Direct-to-consumer Strategies to Promote Deprescribing in Primary Care: A Pragmatic Feasibility Study

Background – Deprescribing, or the intentional discontinuation or dose-reduction of medications, is a patient-centered approach to reduce harms associated with inappropriate medication use. We sought to determine how direct-to-patient educational materials impacted patient-provider discussion about and deprescribing of potentially inappropriate medications. Methods – We conducted a pre-post with historical control group pilot feasibility trial at an urban VA medical center. We included patients in one of two medication-based cohorts: 1) proton pump inhibitor (PPI), dened as use of any dose for 90 consecutive days, or 2) hypoglycemia risk, dened by diabetes diagnosis; prescription for insulin or sulfonylurea; hemoglobin A1c <7%; and age >65 years, renal insuciency, or cognitive impairment. The intervention consisted of mailing medication-specic patient-centered EMPOWER (Eliminating Medications Through Patient Ownership of End Results) brochures, adapted to a Veteran patient population, two weeks prior to scheduled primary care appointments. Our primary outcome – deprescribing – was dened as clinical documentation of target medication discontinuation or dose-reduction. Our secondary outcome was documentation of a discussion about the target medication (yes/possible vs. no/absent). Covariates included age, sex, race, specied comorbidities, medications, and utilization. We used chi-square tests to examine the association of receiving brochures with each outcome. Results – The 348 subjects (253 intervention, 95 historical control) were primarily age >65 years, white, and male. Compared to control subjects, intervention subjects were more likely to have deprescribing (36 [14.2%] vs. 4 [4.2%], p=0.009) and discussions about the target medication (31 [12.3%] vs. 1 [1.1%], p=0.001).


Background
Inappropriate medication use, when the potential risk outweighs potential bene t, has multiple negative consequences. Patients experience adverse drug events and associated sequelae, increased out-of-pocket expenses, and pill burden. (1)(2)(3) Harms from medications can lead to increased healthcare system utilization and overall costs. (4)(5)(6)(7) Efforts to reduce inappropriate medications traditionally have focused on reducing initial prescriptions (e.g., antibiotics for probable viral infections);(8) however, another effective approach is deprescribing. (9,10) This alternative strategy may be more widely applicable given the frequency with which inappropriate medications are currently used by patients. (11,12) Deprescribing is the intentional discontinuation of medications, with the decision occurring within the context of a patient's overall clinical status and integrating their goals and values for healthcare treatment.(10) Potential bene ts include improved medication adherence, patient satisfaction, and decreased costs to the patient and healthcare system. (13,14) While there are few identi ed harms from discontinuing medications and adverse drug withdrawal events are rare, (15,16) many patient and provider barriers to deprescribing preclude it from occurring as often as might be bene cial in clinical practice. (17)(18)(19)(20)(21) Patients may have concerns about symptom return or be reluctant to change medications, especially in the context of multiple clinicians or having been instructed about the importance of medication adherence. Meanwhile, providers often have limited time or feel pressures to meet performance metrics that disincentivize deprescribing, such as achieving strict hemoglobin A1c (HbA1c) goals in patients with diabetes. As a result, efforts to facilitate deprescribing have included interventions targeting healthcare systems, providers, patients, and combinations thereof. (22,23) Including the patient in deprescribing interventions frequently yields increased action to reduce medications. One mechanism to do so is analogous to direct-to-consumer advertising, with outreach targeted to patients. The EMPOWER (Eliminating Medications Through Patient Ownership of End Results) brochures exemplify this approach, capitalizing on cognitive dissonance and adult learning theory to educate and activate patients to discontinue medications under the guidance of a healthcare provider. (24,25) Direct-to-consumer mechanisms also reduce the onus on providers, who are frequently responsible for initiating clinical activities.
Two medication classes addressed by EMPOWER brochures are proton pump inhibitors (PPIs) and sulfonylureas. While PPIs are a potent and effective gastric acid reducing treatment for conditions such as Barrett's esophagus, peptic ulcer disease, and gastroesophageal re ux disease (GERD), they are often used at higher doses or for longer duration than indicated. (26,27) Similarly, sulfonylureas successfully reduce hyperglycemia in patients with diabetes; however, relative to other treatment options, they also place patients at higher risk of hypoglycemia and its potential harms (e.g., falls, loss of consciousness, seizures), especially among older adults for whom guidelines recommend less aggressive treatment targets. (28)(29)(30)(31)(32) With the ultimate goal of preventing medication-associated harm, we sought to determine how medicationspeci c EMPOWER brochures, adapted to a Veteran patient population, impacted discussion about and deprescribing of potentially inappropriate medications (PIMs).

