Andrea B. Troxel, ScD

 To describe the extent of and longitudinal changes in physician practice variation with respect to implant costs, institutional postacute care (PAC) provider utilization, and total episode payments, as well as to evaluate the association between physician volume and quality and these outcomes.

We combined claims and internal hospital cost data for

34 physicians responsible for 3614 joint replacement episodes under bundled payment at Baptist Health System (BHS). Multilevel multivariable generalized linear models were employed and the intraclass correlation (ICC) was used to quantify between-physician variation.

 There was significant between-physician variation in implant costs, institutional PAC provider utilization, and total episode payments not explained by observable variables (

<.001 for all). Over 5 years, the ICC decreased from 0.26 to 0.06, 0.15 to 0.13, and 0.12 to 0.10 for implant costs, institutional PAC provider utilization, and total episode payments, respectively, but differences were not statistically significant. Both higher physician case volume and quality were associated with decreased total episode payments and institutional PAC provider utilization, but not with changes in implant costs.

practice variation was observed under bundled payment at BHS

and decreased to a greater degree for implant costs than institutional PAC provider utilization or total episode payments. Institutional PAC provider utilization and total episode payments were associated with physician volume and quality. Although some organizational strategies achieve gains by reducing physician practice variation, variation reduction is not an absolute requisite for success under bundled payment.

Amid the proliferation of orthopedic bundles, organizations can benefit from understanding whether strategies must reduce physician practice variation in order to succeed. Analyzing the experience of a long-standing participant in Medicare orthopedic joint replacement bundles, we found that although some strategies achieve gains by reducing physician practice variation, variation reduction is not an absolute requisite for success under bundled payment.

There was significant physician practice variation in implant costs, institutional postacute care (PAC) provider utilization, and total episode payments.

Over time, physician practice variation decreased to a greater degree for implant costs than for institutional PAC provider utilization or total episode payments.

To increase value and address substantial cost and quality variation in care, CMS has rapidly expanded orthopedic bundled payment initiatives. By receiving a single “bundled” amount for an entire episode of care, hospitals and physicians are accountable for quality and costs under bundled payment.

After early success bundling acute care among a small group of providers in the voluntary Acute Care Episode (ACE) demonstration,

 CMS extended its focus across the care continuum. In 2012, it launched the Bundled Payments for Care Improvement (BPCI) initiative,

 a large national program through which many hospitals voluntarily accept bundled payment for joint replacement episodes encompassing acute and postacute care (PAC). More recently, in April 2016, CMS used the BPCI framework to initiate the Comprehensive Care for Joint Replacement model,

 a mandatory program that initially held approximately 800 hospitals in 67 metropolitan areas financially accountable for the costs and quality of episodes spanning hospitalization through 90 days of PAC.

Early evidence from BPCI suggests that bundled payment can reduce episode costs and produce savings for both CMS and hospitals while possibly increasing quality.

 Because physician practice style contributes to large variation in hospital and postacute spending for surgeries such as hip replacement,

 variation reduction is frequently championed in bundle payment strategy.

 However, there is little empirical evidence about whether organizational strategies must reduce physician practice variation in order to succeed.

In this study, we describe physician practice variation at Baptist Health System (BHS), a continuous participant in Medicare joint replacement bundles since 2009. BHS achieved high performance—demonstrating notable reductions in total episode payments and hospital savings during a period when joint replacement expenditures rose nationwide

—by engaging physicians to reduce surgical implant costs and postdischarge utilization of institutional PAC providers.

 We evaluate the extent of and longitudinal changes in physician practice variation with respect to implant costs, institutional PAC provider utilization, and total episode payments, as well as the associations among physician volume and quality and these 3 outcomes.

The study was divided into 4 periods. The first was the baseline period (July 2008-December 2008), when BHS received nonbundled fee-for-service payments prior to ACE participation. The second was the ACE period (July 2009-June 2012), during which BHS implemented joint replacement bundles for acute hospitalization. The ACE period consisted of 3 years: ACE year 1 (July 2009-June 2010), ACE year 2 (July 2010-June 2011), and ACE year 3 (July 2011-June 2012). The third was the transition period (July 2012-September 2013), when BHS prepared to transition to BPCI Model 2 and did not receive bundled payment. The fourth was the BPCI period (October 2013-June 2015), in which BHS implemented joint replacement bundles for episodes spanning acute hospitalization through 30 days of PAC. The BPCI period consisted of 2 years: BPCI year 1 (October 2013-June 2014) and BPCI year 2 (July 2014-June 2015).

We obtained both Medicare claims and internal hospital cost data directly from BHS.

 Using Medicare data, we constructed care episodes that were consistent with the health system’s BPCI arrangement with CMS and encompassed acute hospitalization plus 30 days of PAC.

 Data were not available during the transition period between ACE and BPCI when BHS was not paid by CMS under bundled payment.

