Rationale and design of a pragmatic trial aimed at improving treatment of hyperlipidemia in outpatients with very high risk atherosclerotic cardiovascular disease: A pragmatic trial of messaging to providers about treatment of hyperlipidemia (PROMPT-LIPID)

Abstract

Background

Despite guideline recommendations to optimize low-density lipoprotein cholesterol (LDL-C) reduction with intensification of lipid-lowering therapy (LLT) in patients with atherosclerotic cardiovascular disease (ASCVD), few of these patients achieve LDL-C < 70 mg/dL in practice.

Purpose

We developed a real-time, targeted electronic health record (EHR) alert with embedded ordering capability to promote intensification of evidence based LLT in outpatients with very high risk ASCVD.

Methods

We designed a pragmatic, multicenter, single-blind, cluster randomized trial to test the effectiveness of an EHR-based LLT intensification alert. The study will enroll about 100 providers who will be randomized to either receive the alert or undergo usual care for outpatients with high risk ASCVD with LDL-C > 70 mg/dL. Total enrollment will include 2,500 patients. The primary outcome will be the proportion of patients with LLT intensification at 90 days. Secondary outcomes include achieved LDL-C at 6 months and the proportion of patients with LDL-C < 70 mg/dL or < 55 mg/dL at 6 months.

Results

Enrollment of 1,250 patients (50% of goal) was reached within 47 days (50% women, mean age 72, median LDL-C 91). At baseline, 71%, 9%, and 3% were on statins, ezetimibe, or proprotein convertase subtilisin/kexin type 9 inhibitors, respectively.

Conclusions

PRagmatic Trial of Messaging to Providers about Treatment of HyperLIPIDemia has rapidly reached 50% enrollment of patients with very high risk ASCVD, demonstrating low baseline LLT utilization. This pragmatic, EHR-based trial will determine the effectiveness of a real-time, targeted EHR alert with embedded ordering capability to promote LLT intensification. Findings from this low-cost, widely scalable intervention to improve LDL-C may have important public health implications.

Background

Atherosclerotic cardiovascular disease (ASCVD) including coronary heart disease, stroke, and peripheral artery disease is the leading cause of death in men and women in the United States and most developed countries worldwide.¹ Elevated low-density lipoprotein cholesterol (LDL-C) is a causal risk factor for ASCVD, and reductions in LDL-C with statins, ezetimibe, and proprotein convertase subtilisin/kexin type 9 inhibitors (PCSK9i) have been associated with reductions in major adverse cardiovascular events (MACE). Across these interventions each 1-mmol/L (38.7 mg/dL) reduction in LDL-C results in a 23% relative reduction in MACE. ^2–7 Among patients with very high risk ASCVD, defined as having had multiple prior ASCVD events or one ASCVD event and multiple comorbid conditions such as diabetes, hypertension, or age > 65 years, recent professional society guidelines endorse use of high-intensity statin therapy as the cornerstone of lipid lowering therapy (LLT) (class I recommendation) with subsequent addition of non-statin therapies including ezetimibe and PCKS9i if LDL-C remains ≥ 70 mg/dL (class IIa recommendations). ⁸ Stepwise LLT intensification has the potential to reduce LDL-C to < 70 mg/dL in 99% of eligible adults and reduce ASCVD events by up to 21%. ^{9, 10}

Implementation of guideline recommended LLT has been suboptimal across health care systems. Data from registries in the United States, Canada, and Europe show that < 30% of patients with ASCVD had been initiated on high intensity statins on index presentation and < 35% of eligible patients attain LDL-C < 70 mg/dL. ^11–18 Even among patients started on statins, evidence suggests that LLT is often not adhered to or intensified, and is often prematurely discontinued or de-intensified. ^{12, 19–26} Since the benefits of lipid-lowering treatment attained in clinical trials and recommended in professional society guidelines can only be fully realized if all eligible patients are treated appropriately, there is a strong need to identify and implement strategies that enable improved adoption of these recommendations.