Methods
Intervention -We used EMPOWER brochures, designed by the Canadian Deprescribing Network.(33) These brochures provide detailed information about the medication, allow patients to test their knowledge of the medication and indications for use, re ect on their own experiences with potential side effects, discuss alternative therapies (medication and nonpharmacologic options), and include a vignette of a patient who successfully discontinued the medicine. The decision aids were written at a 6th grade reading level and were based upon theories of patient activation (e.g., skills and con dence to enable patient engagement), adult learning (e.g., knowledge acquisition is immediately applicable), and cognitive dissonance (i.e., contradictory behaviors and/or beliefs). (24,(34)(35)(36) The brochures repeatedly emphasize that patients should not make any medication changes without rst consulting with their health care provider. We tailored both study brochures for the Veteran population; for example, we adapted images and only used generic medication names. We further expanded the sulfonylurea brochure to include insulin to align with the VA Hypoglycemia Safety Initiative efforts. (37) Lastly, we included items to promote re ection on personal health goals.

Study setting, population, and design
We conducted a pre-post feasibility trial of a patient-centered intervention at one campus of an urban VA medical center (VAMC). For this pilot study, we identi ed eligible subjects who met inclusion criteria for one of two medication-based cohorts and were scheduled for appointments two weeks in the future (index visit) in either Primary Care Clinics (n = 14 attending primary care providers [PCPs], 17 resident PCPs) or Women's Health Clinics (n = 3 PCPs).
The rst cohort comprised patients at risk for hypoglycemia due to potential overtreatment of diabetes. Inclusion criteria were 1) a diagnosis of diabetes based on International Classi cation of Diseases, Tenth Revision, Clinical Modi cation (ICD-10-CM) codes, 2) most recent HbA1c < 7%, and 3) prescription for either insulin or sulfonylurea, as well as 4) one or more of the following: age 65 or older; renal insu ciency, de ned as creatinine > 2 mg/dL; or cognitive impairment, de ned by a diagnosis of cognitive impairment or dementia and/or prescription for an acetylcholinesterase inhibitor (e.g., donepezil).
The second cohort included patients of any age prescribed a proton pump inhibitor (PPI) at any dose for 90 consecutive days. We did not exclude patients who may have had an indication to remain on a PPI (e.g., diagnosis of Barrett's esophagus), due to the di culty of determining PPI appropriateness from administrative data and because our study procedures promoted discussion of deprescribing rather than directly discontinuing the medication.
If a patient met inclusion criteria for both the hypoglycemia and PPI cohorts, s/he was placed in only the hypoglycemia risk cohort because the short-term risks from hypoglycemia are usually greater than those from chronic PPI use. Because we erred on the side of including patients who might be candidates for medication deprescribing, we acknowledge that the goal deprescribing rate would be less than 100%, since medication continuation would be appropriate for some study subjects.
Subjects were identi ed weekly during the 3-month intervention window (October 16, 2017 to January 14, 2018) and mailed the applicable EMPOWER brochure. No additional contact was made with the subject prior to the index visit. Subjects were eligible to participate in the study once, regardless of additional PCP visits during the study. Patients who did not attend the index visit were excluded from analyses. We identi ed an historical control group of patients seen in primary care in the month prior to the intervention (September 18, 2017 to October 15, 2017) who met study eligibility criteria; they received usual care.
Prior to the intervention window, PCPs received a brief informational overview of the study, along with educational material about deprescribing the targeted medications.

Study measures
We conducted a chart review of completed progress notes to identify our primary outcome of deprescribing, de ned as the clinical documentation that the dose of the target medication was decreased or the medication was discontinued completely. In contrast, deprescribing was considered absent when the target medication was unchanged or its dose increased. We also conducted chart reviews to assess our secondary outcome, namely the occurrence of patient-provider discussion about the target medication, operationally de ned as documentation of discussion about the target medication (yes or possible vs. no or absent).
Additional variables extracted from administrative data in the VA Corporate Data Warehouse included age; sex; race; and speci ed comorbidities, medications, and utilization in the year prior to or 20 days following the index visit that may have been an indication or contraindication for the target medication (e.g., use of antiplatelet agents may indicate a need for a PPI).(38)

Analyses
To determine the presence of non-response bias (i.e., subject non-attendance at index PCP visit), we began by evaluating differences between the intervention subjects in our study population (i.e., received brochure and attended index visit) with those excluded from analyses (i.e., received brochure but did not attend their index PCP visit). We then examined patient characteristics and compared the intervention to control subjects, using descriptive statistics and chi-square tests.
We used chi-square tests to examine the association of receiving brochures with our primary and secondary outcomes of deprescribing and discussions, respectively. We assessed differences in deprescribing by PCP with one-way analysis of variance (ANOVA). We then conducted strati ed analyses by medication cohort for the primary outcome, where we additionally examined patient characteristics associated with deprescribing.
All analyses were conducted with SAS statistical software, version 9.4.