Our analytic sample consisted of 3614 patient episodes from a panel of 34 physicians who performed at least 1 joint replacement surgery in both the ACE and BPCI periods. Overall, 51 orthopedic surgeons affiliated with BHS performed a combined total of 3725 joint replacement surgeries on Medicare beneficiaries under Medicare Severity Diagnosis Related Group 470, Major joint replacement or reattachment of lower extremity without major complications or comorbidities. We excluded physicians who performed surgeries only in the ACE period (n = 9) or BPCI period (n = 8).

Patient demographics and clinical data were calculated from Medicare claims, and patient illness severity was estimated using the Elixhauser Comorbidity Score.

 Medicare data were also used to calculate 2 quality of care measures, 30-day readmission and emergency department (ED) visit rates, based on CMS demonstration project specifications.

 Characteristics of operating physicians (gender, years of experience, allopathic vs osteopathic degree, graduation from US or non-US medical school, and board certification status) were obtained from publicly available information on consumer websites.

We evaluated variation in 3 outcomes of interest: implant costs, total episode payments, and institutional PAC provider utilization. Internal hospital cost data were used to quantify the costs of orthopedic implants (implant costs), while BHS used a time-driven activity-based costing approach

 to quantify other nonsupply hospital costs. Medicare data were used to calculate the sum of all Medicare payments for an episode of care (total episode payments) by combining claims for all acute hospital facility payments, physician fees during hospitalization, and total PAC payments through 30 days post discharge. Total PAC payments consisted of those corresponding to outpatient visits, ED visits, readmissions, physician fees, durable medical equipment, and utilization of PAC providers, such as skilled nursing facilities (SNFs), inpatient rehabilitation facilities (IRFs), and home health agencies (HHAs).

We used Medicare data to define a dichotomous variable (institutional PAC provider utilization) as discharge from hospitalization to either SNFs or IRFs as institutional PAC providers, compared with discharge home or home with HHA services. Patients were discharged to 1 of 4 destinations in nearly all (99.4%) episodes.

Finally, we defined several variables to evaluate associations between physician practice characteristics and our outcomes. We assessed practice quality by calculating the proportion of episodes for each physician with prolonged length of stay (PLOS), a validated measure of complications of orthopedic lower extremity joint replacement.

 PLOS evaluates the pattern of patient discharges as a function of length of stay (ie, number of days in the hospital) and identifies a point at which discharge is less likely than continued hospitalization, representing a complication. We assess practice volume using total case volume over the study period. Due to data skewness, we analyzed volume as a continuous variable, as well as a dichotomous variable with the top 10 physicians by volume (each with more than 100 cases and representing more than 70% of total cases) defined as high-volume physicians.

Descriptive statistics were reported using means and SDs for continuous variables and percentages for categorical variables. Multilevel generalized linear models, adjusted for study year and patient and physician characteristics, were employed to evaluate the extent of physician variation in the 3 outcomes of interest. Models were clustered at the physician level and included a physician random effect. To account for multiple patients per operating physician and heteroscedasticity, all models utilized clustered standard errors with the Huber-White correction.

We utilized linear random effects models with a log link and gamma distribution for the outcomes of implant cost and total episode spending.

 To evaluate institutional PAC provider utilization, we used a random effects model with a logit link and binominal distribution to estimate the logit of the probability of discharge to institutional PAC providers as a linear function of covariates and operating physician random effect.

The significance of between-physician variation was determined by testing whether the variance of the physician random effect differed significantly from 0 and was reported using the intraclass correlation (ICC). The ICC is a measure that reflects the proportion of overall variance explained by variation between, rather than within, clusters of individual observations. In this study, the calculated ICC corresponds to the proportion of variation explained by differences in practice patterns among physicians (ie, between-physician variation), as opposed to among cases for individual physicians (ie, within-physician variation). A linear threshold method was used to calculate ICC for institutional PAC provider utilization.

 To compare changes in the ICC across the study period, we used bootstrapping to calculate standard errors for ICCs in ACE year 1 and BPCI year 2 before using the 

 statistic and pairwise tests for each outcome to evaluate whether there were significant differences between the 2 years.

Random effects models were also used to calculate and compare between-physician variation and ICC for implant costs and total episode payments across our study period (ACE year 1 vs BPCI year 2). Because bundle definitions did not include PAC until the BPCI program period, we compared between-physician variation and ICC for PAC provider utilization among ACE year 1, ACE year 3, and BPCI year 2.

Separately, we used random effects models to test the associations among physician quality and case volume and our 3 outcomes of interest. All cost and spending figures were adjusted for inflation and reported in 2015 US$ equivalents. Implant cost data were aggregated at the level of individual physician per quarter and weighted by episode volume. Analyses were conducted using SAS version 9.4 (SAS Institute; Cary, North Carolina). All tests of significance were 2-tailed and considered statistically significant at an α of .05. The University of Pennsylvania Institutional Review Board approved the study.

The overall patient and program characteristics across the BHS bundled payment program have been described previously.

 From that sample, we included 34 physicians in this analysis, 97% of whom were male, 100% whom had obtained allopathic medical degrees, 91% of whom had graduated from US medical schools, and 82% of whom were board certified (

). On average, physicians in this cohort had 26 years of experience and a total volume of 106 cases across our study period, ACE year 1 to BPCI year 2.

