Judith A. Long, MD

To assess physician views and perceived adoption of overtreatment guidelines and measure whether adoption of these guidelines influenced the recommendation of a targeted service.

A cross-sectional survey mailed from July 2014 to January 2015 to 902 internists who completed residency between 2003 and 2013, randomly selected from the American Medical Association Masterfile.

Poisson regression was used to model the rate of recommending a targeted service included in the guidelines, based on the level of guideline adoption. 

A total of 456 physicians responded (51% response rate). Most expressed familiarity with overtreatment guidelines (88.5%), a comfort level with discussing these guidelines with patients (79.9%), and described overtreatment guidelines as a useful tool in their practice (81.6%). Physicians in the highest tertile of guideline adoption reported double-digit rates of recommending antibiotics for sinusitis (29.7%), mammogram at end of life (16.5%), and electrocardiogram testing for asymptomatic patients (11.0%). Physicians in the bottom tertile of guideline adoption reported lower rates of recommending x rays (—12.0%; 95% confidence interval [CI], –19.4% to –4.5%; P = .002), magnetic resonance imaging for lower back pain (–4.8%; 95% CI, –8.1% to –1.5%; P = .004), and cardiac testing for asymptomatic patients (–10.2%; 95% CI, –18.9% to –1.5%; P = .02). 

US internal medicine physicians who completed residency between 2003 and 2013 reported high levels of adoption of overtreatment guidelines. Physicians who reported the highest levels of guideline adoption reported recommending services targeted by these guidelines in their practice.

We surveyed US internists who graduated residency in 2003 through 2013 about their views of overtreatment guidelines or recommendations against the use of potentially unnecessary tests and procedures. Key findings: 

US internists reported high levels of awareness, agreement, and use of overtreatment guidelines. 

Even physicians who reported the highest levels of guideline adoption, however, reported recommending services targeted by the guidelines in their practice. 

This research highlights the challenge of evidence-based de-implementation of medical tests and treatments in everyday practice.

Overtreatment in medicine, defined as “the waste that comes from subjecting patients to care that, according to sound science and the patients’ own preferences, cannot possibly help them,”

 is estimated to account for nearly 30% of healthcare spending.

 Increasing awareness that diagnostic and therapeutic interventions that physicians order are in some instances unnecessary

 has culminated in widely disseminated overtreatment guidelines, such as the Choosing Wisely campaign.

 Introduced in 2012, Choosing Wisely, an initiative of the American Board of Internal Medicine (ABIM) Foundation, has partnered with Consumer Reports and other medical organizations to provide physicians and patients with lists of potentially avoidable tests, treatments, and procedures.

In a 2014 telephone survey of 600 physicians, 38% of respondents reported having seen or heard about the Choosing Wisely campaign and 81% reported feeling “very comfortable” about “talking to patients about why they should avoid an unnecessary test or procedure.”

Despite the positive response from practicing physicians, there is little evidence that guidelines alone influence physicians’ ordering decisions. In fact, a recent report using commercial health plan claims data to evaluate the utilization of 7 services targeted by the guidelines failed to detect a meaningful decline in their use.

 However, the study looked at global use of services by health plan beneficiaries without accounting for physician characteristics. For example, a 2012 study of Massachusetts health plan data revealed that physicians with fewer than 10 years’ experience had the highest cost profiles compared with those of more senior physicians.

 Whether physicians’ awareness of overtreatment guidelines reduces the propensity to recommend a targeted service remains unknown.

To explore possible explanations behind the higher cost profiles of less experienced physicians, we surveyed recent internal medicine residency graduates about their adoption of overtreatment guidelines. Our specific objectives were to: 1) assess physician views of overtreatment guidelines using a novel 5-item scale, 2) estimate self-perceived practices according to select guidelines using hypothetical patient presentations, and 3) measure whether perceived adoption of the guidelines correlates with the likelihood to recommend a targeted service.

A literature review revealed no previously validated instruments evaluating physician attitudes toward overtreatment guidelines. To identify potential items for cognitive testing, we reviewed the literature, combed references from previously reported studies of physician views,

 and interviewed experts. Items were developed to assess: 1) physician awareness, agreement with, and use of overtreatment guidelines; 2) self-perceived propensity to recommend a service targeted by the guidelines; and 3) other potential confounders of physician practice identified in prior studies. We conducted 2 cycles of cognitive pilot testing to calibrate the wording to detect differences among physicians about these topics. After initial “think aloud” reviews with local (ie, Pennsylvania) practicing physicians, followed by revisions, we performed broader pilot testing with 100 internal medicine physicians randomly selected from the American Medical Association (AMA) Masterfile.

The final survey included questions about physician demographics, practice characteristics, attitudes known to influence overtreatment, views on overtreatment guidelines (awareness of, agreement with, and usefulness in practice), and self-reported practice in specific clinical scenarios related to the guidelines (eAppendix A [

]). Self-reported practice was assessed using fill-in-the-blank questions asking physicians to estimate the percentage of their patients they recommended for a specific test or treatment. Specifically, respondents were presented with brief descriptions of patient presentations for: 1) low back pain, 2) acute sinusitis, 3) cancer screening with a life expectancy of fewer than 10 years, 4) cardiac screening in asymptomatic routine care, and 5) a low pretest probability of venous thromboembolism (VTE). For example, when asked “For what percentage of patients with acute lower back pain do you order the following?” the respondent would fill in a percent for x-ray, magnetic resonance imaging (MRI), physical therapy, acetaminophen/anti-inflammatories, and opioids. The fill-in-the-blank questionnaire regarding treatment decisions has been shown to have high criterion validity (ie, it correlates with actual practice on similar patients) in prior studies.

To establish content validity, these items were tested by 11 clinical and survey design experts, including practicing primary care clinicians, researchers, and experts in survey design.

Using the AMA Masterfile, we pre-screened 2170 randomly selected internal medicine physicians who completed training within the last 10 years to confirm qualifying specialty, mailing address, and that the physician was actively seeing patients at least 20 hours a week. The final sample included 902 internal medicine physicians who were mailed a paper survey between July 2014 and January 2015 using a modified Dillman method.

 The initial mailing was done by first class mail accompanied by a $2 bill and followed by 2 reminder mailings approximately 6 weeks apart.

A set of 9 questions comprised the Overtreatment Guidelines Adoption (OGA) Scale, which assessed physicians’ attitudes toward overtreatment guidelines and cost containment in general. Six questions focused on awareness of, agreement with, and perceived usefulness of overtreatment guidelines; comfort denying patient requests for tests or treatments; comfort discussing costs with patients; and self-perception of cost consciousness. These were assessed using a 4-point Likert scale, ranging from strongly disagree to strongly agree. A second set of 3 questions assessed how frequently physicians discussed costs, used the overtreatment guidelines in practice, and found these guidelines to be useful. These were measured using a 5-point Likert scale of frequency.

To summarize overtreatment guidelines, their adoption, and measure physician attitudes toward guidelines separately from general attitudes toward cost containment, we developed 2 subscales: a 5-item OGA subscale and a 4-item cost-containment subscale, using standard factor analysis techniques (eAppendix B). The OGA subscale possible values ranged from 5 to 22; higher scale scores reflected a higher degree of adoption of overtreatment guidelines. The OGA subscale had high internal consistency with Cronbach alpha of 0.82 and rotated loadings of 0.44 to 0.75. Principal components analysis supported a separate cost-containment subscale of 4 questions related to costs (Cronbach alpha, 0.76; rotated loadings, 0.51-0.70).

