Testing interventions to reduce clinical inertia in the treatment of hypertension: rationale and design of a pragmatic randomized controlled trial

Abstract

Background:

Clinical inertia, or failure to intensify treatment when indicated, leads to suboptimal blood pressure control. Interventions to overcome inertia and increase antihypertensive prescribing have been modestly successful in part because their effectiveness varies based on characteristics of the provider, the patient, or the provider-patient interaction. Understanding for whom each intervention is most effective could help target interventions and thus increase their impact.

Methods:

This three-arm, randomized trial tests the effectiveness of two interventions to reduce clinical inertia in hypertension prescribing compared to usual care. Forty five primary care providers (PCPs) caring for patients with hypertension in need of treatment intensification completed baseline surveys that assessed behavioral traits and were randomized to one of three arms: 1) Pharmacist e-consult, in which a clinical pharmacist provided patient-specific recommendations for hypertension medication management to PCPs in advance of upcoming visits, 2) Social norming dashboards that displayed PCP’s hypertension control rates compared to those of their peers, or 3) Usual care (no intervention). The primary outcome was the rate of intensification of hypertension treatment. We will compare this outcome between study arms and then evaluate the association between characteristics of providers, patients, their clinical interactions and intervention responsiveness.

Results:

Forty-five primary care providers were enrolled and randomized: 16 providers and 173 patients in the social norming dashboards arm, 15 providers and 143 patients in the pharmacist e-consult arm, and 14 providers and 150 patients in the usual care arm. On average, the mean patient age was 64 years, 47% were female, and 73% were white. Baseline demographic and clinical characteristics of patients were similar across arms, with the exception of more Hispanic patients in the usual care arm and fewest in the pharmacist e-consult arm.

Conclusions:

This study can help identify interventions to reduce inertia in hypertension care and potentially identify the characteristics of patients, providers, or patient-provider interactions to understand for whom each intervention would be most beneficial.

INTRODUCTION

Hypertension is highly prevalent in the United States (US) and is a leading cause of myocardial infarction, stroke, and death.^1–3 Despite its importance and the widespread availability of medication treatment, less than half of patients with hypertension in the US have blood pressures that are adequately controlled.⁴ The failure of providers to initiate or intensify treatment regimens when clinically indicated, referred to as clinical inertia,^5–8 is a large contributor to poor blood pressure control.^9–11

Many interventions have been tested to help providers overcome clinical inertia and intensify antihypertensives when indicated. However, even the most effective of these interventions have only modestly improved blood pressure control. For example, in one pooled analysis of randomized controlled trials, only 55% of patients assigned to interventions (such as reminders, ambulatory blood pressure monitoring, or educational interventions) achieved blood pressure control compared with 45% of patients randomized to control groups.⁶

One hypothesis for these modest results is that the effectiveness of different interventions varies by characteristics of the provider, the patient, or the provider-patient interaction.¹² For example, response to social norming interventions may vary based on the clinical role (for example, physician, nurse, or other healthcare worker),¹³ care setting (i.e home, outpatient, or inpatient),¹³ and provider self-perception of where their performance lies relative to their peers.¹⁴ Additionally, there may also be variation in intervention effectiveness based on patient characteristics. For example, pharmacist outreach to a busy physician, who is experiencing burnout, with recommendations on how to intensify a patient’s regimen might be very helpful before a visit with a patient with multiple comorbidities and antihypertensive allergies, but it may be less helpful before a visit with a patient who has several reasonable antihypertensive choices.

Knowing which intervention is most likely to work for an individual physician or clinical encounter could allow for targeted intervention delivery and an increase in overall impact. This is especially true if responsiveness to each intervention could be predicted using routinely available characteristics and interventions could be delivered accordingly.

To explore this idea, we designed and launched a 3-arm, randomized controlled trial that tested the ability of two provider-facing behavior change interventions, an electronic consult from a pharmacist utilizing concepts from academic detailing (pharmacist e-consult) and a social norming dashboard intervention, to reduce prescribing inertia for patients with poorly controlled hypertension. These interventions were selected as they are likely to be effective in this context, and variability in their effectiveness may be in part due to characteristics of providers or patients.^13–15 At trial enrollment, primary care providers completed questionnaires with basic demographic and practice characteristics as well as validated measures of several personality traits, and patient demographic and clinical characteristics were extracted from the electronic health record (EHR). After trial completion, we will conduct exploratory analyses to determine if intervention effectiveness varies as a function of physician traits, patient characteristics, and the nature of the patient-physician interaction.