The mean age of patients cared for by these physicians was 72 years. They were predominantly female, and the mean Elixhauser Comorbidity Score was 1.1, with the 5 most common coexisting clinical conditions being hypertension (76%), diabetes (21%), hypothyroidism (21%), obesity (19%), chronic obstructive pulmonary disease (16%), and depression (10%). Readmissions and ED visits occurred in 5% and 6% of cases, respectively.

Overall, mean implant costs decreased 5.8% across the study period, from $5026 in the first year of bundled payment (ACE year 1) to $4732 through 5 years (BPCI year 2). There was also an overall downward shift in distribution of implant costs among physicians. Compared with ACE year 1, for example, more physicians in BPCI year 2 had average implant costs between $4500 and $5000 (22 vs 9) and fewer physicians had average implant costs of $5550 or greater (1 vs 10) (

After accounting for patient and physician characteristics, higher implant cost was associated with younger patient age and earlier study year. There was a decline in between-physician variation across study year and program period (

]), from an ICC of 0.26 in ACE year 1 to 0.06 in BPCI year 2, a difference that was not statistically significant on pairwise testing (

= .26). Adjusted random effects analysis demonstrated statistically significant between-physician variation in implant costs (overall ICC = 0.11) unexplained by observable variables (

The proportion of patients discharged to institutional PAC providers remained relatively steady during ACE: 42% in ACE year 1 and 40% in ACE year 3 when PAC was not included in bundled payment. When PAC was included in bundles under BPCI, however, discharge to institutional PAC providers decreased to 31% in BPCI year 1 and to 25% in BPCI year 2. Across the ACE program period, most physicians discharged at least half of their patients to institutional PAC providers (

). In comparison, by BPCI year 2, most physicians discharged at least half of their patients to home or HHAs (80% of physicians in BPCI year 2 compared with 38% and 42% in ACE year 3 and ACE year 1, respectively).

After multivariable adjustment, higher institutional PAC provider utilization was associated with earlier study year, female patient sex, younger patient age, higher Elixhauser Comorbidity score, and readmissions. Between-physician variance in institutional PAC provider utilization decreased over the study period from 0.59 in ACE year 1 to 0.48 by BPCI year 2, with ICCs of 0.15, 0.14, and 0.13 in ACE year 1, ACE year 3, and BPCI year 2, respectively, without a statistically significant overall change (

= .86). Overall, the extent of between-physician variation (overall ICC = 0.09) unexplained by observable variables was also statistically significant (

Over the 5 years of the study, mean total episode payments decreased 21% from $23,634 in ACE year 1 to $18,699 in BPCI year 2. Compared with ACE year 1, there was an overall narrowing and downward shift in distribution of total episode payments among physicians. For example, the number of physicians with total episode payments less than $20,000 increased from 6 to 21 and there were fewer physicians with total episode payments greater than $27,500 (9 vs 2) (

After accounting for readmissions and ED visits, as well as patient and physician characteristics, higher total episode payment was associated with younger patient age and earlier study year.

Between-physician variance in total episode payments increased slightly over time (

), and the ICC decreased by a statistically nonsignificant amount, from 0.12 in ACE year 1 to 0.10 in BPCI year 2 (

= .81). The extent of between-physician variation (overall ICC = 0.07) unexplained by observable variables was also statistically significant (

Association Between Physician Practice Characteristics and Outcomes of Interest 

In multivariable analyses, increasing physician clinical practice volume was associated with small but significant decreases in total episode payments (coefficient, —0.00022; 

<.001) and institutional PAC provider utilization (coefficient, —0.00020; 

<.001), but not changes in implant costs (coefficient, 0.00015; 

= .32). Results from analysis using a dichotomized volume indicator did not yield qualitatively different results. Similarly, lower practice quality, defined as increasing proportion of episodes with PLOS, was associated with increased total episode payments (0.2086; 

= .002) and increased institutional PAC provider utilization (2.2347; 

= .001) but not changes in implant costs (0.1695; 

This study demonstrates that bundled payment strategy can reduce costs and generate savings by decreasing variation in some practices but not others. Three conclusions are particularly notable.

First, our analysis demonstrates the presence of and change in physician practice variation in implant costs, institutional PAC provider utilization, and total episode payments. In ACE year 1, practice variation between physicians accounted for 15% to 26% of overall variation in these outcomes. The decreases in between-physician variation by BPCI year 2 demonstrate the role of practice variation reduction in decreasing costs and utilization.

This observation is illustrated by the downward shift and narrowing in distributions of our outcomes: Over time, more physicians achieved lower mean implant costs and total episode payments, and fewer physicians discharged a significant portion of patients to institutional PAC providers. Furthermore, the presence of statistically significant physician variation across the study period that was unexplained by observable variables suggests that attention to individual physician practice patterns may help other organizations identify cost and utilization reduction strategies under bundled payment.