Our main outcome measures were self-reported percentages of patients who were advised to elect 8 services targeted by 5 overtreatment guidelines. The guidelines were selected because they described common clinical scenarios in internal medicine, had been released at least 2 years prior to our survey, and were endorsed by multiple medical groups (

). We asked physicians to fill in the blank with the percentage of their patients they advised to receive a particular service. The options included services targeted by overtreatment guidelines, as well as other management options commonly offered to patients in each clinical context. The following tests and treatments were measured: x-ray and MRI imaging for acute low back pain; antibiotics for mild to moderate sinusitis; breast, prostate, and colon cancer screening for patients with a life expectancy of fewer than 10 years; electrocardiogram (EKG) testing for asymptomatic patients; and computerized tomography scan as the initial test for low-risk patients with possible VTE. Services that were recommended to less than 5% of patients (eg, Papanicolaou test for cervical cancer and stress test for cardiac testing in asymptomatic patients) were excluded from analysis.

Physician demographics, attitudes, reimbursement, and practice characteristics that may confound the relationship between physician views of, and practice according to, overtreatment guidelines, were included in the analysis. Physician demographics included age, gender, and race. Other physician characteristics included practice region, type of practice, compensation type, financial incentives (eg, quality, patient satisfaction, utilization review, productivity), insurance mix (eg, patients with Medicaid insurance), and attitudes (eg, comfort with clinical uncertainty, satisfaction with the practice of medicine in general, malpractice concerns). These items were either drawn from the AMA Masterfile (ie, age and gender) or included questions drawn from previously validated surveys of physicians.

Responses were entered in the REDCap electronic data capture tool (Harvard Catalyst; Boston), hosted at the University of Pennsylvania.

 Ten percent of entries were double-entered with perfect concordance. The data were exported in, and all analyses were conducted, using STATA version 13.0 (StataCorp; College Station, Texas).

We used the American Association for Public Opinion (Research Response Rate 2) definition.

Nonresponse bias was assessed by comparing respondents with nonrespondents and early to late respondents using the Pearson χ

The reported percentages of patients who were recommended for a particular test or treatment indicated a discrete number of events over a constrained range (0%-100%) and were positively skewed. Thus, the reported percentages were converted to a count variable based on a denominator of 100 (ie, 10% was converted to 10 out of 100) and modeled using Poisson regression. The independent variable of interest was a trichotomized OGA scale. Other variables in the model included a scale of physician attitudes toward cost containment in general (measured using the cost-containment subscale), physician demographics, practice characteristics, and attitudes previously shown to be associated with overuse (see “Other Variables” section of text). The predicted percentage of patients recommended for a particular test or treatment were estimated. Bootstrapping with 1000 iterations was used to estimate 95% confidence intervals (CIs). This study was reviewed and approved by the University of Pennsylvania Institutional Review Board.

Of the 902 potential respondents, 456 (51%) returned a completed survey. No differences between respondents and nonrespondents were observed by age, gender, region of current practice, or practice setting (

). Aside from Asian or Asian American respondents being overrepresented among late responders, there were no differences between early and late responders regarding gender, primary compensation, organization or setting of practice, or self-reported attitudes or satisfaction with medicine as a practice (eAppendix Table 1). Nearly half of the respondents self-characterized primary compensation type as salary with bonus (49.5%), followed by billings (28.1%) and salary only (20.9%), and the majority reported compensation linked to quality of care (62.9%) or productivity (65.1%) (

). Less than 5% of respondents (4.2%) completed residency in 2013. Other characteristics of the respondents’ practices are reported in Table 3.

Respondents’ attitudes toward cost containment are shown in 

. Most (88.5%) considered their practice style to be cost conscious. One in 4 (25.1%) reported discomfort discussing costs of care with patients, and 34.7% said they would not feel comfortable making a patient unhappy by denying a request for unnecessary care.

Respondents generally reported high levels of awareness, familiarity, and use of overtreatment guidelines (Table 4). Most (88.5%) reported being familiar with overtreatment guidelines in their specialty, 81.6% reported that the guidelines were useful in their practice, and 79.9% said they felt comfortable bringing up overtreatment guidelines in discussions with patients. However, less than 30% of respondents rated their agreement with these statements as “strong.” Respondents generally reported using overtreatment guidelines in practice with high frequency: 30.9% reported bringing up the guidelines in discussions with patients “frequently” or “always” and 44.2% reported bringing up the guidelines “occasionally.” Approximately 40% of respondents (41.1%) found the guidelines useful in practice “frequently” or “always,” and 42.4% found them “occasionally” useful. When individual responses were combined in the 5-item OGA subscale, the mean scale score was 15.6 (SD = 3.0) and the median 16 (interquartile range [IQR] = 14-18; observed range = 5-22).

In the fully adjusted models, respondents in the middle or top third of OGA subscale scores reported lower rates of recommending a test or treatment targeted by the guidelines for imaging for lower back pain, antibiotics for sinusitis, and cardiac testing for asymptomatic patients compared with the respondents in the bottom third of OGA scores (

). Physicians in the highest tertile of guideline adoption reported double-digit rates of recommending antibiotics for sinusitis (29.7%), mammogram at end of life (16.5%), and EKG testing for asymptomatic patients (11.0%). Physicians with OGA scores in the top third had significantly lower predicted rates of recommending x-rays (—12.0%; 95% CI, –19.4% to –4.5%; 

= .002) or MRI (—4.8%; 95% CI, –8.1% to –1.5%; 

= .004) for lower back pain and EKG for asymptomatic patients (—10.2%; 95% CI, –18.9% to –1.5%; 

= .02) compared with physicians in the bottom third of OGA scores. Physicians with OGA scores in the middle third also had lower predicted rates of recommending antibiotics for sinusitis (—6.9%; 95% CI, –13.0% to –0.8%; 

= 0 .03) and EKG for asymptomatic patients (—8.7%; 95% CI, –15.9% to –1.4%; 

= .02) compared with physicians in the bottom third of OGA scale scores. The differences in predicted probabilities across the tertiles of OGA scale scores were not significant for cancer screening and imaging as the initial test for patients at a low risk of VTE (Figure).

The association between physician cost consciousness and the percentage of patients recommended for a test or treatment targeted by the guidelines was not consistent: physicians in the top third of cost-consciousness scale scores reported lower rates of prescribing antibiotics for sinusitis and recommending mammography at the end of life, but this association was not observed for the other guidelines (eAppendix Table 2). Other factors associated with recommending services targeted by the guidelines were physician age, practice region, type and setting, practice that treated patients with Medicaid, and satisfaction with medicine as a profession.

In this survey study of physician views of overtreatment guidelines, internal medicine physicians generally reported high levels of awareness, agreement, and use of the guidelines in everyday practice, and their attitudes toward the guidelines were distinct from their attitudes toward cost containment. In addition, physicians who reported greater adoption of overtreatment guidelines recommended fewer tests or treatments targeted by some overtreatment guidelines, even after accounting for their overall cost consciousness. Physicians who reported the highest levels of guideline adoption, however, also reported recommending services targeted by the guidelines in their practice.