METHODS

Overall study design

This is a pragmatic, randomized controlled trial testing two interventions targeting clinical inertia in hypertension compared to usual care, followed by analyses to identify factors associated with intervention responsiveness. Forty-five primary care providers (PCPs) caring for 466 patients with hypertension in need of treatment intensification were enrolled and completed baseline validated surveys to measure behavioral characteristics. Physicians were randomized in a 1:1:1 ratio to one of three arms: 1) Pharmacist e-consult, 2) Social norming dashboards, or 3) usual care (no intervention). Enrollment of PCPs began in October 2021 and continued through January 2022. Intervention delivery to PCPs and accrual of patients started in October 2021 and was completed in August 2022. Collection of follow-up data was completed April 2023, and analyses are underway.

This research is supported by the National Institute on Aging of the National Institutes of Health under Award Number P30AG064199 to BWH (Choudhry PI). Dr. Lauffenburger (K01HL141538) and Dr. Haff (K23HL161480) are supported in part by career development grants from the NIH. This study was approved by the institutional review board at Mass General Brigham and is registered on clinicaltrials.gov (NCT04603560).

Study setting and participants

This study was conducted within primary care practices at Massachusetts General Hospital (MGH). MGH has more than 200 PCPs providing care for over 200,000 patients in the Boston area. MGH has several routine system-wide population health and quality initiatives that all PCPs receive, and which continued through the course of the trial. These included quarterly dashboards sent to PCPs by email containing their performance metrics for several chronic disease and preventive health measures, including hypertension, as well as population health coordinators conducting outreach directly to patients to engage them in chronic disease care items that are due. These activities are coordinated centrally across all practices and were delivered equally across study arms, including the usual care arm.

Patient eligibility

We used data from the electronic health record (EHR) to identify patients aged 18–79 years with upcoming visit with their PCP during the intervention period (target visit) at which there was an opportunity for antihypertensive treatment intensification. A patient was considered to have an intensification opportunity if: (1) their blood pressure was ≥140/90 for patients 18–59 years or ≥150/90 for patients 60–79 years of age at their most recent outpatient visit (the “index visit”) with any provider before their target visit date and (2) their hypertension treatment regimen had not been intensified since that time (Table 1). Outpatient visits included in-office and virtual visits that had vitals recorded in the EHR the same day. If there were multiple blood pressures recorded on the same day, we selected the last measurement. Because the intervention was deployed several days prior to upcoming visits, the blood pressure at the target visit could not be used to determine eligibility.

Table 1:

Criteria for inclusion and exclusion of providers

Inclusion Criteria	Primary Care Providers caring for at least 2 patients:  1. Aged 18–79  2. Whose most recent outpatient blood pressure is above goal  3. Who did not have treatment intensification since that time  4. Who are not excluded from the hypertension registry, pregnant, or receiving hospice care
Exclusion Criteria	Primary care panel with fewer than 100 patients Practice less than one session per week

Blood pressure thresholds used for inclusion in the study were higher than contemporary blood pressure treatment goals, which recommend treating to less than 130/80 for most patients.^16,17 Blood pressure values used for study inclusion were measurements at outpatient visits, which in our system are typically blood pressures checked by a medical assistant using an automated blood pressure cuff and performed prior to the physician entering the room. If the initial reading is high, the physician will often recheck using either an automated or manual cuff later in the visit and enter the second check in the vitals field in the EHR. Unattended automated office blood pressure cuffs are not routinely used. Additionally, though in our system patients often check blood pressures using automated cuffs at home, these values are typically shared with providers in free-text patient portal messages, message attachments, or pen-and-paper home blood pressure logs brought to visits. Entry of this information into structured EHR fields is highly variable, which limited our ability to utilize home blood pressure measurements for study inclusion. Given these limitations to blood pressures recorded in the EHR, we decided to use blood pressure thresholds higher than typical treatment targets for study inclusion to specifically identify patients for whom intensification of medication treatment would very likely be indicated.

