Protocol of a Patient Randomized Clinical Trial to Improve Medication Adherence in Primary Care

Abstract

Background:

Enhanced awareness of poor medication adherence could improve patient care. This article describes the original and adapted protocols of a randomized trial to improve medication adherence for cardiometabolic conditions.

Methods:

The original protocol entailed a cluster randomized trial of 28 primary care clinics allocated to either (i) medication adherence enhanced chronic disease care clinical decision support (eCDC-CDS) integrated within the electronic health record (EHR) or (ii) usual care (non-enhanced CDC-CDS). Enhancements comprised (a) electronic interfaces printed for patients and clinicians at primary care encounters that encouraged discussion about specific medication adherence issues that were identified, and (b) pharmacist phone outreach. Study subjects were individuals who at an index visit were aged 18–74 years and not at evidence-based care goals for hypertension (HTN), diabetes mellitus (DM), or lipid management, along with low medication adherence (proportion of days covered [PDC] <80%) for a corresponding medication. The primary study outcomes were improved medication adherence and clinical outcomes (BP and A1C) at 12 months.

Protocol adaptation became imperative in response to major implementation challenges: (a) the availability of EHR system-wide PDC calculations that superseded our ability to limit PDC adherence information solely to intervention clinics; (b) the unforeseen closure of pharmacies committed to conducting the pharmacist outreach; and (c) disruptions and clinic closures due to the Covid-19 pandemic.

Conclusion:

This manuscript details the protocol of a study to assess whether enhanced awareness of medication adherence issues in primary care settings could improve patient outcomes. The need for protocol adaptation arose in in response to multiple implementation challenges.

Background

Suboptimal medication adherence is a widely recognized barrier to adequate chronic disease care (CDC)^1–4 and worsens clinical outcomes.^2,5–9 Approximately 50% of individuals with chronic diseases do not take medications as prescribed,^2,10 leading to increased morbidity, hospitalizations, mortality, and increased costs of care.^3,7,11–19

Determinants of poor medication adherence encompass factors related to care delivery, medical conditions, socioeconomics, intentional patient behaviors, as well as nonintentional behaviors such as forgetfulness.^20–28 As a response to this complexity, various theoretical frameworks have been utilized to develop interventions, including social cognitive theory focusing on the importance of self-efficacy, self-regulation and coping mechanisms, transtheoretical models targeting motivation and readiness to change, and health belief models.²⁸ While many of these theoretical constructs have shown some degree of success in small settings,²⁸ the information, motivation, and belief (IMB) model^29,30 was chosen for this study intervention because it integrates many theoretical constructs, has demonstrated efficacy in enhancing adherence for conditions such as HIV and asthma,^29,31–33 and has gained endorsement from the World Health Organization.²⁰

When implementing an intervention in primary care, time constraints present a challenge for practicing clinicians, a predicament that can be exacerbated by administrative tasks and extensive electronic health record (EHR) documentation requirements.^34,35 Time constraints can also affect the communication of critical information when new medications are initiated which can contribute to patient failure to take medications as directed.³⁶ In addition, studies on time allocation reveal that family physicians address an average of 4–6 issues per visit for older patients, potentially reducing attention to chronic diseases.³⁷ However, brief opportunistic PCC messaging has proven effective for smoking cessation and select chronic diseases.^38–41

Leveraging EHR based clinical decision support (CDS) systems beyond standard EHR dashboards can promote PCC messaging and holds promise in primary care settings to improve chronic disease outcomes,^42–44 but these CDS systems have yet to be incorporate adherence data. Many EHR’s now provide data driven calculations called proportion of days covered (PDC) using methodology endorsed by the Pharmacy Quality Alliance⁴⁵ and used by Centers for Medicare & Medicaid Services (CMS). PDC information can be leveraged within CDS systems without collecting additional adherence data from patients or disrupting care.