Although most physicians generally reported agreement with overtreatment guidelines, only about one-third of the respondents rated their agreement as strong or reported using the guidelines frequently, suggesting considerable ambiguity in their attitudes toward overtreatment. Consistent with this finding, recommended rates of some of the services targeted by the guidelines (eg, x-rays for lower back pain, antibiotics for acute sinusitis) were high even for physicians in the top third of overtreatment guidelines adoption. On the other hand, most respondents (88.5%) assessed their practice style as cost-conscious. These findings suggest that even among physicians who generally had positive attitudes toward cost containment, perceptions of the use of overtreatment guidelines were poor, potentially limiting their impact on physician behavior.

Considering these findings, the lack of a consistent decrease in the use of tests and treatments targeted by the Choosing Wisely campaign is not surprising.

 Of note, although some of the guidelines (eg, cancer screening) categorically recommend against testing when patients meet certain criteria, many guidelines implicitly or explicitly allow for exceptions (eg, for worsening symptoms or prolonged duration of acute sinusitis). These important distinctions were difficult to capture in a survey question that did not ascertain how frequently physicians saw patients that meet the exclusion criteria in the guidelines. Nevertheless, it is unlikely that the variation and high rates of targeted services reported by some of the respondents would be fully explained by variation in case mix.

The 4 services for which we did not observe an association with the OGA scale scores (ie, mammography, colonoscopy, prostate cancer screening, and imaging for VTE) were targeted by guidelines that did not include exceptions in certain patient presentations or for duration of symptoms. This contrasts with the other 4 services that were targeted by guidelines that were worded to include exceptions for certain patient presentations (ie, antibiotics for acute sinusitis, which recommended against ordering antibiotics unless symptoms lasted longer than 7 days or worsen, or for acute lower back pain that is nonspecific). This suggests that guidelines that are more categorically worded may be less likely to influence physician behavior. However, our study was not powered to determine the significance of this pattern. Future research should evaluate the effect of guideline wording on physician behavior.

The 8 tests and treatments evaluated in this study were selected to correspond to recommendations of the Choosing Wisely campaign, which had advantages, including widely disseminated endorsement by multiple professional physician organizations. All recommendations included in the study were proposed by 3 or more specialty groups. The Choosing Wisely campaign leaves the mechanism of endorsement up to the group, emphasizing the grassroots characteristics of the campaign. Specialty groups play a lead role in developing the lists of recommendations, an approach designed to appeal to physician professionalism and establish specialty-endorsed norms of care. However, a review of the recommendations by the first 25 professional groups that participated in the Choosing Wisely program raised concerns that groups may be reluctant to endorse recommendations limiting the use of services that are highly lucrative to the specialty.

 Furthermore, the extent of regional and local professional groups’ involvement in the development of national specialty societies’ Choosing Wisely recommendations is not clearly mandated by the campaign. Hence, regional variation in the propensity of physicians to recommend some services may be less responsive to guidelines endorsed at the national level. Although practice region was significantly associated with only 1 of the 8 services evaluated (EKG for asymptomatic patients), local costs may influence physicians’ recommendations of specific tests and treatments. Future studies should assess how physician perceptions of costs influence their recommendations of services targeted by overtreatment guidelines.

Even in cases in which relatively strong consensus exists regarding the evidence base for optimal care, such as the overtreatment guidelines evaluated in this study, a complex interplay of working environment and personal factors plays a role in physician recommendations.

 Whereas overtreatment guidelines target intrinsic motivation in practicing evidence-based care, policy-level interventions typically focus on extrinsic motivators, such as value-based payments, bundling of payments, or other types of monetary incentives.

 Our findings provide empiric evidence supporting the importance of evaluating the effect of intrinsic and extrinsic motivators on physician behavior within the context of practice environment and physician characteristics.

While a mix of incentives could be calibrated to achieve value-based care in theory, in practice, these factors are in flux and conflict with each other at times. Although the current study evaluated the adoption of overtreatment guidelines within the context of environmental (ie, treatment facility), practice, and physician-level factors,

 we were unable to evaluate actual physician practice or compare the relative effect of alternative motivators. Behavioral theory suggests that getting physicians to “de-adopt” practices is more challenging than the adoption of healthcare innovations.

Moreover, physicians often lack self-awareness of the nonclinical factors that may influence their behavior.

 Although overtreatment guidelines that are evidence based and disseminated in a transparent way may be successful in engaging physicians to consider these issues, the sheer magnitude of factors that influence physician behavior suggests that overtreatment guidelines alone are unlikely to produce a sizeable impact on overuse.

Alternative explanations of the observed associations between overtreatment guideline adoption and the rate of recommending targeted services include patient case mix, social desirability bias (ie, underreporting undesirable behaviors, such as the use of services targeted by the guidelines in our study), and recall bias. Although case vignettes with open-ended answer options have high criterion validity (ie, they correlate with actual practice on similar patients),

 reported practices may not represent actual physician recommendations to their patients. Furthermore, the proximity of questions about practice patterns and overtreatment guidelines in the questionnaire may have primed respondents to underreport overtreatment.

Concerns about priming and desirability bias suggest that the rates of recommending services targeted by the overtreatment guidelines may be underestimated in this study. Although we obtained a relatively high response rate for a physician survey, and our respondents were similar to the general population (ie, nonrespondents and early and late respondents), the potential for response and selection bias remains.

Lastly, despite efforts to confirm physician eligibility during prescreening, specialty and contact information in the AMA Masterfile may be inaccurate, introducing respondents who fall outside our target sample. A recent comparison of AMA Masterfile physician contact information with other databases found that only 37% were accurate.

In a national survey, the majority of US internal medicine physicians reported positive attitudes toward overtreatment guidelines in their specialty. However, physicians’ recommendations in guideline-specific standardized patient cases varied. Physicians’ propensity to recommend low-value services was explained in part by physician and practice characteristics. The complexities of physician decision making may explain an observed lack of reduction in the utilization of tests and treatments targeted by such widely disseminated overtreatment guidelines as the Choosing Wisely initiative. Guidelines or similar broad educational interventions by physician organizations are unlikely to reduce physician-level variation in the utilization of low-value services. Furthermore, interventions to reduce low-value care should be evaluated within the context of health system-, practice-, and physician-level factors to avoid unanticipated effects.&ensp;

 Division of General Internal Medicine (KLR, RMW, JAS, JAL), and Leonard Davis Institute of Health Economics (KLR, RMW, JAS, JAL), University of Pennsylvania, Philadelphia, PA; Accreditation Council for Graduate Medical Education (ESH), Chicago, IL; American Board of Internal Medicine (EB), Philadelphia, PA; Corporal Michael J. Crescenz VA Medical Center (RMW, JAL), Philadelphia, PA.

 This project was supported by the Health Resources and Services Administration (HRSA) of the US Department of Health and Human Services (HHS) under the Ruth L. Kirschstein National Research Service Award (NRSA) T32-HP10026. Content and conclusions are those of the author and should not be construed as the official position or policy of, nor should any endorsements be inferred by HRSA, HHS, or the government.