Treatment intensification was ascertained from prescription data in the EHR and was defined as an increase in medication dose or the addition of a new medication since the most recent outpatient visit with an elevated BP. Medication name and total daily dose were extracted from the EHR. If instructions for medication taking were listed in free text rather than in structured EHR fields, two study authors reviewed the free-text instructions and manually calculated the total daily dose. Any disagreement was adjudicated by the first author. When total daily dose could not be determined due to absent or inconsistent instructions, patients were excluded from the study due to an inability to ascertain if intensification had occurred. Once free-text instructions were manually mapped to total daily doses in this way, this mapping was subsequently used for identification of intensification without requiring additional manual review. If a patient had an increase in dose or a new antihypertensive medication added after the index visit but before their upcoming (target) visit, their treatment was considered to have been intensified, and they were excluded from the cohort.

We also excluded patients who, based upon EHR data, were pregnant, receiving hospice care, or who had been excluded by a provider from the hospital’s hypertension registry for having a clinical reason for non-standard blood pressure control goals (e.g., late-stage cancer). Though results from the SPRINT¹⁸ and HYVET¹⁹ trials suggest that intensive blood pressure control for patients aged 80 and older is safe, in this age group clinical scenarios often arise which may warrant different blood pressure goals. Using only structured EHR data we were not able to ascertain these types of nuanced decisions around blood pressure goals for older patients, and so we chose to exclude patients aged 80 and older from this study.

Provider Eligibility

Physicians with at least 2 patients meeting these criteria at baseline were potentially eligible for the study. To ensure that PCPs had sufficient practice volume to trigger receipt of the study interventions, we excluded PCPs with fewer than 100 patients on their primary care panel or who practiced less than one half-day session per week. Advanced practice providers working in primary care did not routinely carry independent patient panels at the time of the study and so were not eligible for inclusion.

Enrollment and Randomization

Data on the number of eligible patients with visits in the subsequent three months for all study eligible PCPs were obtained. The 90 PCPs with the most eligible patients were selected, and from that list PCPs were randomly selected for invitation to enroll in the study on a rolling basis until the target sample size of 45 PCPs was reached. Eligible PCPs were invited to participate by email, informed about the study, and asked to agree to participate using an electronic form. They then completed a set of surveys that included self-reported age, gender, race/ethnicity, years in practice, and number of sessions of practice per week, as well as several behavioral scales designed to capture personality traits and behavioral tendencies that, based on our hypotheses and prior research,^13–15,20 might elucidate potential mechanisms behind differential responsiveness to each intervention. These included validated measures of risk aversion (Physicians’ Reactions to Uncertainty Scale), ²¹ long and short term orientation (Individual Cultural Values Scale),²² conscientiousness (Ten-Item Personality Inventory),²³ need for cognition (Need for Cognition Scale),²⁴ resistance to change (Resistance to Change Scale),²⁵ susceptibility to social norms (Social-Norm Espousal Scale),²⁶ burnout (Professional Fulfilment Index),²⁷ and habit automaticity (Self-Report Behavioural Automaticity Index).²⁸ Survey results will be used in analyses but were not used to adjust interventions and providers were not informed of their survey results.

We randomized at the physician level to avoid contamination from study interventions across arms. We used stratified permuted block randomization to help achieve balanced numbers of providers and patients in each arm.²⁹ Strata were based on the number of half-day sessions physicians worked per week (small [1–2 sessions], medium [3–4 sessions], or large [5 or more sessions]). Then, within each stratum, a random number generator was used to create a randomly ordered allocation within blocks of 3. Physicians were then allocated within each block and each stratum so that there was an equal probability of being assigned to each of the study arms.

Intervention and Usual Care Arms

After randomization, PCPs received the interventions or usual care for six months. In the usual care arm, there was no additional contact with the PCPs.