Evidence further supports the effectiveness of using pharmacists on the care team to support individuals with chronic conditions,^46–53 particularly when they systematically address the determinants of poor adherence through the IMB model.^24,30 The integration of pharmacists into the primary care team is expanding nationwide, backed by many rigorous demonstrations of their clinical and economic value.^54–58

This study aimed to develop and evaluate an adherence intervention informed by the IMB model targeting enhanced patient adherence, utilizing a succinct and nondisruptive PCC intervention using a CDS system and asynchronous pharmacist phone outreach for individuals with persistent adherence issues. A secondary aim was to assess cost effectiveness, providing justification for widespread dissemination if positive study results are obtained. The study outcomes will offer crucial insights into whether integrating adherence data into primary care CDS systems, alongside pharmacist intervention addressing adherence determinants, can ameliorate poor medication adherence and improve clinical outcomes.

Methods

2.1. Overall study design

The original protocol (Figure 1) outlined a clinic cluster randomized trial comparing the adherence intervention to usual care. Designed as a pragmatic trial, it aimed to assess the effectiveness of the intervention in real-life routine practice conditions, as opposed to assessing it in optimal conditions.^59,60 The study was conducted at HealthPartners, a nonprofit integrated health system in Minnesota and western Wisconsin serving approximately 1.2 million patients. It encompassed a multispecialty group practice with more than 1800 physicians, 25 Medical Therapeutics Management (MTM) pharmacists, 8 hospitals, and 55 primary care clinics. The HealthPartners care group accepted all forms of commercial insurance, Medicaid, and Medicare, and served a diversity of patients in terms of age, race/ethnicity, and socioeconomic status.

Figure 1:

Cluster Randomized Study Design (Original Protocol)

The usual care arm during the study period accessed a CDC-CDS tool previously implemented, offering treatment considerations for HTN, DM, and dyslipidemia, weight, tobacco, and aspirin/anticoagulation but did not include any medication adherence information. For patients identified at encounters with high cardiovascular (CV) risk, the system prompted rooming staff to print the CDS tools for patients and PCCs. The CDS system was HER-integrated, Web-based, and deployed by rooming staff within the primary care visit workflow at an average of 75% of targeted encounters for high CV risk patients. In a previous randomized trial, the CDC-CDS system reduced CV risk in the targeted population and has been described in detail in previous publications.^61–63

Due to unforeseen circumstances, the study protocol was adapted just before intervention implementation. This was driven by the closure of two of the three pharmacies involved in the study in late 2019, and significant care disruptions caused by the Covid-19 pandemic beginning in March 2020 that lead to multiple clinic closures, dedication of some clinics to patients with respiratory problems, and a shift to preferred virtual primary care encounters for patient care when feasible. Consequently, in collaboration with the funding agency, Internal Review Board (IRB), and Data Safety and Monitoring Board (DSMB), the original clinic randomized trial was transformed into a patient randomized study design and MTM pharmacists were used to deliver the pharmacist phone outreach. The MTM pharmacists were not part of the study team but were established health care professionals within the care system. These pharmacists had completed a recognized medication therapy management education program with both clinical and didactic elements approved by the Board of Pharmacy of the state(s) where the pharmacist practiced that was recognized by the Accreditation Council for Pharmacy Education. Additional qualifications for MTM pharmacists were at least two years of clinical pharmacy experience in ambulatory care as part of a multidisciplinary team or successful completion of one-year specialized post-doctoral residency. The rational for all modifications of the original protocol that were incorporated into the adapted protocol are outlined in detail in Table 1.

Table 1:

Study Design Modifications and the Rationale

Original Protocol	Adapted Protocol	Reason
Clinic level randomization	Patient level randomization	Clinic closures and increased virtual encounters led to widespread patient migration to various clinics.
Intervention delivered at in-person office visits	Intervention delivered at in-person and virtual visits	Due to increases in telehealth/virtual visits during Covid-19, the team adjusted programming to make CDS available in virtual visits
12-month patient accrual period	6-month accrual period	With the patient randomization, a smaller analytic sample was needed for power estimates. The shortened accrual period also helped keep the project on time.
Pharmacist phone outreach to be done by clinic affiliated retail pharmacies	Pharmacist phone outreach to be done by MTM pharmacists affiliated with the care system	Unexpected closure of clinic-based pharmacies caused this change. The MTM’s already had established relationships with PCCs, allowing this to be a smooth transition
Primary outcome 12month post-index	Primary outcome 12month post-index. Secondary 18-month outcome.	Shortened accrual period allowed secondary analysis of 18-month follow up. Fewer encounters were expected due to the pandemic, and a longer follow up time could increase intervention exposure.