 Dr Holmboe receives royalties from a textbook on assessment. The remaining authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

 Concept and design (KLR, ESH, EB, JAS, JAL); acquisition of data (KLR, JAL); analysis and interpretation of data (KLR, ESH, RW, JAS, JAL); drafting of the manuscript (KLR, ESH, EB, JAS, JAL); critical revision of the manuscript for important intellectual content (KLR, ESH, EB, RW, JAL); statistical analysis (KLR, RW, JAL); provision of patients or study materials (KLR, JAS, JAL); obtaining funding (KLR); administrative, technical, or logistic support (KLR, ESH, EB, JAL, JAS); and supervision (JAL).

 Kira Ryskina, MD, MS, 12-30 12th Fl, Blockley Hall, 423 Guardian Dr, Philadelphia, PA 19104. E-mail: ryskina@mail.med.upenn.edu. 

1. Berwick DM, Hackbarth AD. Eliminating waste in US health care. JAMA. 2012;307(14):1513-1516. doi: 10.1001/jama.2012.362.

2. Smith MD, Saunders R, Stuckhardt L, McGinnis JM, eds; Committee on the Learning Health Care System in America, Institute of Medicine. Best Care at Lower Cost: The Path to Continuously Learning Health Care in America. Washington, DC: National Academies Press; 2012.

3. Yong PL, Saunders RS, Olsen L; Institute of Medicine (US); Roundtable on Evidence-Based Medicine. The Healthcare Imperative: Lowering Costs and Improving Outcomes: Workshop Series Summary. Washington, DC: National Academies Press; 2010.

4. Choosing Wisely: an initiative of the ABIM Foundation. Choosing Wisely website. http://www.choosingwisely.org/. Accessed October 2015.

5. PerryUndem Research/Communication. Unnecessary tests and procedures in the health care system: what physicians say about the problem, the causes, and the solutions. Choosing Wisely website. http://www.choosingwisely.org/wp-content/uploads/2015/04/Final-Choosing-Wisely-Survey-Report.pdf. Published May 1, 2014. Accessed June 2016.

Articles

In a national survey, US internists reported high levels of adoption of overtreatment guidelines; despite this fact, they also reported recommending services targeted by the overtreatment guidelines.

US Internists' Awareness and Use of Overtreatment Guidelines: A National Survey

Value-based insurance designs are being widely used. We undertook this study to examine whether a financial incentive that lowered co-payments for blood pressure medications below $0 improved blood pressure control among patients with poorly controlled hypertension.

Participants from 3 Pennsylvania hospitals (n = 337) were randomly assigned to: a) be paid $8 per medication per month for filling blood pressure prescriptions, b) a computerized behavioral intervention (CBI), c) both payment and CBI, or d) usual care. The primary outcome was change in blood pressure between baseline and 12 months post enrollment. We also measured adherence using the medication possession ratio in a subset of participants.

There were no significant interactions between the incentive and the CBI interventions. There were no significant changes in medication possession ratio in the treatment group. Blood pressure decreased among all participants, but to a similar degree between the financial incentive and control groups. Systolic blood pressure (SBP) dropped 13.7 mm Hg for the incentive group versus 10.0 mm Hg for the control group (difference = —3.7; 95% CI, –9.0 to 1.6; 

 = .17). The proportion of patients with blood pressure under control 12 months post enrollment was 35.6% of the incentive group versus 27.7% of the control group (odds ratio, 1.4; 95% CI, 0.8-2.5; 

 = .19). Diabetics in the incentive group had an average drop in SBP of 12.7 mm Hg between baseline and 12 months compared with 4.0 mm Hg in the control group (

 = .02). Patients in the incentive group without diabetes experienced average SBP reductions of 15.0 mm Hg, compared with 16.3 mm Hg for control group nondiabetics (

Among patients with poorly controlled blood pressure, financial incentives—as implemented in this trial&mdash;did not improve blood pressure control or adherence except among patients with diabetes.

Improving medication adherence through small rewards that lower co-payments below $0 may be effective in subgroups of the population, but may not improve outcomes overall. Consideration should be given to further testing of this approach in carefully selected high-risk populations in which increased adherence could have significant economic and health benefits.

Insurers are widely adopting value-based insurance design (V-BID), which is based on the premise that reducing costs will significantly increase the use of beneficial and cost-effective services. These approaches are seen as a way of trying to address widespread problems with adherence to medication for chronic diseases, as there is strong evidence that medication adherence for such chronic diseases as hypertension and hypercholesterolemia is low,

 limiting the potential for medications with high efficacy in randomized controlled trials to improve the health of the population. V-BID designs are also being used more broadly to increase the use of high-value services and reduce the use of low-value services through differential co-payments tied to the value of services. However, although observational studies have consistently shown that increases in co-payments are associated with both decreases in utilization of medications and worse outcomes,

 the impact of decreasing co-payments seen in observational studies has been more modest

; the underlying psychology of how people process changes in payments as losses compared with gains suggests that increases and decreases in co-payments may not be equivalent.

Nearly two-thirds of Americans with hypertension have poor control,

 which puts them at risk for substantial morbidity and mortality; poor adherence is an important factor in poorly controlled hypertension. Taking the logic behind V-BID initiatives that lower financial burden might improve adherence and patient outcomes one step further; we examined whether providing incentives of $8 per medication per month (a type of “negative co-payment”) for all antihypertensive medications would significantly improve blood pressure control among patients with poorly controlled blood pressure at 3 medical centers in Pennsylvania.

Study participants were drawn from patients at 3 hospitals in Pennsylvania: the Philadelphia Veterans Affairs Medical Center (PVAMC), the Veterans Administration Pittsburgh Health Care System (VAPitt), and the PinnacleHealth clinic in Harrisburg, with recruitment occurring between March 2005 and July 2007. (

 shows the study flow.) Potential participation was elicited by sending letters to patients who met eligibility criteria based on electronic or manual screening of records. Eligible patients were aged ≥21 years, with 1 or more active prescriptions for an antihypertensive medication, and systolic blood pressure (SBP) of at least 140 mm Hg (130 mm Hg in patients with diabetes), and with eligibility to receive medications without co-payments (due to low income or disability). Exclusion criteria included participation in another experimental study, markedly shortened life expectancy (due to diagnosis of metastatic cancer, end-stage renal disease on dialysis, New York Heart Association class IV congestive heart failure, or dementia), or atrial fibrillation (because of concerns with accuracy of blood pressure measurement). We enrolled 337 of the 1253 potentially eligible participants in the study (see Figure 1); approximately 20% of screened potentially eligible patients were excluded due to ineligibility.

The protocol was approved by the Institutional Review Boards of the PVAMC, VAPitt, PinnacleHealth, and the University of Pennsylvania, and all participants provided written informed consent prior to randomization. The study was registered at clinicaltrials.gov as Collaboration to Reduce Disparities in Hypertension, ID #NCT00133068. Participants were randomized to receive either: a) a financial incentive of $8 per medication prescription filled per month, b) a computerized behavioral intervention (CBI) provided at enrollment and at the 6-month follow-up visit, c) both the financial incentives and the CBI, or d) usual care (with study follow-up visits every 3 months). After an initial visit, participants were requested to return for follow-up blood pressure readings and surveys at 3, 6, 9, and 12 months; financial rewards were paid at each follow-up visit after confirmation that each prescription was filled using either the VA’s computer records, a prescription bottle, or a receipt. The CBI was designed to help study participants learn the basics about blood pressure control and its impact. Participants watched a video testimonial from a patient who had had multiple strokes due to medication nonadherence, and who felt he had learned the importance of medication adherence from that experience. This helped participants improve self-efficacy by reviewing techniques that could be used to enhance adherence.