Second, this study suggests that reducing physician practice variation may not be an absolute requisite for succeeding in joint replacement bundles. Although physician practice variation only accounted for a portion of overall implant costs and differences in ICC were not statistically significant, we nonetheless observed a steady decrease in between-physician variation over time. In the first year of the bundled payment program, BHS leaders used quality and cost data transparency to leverage physician engagement and lower implant costs program-wide over several rounds of negotiations.

 The observed trend in physician variation reflects the result of this strategy: more consistent pricing across available implant options and systematic cost reductions across physicians.

Articles

This study demonstrates that variation reduction is an important, but not requisite, component of organizational success under orthopedic bundled payment.

Physician Practice Variation Under Orthopedic Bundled Payment

Clinical trials are increasingly testing the effectiveness of paying patients’ financial incentives for achieving desired clinical outcomes. Some researchers and providers are concerned that patient financial incentives will harm the doctor—patient relationship. How patients feel about these approaches, and these trials, is largely unknown. This study examined patients’ perceptions of a compound behavioral and financial incentive intervention used in a large multicenter trial to lower low-density lipoprotein cholesterol (LDL-C), including their perceptions of benefits and challenges and the study’s effect on patients’ relationship with their primary care physicians (PCPs). 

Semi-structured telephone interviews with patients post intervention. 

PCPs from 3 primary care practices in the northeastern United States were randomized to 1 of 4 arms: physician financial incentives, patient financial incentives, shared incentives between physicians and patients, and a control arm. Within each arm, 10 high, 10 medium, and 10 low performers in LDL-C reduction were interviewed. Interviews targeted reasons for enrolling in the study, the specific intervention elements that helped them reach the goal (incentives, engagement, monitoring), challenges faced in reducing cholesterol, and the impact of study participation on their relationship with their PCP.

Patients reported positive experiences with the study: 65% described personal changes to improve health and 61% reported increased awareness. Views about financial incentives varied: 71% clearly found them motivating and 36% claimed they made no difference. Patients noted that changing lifestyle (36%) and diet (65%) was difficult. Patients who substantially lowered their LDL-C revealed themes similar to those who did not. 

Overall, behavioral interventions with financial incentives appear to be socially acceptable to patients who participate in them. Both adherence monitoring and financial incentives were well received, with little effect on the physician-patient relationship.

Patients who enrolled in a trial to lower low-density lipoprotein cholesterol (LDL-C)—one that included financial incentives to patients or to physicians or were shared between patients and physicians&mdash;spoke positively of the multifaceted intervention. One-on-one interviews revealed:  

Wireless pillboxes for monitoring served as a reminder to take medications. 

Blood draws every 3 months were reinforcing and informative. 

Financial incentives were appreciated but not viewed as particularly motivating. 

Patients who substantially lowered their LDL-C revealed themes similar to those who did not. 

Financial incentives were rarely discussed; they did not disrupt the patient—physician relationship. 

Overall, financial incentives appear to be effective and socially acceptable.

Patient nonadherence to recommended treatment is a barrier to improving outcomes, but the use of financial incentives to overcome this barrier is increasing.

 Eighty-five percent of large employers use financial incentives to improve health behaviors in realms ranging from smoking and other addictions to weight control and medication adherence.

 In a recent randomized controlled trial focused on reducing low-density lipoprotein cholesterol (LDL-C) among patients with high cardiovascular risk or established cardiovascular disease, incentives based on principles of behavioral economics were used to motivate both patients and providers.

 In the parent study, 340 primary care physicians (PCPs) from 3 health systems and about 15,030 of their patients were enrolled into a 4-arm trial to test alternative incentive strategies to lower LDL-C. PCPs were randomized to 1 of 4 arms: physician financial incentives, patient financial incentives, shared incentives between physicians and patients, and a control arm. Progress was monitored through quarterly LDL checks and medication adherence via pill bottle opening. The primary outcome was LDL reduction over 12 months. The intervention was a compound of multiple behavioral economic strategies.

The main trial showed significant benefits for those in the shared incentive arm, but the effectiveness of the intervention ranged across individuals. The purpose of this current study was to investigate patients’ perceptions of the intervention and learn from their stories of how the intervention affected their behaviors. The specific objectives were to investigate patients’ perceptions of financial incentives and of multiple components of the behavioral intervention, including overall thoughts about benefits and challenges and how being in the study affected their relationship with their physician. We were particularly interested in how perspectives differed between patients who succeeded in lowering their LDL-C and those who did not. Secondarily, we examined if narratives about incentives differed among patients randomized to different arms of the study.

Overview/Summary of the Underlying Intervention

The protocol for the main trial is summarized briefly here.

 A total of 1503 patients with high cardiovascular risk or established cardiovascular disease were enrolled; they were nested within 340 PCPs from the University of Pennsylvania (Penn), Geisinger Health System (GHS), and Harvard Vanguard Medical Associates (HVMA). The study compared 4 approaches to reduce LDL-C: physician incentives, patient incentives, shared physician—patient incentives, or control (no financial incentive or regular feedback). For the patient and physician incentive arms, the maximum expected value of the incentive was $1024 per enrolled patient who met LDL-C goals. PCPs and patients in the shared arm shared those possible financial incentives. Patients also received surveys, quarterly LDL-C checks, and wireless pill bottles (Vitality GlowCaps) which were used to store their cholesterol-lowering medication and electronically monitor bottle openings.