Population description
We mailed brochures to 317 potential intervention subjects; of these, 253 attended the index PCP visit and were included in the study population. There were no statistically signi cant differences between subjects who were included vs. excluded from analyses (data not shown) with the exception of age ≥ 65 years (80.5% of included subjects vs. 67.9% of excluded subjects; p = 0.037), receipt of gastroenterology procedure (7.3% of included subjects vs. 16.1% of excluded subjects; p = 0.03), and diagnosis of esophagitis (0.4% of included subjects vs. 3.6% of excluded subjects; p = 0.03).
The entire study population of 348 patients, which included the 253 intervention subjects and 95 control subjects, was predominantly 65 years of age or older, white, and male (Table 1). There were no statistically signi cant differences between the intervention and control subjects with respect to clinical providers (e.g., trainee or women's health clinician) or select diagnoses and medication use related to the target medication. There were 296 subjects in the PPI cohort (80 control, 216 intervention) and 52 subjects in the hypoglycemia cohort (15 control, 37 intervention). Complete data were available for deprescribing outcomes; however, because not all progress notes were available for review, we could not determine for all subjects the occurrence of a patient-provider discussion about deprescribing the target medication.

Deprescribing and Discussion Outcomes
Among the entire population of 348 subjects, deprescribing was more common among intervention subjects compared to controls, with an absolute magnitude difference of 10% (36 [14.2%] vs. 4 [4.2%], p = 0.009, Fig. 1). Deprescribing rates differed between PCPs (p = 0.03); however, there were no differences between resident clinicians compared to attending physicians (p = 0.83) nor between Women's Health PCPs and general PCPs (p = 0.19) (data not shown). Deprescribing was more common among subjects who had a documented discussion about their medications compared to those who did not ( In strati ed analyses (Fig. 1 In strati ed cohort analyses, deprescribing was more common in PPI subjects with a diagnosis of gastroesophageal re ux disease (GERD) compared to those without documented GERD (19.2% vs. 6.3%, p = 0.0006). No other patient characteristic in the PPI cohort, nor any patient characteristic in the hypoglycemia cohort, was associated with deprescribing.

Discussion
Distributing EMPOWER brochures directly to patients in advance of an already-scheduled primary care visit successfully increased both deprescribing and documentation of patient-provider discussions of the target medication in a Veteran population. Given the common use of proton pump inhibitors and the frequency with which hypoglycemia-risk medications result in emergency department visits, a 10% absolute magnitude reduction in prescriptions at the population-level could potentially mitigate large numbers of adverse drug events and associated healthcare costs. EMPOWER brochures have demonstrated success to reduce potentially inappropriate use of multiple medication classes. The EMPOWER trial, which addressed benzodiazepines, and the D-PRESCRIBE trial, which targeted sedative-hypnotics, rstgeneration antihistamines, glyburide, or nonsteroidal anti-in ammatory drugs, found community pharmacy-based distribution of these educational materials to older adults led to signi cant reduction in prescriptions for a potentially harmful medication in the subsequent six months. (24,25) While we did not achieve the same absolute magnitude of deprescribing as these landmark trials, we used a different mechanism of information distribution, shorter follow-up time, and medications that may be viewed as having lower risk.
Our choice to send the material directly to patients highlights the value of providing patients with the knowledge and encouragement to be active participants in their health care decisions. (39) With increasing recognition of the importance of shared decision-making,(40) equipping patients to have thoughtful dialogue with their clinical providers is essential to providing patient-centered care. (41) The history of educational materials and decision aids to support thoughtful selection among equal options is well established for decisions across the healthcare continuum.(42) Acquisition of con dence and communication skills from medication-speci c EMPOWER brochures may foster patients to remain active agents in other healthcare decisions, consistent with the theories upon which their development was based. However, clinicians must also possess the knowledge, skills, and attitudes to respond to patient-initiated discussions of clinical decisions. In the context of safe medication use, this competency includes knowledge of safe deprescribing. Multi-faceted efforts that include components to overcome provider-level barriers would likely have even greater success. For example, embedding pharmacists in a primary care setting to provide face-to-face medication reviews resulted in a 20% acceptance rate of pharmacist recommendations for deprescribing. (43) Shared decision-making regarding medications, diagnostic testing, and treatments for chronic or non-urgent conditions does not necessarily require an immediate choice. Decisions in these circumstances could -and perhaps should -occur over multiple visits and time to allow for patients to process the information and consult with family and friends. To that end, receipt of EMPOWER brochures may have initiated a chain of events that would subsequently lead to deprescribing via provision of knowledge. Longitudinal interactions with primary care and other clinicians provide multiple opportunities for discussion and decision-making. (44) While brochures in our study were temporally linked to a scheduled visit in Primary Care clinics, there may have been more urgent or higher priority needs at that visit, deferring deprescribing of the target medication to a subsequent encounter. Understanding the effect of multiple opportunities to discuss and deprescribe on reduction of potential inappropriate medications would allow for identi cation of ideal timing of content delivery.
As evidence accumulates in support of EMPOWER brochures, the collection has expanded to include multiple different classes of medications.(33) The perception of danger or concern imparted with the brochure likely differs by medication class. Our choice of PPIs and medications with high risk of causing hypoglycemia may not be viewed by either patients or providers in the same manner they view benzodiazepines or other high-risk medications. While Veterans often obtain PPIs from the VA for cost reasons, their availability without a prescription potentially diminishes perceptions of risk.(45) On the other hand, the presence of performance metrics that evaluate providers on management of their patients with diabetes (e.g., glycemic control as measured by HbA1c) could disincentivize providers from reducing or discontinuing insulin or sulfonylureas, despite guidance for relaxed treatment targets.(46-48) Understanding the interaction of medication class with these complex decisions can yield more precise targeting of medications for which EMPOWER brochures foster deprescribing.
Several limitations should be noted. Because we mailed EMPOWER brochures to patients, we were unable to determine whether patients reviewed the information; however, this delivery mode was intentionally selected to minimize any burden on clinic staff or providers. Our outcome assessment was conducted by chart review, which was hindered by incomplete notes and poor documentation, as would be expected in any healthcare system. It is possible that patients were verbally instructed to reduce or stop medications without the change re ected in the medication ordering section of the electronic health record. Similarly, written progress notes may have omitted documentation of discussions. Conversely, if the provider took the time and effort to document a clinic-based conversation (without the use of templated text or clinical reminders), it seems reasonable that this documentation accurately re ects provider perceptions of the activities of the primary care visit.
Taken together, we likely had under-ascertainment of our outcomes, biasing our ndings toward the null. We only assessed outcomes up to 4 weeks after the visit; a longer follow-up period may have identi ed additional deprescribing events or, for those who had deprescribing, resumption of the target medication. Future studies should evaluate outcomes after a longer time interval to better understand the stability of deprescribing actions. Our ndings may not be generalizable to non-VA settings, as PPIs are lower cost in VA than when purchased over the counter (OTC), and guidelines from VA and the Department of Defense recommend relaxed glycemic targets compared to some professional societies.(48) However, the nancial consequences for patients paying out-of-pocket for OTC PPIs may increase the effect of the EMPOWER brochures in non-VA settings.