There are many potential contributors to the gap between evidence and practice including knowledge, clinical inertia, patient choice, and cost, among others. The electronic health record (EHR) is now in use in over 90% of physician offices and can enable real-time alerting of important clinical conditions and recommended practices using patient-specific factors. ²⁷ The EHR thus may be a readily scalable, low-resource tool to influence prescribing patterns via provider-focused EHR interventions. The PRagmatic Trial of Messaging to Providers about Treatment of HyperLIPIDemia (PROMPT-LIPID) will test the hypothesis that real-time, tailored EHR alerts with embedded ordering of guideline-recommended LLT will increase rates of LLT intensification for patients with high risk ASCVD when compared to usual care.

The PROMPT-LIPID clinical trial

The full study protocol is accessible at www.theprompttrials.org (NCT04394715). The study design is illustrated in Figure 1. PROMPT-LIPID is a pragmatic, randomized, open label intervention trial in outpatient clinics that will test the effectiveness of a real-time, tailored, targeted EHR-based alert system on improving the proportion of patients with very high risk ASCVD who have intensification of LLT 90 days post randomization compared to usual care. Subjects were automatically enrolled from outpatient internal medicine and cardiology clinics within the Yale New Haven Health System if they were seen by an enrolled provider and met the inclusion criteria: age ≥ 18, very high risk ASCVD (based on the 2018 Multisociety Guideline on the Management of Blood Cholesterol) ⁸ and LDL-C > 70 mg/dL. Very high risk ASCVD criteria were identified using validated International Classification of Diseases 10 (ICD-10) codes in the patient’s EHR Problem List or past medical history. Additional inclusion and exclusion criteria are described in Table I. The protocol conforms to the principles of the Declaration of Helsinki and was approved by the Yale University Institutional Review Board, which deemed it minimal risk, thus allowing for a waiver of informed consent at the patient level; providers provided informed consent. The trial protocol is in adherence with the SPIRIT guidelines ²⁸. PROMPT-LIPID was designed and conducted by Yale University investigators (Desai, Wilson, Ahmad). The authors are solely responsible for the design and conduct of this study, all study analyses, the drafting and editing of the manuscript, and its final contents.

Figure 1.

Study design and overview of study flow. The informational alert intervention is randomized at the provider level for eligible patients with very high risk ASCVD.

Table I.

Patient inclusion and exclusion criteria

Inclusion criteria
1.	Age ≥ 18 y old
2.	Currently being managed as an outpatient within the Yale New Haven Health System
3.	Very high risk ASCVD
4.	LDL-C > 70 mg/dL
Exclusion criteria
1.	Heart transplant recipient
2.	Left ventricular device recipient
3.	Hospital inpatient status
5.	Pregnancy
6.	Have previously opted out of clinical research within EHR
7.	No change in lipid management since most recent LDL-C measured.

Provider selection and enrollment

We retrospectively identified providers (physicians and advanced practice providers) within the Yale New Haven Health System outpatient cardiology and internal medicine clinics with the highest volume of patients meeting eligibility criteria in 2020. Eligible providers were contacted by the study team via e-mail and Epic (Wisconsin, MI) “In Basket” messaging with an overview of the study along with a hyperlink to consent on REDCap (Research Electronic Data Capture), hosted by Yale New Haven Hospital. ²⁹ Consenting providers were asked to complete a pre- and post-trial survey (Supplement), aimed at assessing knowledge and comfort levels with lipid-lower ing guidelines. The post-trial survey also includes questions that assess provider opinions on user experience with the alert. The survey responses will help appraise helpful aspects of the alert and uncover areas in need of improvement that may represent barriers to implementation.