In this study, semi-structured telephone interviews were conducted with patients post intervention from February 2014 through July 2014. The methods and script were approved by the University of Pennsylvania Institutional Review Board.

Purposive targeted sampling was used. Within each of the 4 arms, the goal was to interview 10 high, 10 medium, and 10 low performers in LDL-C reduction, stratified across the 3 study sites. Within each of the 12 cells, 5 patients were targeted from GHS, 4 from Penn, and 1 from HVMA, proportional to the total patient sample at each site. Patients were interviewed in multiple waves, formed as groups of patients completed their 15-month visits. The final number of interviews is shown in 

. For each wave of completed data, usually for 60 to 80 patients, the mean change in LDL-C (baseline to 12-month) was calculated, then sorted by site within arms. Within each arm, the distribution was inspected and the 4 to 5 patients with the most improved, most worsened, and least changed LDL-C were prioritized, choosing only 1 patient per clinical site per wave. Therefore, “high” and “low” do not have a constant cut-off. However, across the selection waves, median changes in LDL-C among targeted patients were similar.

The interview script was developed by a study team of health economists, physicians, and qualitative researchers. They drew on the literature and their expertise to develop conversational questions related to the study elements and the potential impact of behavioral interventions. Initial scripts were piloted and revised as necessary prior to the study and were constant for the duration of active interviewing. Revisions were based on queries about unclear questions, and the order of the questions was changed to improve flow. Pilot interviews are not included in the analyses. Key questions targeted reasons for enrolling in the study, parts of the study that helped them reach the goal (surveys, wireless pill bottle, cholesterol check every 3 months, and incentives, for patients in arms that received them), challenges faced in reducing cholesterol, and the effect of the study on their relationship with their doctor.

Two trained interviewers conducted the interviews. Recordings of the interviews were transcribed by an independent transcription agency. De-identified transcripts were imported into NVivo version 10 (QSR International Pty Ltd) for coding and analysis.

Four trained coders analyzed the data using elements from grounded theory. The team first identified broad themes based on the first 6 transcripts and then applied open codes to the transcripts. The initial set of codes was tested on each subsequent interview transcript to refine the codebook. Codes were revised iteratively by constant comparison between and within transcripts, and more content-driven axial codes were applied. A final coding scheme was established and applied to all transcripts. In a post hoc analysis, the comments about financial incentives were grouped by arm, blinded, and reviewed by the senior author, with a goal of understanding the main trial results. Inter-rater reliability was assessed on samples exceeding 20% of transcripts. The research team met weekly to resolve discrepancies via consensus.

In the following section, results are organized around the major research questions, examining perceived benefits and facilitators of study enrollment and components, barriers to study participation and meeting goals, and how study participation affected their relationships with their doctors. Results focus on themes mentioned by more than 20% of the sample. The median kappa was 0.92, with a range of 0.64 to 1.0.

A total of 110 of 120 planned interviews were completed (Table 1). They came from a targeted list of 163 patients, representing a minimum 67.5% response rate; 9 patients refused and the remainder were not reached prior to saturation. Patients providing interviews were demographically and clinically similar to those in the parent trial (

Prevalent themes are presented in italics, with quotations summarized in the 

). There were few differences between patients whose LDL-C improved greatly and those who did not improve. Similarly, meaningful differences among study arms were not prevalent. When substantial differences were discovered, they are noted at the end of each section.

Patients expressed a range of reasons for joining the study. First, the 

 of a doctor’s recommendation to join the study was a motivating factor for 40 (36%) patients. Second, 34 (31%) patients stated that they joined the study with the 

 to their high cholesterol. A third reason for joining was 

n = 30; 27%). Other relatively common motivators were 

(n = 27; 25%), including free and more frequent monitoring, and the 

 of helping with research (n = 25; 23%). The subgroup of interviewees whose LDL-C worsened gave more responses to this question, but the nature of what they said was not qualitatively different. The subgroup whose LDL-C improved often mentioned enrolling to facilitate healthy behaviors and to learn more about cholesterol issues.

Patients offered a wide range of responses to queries about the benefits of enrolling in the study. Seventy-one (65%) patients described 

 that affected the way they manage their health and their understanding of health-related topics. Commonly referenced topics were diet choice and exercise regimens. The study made patients think more about the foods that they were eating and the necessity of changing eating habits. These changes included reading labels before buying food, shopping for healthier food options, and portion control. The study also inspired patients to engage in more 

. Some learned that they were able to control their cholesterol with diet and exercise and others learned the necessity of medication in controlling their cholesterol.