2.2. Study Participant Accrual and Randomization

The study included 26 clinics utilizing an EHR with Epic software by Epic Systems Corporation, Verona, WI [epic.com]. Patients were randomized 1:1 to intervention or control at the time of their index visit, with randomization determined by a CDC-CDS computerized allocation system based on assigned study IDs. Accrual occurred at primary care encounters over six months (from August 19, 2020 through February 18, 2021). The IRB waived the need for patient consent. Patients did not receive information about the study in advance. Blinding was not feasible due to the nature of the intervention. However, data extraction and analysis were conducted while blinded to randomization.

2.3. Study Participant Inclusion and Exclusion Criteria

The CDS system extracted EHR patient data at primary care encounters at the 26 clinics and assessed study eligibility with algorithms programmed within the CDS system. Patients were eligible for the study if they had a clinical encounter (index visit) with a PCC defined as a physician or advanced practice provider within a primary care department (family medicine, internal medicine, combined pediatric/internal medicine) and met cut-off thresholds for both suboptimal care and poor adherence as defined in Table 2. The thresholds for determining suboptimal care were an A1C ≥ 8% or a BP ≥ 140/90 mmHg in alignment with national guidelines and quality measurements for Healthcare Effectiveness Data and Information Set (HEDIS) for diabetes care and blood pressure control.

Table 2.

Study Eligibility by Cohort

STUDY COHORT	Suboptimal Care Criteria	Rational for Suboptimal Care Threshold	Adherence Thresholds
Hypertension	Index date and the prior encounter date with systolic BP (SBP) ≥140 mmHg or diastolic BP (DBP) ≥ 90 mmHg	Consistent with the ACC/AHA Blood Pressure Guideline Consistent with HEDIS measure for controlling High Blood Pressure	PDC <80% with moderate or high data confidence for all medications in at least one class of antihypertensive medications. Classes of hypertensive medications included (a) diuretics (thiazides or loop), (b) angiotensin converting enzyme inhibitors or angiotensin receptor blockers, (c) calcium channel blockers and (d) beta blockers.
DM	Met criteria for suboptimal glycemic control defined as most recent A1C ≥8% in the 12 months prior to index	Consistent with American Diabetes Association guideline Consistent with A1C component of HEDIS measure for Comprehensive Diabetes Care	PDC <80% with moderate or high data confidence for all medications in at least one class of active non-insulin glycemic medications. Classes of non-insulin glycemic medications included (a) biguanides (e.g., metformin), (b) Insulin secretagogues (e.g. sulfonylureas or meglitinides), (c) glucagon-like peptide 1 receptor agonists, (d) sodium-glucose co-transporter 2 inhibitors, (e) dipeptidyl peptidase-4 inhibitors and (f) thiazolidinediones.
Lipid	Met one of the high-risk categories for recommending statin usea Patients with most recent LDL < 100 mg/dL in the last two years were excluded from the lipid cohort because they could be at a recommended LDL goal despite potential statin nonadherence.	Categories for statin usea and LDL goals were based on the ACC/AHA lipid guideline.73 HEDIS measure for statin therapy for patients with cardiovascular disease or diabetes	PDC <80% with moderate or high confidence (for all currently prescribed statins).

^aACC/AHA Categories for statin use included:

a. Aged >21 years with atherosclerotic cardiovascular disease (ASCVD) identified by an ASCVD diagnosis on the problem list or two or more International Classification of Diseases 10^th revision (ICD10)⁷⁴ codes for ASCVD conditions in the last 2 years.

b. Aged >21 years and LDL >190 mg/dL.

c. Aged 40 to 75 years AND diagnosis of DM. DM was determined from DM on the problem list or two or more encounter IDC10 codes for DM in the last 2 years.

d. Aged 40 to 75 years with 10-year ASCVD Risk Score ≥7.5% based on the ACC/AHA 10-year Pooled Cohort risk equations.⁷⁵

All BP and A1C values were obtained from the EHR. The clinics used standardized procedures for obtaining BP using automated devices. If the initial encounter BP was > 140/90 mmHg, the BP was to be repeated after at least 5 minutes of quiet rest using the average of two or more BP readings taken one minute apart. A1C and lipid testing frequency and goals were individualized based on clinician preferences. A1C was recommended at least yearly and done for more for 90% of patients with diabetes. The eCDC-CDS used the last recorded BP in encounters, and the most recent A1C and lipid test values to assess the study eligibility criteria and for making CDS clinical suggestions.