Pharmacist e-consult

In this arm, PCPs received individual feedback and recommendations generated by a clinical pharmacist delivered as a message through the EHR. E-consults are an increasingly widely used service, typically to provide expanded access to specialty care.³⁰ In this intervention, e-consults were triggered by upcoming visits with eligible patients (not solicited by providers), and thus served as a reminder to consider antihypertensive intensification in the visit. They applied concepts from academic detailing (i.e., synthesized, up-to-date information, packaged clearly and concisely, and delivered from a trusted source) to increase effectiveness. Academic detailing is a process by which trained clinicians, typically physicians, pharmacists or nurses, provide evidence-based outreach education, with the goal of improving guideline-concordant prescribing.^31,32 Evidence supports the ability of academic detailing to change prescribing behavior,³³ including in antihypertensive prescribing.^34–38 However, academic detailing can be resource intensive because it requires an academic detailer to have 1:1 contact with each provider, usually in person. In addition, it is not specific to a provider’s patients. Less intensive versions of academic detailing, electronically delivered by a pharmacist, are a potentially impactful alternative, can be specific to patients currently under the provider’s care, and may help overcome some of the barriers to traditional academic detailing.^15,39

Eligible patients with an upcoming (target) visit with their PCP were identified using EHR data, as above. A clinical pharmacist reviewed each patient’s chart and generated a personalized recommendation for how to modify the specific patient’s antihypertensive regimen. Pharmacists followed a hypertension collaborative drug therapy management algorithm that was developed by the hospital based on the 2017 ACC/AHA hypertension management guidelines and were empowered to modify recommendations based on their clinical review and assessment of the patient’s medical record.¹⁷ For example, the clinical pharmacist might recommend adding a medication based on the patient’s comorbid conditions and could suggest a starting dose and timeframe for dose escalation (example message in Appendix A). These recommendations were then sent to PCPs as InBasket messages 1–3 days in advance of an upcoming visit. Messages were also stored in a secure study database (REDcap). PCPs could reply to the messages to reach the pharmacist with any questions or comments. If this occurred, the pharmacist replied and documented the additional contact in REDcap.

Social Norming Dashboards

Social norming is a construct that aims to change the behavior of an individual by presenting them with information about the values, behavior, or beliefs of a group or another person.⁴⁰ This is often done by graphically comparing an individual person’s performance on a metric to that of their peers. It has been demonstrated to be effective across multiple contexts including prescribing,^13,41 ordering of medical tests,¹³ and primary care quality metrics,⁴² although its effectiveness may vary based on provider characteristics.^13,14

In this arm, rates of blood pressure control across all patients with hypertension cared for by each PCP were obtained from a system-wide performance dashboard, and reports were generated that displayed the physician’s hypertension control rates compared to their peers. Targets chosen for comparison influence the effectiveness of social norming interventions.^13,43 To increase effectiveness of this intervention, PCPs were compared to targets that they could reasonably attain, and so a different comparison was used based on the PCP’s current rate of hypertension control. If the PCP’s rates were below the practice average, the practice average was used for comparison. If the PCP’s rate was above the practice average, the rate of the top performer in the practice was used. If they were the top performer in the practice, the hospital quality target was presented. PCPs were sent secure emails once per week that contained a list of their patients with upcoming visits the following week who had poorly controlled blood pressure and evidence of inertia in their hypertension care, as defined above (example dashboard in Appendix B). Because overall rates of blood pressure control change slowly, the social norming report was included with this email once every 3 weeks.

Outcomes

The primary outcome was whether physicians intensified treatment at the target visit. Intensification was defined as adding a new antihypertensive medication or increasing the total daily dose of an existing one. This was measured by identifying all prescriptions for antihypertensives in the EHR on the day of the target visit and comparing the medication name and total daily dose to antihypertensives prescribed prior to the target visit date, following the same procedure used to identify intensification during the eligibility assessment, described above.

The secondary outcome was the change in systolic blood pressure over 6 months of follow-up in each intervention arm compared to usual care. The baseline blood pressure was the systolic blood pressure at the index visit that qualified the patient for inclusion in the study. The follow-up blood pressure was the blood pressure available in the EHR closest to 6 months after the target visit and within 3 to 9 months after the target visit. Blood pressures were identified from outpatient visits using the EHR.