Intervention

The intervention is a clinical decision support alert that displays upon opening the order entry screen of a patient’s electronic medical record on clinical desktop computers or remote clinical computer workstations. The alert opens only for eligible patients upon each outpatient visit during the study period with a provider in the intervention group (Figure 2). The alert was created and modified after targeted sessions with provider focus groups to elicit feedback on design, user experience, and hindrance to workflow. The alert was designed to appear in real-time and embedded within the providers’ usual workflow to minimize workflow inter ruption and alarm fatigue. It incorporates patient-specific clinical data to notify providers in the intervention group that a given patient has very high risk ASCVD, while displaying their most recent lipid profile to aid diagnosis. Tailored recommendations for subsequent LLT are also displayed, such that therapies already prescribed to the patient are not recommended again and therapies not already provided are recommended (Figure 3). If the patient has a documented allergy or other contraindication to a recommended medication, the statement “Patient has an allergy and/or intolerance to” is listed alongside that medication. In order to avoid a recurrent change in therapy without an interval measure of response after the initial change, the alert does not activate if a change to LLT had been made since the most recent lipid profile was ordered. The alert also contains a link to a smart order set that is customized to the patient and displays all the recommended lipid-lowering medications the patient is not currently prescribed, excluding those to which the patient had a documented allergy or contraindication. A hyperlink to current multi-society treatment guidelines for cholesterol management is also provided within the alert, which includes an option to obtain continuing medical education credits. The alert closes once the provider acknowledges the alert by selecting “I will adjust medications,” “Med changes not clinically indicated,” or “Defer for other reason (comment).” Finally, the alert was run silently (without displaying itself to providers) prior to the launch of PROMPT-Lipid to verify that patients with very high risk ASCVD were appropriately identified and that no changes in LLT were made in the interval period.

Figure 2.

Clinical decision support alert generated among the intervention group of providers for eligible patients with very high risk ASCVD.

Figure 3.

Clinical decision support smart order set. 1. Sections divided by statins, ezetimibe, and PCSK9i drugs. 2. Each section has wording to provide user with typical usage scenarios. 3. Each medication has default dosage, route, quantity, and frequency. 4. Orders have a default diagnosis of ‘hyperlipidemia,’ but can be changed to an alternative diagnosis by the ordering provider. 5. A follow up lab order for a lipid panel 3 months in the future is selected by default.

For providers in the control group, eligible patients identified by the EHR do not generate a visible alert for the provider, and these patients will receive usual care. However, a “silent alert” is generated to log the patient as a trial participant within the control group to the study team without notifying the provider in any manner.

Once a patient is seen by an enrolled provider, that patient will only generate future alerts for providers within the intervention arm but not for providers in the control arm, to avoid contamination between study groups (Table II). Alerts in the intervention group only display once per provider per 24 hours, during each future encounter in which the patient is eligible, and only when the provider enters the order entry interface.

Table II.

Alert behavior under various conditions

First provider	Alert firing	Subsequent provider	Alert firing
Randomized to alert	✓	Randomized to alert	✓
Randomized to alert	✓	Randomized to usual care	✕*
Randomized to usual care	✕	Randomized to usual care	✕
Randomized to usual care	✕	Randomized to alert	✕*

^*This logic is intended to avoid contamination between study arms.

Randomization

Randomization will occur at the level of the provider. All included providers are randomized within the EHR to either the electronic automated alert (intervention) group or the usual care (control) group via a permutated block randomization scheme to ensure a similar number of providers in each study arm. One hundred clusters (based on providers) will be created to which eligible patients are assigned upon their initial outpatient visit with these providers during the study period. Identification and enrollment of patients among each cluster in the study population is performed automatically within the EHR using eligibility criteria, such that patients are randomized to each group based on the randomization of their provider.

Since it is likely that providers may internalize best practice recommendations once repeatedly exposed to the alert, cluster randomization at the level of their provider is intended to ensure that providers consistently receive an alert or not, which would reduce contamination between study arms. To avoid disproportionate distribution of patients to any single provider, we limited the number of patient assignments to 50 patients per provider. The study team was blinded to the randomized group assignment until the end of the trial period.

Outcomes and follow-up

Covariates of interest will include patient age, sex, race, ethnicity, baseline lipid profile, baseline lipid-lowering medication use, and comorbid conditions. Comorbidities will be determined using ICD-10 codes extracted from the patient’s EHR Problem List.