Just over half (n = 61; 55%) of the patients spoke of appreciating 

. Several patients explained that being in the study made them more aware of the severity of their health condition; 34 (31%) specifically mentioned improved health. The subgroup of interviewees whose LDL-C got better more often gave examples of how they changed a health behavior and more frequently spoke of more consistent use of medication.

Benefits and Facilitators: Intervention Components

 More than half (n = 69; 63%) of the patients mentioned the benefits of receiving blood test results every 3 months; they 

, and tracking enabled them to make appropriate adjustments as well as understand how medication use affected their cholesterol levels. It was 

 to see how diet and exercise changes affected their cholesterol. They also appreciated the increased access to services (n = 48; 44%) and did not find the visits inconvenient (n = 47; 43%)

 Patients generally agreed that the wireless devices were a facilitator in lowering their cholesterol. The devices served as a 

to take medication (n = 39; 35%). Many thought the devices were easy to use and expected to continue to use them after the conclusion of the study. They liked being monitored (n = 33; 30%)

 Patients did not say much about the surveys. The most prevalent comment was about 

(n = 40; 36%). The surveys provided a friendly reminder of the appropriate steps that patients should be taking to keep their cholesterol under control.

 Seventy-five patients provided comments about financial incentives. Many (n = 53; 71%) described the 

 that the incentives had on their behavior, offering at least 4 different examples: 1) several thought it was good that the study provided them with 

 and have some extra money to spend on other things, 3) multiple patients stated that the incentives made the other inconvenient aspects of the study more 

, and 4) several patients felt the positive reinforcement 

 them to do the necessary things to lower their cholesterol and continue with the study. Most of the 42 (56%) who commented felt that the 

At the same time, 27 (38%) patients said that the incentives 

 on their performance in the study; positive results were what motivated them to keep up their efforts. They were involved in the study to get healthier and learn about themselves. The blinded review of the comments about incentives yielded nothing to distinguish among the treatment arms with patient incentives.

Only 11 (15%) patients recalled talking about the incentives with their doctors; these conversations were brief affirmations of their participation in the program and encouragement to continue with their progress. Overall, 38 (51%) patients made some commentary about physicians and incentives. Although there was no prevalent theme, the tone was generally indifferent (eg, it would not affect the patient’s own performance) or supportive (eg, they were happy for their doctors to receive payment as part of the study). Few feared that physician incentives would affect the physician’s impartiality.

There were no differences among subgroups of patients related to discussions about the 3-month visits, wireless medication devices, or the surveys. The subgroup of interviewees whose LDL-C improved more often spoke about the benefits of financial incentives.

 In contrast to the many experienced benefits, some barriers to improving cholesterol remained. The main downfall of 72 (65%) patients was their inability to stay away from food that is believed to contribute to higher LDL-C levels, and for others (n = 40; 36%), it was their inability to maintain a healthy lifestyle. A host of reasons were given, including

 who regularly ate or prepared unhealthful meals, 

. Comments about inadequate activity were often mentioned in the same context as 

. Multiple reasons were provided, often related to the weather, amount of time required for work, travel, or 

. Issues regarding the difficulties of maintaining a healthful diet and lifestyle were more prevalent among the subgroup whose LDL-C worsened over the study period.

 For more than half (n = 67; 61%) of the patients and for a number of different reasons, the wireless medication device used in this study (Vitality GlowCap) was seen as a nuisance, but usually temporarily. Patients on multiple medications did not like the necessity of having 

such as finding an appropriate outlet and connectivity difficulties; but for most, these were transitory issues amid an overall positive experience. The subgroup that did particularly well in improving its LDL-C more often noted issues with the wireless medication devices, but also that the issues were infrequent and/or solved.

 Numerous patients (n = 49; 45%) referenced regular medication use as a challenge. A significant part of the commentary was about 

, such as muscle weakness and pain and liver complications. Side effects, such as feelings of illness, muscle and joint pain, and “fuzzy memory,” were cited as reasons for dropping the medication altogether. Other patients 

 to take their medicine on a daily basis, and a significant number of patients stated they were 

 in taking any medication at all. The frequency and content of comments was similar between the high and low performers.

Only 1 patient specifically noted a change in the relationship between the patient and physician. However, 29 (26%) indicated that the study led to more conversations with their doctor—typically about the 

, especially if the labs demonstrated “poor results.” Some noted that they saw their physician more often due to participating in the study.

enrolled in the study because their doctor suggested

 it to them. Many of these were from the group who did well. Most of these comments noted that a brief conversation took place during the course of an office visit, in which the doctor brought up the study and recommended participation, and the patients subsequently agreed to join.

This qualitative process evaluation of a large randomized trial has 4 main findings. First, patients reported positive experiences with the study; they enrolled based on a genuine desire for a healthier life. The structure of the study was similarly favorable: blood draws every 3 months were informative, and the wireless medication devices—although they not without their glitches&mdash;provided valuable reminders to take one’s medication. Second, views about financial incentives were varied. Some clearly found them motivating, whereas others claimed they made no difference. Importantly, no patients found them negative. In general, discussion of the study with their physicians, including financial incentives, was not extensive. Third, patients noted that changing behavior and diet was difficult. Food preferences, habits, and social times interrupted plans to make positive changes. There was also ambivalence about taking medication, and for some, there was a clear desire to avoid medication. Fourth, patients who substantially lowered their LDL-C revealed themes similar to those who did not.