Conclusions
This single-site, time-limited study illustrates the feasibility and preliminary effectiveness of using direct-to-patient methods to promote deprescribing in VA settings. Targeted mailings of EMPOWER brochures temporally linked to a scheduled visit in primary care clinics is a low-cost, low-technology method to impart knowledge and leverage the ability of patients to drive change within clinical encounters and reduce prescribing of medications that may be unnecessary or place patients at harm. Discontinuing or reducing the dosage of risky medications has potential to reduce drug burden, decrease adverse drug effects and harms, and improve overall patient safety.

Declarations
Ethics approval and consent to participate -The study was approved by the VA Boston Healthcare System Institutional Review Board, who determined the study was minimal risk. They also approved a waiver of informed consent for recruiting subjects and a waiver of Health Insurance Portability and Accountability Act (HIPAA) authorization for data acquisition from electronic health records (i.e., chart review). Informed consent to participate was only applicable to components of the study not described in this manuscript. Had a GI procedure n/a n/a n/a 5 (6.3) 17 (7.9) 0.64 n/a n/a n/a Was seen in a GI clinic n/a n/a n/a 19 (23.8) 76 (35.2) 0.061 n/a n/a n/a Diagnosis of GI condition n/a n/a n/a 29 (36.3) 106 (49.1) 0.049 n/a n/a n/a Glucocorticoid use n/a n/a n/a 3 (3.8) 19 (8.8) 0.21 n/a n/a n/a NSAID use n/a n/a n/a 45 (56.3) 111 (51.4) 0.46 n/a n/a n/a Clopidogrel use n/a n/a n/a 3 (3.8) 11 (5.1) 0.77 n/a n/a n/a Insulin (rather than sulfonylurea) use n/a n/a n/a n/a n/a n/a 10 (66.7) 15 (40.5) 0.088 Diagnosis of renal Impairment (Creatinine >2mg/dL) n/a n/a n/a n/a n/a n/a 6 (40) 5 (13.5) 0.034 Diagnosis of cognitive impairment or dementia n/a n/a n/a n/a n/a n/a 1 (6.7) 5 (13.5) 0.66 Visit to an endocrine clinic n/a n/a n/a n/a n/a n/a 7 (46.7) 7 (18.9) 0.041 History of hypoglycemia n/a n/a n/a n/a n/a n/a 0 0 --*All diagnoses, clinic visits, and procedures identi ed in the year prior to the index visit. BMCFPEmpowerPilotSTROBEchecklistv4.docx