Statistical considerations

Power and sample size

Sample size was estimated prior to study start. Because we specifically included patients with evidence of clinical inertia in their hypertension care, we expected the usual care arm rate of medication intensification at the target office visit to be approximately 10%. We also estimated that each physician would see 11 unique patients over the study period, for a total of 495 potential patient subjects. Assuming an intracluster correlation coefficient of 0.05, a Type I error rate of 5%, and 15 physicians per arm, this sample should provide more than 80% power to detect a 14 percentage-point difference in treatment intensification between each intervention group and usual care. Prior trials to reduce clinical inertia that utilized a similar outcome showed 15–20 percentage-point differences in intensification between intervention and usual care.^6,44

If the usual care arm intensification rate is higher than the 10% we anticipate, we will still be sufficiently powered for reasonable effect sizes. For example, if the rate of the primary outcome in the usual care arm is 15%, we will still have 80% power to detect an effect size of 16 percentage points or more.

Analytic plan

Baseline demographic and clinical characteristics of patients were determined from the EHR using structured fields for demographic, laboratory, and biometric data as well as ICD 10 codes to assess for comorbid conditions and calculate a Combined Comorbidity Score⁴⁵ and Frailty Index.⁴⁶ Provider behavioral characteristics were collected using validated surveys, as above, and scores were calculated and validated cutoffs were applied when available. Due to the study design, we were not able to directly measure behavioral characteristics of patients. Descriptive statistics were used to report the means and frequencies of baseline variables for eligible patients and physicians.

For the primary outcome, our primary analysis will include antihypertensive intensification at the first target visit for each eligible patient during the intervention period. We will evaluate the primary outcome using a logistic regression model with generalized estimating equations (GEE) to adjust for clustering of patients within providers; these models will also use fixed-effects to adjust for the block-randomized design. If there are strong predictors of the outcomes not balanced by block randomization, we will additionally adjust for these in the primary analyses. We will conduct analyses using intention-to-treat principles, and each intervention arm will be compared to usual care.

We will conduct several sensitivity analyses to evaluate the robustness of our results. Because patients who do not attend the planned target visit would be unlikely to have meaningful exposure to the intervention effects, we will conduct an as-treated analysis for patients who attended the target visit. Due to natural fluctuation in blood pressure visit to visit, we expect that some patients will have a normal blood pressure at the visit targeted by the intervention. To address this, we will also repeat this as-treated analysis restricting to patients whose blood pressure was >130/80 at the target visit. This last group who attended the visit and had evidence of poor BP control are those for whom intensification is most likely to be indicated. Because intensification may occur after the visit day, if for example a provider obtains labs before adding a new medication, we will also conduct a sensitivity analysis in which we widen the window to observe antihypertensive intensification to include the date of the target visit and the 14 days after. We will additionally compare the overall rate of intensification for each patient over follow-up across arms. For sensitivity analyses, we will again use GEE to account for clustering of patients within providers and the block randomized design.

The secondary outcome of this trial is the change in systolic blood pressure control in each arm over 6 months. Based on prior studies in a similar local healthcare system, we anticipate approximately 20% missingness rate for the follow-up blood pressure within 6 months. This is typically due to patients not having a follow-up visit where blood pressure was measured and entered into the EHR. We will use multiple imputation to handle missing data, specifically 20 imputations with Proc MI in SAS to impute any estimated values using fully conditional specification. It is possible that missingness will differ among patients who did and did not receive intensification, which could bias these results, and so we will evaluate for apparent differences in missingness between study arms and among patients who did and did not receive intensification. Analyses will then be conducted on each imputed dataset and combined using Rubin’s rules. ⁴⁸ We will then use GEE with an identity link function and normally-distributed errors to account for clustering of patients within providers and adjust for the block-randomized design. As in the primary analysis, we will adjust for provider and patient-level covariates if there are imbalances across arms. We will conduct analyses using intention-to-treat principles and each intervention arm will be compared to usual care. We will also perform a complete case analysis as a sensitivity analysis.

Qualitative Analyses

After completion of the study interventions, all PCPs who consented to the study were additionally invited to participate in follow-up qualitative interviews as part of an explanatory sequential study design.⁴⁹ Interviews were conducted by one author and asked about factors contributing to clinical inertia in hypertension care and feedback on study interventions. Interviews were transcribed, and coding is being conducted using the DeDoose software,⁵⁰ separately by two authors, with discussion to reconcile any areas of disagreement until a final coding scheme is reached. Analyses will be conducted using an immersion crystallization approach.^51,52 Findings will be used to explore reasons behind the trial results as well as to identify potential factors important to include in our exploratory predictive modeling approach.