Randomization was carried out using a random number generator and via permuted block randomization with a block size of 4. Randomization was stratified by site, income (<100%, 100%-200%, 200%-300%, and >300% of the federal poverty level [FPL]), and baseline blood pressure (SBP <160 mmHg or SBP ≥160 mm Hg). Randomization was performed after signed written consent forms were received. Allocation assignments were concealed, with staff unable to access randomization assignment for each subject until all eligibility criteria were entered in an electronic tracking system and consent forms were completed. Neither staff nor study participants could be blinded due to the nature of the intervention; investigators and analysts, however, remained blinded to intervention assignments until unblinding occurred, in coordination with the Data Safety Monitoring Board, once follow-ups were nearly complete.

The primary outcome variable was change in blood pressure from enrollment to 12 months post enrollment. Secondary outcome variables included change in blood pressure 6 months post enrollment, the percentage of patients with blood pressure in control at 6 and 12 months post enrollment, self-reported medication adherence, and prescription refill data from the VA electronic medical record system. Blood pressure control was defined as SBP below 140 mm Hg and diastolic blood pressure (DBP) below 90 mm Hg for patients without diabetes; for participants with diabetes, blood pressure control was defined as SBP below 130 mm Hg and DBP below 85 mm Hg.

Measurement of blood pressure was done following a standardized protocol using an automated blood pressure cuff (Omron HEM-90R), ensuring that the correct cuff size was used.

 Participants were instructed to relax for 5 minutes before their blood pressure was taken; then the patient’s arm (the dominant arm unless the patient expressed a preference to use the other arm) was supported on a chair or desk and the blood pressures were measured while the patient was sitting. Three measurements were taken, 2 minutes apart, and averaged. The blood pressure measurements were not revealed to the study participants. Although the study nurses could not be blinded to the randomization, the use of an automated blood pressure cuff and a standardized protocol protected against systematic differences among groups in the way blood pressure was measured.

Medication adherence was measured using self-reporting based on the Hill-Bone Scale,

 with supplemental assessment using electronic prescription fill records where available. For these records, we calculated medication possession ratios (MPRs; number of days a patient had a filled prescription divided by 360 days) and gap ratios of 30, 60, and 90 days (percentage of patients who had gaps in filled prescriptions of at least 30, 60, or 90 days).

Baseline blood pressure levels were assessed with other factors including height, weight, and creatinine level. We also collected information on income, baseline health status, health history, medication use, age, gender, and self-reported race or ethnicity. We used information on income and family size to calculate income as a percent of the FPL.

To evaluate the similarity of the treatment groups with respect to baseline covariates, we compared groups using Student’s 

 test for categorical variables, with Fisher’s Exact tests used for analyses with 5 or fewer subjects per cell. The trial was designed to test the effects of the 2 interventions individually; we did not assume an interaction, but powered the study adequately for individual comparisons of the 3 active arms against control. The primary analysis of the financial incentive began with a formal statistical test of interaction between the incentive and CBI; there was no evidence of this, so we collapsed across the CBI arms to assess the effect of the financial incentive alone.

The primary unadjusted analyses tested the mean differences in the degree of change in SBP and DBP between the incentive and control groups from baseline to 12 months post enrollment using Student’s 

 test. We similarly calculated differences in change in blood pressure from baseline to 6 months post enrollment. Some participants had missing blood pressure values at 6 and 12 months; we compared baseline data between participants with and without follow-up values. Missing values for 6-month and 12-month SBP and DBP readings were handled using the Markov Chain Monte Carlo multiple imputation method, utilizing 10 imputations after verification that this yielded relative efficiencies of 95% or higher; we also checked that imputed values were within the range of observed values.

 Separate imputation regression models were implemented for SBP and DBP. The primary analyses were conducted on each of the 10 imputed data sets and the results combined using the standard approach to yield a single result. Unadjusted odds ratios (ORs) for achieving in-control blood pressure were estimated via logistic regression using the imputed data in the same manner. We compared the results using multiply imputed data with models using a baseline observation carried-forward approach and with a last observation carried-forward approach.

Regression coefficients and their 95% confidence intervals (CIs) were estimated from an unadjusted linear regression model that incorporated only a factor indicating receipt of incentives versus control; these were compared with regression coefficients estimated from a model adjusted for the stratification variables (site, high SBP, and income), in all cases using the imputed data. In addition to prespecified subgroup analyses on race and income, we examined changes in SBP and DBP in a priori subgroups defined by study site, initial SBP (≥160 mm Hg vs <160 mm Hg), presence of diabetes, and education level (high school or lower, some college or college degree, beyond college). Homogeneity of the association between treatment groups and blood pressure change across subgroups was tested by assessing the significance of appropriate interaction terms included in the linear regression models described above.

The trial was powered to ensure that clinically meaningful differences in SBP of 10 mm Hg and in DBP of 5 mm Hg

 could be detected in any of the contrasts discussed above, assuming an interaction between the CBI and incentive interventions. We used an alpha of 0.05 and standard deviations of change in SBP and DBP of 20 mm Hg and 10 mm Hg, respectively (based on the upper limit of standard deviation directly measured in clinical trials).

 Based on these estimates, we projected we would need 63 subjects per arm to detect the clinically meaningful difference in BPs discussed above. To accommodate an estimated 20% loss to follow-up, recruitment goals for each arm were increased to 79 subjects, for a total of 316 subjects. The trial was not explicitly powered to detect effects on adherence; post hoc calculations indicate approximately 80% power to detect a difference of 8 percentage points in the MPR.

Study characteristics were generally balanced across the arms of the study (

); exceptions are noted below. The average age was 61 years, with approximately 81% males, 5% Hispanic, and 61% black; there were significantly more blacks in the control group (

 = .01). About 45% had incomes below 100% of the FPL, 26% at 100% to 200% FPL, 12% at 200% to 300% FPL, and the remainder above 300% FPL. Baseline SBP and DBP readings averaged 154 and 84 mm Hg, respectively.

Follow-up rates were slightly higher among incentive arm participants at 12 months, but the difference in follow-up rates was not significantly different (84% in incentive arms, 76% in nonincentive arms; 

 = .10) (Figure 1). Baseline SBP, number of antihypertensive medications, or number of medications overall did not differ between those who were lost to follow-up and those participants for whom we had data at 12 months.

Mean changes in SBP were —9.8 mm Hg (95% CI, –15.0 to –4.5) in the control group, –12.6 (95% CI, –18.0 to –7.1) in the co-payment reduction group, –10.2 (95% CI, –15.4 to –5.1) in the CBI group, and –14.8 (95% CI, –19.9 to –9.7) in the combined co-payment/CBI group. There were no significant interactions between incentive payments and receipt of the computerized behavioral intervention with respect to 12-month outcomes (

 = .76). Therefore, the primary and all subsequent analyses are collapsed across CBI status to focus on the impact of $8 incentives on blood pressure and outcomes (hereafter comparison of “incentive” vs “control”).