Poor adherence was defined as a PDC < 80% to be consistent with optimal cut points for poor adherence⁶⁴ and thresholds used by the Pharmacy Quality Alliance⁴⁵ and Medicare adherence measures for hypertension, diabetes, and cholesterol. The PDC calculation in Epic compared a patient’s medications on the current medication list with the number of times a medication had been dispensed. For example, if a metformin prescription was active for 3 months but was only filled once with a 30-day supply in the last 3 months, the PDC score would be roughly 33%. Epic also supplied a confidence rating of high, medium, or low for each adherence score based on the reliability of the source dispense data. If a patient got all their prescriptions from a pharmacy that reported dispense data, then the Epic system assigned a high confidence rating. If the patient did not meet these criteria, but claims data was available, then the Epic system assigned a medium confidence rating. Claims data may not be available if a patient paid out of pocket or used a different insurance. If not all prescriptions for the same medication were dispensed by pharmacies that sent dispense data, and no claims data was available, the Epic system assign a low confidence rating. The eCDC-CDS intervention evaluated currently prescribed medications and Epic-based PDC calculations for all BP, noninsulin glucose lowering, and statin medications. PDC values with low confidence were excluded from determinations of the low adherence eligibility criteria. Additional eligibility criteria included age between 18–74 at index, no nursing home or hospice care, no active cancer, and no pregnancy in the prior 12 months.

2.4. Description of the Intervention

The intervention aimed to improve patient medication adherence and outcomes by delivering point-of-care CDS medication adherence information to PCCs and patients at clinical encounters when potential adherence issues were identified. Existing workflow prompted rooming staff (nurses and medical assistants) to click on a Best Practice Advisory (BPA) that appeared for study eligible patients that would display and print the two separate interfaces. One interface was a low literacy engagement tool used for patients, and the other was an information rich tool used by PCCs. A display of the eCDC-CDS output with more detailed mediation dispense information was also viewable within EHR. Figure 2 shows examples of the interfaces and EHR display. Repetitive eCDC-CDS use over subsequent encounters during the study period offered an efficient and powerful tool for ongoing surveillance and management of adherence.

Figure 2:

Examples of eCDC-CDS displays: (a) printed for study eligible patients, (b) printed for PCCs, and (c) viewable by PCCs within the EHR

A one-time pharmacist phone outreach was conducted approximately 6 months after the index visit (the first encounter during the intervention in which study inclusion criteria were met) if adherence issues persisted. The pharmacists used scripted templates to assess and address barriers to medication adherence related to the IMB framework. Action(s) for patients to consider were individualized but could involve education, recommending lower cost alternative medications or combination medications, addressing side effects, using pill boxes, modifying pill-taking schedules and/or using reminder systems, or simplifying the refill process. Pharmacists had full read/write access to the EHR and could make medication changes using established care protocols and/or communicate with PCCs or other prescribers. To manage the pharmacist outreach process, study subjects were automatically re-evaluated 6 months after the index visits by CDS algorithms and those with persistent suboptimal care thresholds and adherence issues were added to an electronic registry in the EHR. Pharmacists used the registry to call patients and track contact attempts. When they reached patients, they identified themselves as part of the care team working with the PCCs in their clinic. Figure 3 shows details about the recommended team-based workflow for using the e-CDC-CDS and to direct proactive pharmacist outreach.

Figure 3:

Adherence Intervention Workflow

2.5. Pilot

The eCDC-CDS and the pharmacist outreach interventions were developed collaboratively by primary care clinicians and pharmacists. Prior to the go-live date for the study, we pilot tested the eCDC-CDS and the pharmacist phone outreach protocol at 3 clinics. We asked for feedback from PCCs and pharmacists and refined algorithms, CDS materials, script templates, and registry workflows based on this feedback.