Predictive Modeling Approach

We will use subgroup analyses as preparatory work for our predictive modeling. We will first use multivariable logistic regression to explore whether there are differential effects of each intervention based on characteristics of providers (for example practice volume and surveyed behavioral tendencies) or patients (such sex and race/ethnicity). For example, we will evaluate whether a provider’s level of risk aversion is associated with their likelihood of responding to the pharmacist e-consult or social norming.

We hypothesize that a combination of characteristics of providers, patients, and their clinical interactions will influence responsiveness to each intervention (Table 2). We will explore these potential interactions using random forest classification models. Random forest is a common machine learning technique that constructs and combines multiple decision trees to handle classification and regression tasks.⁵³ As a secondary approach, we will use boosted regression, a different machine learning method that is robust to multi-collinearity and overfitting and can be used to evaluate prediction models.⁵⁴ We will use the provider behavioral survey results to help interpret our findings and to explore potential behavioral mechanisms that might underly associations between provider and patient characteristics and intervention responsiveness. We expect these models to be helpful in indicating which characteristics of providers and patients are the most influential in predicting responsiveness to each intervention. If the sample size limits the number of variables we can simultaneously include in a model, we will focus on simpler models that include fewer variables (e.g., one pair at a time) and evaluate which characteristics provide higher increments of predictive performance.

Table 2:

Hypothesized influence of provider behavioral characteristics on intervention responsiveness and potential for differential response by patient characteristics.

Provider Characteristic	Definition	Hypothesized influence on intensification without intervention	Hypothesized influence on intervention responsiveness	Potential for differential response by patient characteristics
Risk averse	Higher Physicians’ Reactions to Uncertainty Scale	Less likely to intensify	More responsive to pharmacist e-consult	Particularly for patients at risk for adverse effects (higher frailty index)
Conscientious	Higher conscientiousness from Ten-Item Personality Inventory	More likely to intensify	Responsive to both interventions	Likely not differential
High need for cognition	Higher score on Need for Cognition Scale	Neutral	Responsive to both interventions	Particularly responsive for complex patients (higher comorbidity score)
Resistant to change	Higher score Resistance to Change Scale	Less likely to intensify	More responsive to pharmacist e-consult	Particularly responsive among patients already on a med with room for dose increase
Susceptible to social norms	Higher score Social-Norm Espousal Scale	Neutral	More responsive to social norming dashboards	Likely not differential
Experiencing burnout	Score classified as burnout on Professional Fulfilment Index	Less likely to intensify	More responsive to pharmacist e-consult	Particularly for complex patients (higher comorbidity score)

RESULTS

Forty-five primary care providers were enrolled, completed baseline questionnaires, and were randomized to study interventions or usual care. This resulted in 16 providers and 173 patients in the social norming dashboards arm, 15 providers and 143 patients in the pharmacist e-consult arm, and 14 patients and 150 patients in the usual care arm (Figure 1). Overall, the mean age of the patient sample was 64 years, 47% were female, and 73% were white. Other than Hispanic ethnicity, for which there were more Hispanic patients in the usual care arm and fewest in the pharmacist e-consult arm, baseline demographic and clinical characteristics of patients were well balanced across arms (Table 3).

Figure 1:

CONSORT diagram

*Anticipating a 50% acceptance rate, 90 PCPs with the highest number of eligible patients were invited to participate, of whom 45 ultimately agreed

Table 3:

