and was particularly robust among Black patients. The findings that Black race, male gender, older individuals, Medicaid coverage, lower income patients, and clinics not associated with hospitals demonstrated significant risk reductions adds to the case for generalizability.

Some study limitations need to be considered. Forty‐seven practice sites withdrew prior to starting the study (verbally agreed but did not sign MOU) and twenty‐six participated but did not provide data to the HHN registry. These practices were similar in size, rurality, and payer mix to those that participated fully but may have had intrinsic structural or organizational factors that would have systematically blunted their progress. Also of note, we focused on patients with ASCVD 10‐year risk scores that were 10 percent or higher because this group was eligible for medication for primary prevention. Because of the high starting point, some of the improvement seen might have been simply due to regression to the mean; the fact that the stepped‐wedge control would regress similarly argues against this reasoning. In addition, LDL‐C measurements were limited to a subgroup of patients who happened to have repeat measures ordered by their clinician and not by specific study protocol. Therefore, we could only confirm LDL‐C improvements as a concrete measure of adherence in a small proportion of patients prescribed statins. Finally, we did not use tobacco cessation in risk score reduction calculations because we only had data on smoking cessation counseling—not actual cessation. Therefore, we might have underestimated the intervention effect.

In conclusion, we observed a substantial reduction in 10‐year ASCVD risk score from baseline to follow‐up. This reduction was large, clinically significant, and unlikely due to secular trend or other factors that may have changed practice at study sites. We believe the observed translation of new evidence into practice, in particular, guidelines promoting formal risk stratification to trigger prescribing was largely attributable to combining PF with digital support. The policy implications are that adaptation of this approach could lead to gains in dissemination and implementation of other new evidence in the primary care environment, especially in rural and other under‐resourced practices. Building a shared services structure using PF and informatics tools to regularly support these practices is likely to overcome barriers to implementation that are traditionally observed in this environment. A formal cost‐benefit analysis is beyond the scope of this report. However, given the 4 percent attributable absolute risk reduction to the intervention among over 146 000 patients, it suggests 5800 cardiovascular events avoided over 10 years. Using a conservative estimate of $11 000 per hospitalization, the estimated cost savings would be $63.8 million compared with the $13.5 million cost of the salary and benefits of 18 practice facilitators over the same period.

Supporting information

Author Matrix

Cykert S, Keyserling TC, Pignone M, et al. A controlled trial of dissemination and implementation of a cardiovascular risk reduction strategy in small primary care practices. Health Serv Res 2020;55:944–953. 10.1111/1475-6773.13571