2.6. Intervention Training

Live one-hour training webinars were conducted by the study team for PCCs and for clinic staff involved in rooming patients to familiarize them with the intervention. Training reinforced the existing usual care CDC-CDS workflow and focused on the eCDC-CDS enhancements to address medication adherence. PCCs were also oriented to the additional medication dispense information on the eCDC-CDS display in the EHR. PCCs could ask questions or give feedback about the CDS at any encounter by clicking a “feedback” button on the EHR display. The pharmacists involved in conducting outreach were trained separately by the pharmacist investigator on the research team. The training included information about the IMB framework, how to use the outreach registry for contacting and tracking call attempts, helpful scripting for talking to patients, and how to complete the outreach documentation templates within the EHR.

2.7. Intervention Start Date

The study went live in 28 clinics under the original protocol on March 16, 2020, the day before Covid-19 was declared a national pandemic. As a consequence of the immediate pandemic-related care disruptions, the study was suspended one week later. The disruptions included clinic closures, repurposing of several clinics for respiratory and Covid-19 care, shifts to virtual encounters, and increased stress on the care delivery system. The intervention resumed under the adapted protocol 6 months later in August 2020 in the 26 clinics that remained open, with modifications as previously described.

2.8. Data Sources

Data for the primary outcomes were sourced from the EHR Clarity database, a large subset of Epic data transferred to a server that allows for retrieval of EHR information without affecting the production environment. Data reports were created by trained programmers with organizational permission for EHR access who performed quality checks to ensure data completeness and accuracy. The EHR based PDC was calculated at patient encounters but was not automatically saved within the EHR Clarity database. Because the PDC data was also necessary for this research analysis, the PDC calculation at each encounter was stored by the eCDC-CDS system in a separate secure analysis data repository. Additional data for the secondary cost analysis was obtained from insurance claims and Surescripts data. Surescripts is an information technology company that supports e-prescription and the electronic transmission of prescriptions and general health information exchange between health care organizations and pharmacies [surescripts.com].

In addition, we collected data from the pharmacist outreach intervention. At each contact attempt, the pharmacist completed an electronic form within Epic that included discrete data on contact attempts, contact success, confirmation of adherence issues, and reasons for poor adherence that included cost, forgetfulness, side effects, health beliefs, and need for knowledge/information. Action steps to address poor adherence such as pill boxes, education, switching to mail order, were also recorded. Qualitative information was obtained from the MTMs to describe success stories and challenges.

2.9. Study Outcomes

The study evaluated the intervention’s impact at 12 months on (a) medication adherence to antihypertensive, oral hypoglycemic, and statin medications and (b) blood pressure and A1C control (Table 3). The denominator for outcome measures were study eligible patients enrolled into hypertension, diabetes, and lipid cohorts at the index visit. Patients could be enrolled in more than one cohort. Because the outcomes were obtained from routine clinical care, there was no attempt by the researchers to influence patient visit or testing frequency. If patients were on multiple medications for hypertension or diabetes, PDC scores were calculated for individual medications but adherence levels were assessed by medication class. Classes of hypertensive medications include (a) diuretics (thiazides or loop), (b) angiotensin converting enzyme inhibitors or angiotensin receptor blockers, (c) calcium channel blockers and (d) beta blockers. Classes of non-insulin glycemic medications include (a) biguanides (e.g. metformin), (b) Insulin secretagogues (e.g. sulfonylureas or meglitinides), (c) glucagon-like peptide 1 receptor agonists, (d) sodium-glucose co-transporter 2 inhibitors, (e) dipeptidyl peptidase-4 inhibitors and (f) thiazolidinediones. A positive adherence outcome was defined as a PDC >80% for at least one medication in each prescribed class. For example, if a patient was prescribed medications from several different classes to manage hypertension, a positive outcome was a PDC >80% for at least one medication in each of those classes.