Baseline and follow-up MPR values were available for 79 control group participants and 69 incentive group participants. There was no relative difference in the change in MPR between baseline and 12 months (—0.35% and 3.08% in the control and incentive groups, respectively

 = .36). Changes, as measured by gaps in the MPR of 30 days (OR, 0.7; 95% CI, 0.3-1.4; 

 = .30), 60 days (OR, 1.5; 95% CI, 0.6-3.6

= .39), or 90 days (OR, 1.7; 95% CI, 0.6-5.0; 

 = .32) indicated no differences between the control and incentive groups.

We found no significant difference in blood pressure reduction between the incentive and control groups (

). The incentive group lowered their SBP by 13.7 mm Hg on average versus 10.0 mm Hg for the control group (

 .17). The drop in DBP was 6.8 mm Hg for the incentive group compared with 4.1 mm Hg for the control group (

 = .07). At the end of 12 months, 35.6% of the incentive group had their blood pressure in control versus 27.7% for the control group (OR, 1.4; 95% CI, 0.8-2.5; 

 = .19). Results of sensitivity analyses in which we assumed that the blood pressure in all patients lost to follow-up was equal to their last measured value or their baseline blood pressure (as opposed to the imputed blood pressure value) were qualitatively similar. Adjusted estimates of changes in SBP indicated no significant differences between incentive and control groups in the degree of change in blood pressure (

The degree of change in SBP between the incentive and control groups was compared among several a priori subgroups of the populations in the study, including those with and without diabetes, those with a baseline SBP above or below 160 mm Hg, and subgroups determined by race, income, and education. The subgroup of patients with diabetes showed a significant difference between the incentive and control groups (

 for interaction = .04). Specifically, patients with diabetes in the incentive group had an average drop in SBP of 12.7 mm Hg between baseline and 12-month follow-up, while patients with diabetes in the control group had an average SBP reduction of only 4.0 mm Hg (

 for the difference = .02). This difference was driven primarily by continued improvement in blood pressure between months 6 and 12 in the incentive group compared with a lack of such improvement in the control group (

). Patients without diabetes experienced an average SBP reduction of 15.0 mm Hg compared with an average reduction of 16.3 mm Hg for individuals without diabetes in the control group (

 value for the difference = .71.) None of the other subgroups experienced any significant differences.

In the first randomized controlled trial to provide financial incentives for antihypertensive medication prescription filling among patients with poorly controlled hypertension, we found no overall improvement in blood pressure control. We did see a relative improvement in blood pressure among patients with diabetes, but not among other subgroups.

These findings are important to ongoing discussions about V-BID, and specifically efforts to improve patient outcomes through reducing financial burdens for high-value prescription medications.

 Recent efforts to reduce the degree of patient cost sharing based on the value of prescriptions have garnered extensive interest among payers and employers.

Whereas increases in prescription co-payments have been associated with decreases in medication adherence and worse outcomes in numerous studies,

 only 2 clinical trials have examined the question of to what degree decreasing co-payments improves outcomes. The most definitive randomized trial on the impact of cost sharing on healthcare utilization—the RAND Health Insurance Experiment (HIE)&mdash;found sizable effects of patient cost sharing on use and expenditures, but more modest effects on health status. This was also performed more than 25 years ago when less effective medications were available; in addition, it excluded the elderly, more than 80% of the subjects were Caucasian, and it included a population with relatively few comorbidities

—a population for whom prescription drug coverage is most likely to be cost-effective. Of note, among low-income persons with high blood pressure, free care resulted in significant improvements in blood pressure. However, the RAND HIE randomly assigned roughly 2000 families to 1 of 14 experimental health plans that varied in their cost-sharing arrangements for all medical goods and services. The Post-Myocardial Free Rx Event and Economic Evaluation Trial (MI-FREEE) demonstrated that making medications free post acute myocardial infarction increased MPRs by about 5 to 6 percentage points (control group average MPR, 38.9%), which was associated with reductions in the rate of total major vascular events or revascularization without any increase in total healthcare spending.

Several observational studies have now indicated that lowering co-payments in a V-BID framework is associated with increases in MPR of about 3 to 5 percentage points on a base of about 60% to 80%.

 While such effects are statistically significant in large populations, these studies did not measure clinical outcomes and it seems unlikely that changes of this magnitude in MPR would have large clinical effects within a population. The findings of this study suggest that small reductions in co-payment below $0 do not significantly improve blood pressure or MPRs.

There are a number of reasons why financial incentives may have less of an impact on health than increases in co-payments. First, increases in co-payments affect utilization primarily among adherent patients, whereas financial incentives are targeted at affecting utilization among nonadherent patients in whom a change in cost of a given magnitude (ie, from $0 to —$8) is likely to have less impact. Second, increases in co-payments are likely processed as a loss by patients and behavioral economists have demonstrated that losses are felt much more strongly than equivalent gains.

 Third, incentives may be a bit like “the dog that didn’t bark.” For a nonadherent patient who doesn’t come to the pharmacy or fill prescriptions, communications about a financial incentive may be largely ignored. Fourth, we limited eligibility for the study to individuals who paid no co-payments at baseline; such individuals may have been less attuned to co-payments in relation to prescriptions since, at baseline, they would have had less reason to pay attention to co-payments.

Because studies have indicated that unbundling rewards from other payments make the rewards more effective

 and due to the fact that the PinnacleHealth system patients filled prescriptions at a large number of pharmacies in the Harrisburg area in which we could not control point-of-service pricing, we provided post hoc rebates rather than incentives at point of service. The lack of an overall effect on blood pressure or adherence could be due to this design feature of the trial. This had the disadvantage of introducing time delays in feedback after the incented behavior occurred. Other limitations include the incentive magnitude, as it may have been too small to induce changes in behavior, though previous work suggested that in low-income populations, co-payment increases as low as $0.50 to $1 per prescription (approximately $2-$3 adjusting for inflation) can reduce drug utilization.

 The study was conducted primarily among veterans at 2 VA hospitals, and thus had primarily male participants, although there are no obvious reasons to believe that the interventions would be less effective in men than women.

Financial incentives for blood pressure medications had little impact on blood pressure control except, perhaps, among patients with diabetes. Because this was an isolated finding in 1 subgroup, we conclude this intervention did not systematically improve patient outcomes. Further initiatives should examine the comparative effectiveness of different ways of delivering such incentives, the relationship between the magnitude of incentives and effectiveness, and the impact on different populations.

Other coauthors (included here rather than the byline due to space constraints) include: Jingsan Zhu, MBA; Yuanyuan Tao, MS; Wei Yang, PhD; John H. Holmes, PhD; Dominick L. Frosch, PhD; Katrina Armstrong, MD, MSCE; Shiriki Kumanyika, PhD; Kjell Enge, PhD; Raymond R. Townsend, MD; and Nirmal Joshi, MD. 

 Center for Health Equity Research & Promotion, Philadelphia Veterans Affairs Medical Center (KGV, JAL, SAI), PA; Center for Health Incentives and Behavioral Economics, Leonard Davis Institute, Penn CMU Roybal P30 Center in Behavioral Economics and Health (KGV, ABT, JAD, SEK), Philadelphia, PA; Department of Medicine (KGV, JAL, JAD, SEK, SAI) and Center for Therapeutic Effectiveness Research (SEK), Perelman School of Medicine of the University of Pennsylvania, Philadelphia; Department of Health Care Management, The Wharton School (KGV), and Center for Clinical Epidemiology and Biostatistics and Department of Biostatistics and Epidemiology (ABT, DA, JJ, SEK), University of Pennsylvania, Philadelphia; Center for Health Equity Research & Promotion, Pittsburgh Veterans Affairs Medical Center (SAI), PA; University of Pittsburgh School of Medicine (SAI), PA; Department of Psychology, Cheyney University (JOS), Cheyney, PA; Department of Psychology, Dickinson College (MH-L), Carlisle, PA.