Baseline demographic and clinical characteristics of patient sample

	Usual care (N=150)	Social Norming Dashboard (N=173)	Pharmacist E-consult (N=143)	p-value
Age, mean (SD)	64.11 (12.83)	64.35 (11.69)	64.94 (10.72)	0.83
Sex				0.14
Male	75 (50.0%)	102 (59.0%)	70 (49.0%)
Female	75 (50.0%)	71 (41.0%)	73 (51.0%)
Race				0.10
Asian	10 (6.7%)	6 (3.5%)	13 (9.1%)
Black or African American	15 (10.0%)	20 (11.6%)	17 (11.9%)
White	104 (69.3%)	132 (76.3%)	105 (73.4%)
Other/Unknown/Not reported	21 (14.0%)	15 (8.7%)	8 (5.6%)
Hispanic				0.003
No	119 (79.3%)	137 (79.2%)	129 (90.2%)
Yes	20 (13.3%)	13 (7.5%)	4 (2.8%)
Unknown	11 (7.3%)	23 (13.3%)	10 (7.0%)
Partnered	86 (57.3%)	90 (52.0%)	67 (46.9%)	0.20
Primary insurance				0.62
Medicare	54 (36.24%)	70 (40.46%)	55 (38.46%)
Medicaid	10 (6.71%)	16 (9.25%)	11 (7.69%)
Self	3 (2.01%)	4 (2.31%)	2 (1.4%)
Commercial	81 (54.36%)	81 (46.82%)	70 (48.95%)
Others	1 (0.67%)	2 (1.16%)	5 (3.5%)
Systolic blood pressure, mean (SD)	148.13 (10.89)	148.97 (11.06)	149.48 (9.02)	0.53
Diastolic Blood Pressure, mean (SD)	82.78 (8.89)	81.57 (9.34)	81.75 (8.82)	0.45
Body Mass Index, mean (SD) (N=414)	30.23 (5.41)	30.08 (6.52)	30.50 (6.64)	0.85
Comorbid conditions
CKD	19 (12.7%)	24 (13.9%)	15 (10.5%)	0.66
COPD	12 (8.0%)	12 (6.9%)	11 (7.7%)	0.93
Chronic ischemic heart disease	19 (12.7%)	28 (16.2%)	15 (10.5%)	0.32
Depression	42 (28.0%)	50 (28.9%)	37 (25.9%)	0.83
Diabetes	47 (31.3%)	50 (28.9%)	45 (31.5%)	0.85
Hyperlipidemia	104 (69.3%)	123 (71.1%)	96 (67.1%)	0.75
Obesity	45 (30.0%)	61 (35.3%)	41 (28.7%)	0.40
Osteoporosis	16 (10.7%)	15 (8.7%)	6 (4.2%)	0.11
Peripheral vascular disease	8 (5.3%)	12 (6.9%)	6 (4.2%)	0.57
Smoking	30 (20.0%)	50 (28.9%)	37 (25.9%)	0.18
Combined Comorbidity Score, mean (SD)	1.56 (2.65)	1.66 (2.96)	1.37 (2.72)	0.64
Frailty Index, mean (SD)	0.15 (0.04)	0.15 (0.04)	0.14 (0.04)	0.17
Unique Drugs, mean (SD)	13.65 (13.60)	12.09 (12.72)	13.66 (15.01)	0.50
Taking at least one antihypertensive, N (%)	117 (78.0%)	145 (83.8%)	124 (86.7%)	0.13
Number of office visits, mean (SD)	5.79 (5.86)	4.90 (5.89)	5.17 (5.49)	0.38
Number of hospitalizations, mean (SD)	0.11 (0.39)	0.10 (0.36)	0.08 (0.39)	0.70
Number of ER visit, mean (SD)	0.44 (1.16)	0.41 (0.97)	0.31 (0.87)	0.54
HbA1C, mean (SD)	6.59 (1.57)	6.45 (1.48)	6.44 (1.28)	0.75
EGFR, mean (SD)	77.71 (20.14)	77.32 (19.48)	74.93 (17.39)	0.48

Note: Number of office visits, hospitalizations, and ER visits were measured in the 365 days prior to and including the patient’s study enrollment date. SD – standard deviation, CKD – Chronic Kidney Disease, COPD – Chronic Obstructive Pulmonary Disease, ER – Emergency Room, HbA1c – Hemoglobin A1c, EGFR – estimated glomerular filtration rate.

DISCUSSION

This study directly tests interventions to reduce clinical inertia in hypertension care and assesses if characteristics of providers, patients, or their interactions are associated with intervention responsiveness and could be used to target interventions to where they are most likely to be impactful.