Table 3:

Study Objectives and Outcomes

OBJECTIVES	OUTCOME MEASUREMENTS
Primary
1. Impact of the intervention on HTN medication adherence at 12 months 2. Impact of the intervention on DM medication adherence at 12 months 3. Impact of the intervention on Lipid medication adherence at 12 months 4. Impact of the intervention on SBP control at 12 months 5. Impact of the intervention on A1C control at 12 months	1. Achievement of PDC ≥80% for at least one antihypertensive medication in each currently prescribed BP medication class at 12 months following the index office visit 2. Achievement of PDC≥80% for at least one medication in each currently prescribed non-insulin glycemic medication class at 12 months following the index office visit 3. Achievement of a PDC ≥80% for a statin medication if currently prescribed at 12 months following the index office visit 4. Differential change in SBP from the index SBP to last SBP in the 12-month period following the index office visit 5. Differential change in A1C value from the index A1C to last A1C in the 12-month period following the index office visit
Secondary
1. Impact of the intervention on overall healthcare costs 2. Predicted long-term health impact and cost-effectiveness of the intervention from the health system (payer) perspective	1. Measure the incremental medical care costs attributable to the eCDC-CDS intervention, defined from the health system perspective, measured using utilization incurred in the 12-month pre- and post-index date periods. 2. If the primary endpoints show effectiveness, we will use a cardiovascular disease microsimulation model to predict the long-term impact of the eCDC-CDS intervention on health outcomes and its cost-effectiveness, defined from the health system perspective, over up to a 30-year prospective time horizon.
Exploratory
1, 2, 3. Impact of the intervention on medication adherence for HTN medication, DM medication and lipid medication at 18 months 4. Impact of the intervention on SBP control at 18 months 5. Impact of the intervention on A1C control at 18 months	1, 2, 3. Adherence endpoints as described above for primary objectives at 18 months following the index office visit for antihypertensive, glucose lowering, and statin medications 4. Reduction in SBP from the index SBP to last SBP in the 18-month period following the index office visit 5. Reduction in A1C value from the index A1C to last A1C in the 18-month period following the index office visit
6, 7. Impact of the intervention on DBP control at 12 and 18 months 8, 9. Impact of the intervention on LDL control at 12 and 18 months	6, 7. Reduction in DBP from the index DBP to last DBP in the 12-month period (6) and 18-month period (7) following the index office visit 8, 9. Reduction in LDL value from the index LDL to last LDL in the 12-month period (8) and 18-month period (9) following the index office visit

The BP outcome was differential change in systolic BP from index SBP to the last measured SBP at 12 months. The A1C outcome was differential change in A1C value from index to the last measured A1C at 12 months. An exploratory analysis was planned to evaluate the longer-term impact on adherence and clinical care at 18 months. Data was collected to evaluate other potentially confounding effects such as interpreter needs, comorbidities, number of visits/exposures to the intervention, and baseline levels of A1C, BP, and LDL.

Additional data was obtained from patient and clinician surveys. Eligible study patients were surveyed shortly after their index visit to assess self-reported adherence. The survey used a validated adherence questionnaire called ASK-12 (Adherence Starts with Knowledge-12).⁶⁵ PCCs from study clinics were surveyed in June of 2022, about 2 years after the start of the intervention. The survey queried PCCs about their experience using the CDS system. It addressed the extent of use at in-person and telehealth encounters, usefulness of the CDS for addressing different clinical issues, overall satisfaction with the tools and the workflow, helpfulness of the various CDS functions, reasons for using or not using the CDS, and suggestions for CDS enhancements. The survey was conducted by the HealthPartners Institute Center for Survey Evaluation via an email invitation with a link to an online Redcap Survey.

The process measure monitored closely by the study team after implementation of the intervention was monthly print rates of CDS interfaces at eligible primary care encounters at the PCC, clinic, and system levels.

2.10. Analysis Plan

The a priori estimates for sample size and power were revised after the transition from the clinic to the patient randomized trial. The expected accrual counts of study eligible patients were 1,066, 832, and 1,694 for the HTN, diabetes, and lipid cohorts respectively. Given these sample sizes, at a probability threshold for statistical significance of α=0.05, we calculated ≥80% power to detect absolute differences in proportions by study arm of patients becoming adherent to medications at 12 months of ≥8.7% (HTN), ≥9.9% (DM), and ≥6.9% (lipids). Additionally, under similar assumptions we estimated ≥80% power to detect differences of ≥3 mmHg in mean SBP and ≥0.3% in mean A1C by study group.