 The work in this paper was primarily supported by a grant from the Commonwealth of Pennsylvania, titled Collaboration to Reduce Disparities in Hypertension, grant number ME-02-382. Supplemental support was received from Pfizer, Inc. The sponsors/funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the manuscript. Drs Volpp and Kimmel had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

 Dr Volpp has received research funding from the Aetna Foundation, Aramark, Discovery (South Africa), Horizon Blue Cross and Blue Shield, Humana, McKinsey, Merck, Hawaii Medical Services Association, Weight Watchers (all unrelated to the topic of this paper), research funding and consulting income from CVS Caremark, and is a principal at the behavioral economics consulting firm VAL Health. Dr Kimmel has done consulting for several pharmaceutical companies, including Pfizer, all unrelated to the topic of this paper. Dr Troxel has received consulting income from VAL Health. Other than what is listed above, there are no known financial conflicts of interest among any of the authors including, but not limited to, employment/affiliation, all grants or funding, honoraria, paid consultancies, expert testimony, and patents filed, received, or pending. Dr Doshi is a consultant for Alkermes, Boehringer Ingelheim, Forest Labs, Ironwood Pharmaceuticals, Merk, and Shire. She has grants pending from Humana and PhRMA, and has previously received grants from Humana, Pfizer, PhRMA, Merck, National Pharmaceutical Council, Amgen; her spouse holds stock in Merck and Pfizer. The remaining authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

 Concept and design (KGV, MH-L, SEK, JOS, JAD); acquisition of data (MH-L, SEK, JOS); analysis and interpretation of data (ABT, MH-L, SEK, JOS, DA, JAD); drafting of the manuscript (KGV, DA, JAD); critical revision of the manuscript for important intellectual content (KGV, ABT, SEK, JAD); statistical analysis (ABT, DA); obtaining funding (MH-L, SEK); and supervision (SEK, JOS).

 Kevin G. Volpp, MD, PhD, Leonard Davis Institute, Center for Health Incentives and Behavioral Economics, Perelman School of Medicine and the Wharton School of the University of Pennsylvania, 1120 Blockley Hall, 423 Guardian Dr, Philadelphia PA 19104-6021. E-mail: volpp70@wharton.upenn.edu.

1. Ho PM, Spertus JA, Masoudi FA, et al. Impact of medication therapy discontinuation on mortality after myocardial infarction. 

2.Oster berg L, Blaschke T. Adherence to medication. 

3. Jackevicius CA, Mamdani M, Tu JV. Adherence with statin therapy in elderly patients with and without acute coronary syndromes. 

4. Haynes RB. Improving patient adherence: state of the art, with a special focus on medication taking for cardiovascular disorders. In: Burke LE, Ockene IS, eds. 

. Armonk, NY: Futura Publishing Company, Inc; 2001:3-21.

5. Ho PM, Rumsfeld JS, Masoudi FA, et al. Effect of medication nonadherence on hospitalization and mortality among patients with diabetes mellitus. 

6. Goldman DP, Joyce GF, Escarce JJ, et al. Pharmacy benefits and the use of drugs by the chronically ill. 

This study extends value-based insurance design concepts in testing the impact on blood pressure control of rewards that provided negative co-payments for blood pressure medication.

A Randomized Controlled Trial of Negative Co-Payments: The CHORD Trial

Efforts to improve adherence by reducing co-payments through value-based insurance design are become more prevalent despite limited evidence of improved health outcomes. The objective of this study was to determine whether eliminating patient co-payments for blood pressure medications improves blood pressure control.

The Collaboration to Reduce Disparities in Hypertension (CHORD) was a randomized controlled trial with 12 months’ follow-up conducted among patients from the Philadelphia and Pittsburgh Veterans Administration Medical Centers. We enrolled 479 patients with poorly controlled systolic blood pressure. Participants were randomly assigned to: a) receive reductions in co-payments from $8 to $0 per medication per month for each antihypertensive prescription filled, b) a computerized behavioral intervention (CBI), c) both co-pay reduction and CBI, or d) usual care. Our main outcome measure was change in systolic blood pressure from enrollment to 12 months post enrollment. We also measured adherence using the medication possession ratio in a subset of participants.

There were no significant interactions between the co-payment interventions and the CBI interventions. There was no relative difference in the change in medication possession ratio between baseline and 12 months (0.05% and —0.90% in control and incentive groups, respectively; 

 = .74) or in continuous medication gaps of 30, 60, or 90 days. Blood pressure decreased among all participants, but to a similar degree between the financial incentive and control groups. Systolic pressure within the incentive group dropped 13.2 mm Hg versus 15.2 mm Hg for the control group (difference = 2.0; 95% CI, —2.3 to 6.3; 

 = .36). The proportion of patients with blood pressure under control at 12 months was 29.5% in the incentive group versus 33.9 in the control group (odds ratio, 0.8; 95% CI, 0.5-1.3; 

Among patients with poorly controlled blood pressure, financial incentives—as implemented in this trial&mdash;that reduced patient cost sharing for blood pressure medications did not improve medication adherence or blood pressure control.

Small rewards that lowered co-payments to $0 did not improve blood pressure control significantly more than in control-group subjects who simply had blood pressure measured. This may be because of the way in which this program was administered—there were delays in providing rebates for co-payments following prescription filling. The exact nature of the program implementation may have a significant impact on program effectiveness and should be carefully considered in value-based insurance design program design and assessment of impact.

Hypertension—especially within socioeconomically disadvantaged communities&mdash;remains a leading cause of cardiovascular morbidity and mortality in the United States, affecting nearly 50 million Americans.

 Although there are effective medications to treat hypertension, nearly two-thirds of Americans with the condition have poorly controlled blood pressure.

 Lack of adherence to antihypertensive medications is considered a critically important factor in blood pressure management. Medication adherence for chronic diseases such as hypertension and hypercholesterolemia is extremely low,

 limiting the potential for highly efficacious medications to improve population health.

Value-based insurance design (V-BID)—an approach based on the premise that reductions in co-payments will significantly increase the use of beneficial and cost-effective services&mdash;is being widely adopted.

 As part of the Affordable Care Act, elimination of cost sharing for preventive services is being mandated to increase utilization of such services. Although observational studies have shown that increases in co-payments are associated with both decreased medication use and worsened health outcomes,

 and studies have generally focused exclusively on medication use rather than measured health outcomes. The underlying psychology of how people process changes in payments as losses compared with gains suggests that increases and decreases in co-payments may not be equivalent.

Only 2 randomized trials examining the relationship between co-payments and health have been published. The first, the RAND Health Insurance Experiment (HIE), was conducted in the 1970s and varied cost sharing for all services, not just those for medications. The second, the recently published Post-Myocardial Infarction Free Rx Event and Economic Evaluation (MI FREEE) study, was a test of co-payment reduction following discharge for myocardial infarction (MI) in an insurer-based intervention.