These findings contrast with some of the concerns the study team had when developing the study, as well as those voiced by others.

 In preparation for study launch, some team members and participating physicians wondered whether explicit financial incentives might adversely affect existing physician—patient relationships. Those concerns seem to have been unjustified. There was also concern that the wireless medication adherence devices would be seen as intrusive, but they were not. Most importantly, different experiences and feelings based on different intervention arms were anticipated; it seemed logical to think that whether a patient or their physician is being paid for outcomes might alter one’s sense of the process. In fact, no differences were found across arms. Nor were there systematic differences based on how well patients did in the intervention arm in terms of improving their LDL-C.

There is little previous research examining patient perceptions of financial incentives to improve their health. One other published paper found nearly equal proportions of a clinic population had favorable or unfavorable opinions of incentives, but the authors noted greater enthusiasm when incentives were framed as a reward rather than a penalty, when they were offered by health insurers as part of benefit design, and when the survey respondent personally had the condition for which the incentives were offered (in the case of smoking or obesity).

 Importantly, from a complementary perspective, this previous study was conducted among clinic patients who had not been enrolled in a trial of financial incentives.

Patients in the underlying study had to consent to participate; thus, by design, only the experiences of those who were willing to engage in the first place are represented in this study. High and low performers were oversampled, so it is possible that this sample was not effectively representative; however, the observation that no differences were observed between the extremes suggests that no differences would be observed among a more distributed sample. This study also has strengths: the team conducting this evaluation was different from the team that designed and implemented the trial, the sampling was systematic, and the sample included 110 patients, which is substantially larger than many process evaluations.

Behavioral interventions that include financial incentives offer promise to improve important health outcomes, but they must overcome 2 important evidentiary hurdles before widespread implementation. First, they need to demonstrate effectiveness. The results of the underlying study in this report suggest that a system of shared incentives offers significant benefit compared with a control incentive. Second, they must be socially acceptable. There are many stakeholders in such acceptance, including those who might finance these interventions, including the incentives, and the general public who might pass judgment upon them. This study's results suggest that patients are willing to engage in these programs and generally felt participating was a positive experience.&ensp;

The authors would like to acknowledge the team from the Mixed Methods Research Lab at the University of Pennsylvania who participated in conducting interviews and coding transcripts: Shimrit Keddem, Trocon Davis, Ebony Easley, Samuel Katz, Jalal Osman, Breah Paciotti, Ashley Ryman, and Richard Sorelle.

 Division of General Internal Medicine, Department of Medicine (JAS, KGV, DAA), Department of Medical Ethics and Health Policy (KGV, DF, DAA), Department of Biostatistics and Epidemiology (ABT), Center for Health Incentives and Behavioral Economics (KGV, KH, DAA), and Center for Health Care Innovation (KGV, DAA), Perelman School of Medicine at the University of Pennsylvania, Philadelphia, PA; Ohio State University Medical School (AA), Columbus, OH; Corporal Michael J. Crescenz 

 (KGV, DAA), Philadelphia, PA; The Wharton School, University of Pennsylvania (KGV, DAA), Philadelphia, PA; Atrius Health (TI), Newton, MA; Harvard T.H. Chan School of Public Health (MR), Boston, MA; Partners Healthcare System (TDS), Boston, MA.

Patients who enrolled in a trial to lower low-density lipoprotein cholesterol spoke positively of the multifaceted intervention: pillbox monitoring and financial incentives were socially acceptable.

Patients' Views of a Behavioral Intervention Including Financial Incentives

Medication adherence is often suboptimal, especially among patients on multiple chronic medications. We examined the association between synchronized medication refill schedules—which typically reduce organizational effort and logistical demands&mdash;and adherence.

Retrospective study among patients enrolled in Medicare Advantage prescription drug plans. 

 We used 2012 pharmacy, medical, and enrollment data linked with consumer meta-data for Medicare patients filling 2 or more maintenance prescriptions for antihypertensives, lipid-lowering agents, antidiabetic agents, antidepressants, and/or antiosteoporotic agents. Medication adherence for the year was measured using the proportion of days covered (PDC) at the drug class level. Patients were deemed adherent if drug class PDC was ≥0.80. Outcomes were compared between 1:1 propensity score-matched patients on synchronized versus nonsynchronized refill schedules for maintenance medications. 

The synchronized refill group showed better adherence than the control group, although the magnitude of effects varied by drug class and specific outcome measure. Mean PDC scores ranged from 0.02 higher for antihypertensives to 0.07 higher for antidepressants in the synchronized refill group relative to the control group (P <.01). Further, compared with the control group, a larger proportion of synchronized refill group members were deemed adherent, ranging from 6 percentage points higher for antihypertensives to 15 percentage points higher for lipid-lowering agents (P <.01). Differences between the synchronized and control groups were larger among exclusive users of retail versus mail order pharmacies for maintenance medications.