The primary analysis was intention to treat and employed generalized linear regression models for adherence endpoints and general linear mixed regression models for clinical endpoints. Models for adherence at 12 months featured a binary dependent variable for adherence to all currently prescribed medication classes. The model also included a binary study arm predictor and baseline covariates for age, sex, and race/ethnicity (chosen a-priori) to potentially improve efficiency of the analysis. Estimates from these models are reported as adjusted odds ratios (binomial distribution, logit link function) comparing the probability of adherence in intervention patients vs. usual care patients, along with 95% confidence intervals (CIs). Models for SBP and A1C featured continuous dependent variables, a binary study arm predictor, time (baseline vs. 12 months) and a treatment by time interaction term used to assess treatment effect. Estimates from these models are reported as beta estimates (difference in means by group; normal distribution, identity link function) with 95% CIs. Supplemental analyses evaluated heterogeneity of treatment effects by patient sex, age group, race/ethnicity, and insurance class by including interaction terms in the models described above. Other heterogeneity effects that were considered include patient interpreter needs, presence of comorbidities such as depression, ASCVD, chronic kidney disease, potential number of exposures or post-index encounters, and baseline levels of care outcomes. Exploratory analyses examined change from baseline to 12 months in DBP and LDL, PDC endpoints at 18 months, and change from baseline to 18 months in clinical measures (SBP, DBP, A1C, LDL).

2.11. Missing Data

All key analytic variables were derived from EHR or health plan databases in which it is extremely rare for care-delivery information to be incompletely recorded. Because all data originates from EHR production tables, the absence of lab values, vital signs or medications were interpreted as care processes or tests that were not performed or medication that was not prescribed rather than missing values. The primary endpoints for differential change in A1C and SBP by study arm were analyzed using the index encounter SBP and the most recent A1C, and analysis was limited to those having a subsequent SBP or A1C in the 12-month follow-up period and additionally at least one week after the index encounter for A1C.

2.12. Evaluation of medical costs

The study’s secondary aim was to assess medical costs from the health system perspective. Medical care costs included the utilization of clinic services—including laboratory and physician services—and pharmacy fills incurred in the 12-month pre- and post-index date periods by participants in each study group. To calculate medical costs, we planned to use Total Care Relative Resource Values, a nationally representative and standardized set of pricing measures derived from CMS relative value units (RVUs).⁶⁶

A mixed-effects generalized linear regression model with a random intercept for each patient was planned to estimate costs by study arm while controlling for time (12 months prior to index vs. 12 months after index), a treatment by time interaction term, and baseline covariates of age, sex, and race/ethnicity (chosen a priori). The marginal effect of being assigned to the intervention group will provide an estimate of the incremental medical cost associated with the eCDC-CDS intervention.

Prediction of long-term health outcomes and cost-effectiveness was planned for using microsimulation analysis if statistically significant effects are found in the primary clinical endpoints (SBP and A1C control at 12 months). The HealthPartners Institute ModelHealth^™: Cardiovascular Disease microsimulation model will be used to estimate long-term outcomes and cost-effectiveness from the health system perspective.⁶⁷

Discussion

In the context of pragmatic trials conducted within busy primary care settings, much of what occurs in clinical care is outside the control of the research team.⁶⁸ While making substantial modifications to study design or intervention isn’t a routine practice, there are situations where such adjustments become necessary to uphold the original aims and hypotheses. This report details major protocol adaptations that were implemented to address unforeseen changes in the care delivery system including the Covid-19 pandemic.

One limitation of this study involves using objective PDC measures to evaluate patient adherence. PDC assesses availability of medications but does not measure actual medication taking behavior. This approach can potentially lead to overestimation of adherence compared to self-reported data by mistakenly classifying patients as nonadherent when there are appropriate clinical reasons for claims gaps.^69,70 PDC can also lead to underestimation of adherence issues if patients participate in automated mail-order refills or engage in “stock piling” of medications, trends that increased during the Covid-19 pandemic.^71,72 Despite the limitations of data driven adherence calculations, we opted for this method because of their availability without causing care disruption and the general acceptance of PDC for quality measurement purposes. Patient surveys conducted to assess self-reported adherence after the index visit could offer insight into the relative accuracy of PDC measures of adherence.