Outcomes of interest are presented in Table III. The primary outcome will be the proportion of eligible patients who have intensification of LLT, defined as new initiation of high-intensity statin (from no statin therapy or from low- or moderate-intensity statin), addition of ezetimibe, or addition of PCSK9i at 90 days after the first alert. The initial 90-day follow up period was chosen to not only allow for LLT intensification at the study visit, but also to allow patients to have an opportunity to obtain and begin taking the ordered medications. The key secondary outcome is the achieved LDL-C level at 6 months from the first alert for any given patient. The LDL-C measured closest to 6 months from first alert (± 3 months) will be used for each patient for this outcome. We note that this secondary outcome is subject to ascertainment bias, such that individuals who receive lipid-lowering agent titration may be more likely to have LDL measured in the future. We will address this first by performing a complete-case analysis, comparing LDL levels between the intervention and usual care arm among those who have measurements. We will further use multiple imputation by chained equations to impute missing LDL values for those who do not have them measured in follow-up to assess the robustness of our findings. Finally, we will develop a model predicting LDL measurement which includes randomization status to quantify the effect of randomization on future LDL measurement. We can then weight this analysis by the inverse probability of having an LDL measured to approach a less biased estimate. Additional secondary outcomes are the proportion of patients with LDL-C < 70 mg/dL or < 55 mg/dL at 6 months from the first alert for any given patient, and the rate of MACE, defined as hospitalization for myocardial infarction, ischemic stroke or coronary or peripheral arterial revascularization within 6 months. All clinical events will be abstracted from the EMR based on ICD-10 codes. Although some loss to follow up is expected, we do not expect this to vary between groups.

Table III.

Patient and provider outcomes

Patients outcomes	Data source
Primary outcome
Percentage of patients with intensification of lipid-lowering therapy, defined as an increase in statin dose/intensification of statin,* addition of ezetimibe, or addition of PCSK9i at 90 d after first alert	Hospital record
Secondary outcomes
Achieved LDL-C level at 6 mo from the first alert for any given patient.	Hospital record
Percentage of patients with LDL-C < 70 mg/dL at 6 mo from first alert	Hospital record
Percentage of patients with LDL-C < 55 mg/dL at 6 mo from first alert	Hospital record
Hospitalization at 90 d from the first alert for myocardial infarction, percutaneous coronary intervention, coronary artery bypass grafting, peripheral revascularization, or stroke	Hospital record
ED visit at 90 d from the first alert	Hospital record
Death at 90 dfrom the first alert	Hospital record
Provider outcomes
Secondary outcomes
Percentage of providers who visit the guideline site	Hospital record (from alert)
Percentage of providers who open the order set within the alert	Hospital record (from alert)
Pretrial survey (evidence-based guidelines knowledge)	REDCap
Post-trial survey (evidence-based guidelines knowledge and alert feedback)	REDCap

^*Intensification of statin refers to changing from no statin therapy or from low- or moderate-intensity statin to high-intensity statin. (Moderate-intensity: atorvastatin 10 to 20 mg, Fluvastatin 80 mg, lovastatin 40 to 80 mg, pitavastatin 2 to 4 mg, pravastatin 40 to 80 mg, rosuvastatin 5 to 10 mg, simvastatin 20 to 40 mg. High-intensity: atorvastatin 40 to 80 mg, rosuvastatin 20 to 40 mg).

We will perform prespecified analysis across the following subgroups of interest to determine differences in the effect of the alert: insurance coverage (private/supplemental vs Medicare/Medicaid), provider type, provider specialty (internal medicine vs cardiology), baseline lipid levels (LDL-C > 100 mg/dL vs LDL-C 70–100 mg/dL), patient age (< 75 and ≥ 75 years), and patient sex.