To examine whether reducing co-payments from $8 to $0 per medication per month for all antihypertensive medications significantly improves blood pressure control, we conducted a clinic-based, randomized controlled trial of elimination of co-payments among patients with poorly controlled blood pressure at 2 medical centers in Pennsylvania.

Study participants were drawn from patients at 2 ambulatory clinics in Pennsylvania: the Philadelphia Veterans Affairs Medical Center (PVAMC) and the VA Pittsburgh Healthcare System (VAPitt), with recruitment occurring between March 2005 and July 2007. Eligible patients were 21 years or older, with 1 or more active prescriptions for an antihypertensive medication, systolic blood pressure (SBP) of at least 140 mm Hg (130 mm Hg in patients with diabetes), and those who paid a co-payment for their medications. Exclusion criteria included: participation in another experimental study; markedly shortened life expectancy (due to diagnosis of metastatic cancer, end-stage renal disease on dialysis, New York Heart Association class IV congestive heart failure, or dementia), or atrial fibrillation (because of concerns with accuracy of blood pressure measurement). A total of 479 participants were enrolled (details on enrollment provided in 

The protocol was approved by the Institutional Review Boards of the PVAMC, VAPitt, and the University of Pennsylvania; all participants provided written informed consent prior to randomization. The study was registered at clinicaltrials.gov as ID # NCT00133068. Participants were randomized to receive 1 of the following: a) a financial incentive equal to their co-payments for all antihypertensive medications that effectively lowered co-payments from $8 to $0 per medication per month (note: on January 1, 2006, 9 months after initiation of enrollment, co-payments were raised from $7 to $8 per month per Veterans Health Administration Directive 2005-052 and incentives were adjusted accordingly); b) a computerized behavioral intervention (CBI) that was provided immediately following enrollment and repeated at the 6-month follow-up visit; c) both the financial incentive and the CBI; or d) usual care, which consisted only of their existing medical care.

The CBI was administered at baseline and at 6 months and was designed to help study participants learn the basics about blood pressure control and its impact. Participants watched a video testimonial from a patient who had had multiple strokes due to medication nonadherence and who felt that he had learned the importance of medication adherence from that experience. This helped to improve self-efficacy by reviewing techniques that could be used to enhance adherence. After an initial visit, all participants were requested to return for follow-up blood pressure readings and to complete surveys at 3, 6, 9, and 12 months. The financial reimbursements were paid as soon as study staff received confirmation of prescription fills using either the Veterans Administration’s (VA’s) computer records, prescription bottles, or receipts.

Randomization was carried out using a random number generator and via permuted block randomization with a block size of 4. Randomization was stratified by site, income (<100%, 100%-200%, 200%-300%, and >300% of the federal poverty line [FPL]), and baseline blood pressure (SBP <160 mm Hg or SBP ≥160 mm Hg). Allocation assignments were concealed, with staff unable to access the randomization assignment for each subject until all eligibility criteria were entered into an electronic tracking system and consent forms were completed. Neither staff nor study participants could be blinded due to the nature of the intervention; investigators and analysts, however, remained blinded to intervention assignments until unblinding occurred, in coordination with the Data Safety Monitoring Board, once follow-ups were complete.

The primary outcome variable was change in SBP and diastolic blood pressure (DBP) from enrollment to 12 months post enrollment. Secondary outcome variables included change in SBP and DBP 6 months post enrollment, the percentage of patients with well-controlled blood pressure at 6 and 12 months post enrollment, self-reported medication adherence, and prescription refill data from the VA electronic medical record system. Blood pressure control was defined as SBP below 140 mm Hg and DBP below 90 mm Hg for patients without diabetes, and as SBP below 130 mm Hg and DBP below 85 mm Hg for patients with diabetes.

 Participants were instructed to relax while seated for 5 minutes before their blood pressure was taken; the patient’s arm was supported on a chair or desk and blood pressures were measured 3 times, each reading 2 minutes apart, and averaged, but not revealed to the study participants. Although the study nurses could not be blinded to the randomization due to the need to administer the intervention, the use of an automated blood pressure cuff and a standardized protocol protected against differences in blood pressure measurement.

Medication adherence was measured using self-report based on the Hill-Bone Scale,

 with supplemental assessment using electronic prescription fill records where available. For patients with these records, we calculated the proportion of days covered (number of days with antihypertensive drug supply on hand divided by the number of days in the observation period).

 In addition, continuous medication gap measures were calculated as at least 1 continuous episode with no antihypertensive medication for a minimum of 30, 60, or 90 days.

Other factors assessed at baseline included height, weight, serum creatinine level, income, health status, health history, medication use, age, gender, and self-reported race or ethnicity. We used information on income and family size to calculate income as a percent of the FPL.

 test for categorical variables, with Fisher’s Exact tests used for analyses with 5 or fewer subjects per cell.

Because of the factorial design, we first assessed whether receipt of CBI affected the impact of incentive payments. The trial was designed to test the effects of the 2 interventions individually; we did not assume an interaction but powered the study adequately for individual comparisons of the 3 active arms against control. The primary analysis of the financial incentive began with a formal statistical test of interaction between the incentive and CBI; there was no evidence of this, so we collapsed across the CBI arms to assess the effect of the financial incentive alone.

 test. We similarly calculated differences in change in SBP and DBP from baseline to 6 months post enrollment. Some participants had missing blood pressure values at 6 and 12 months; we compared baseline data between participants with and without follow-up values. Missing values for 6-month and 12-month SBP and DBP readings were handled using the Markov Chain Monte Carlo multiple imputation method, utilizing 10 imputations after verification that this yielded relative efficiencies of 95% or higher; we also checked that imputed values were within the range of observed values.

 Separate imputation regression models were implemented for SBP and DBP. Unadjusted odds ratios (ORs) for achieving in-control blood pressure were estimated via logistic regression using the imputed data in the same manner. We compared the results using multiply imputed data with models using a baseline observation carried-forward approach and with a last observation carried-forward approach.

We estimated regression coefficients and their 95% confidence intervals (CIs) from an unadjusted linear regression model that incorporated only a factor indicating receipt of incentives versus control. We then compared these with regression coefficients estimated from a model adjusted for the stratification variables (site, high SBP, and income) in all cases using the imputed data.

In addition to prespecified subgroup analyses on race and income, we examined changes in SBP and DBP in subgroups defined by study site, initial SBP (≥160 mm Hg vs <160 mm Hg), presence of diabetes, and education level (high school/GED or lower, some college or college degree, beyond college). Homogeneity of the association between treatment groups and blood pressure change across subgroups was tested by assessing the significance of appropriate interaction terms included in the linear regression models described above.

The trial was powered to ensure that clinically meaningful differences in SBP of 10 mm Hg and of DBP of 5 mm Hg

 could be detected in any of the contrasts discussed above, assuming an interaction between the CBI and incentive interventions. We used an alpha of 0.01 to account for multiple comparisons and standard deviations of change in SBP and DBP of 20 and 10, respectively (based on the upper limit of standard deviation directly measured in clinical trials).

This study tests the impact on blood pressure control of a reward that lowered co-payments for blood pressure medication to $0.

Judith A. Long, MD

Articles

US Internists' Awareness and Use of Overtreatment Guidelines: A National Survey

A Randomized Controlled Trial of Negative Co-Payments: The CHORD Trial

A Randomized Controlled Trial of Co-Payment Elimination: The CHORD Trial