Starting in 2018, Epic PDC calculations obtained from external prescribing and dispense data improved clinician accessibility to PDC calculations for patients across various insurance products. This enhancement broadened the pool of eligible study patients beyond just HealthPartners insured patients with accessible claims data and improved the study power. This also made PDC information more accessible within the EHR for patients in usual care, but viewing it required multiple clicks. In contrast, the eCDC-CDS adherence intervention streamlined the process by alerting PCCs to potential adherence problems and including the PDC information on interfaces without additional clicks or searching. Furthermore, patients in the intervention group were informed of potential medication adherence concerns via the patient eCDC-CDS printed interface, a feature unavailable to patients in usual care who generally remained unaware of medication adherence calculations within the EHR. Consequently, the study outcomes will assess the incremental benefit of using PDC information for CDS beyond what is now typically viewable within EHR.

Originally, the protocol involved retail-based pharmacists for adherence outreach but due to unexpected closure of the involved pharmacies, the adapted protocol engaged clinical MTM pharmacists affiliated with the clinics. These MTM pharmacists brought more experience in chronic disease management and enabled higher quality outreach, but they were also more highly paid. To maintain budget neutrality, we restricted pharmacist outreach to intervention patients who at 6 months post-index visit had persistence of meeting suboptimal clinical thresholds and low medication adherence, and 38% of the study eligible patients met this criteria. The use of MTM pharmacists as part of the care team aligns with growing trends and is unlikely to impede potential scalability of the adherence intervention. A secondary analysis of the pharmacist documentation of determinants for poor adherence and actions taken will offer additional insights into the value of this intervention component.

The adherence intervention had multiple components including the printed interfaces, the EHR display, and the pharmacist outreach. While the multi-component intervention heightened the probability of improved outcomes, it may be difficult to discern which part of the intervention is most crucial if the study is positive. Conducting exploratory analysis with usage data may offer more clarity on the relative significance of the various components. Provider surveys conducted over the course of the study will also add insight into satisfaction with the CDS tools and the perceived value of the different intervention components.

The adaptations made during the study carried both advantages and disadvantages. Shifting from a clinic randomized trial to patient randomized trial guaranteed consistent allocation of the intervention or usual care to individual patients even amidst the clinic disruptions resulting from the pandemic. This transition also enhanced study power allowing for a shorter accrual period and helped to adhere to the original study timeline despite the 5-month suspension needed for adapting the protocol and preparing the eCDC-CDS for use in telehealth encounters.

The impossibility of blinding clinicians and patients to the intervention and the potential for intervention contamination at the PCC level are potential drawbacks of the design that could theoretically lead to improved care for patients in usual care and bias the study towards a null hypothesis. However, this effect would not undermine the results of the study in the case of positive outcomes.

Conclusion

This study was designed to assess whether heightened awareness of adherence to cardiometabolic medications in primary care settings could improve medication adherence and improve the care of patients with HTN, DM, or dyslipidemia. In response to major unforeseen disruptions to the care delivery system, adaptations to the study protocol were necessary to achieve the overall study goals. Remarkably, these adaptations were made within the original study timeline and without requiring additional funding, providing valuable insights into the process of modifying a study protocol while upholding scientific integrity and study aims. The study is poised to provide substantial insights into the augmentation of basic CDS with medication adherence information and the collaborative integration of pharmacists into caring for patients with cardiometabolic conditions.

Abbreviations

A1C

glycated hemoglobin

ACC/AHA

American College of Cardiology/American Heart Association

ASCVD

Atherosclerotic Cardiovascular Disease

BP

Blood Pressure

BPA

Best Practice Advisory

CMS

Centers for Medicaid and Medicare Services

CDC

Chronic Disease Care

CDS

Clinical Decision Support

CI

Confidence Interval

CV

Cardiovascular

DM

Diabetes Mellitus

DBP

Diastolic Blood Pressure

DSMB

Data Safety Monitoring Board

eCDC-CDS

Enhanced Chronic Disease Care Clinical Decision Support

EHR

Electronic Health Record

HEDIS

Healthcare Effectiveness Data and Information Set

HTN

Hypertension

ICD10

International Classification of Diseases 10th revision

ICER

Incremental Cost-effectiveness Ratio

ID

Identification Number

IMB

Information, Motivation, and Behavior

IRB

Institutional Review Board

LDL

Low Density Lipid

MTM

Medical Therapeutics Management

NHLBI

National Heart Lung and Blood Institute

PCC

Primary Care Clinician

PDC

Proportion of Days Covered

SBP

Systolic Blood Pressure