Statistical analysis plan

A retrospective review of outpatient visits during calendar year 2017 revealed 4,451 unique patients with ASCVD and LDL-C > 70 mg/mL, approximately 3,700 of whom were seen by the top 100 providers with the highest rate of patients with very high risk of ASCVD. Preliminary data suggested that 10% of patients would have appropriate up-titration of LLT over the timeframe of the study. A sample size of 2,500 enrolled patients was thus planned to achieve 80% power to detect a difference in intensification of LLT from the baseline rate of 10% overall to 15.5% in the intervention arm. We felt this change represents a clinically meaningful increase given the modest incremental intensification of LLT observed in a contemporary registry cohort. ³⁰

Given the cluster-randomized nature of the study, we will compare baseline covariates between the 2 intervention groups using logistic (for binary categorical variables) and linear (for continuous variables) regression clustered at the provider level and using robust estimates of standard errors. The primary analysis will be conducted using logistic regression where the independent variable (primary endpoint) is the proportion of patients who have intensification of LLT and the primary dependent variable is randomization status. The regression will similarly be clustered at the provider level and employ robust standard errors. Statistical significance will be based on P < .05. For categorical secondary outcomes, a similar generalized linear modelling approach will be used. For continuous secondary outcomes, a generalized linear model with a Gaussian distribution will be used with an identity-link to describe adjusted absolute differences between outcomes. A P -value of <.05 will be used for secondary outcomes.

Funding

PROMPT-Lipid is funded as an Investigator Initiated Study by Amgen. While the PCSK9i evolocumab is marketed by Amgen, this trial included all FDA-approved PCSK9i.

Results

Between 10/28/2019 and 12/13/2019, the trial reached 50% patient enrollment, with 48 providers recruited to the intervention group and 47 providers recruited to the control group (Figure 4). In the alert group (n = 535, median age [IQR] 72 [63, 81], 50% women, median LDL-C 90 [78, 112] mg/dL), 70%, 7%, and 2% were on statins, ezetimibe, or PCSK9i, respectively, at baseline. In the control group (n = 715, median age 72 [62, 79], 52% women, median LDL-C 91 [79, 116] mg/dL), 71%, 10%, and 3% were on statins, ezetimibe, or PCSK9i, respectively, at baseline. Baseline lipid lowering therapies and comorbidities in each arm were similar (Table IV). Among ASCVD risk factors, there was a high prevalence of hypertension in both groups (88% overall), and among ASCVD events, there was a high prevalence of prior myocardial infarction (75% overall).

Figure 4.

Patient enrollment to intervention (alert) arm and control (usual care) arm during trial enrollment.

Table IV.

Baseline patient characteristics

Variables	Intervention (alert) (N = 535)	Control (usual care) (N = 715)	P-value
Demographics:
Age (mean, SD)	72 (63,81)	72 (62,79)	.6220
Sex, female	268 (50.1%)	369 (51.6%)	.6858
Race, Black	48 (9%)	110 (15.4%)	.0321
Ethnicity, Hispanic or Latino	52 (9.7%)	76 (10.6%)	.8124
Lipid profile and treatment:
Cholesterol (mg/dL)	169 (151,194)	170 (152,200)	.0110
HDL-C (mg/dL)	51 (41,61)	50 (42,62)	.7370
LDL-C (mg/dL)	90 (78,112)	91 (79,116)	.0280
Triglycerides (mg/dL)	110 (80,158)	112 (82,155)	.5860
On Statin	376 (70.3%)	509 (71.2%)	.7914
On Ezetimibe	37 (6.9%)	74 (10.3%)	.1748
On PCSK9i	13 (2.4%)	19 (2.7%)	.8061
Comorbidities:
Cardiovascular disease	230 (43%)	335 (46.9%)	.4002
Prior myocardial infarction	405 (75.7%)	526 (73.6%)	.5848
Peripheral arterial disease	179 (33.5%)	286 (40%)	.0518
Prior stroke	258 (48.2%)	396 (55.4%)	.0862
History of obesity	114 (21.3%)	178 (24.9%)	.2748
History of diabetes	172 (32.1%)	242 (33.8%)	.6374
History of hypertension	463 (86.5%)	631 (88.3%)	.4729
Provider assignment:
# of unique providers	48	47
# of patients per provider	8.5 (5,14)	12 (7,22)	.0397

Values represent median (IQR) for continuous variables or n (%) for categorical variables).