This study is innovative in several ways. First, although there is evidence of at least modest success of both pharmacist consult and social norming interventions for physician prescribing, evidence for the application of these strategies specifically to clinical inertia in hypertension is limited. Second, the ability to target interventions to specific patient-provider interactions has not yet been explored and could be a promising way to increase intervention effectiveness. Third, we measured behavioral characteristics of the providers, which may help elucidate behavioral principles underlying inertia behavior and intervention response.

This study has several important limitations. First, this study has a small sample size and was conducted at a single academic medical center; however, if successful, the results could be used to inform larger-scale, more generalizable work. Second, inclusion criteria and study interventions were based on in-office blood pressure measurements. Home blood pressure measurements would be preferable, but in our system, as with many others, these are not consistently available in structured EHR fields. The use of office measurements could reduce PCPs confidence in the feedback provided by the study interventions. However, existing hospital quality metrics, on which providers are evaluated, are also based on EHR-entered blood pressures, so our approach is similar to what providers experience in routine practice. Third, blood pressure thresholds used to determine eligibility were higher than contemporary treatment goals. Though our intention with this choice was to identify patients most in need of medication intensification, this process could also have selected for a cohort of patients whose blood pressures are particularly difficult to control or who have some degree of “white coat” hypertension. Fourth, although we utilized two interventions targeting different behavioral mechanisms and measured a broad array of patient and provider characteristics, we cannot fully capture all underlying behavioral mechanisms that could contribute to inertia behavior. Fifth, our sample size may limit our ability to generate a model that strongly predicts intervention responsiveness. Nonetheless, we expect these models to be helpful in indicating which characteristics of providers and patients are the most influential, and these could be further investigated in subsequent studies. Lastly, due to nuances in blood pressure treatment of older adults we excluded patients aged 80 and older which limits the generalizability of our findings.

Overall, given the modest success of interventions to reduce inertia in prescribing to date and the importance of overcoming inertia to widely achieve good blood pressure control, it is important to test interventions to improve antihypertensive prescribing and to understand how to target future interventions to where they will be most effective.

Appendix:

Appendix A: Example pharmacist e-consult message sent through the EHR

Dear Dr. [physician name],

I am an MGPO pharmacist collaborating with MGH Primary Care. As part of the hypertension study you are enrolled in, I am reviewing the charts of patients with recent blood pressures above goal. You have an upcoming visit with [patient name (patient MRN)]. Here are possible treatment options to consider:

Blood pressure goal: <140/90

Blood pressure status: Most blood pressures in our system in the last 6 months are above goal.

Current treatments: In review of the chart, it appears that this patient is on lisinopril 40mg daily. This medication has been well tolerated, and he had not tried other antihypertensives in the past.

Recommendations: Based on JNC 8 and ACC 2017 guidelines, two medications at moderate or low doses may be more effective than a single agent at maximum dose. Therefore, I would recommend:

1. Initiating a diuretic in addition to the lisinopril. Based on the patient’s renal function, HCTZ at 25 mg daily would be a good start.

2. Checking a BMP in one week

3. Follow-up blood pressures over the course of 2–3 weeks, at home if possible.

This patient was identified in the EHR based on the following criteria: (1) age 18–79, (2) recent blood pressures above goal (3) did not have recent hypertension treatment intensification (i.e. dose increase, new medication, or medication change).

Please note that, because this is an e-Consult, I have not had the opportunity to speak with or examine the patient. The recommendations were based upon the JNC 8 and ACC/AHA 2017 hypertension guidelines, but the above must be interpreted after taking into account any considerations that are not available to me from the electronic record. The ongoing management of this clinical problem is the responsibility of the patient’s care team. Please alert me if there are further questions.

Appendix B: Example social norming dashboard sent by email

Dear Colleague,

Our records indicate that the following patients with recent elevated blood pressures have visits with you next week. Some of these patients may need intensification of their hypertension treatment.

NAME (MRN) – Visit Date

NAME (MRN) – Visit Date

NAME (MRN) – Visit Date

Below is a graph of your hypertension control rates compared to your peers.

This dashboard was generated using data from routine documentation in the electronic health record, and the hypertension control rates are based on MGH quality metrics.