Discussion

A cornerstone of medical management for patients with very high risk ASCVD is lipid-lowering therapy. Despite multisociety guideline recommendations for intensification of LLT to reduce LDL-C levels and risk of MACE, implementation and adherence has been globally suboptimal. We designed a multicenter cluster-randomized clinical trial to test whether a personalized EHR alert to outpatient providers will increase the proportion of patients with very high risk ASCVD who receive evidence-based intensification of lipid-lowering pharmacologic therapy. We rapidly completed 50% enrollment with 1,250 patients within 47 days, demonstrating the efficacy of the trial design on recruitment.

Multiple prior attempts to improve adherence to LLT have been attempted including simplification of drug regimens, patient education and information, intensified patient care, complex behavioral approaches, decision support systems, administrative improvements, and pharmacy-led interventions. ³¹ A recent cluster-randomized pragmatic trial comparing guideline-directed statin prescribing among cardiologists randomized to usual care or passive or active choice EHR alerts found no overall difference in statin prescribing rates among nearly 12,000 patients, with a small increase in statin prescribing rates for the 68% of patients with ASCVD. ³² PROMPT-LIPID further includes multiple non-statin interventions, targets internal medicine clinicians in addition to cardiologists, and only includes patients with very high risk ASCVD. The trial’s multicenter design across a large health system and geographic area, including a mix of care models and socioeconomically diverse patient population, strengthens its generalizability. ³³ Additionally, the large size of this trial with anticipated enrollment of 2,500 eligible patients allows for detection of clinically meaningful changes in our endpoints. The use of a fully automated system, from subject selection and enrollment to randomization and intervention, improves costs, feasibility, generalizability, and rapidity of trial enrollment as observed in the concurrent PRagmatic trial of messaging to providers about treatment of heart failure trial by our group. ³⁴ While the automated alert aims to be as patient-specific as possible, it cannot fully remove the necessity for patient-provider interaction to address patient-specific clinical context. Possible contamination between intervention and trial provider groups may also limit study findings but has been addressed by careful alerting logic as described above.

Given well-established improvement in ASCVD outcomes with LLT intensification, we chose instead to assess the feasible and relevant primary outcome of increases in LLT. The suboptimal baseline utilization of LLT in this cohort is consistent with national estimates yet provides ample opportunity for therapy intensification that may produce a generalizable and feasible approach across other health care systems if the trial is successful. Although this trial focuses primarily on the impact of a targeted, tailored alert on provider prescribing patterns, future studies may allow additional insights on real-world barriers to adherence to evidence-based LLT as well as barriers to engage with the alert. Assessment of the secondary outcome of achieved LDL-C levels at 6 months will allow us to evaluate the magnitude of clinical impact of the intervention, especially since LDL-C lowering is associated with risk improvement. Finally, pre- and postintervention survey findings regarding overall user experience and provider education will guide optimization in the alert design. Given the current suboptimal adherence to guideline-recommended LLT intensification, a simple, pragmatic, and cost-effective automated alert system is an attractive solution to improve care of patients with ASCVD.

If no difference is observed in LLT intensification or LDL-C lowering, the trial will still provide new information regarding implementation barriers. While we will assess changes in provider orders of LLT based on the alert vs usual care, it is also possible that other barriers to success not measured by this trial – such as alarm fatigue, lack of actionable or new information in the alert, costs, or adverse drug effects – may negatively impact LLT intensification. By surveying participating providers, we may elicit these or other barriers to inform future strategies in improving LDL-C management.

Conclusions

With the PROMPT-LIPID randomized controlled trial, we have developed an automated, patient-specific EHR alert of evidence-based lipid-lowering therapies for patients with very high risk ASCVD to increase utilization of these therapies. If proved successful, this approach offers a high-yield, low-resource method of closing the substantial gap in LDL-C management.