Synchronized medication refill schedules were associated with better medication adherence, particularly for patients filling maintenance medications exclusively at retail pharmacies.

	Prescription synchronization programs are popular, but evidence regarding their utility in improving adherence is limited. Our retrospective claims analysis of Medicare Advantage patients receiving multiple maintenance medications (ie, antihypertensives, lipid-lowering drugs, antidiabetic agents, antidepressants, and/or antiosteoporotic agents) compared patients with synced refill schedules to propensity score-matched patients without synced schedules.

		Compared with the nonsynced group, the synced group showed better adherence as measured by proportion of days covered. The magnitude varied by drug class.

		Differences between the synced and nonsynced groups were larger in patients exclusively using retail pharmacies compared with patients exclusively using mail order pharmacies.

		Synchronized prescription refills may facilitate adherence.

Poor medication adherence is a common problem among patients taking medications for a chronic condition.

 For patients receiving multiple medications for the treatment of more than 1 chronic condition, the resulting complexity can be particularly challenging.

 To address some of the practical burdens associated with long-term medication regimens, there has been increasing interest in whether synchronizing medication refill schedules—whereby all of a patient’s maintenance medications are refilled at the same time&mdash;may be a promising, scalable approach to improving adherence.

A recent review of the synchronization literature revealed few major peer-reviewed studies of synchronization.

 Holdford and Inocencio (2013) examined a convenience sample of patients at an independent retail pharmacy chain in the rural midwestern United States who were receiving new prescription medications, from 1 of 6 drug classes, to treat cardiovascular disease and diabetes.

 Intervention patients (ie, those who were participating in an appointment-based medication synchronization program that included synchronization of refills) had higher adherence across all drug classes than a control group of patients receiving usual care. A second study by Holdford (2015) among commercially insured retail pharmacy patients in Ohio who had been using medications for 6 months or more found similar results.

 Further, Choudhry et al (2011) used retrospective claims data from a large pharmacy benefit manager to examine therapeutic complexity in patients receiving refills for cardiovascular medications from a chain of retail and/or mail order pharmacies and found that less consolidated (ie, less synchronized) refill schedules were an independent predictor of worse medication adherence.

 Although valuable, the generalizability of these studies may be limited due to their focus on either specific geographic regions, retail pharmacy users, commercially insured patients, and/or a limited set of drug classes. 

Despite the absence of more generalizable evidence, there has been increasing adoption of medication synchronization programs across the nation.

 According to the American Pharmacists Association Foundation, more than 23 pharmacy chains and approximately 2000 independent pharmacies in the United States have implemented medication synchronization programs.

 An increasing number of states have proposed or recently passed legislation to remove barriers to medication synchronization, such as requiring insurers to pro-rate co-payments for the partial refills that are often needed to synchronize existing prescriptions.

 According to a recent national survey of 1000 adults 40 years or older who were receiving 1 or more medications for long-term medical conditions, about 1 in 8 patients were enrolled in a prescription synchronization program.

 In light of the growing adoption of, and policy interest in, synchronization programs, there is a pressing need for additional evaluations of the effectiveness of synchronizing refill schedules.

In this study, we utilized existing claims data for a sample of individuals enrolled in Medicare Advantage prescription drug plans (MA-PDs) who were receiving antihypertensives, lipid-lowering drugs, antidiabetic agents, antidepressants, and/or antiosteoporotic agents, to examine adherence outcomes in patients with synchronized medications compared with those whose medications were not synchronized. We focused on the Medicare population for several reasons. First, an estimated 85% of US adults 65 years or older have at least 1 chronic health condition, and 60% have at least 2.

 Second, given that forgetfulness and difficulties with transportation may increase with age, Medicare patients may be a particularly appropriate target population for interventions that reduce trips to the pharmacy and make it easier to manage prescription refills. Third, medication adherence for seniors has garnered increasing attention due to the CMS Five-Star Quality Ratings System program, which provides public ratings of Medicare Part D plans and ties performance to bonus payments.

 The ratings incorporate several medication adherence measures and award an increasing number of stars based on the percentage of a Part D plan’s members that meet adherence benchmarks.

Using 2012 pharmacy, medical, and enrollment data from Humana Inc’s top 50 (ie, highest enrollment) MA-PD plans linked with consumer meta-data files from the AmeriLINK Consumer Database,

 we took a 20% random sample of individuals meeting the following study criteria: 1) continuous eligibility for pharmacy and medical benefits in the same plan throughout 2012; 2) filling 2 or more prescriptions for antihypertensives, lipid-lowering drugs, antidiabetic agents, antidepressants, and/or antiosteoporotic agents; 3) a record of at least 2 different prescription fill dates for those medications in calendar year (CY) 2012; 4) exclusive use of either retail or mail order pharmacies to refill maintenance medications during the study year (ie, all prescriptions filled via 1 delivery channel); and 5) an equal number of unique maintenance drug classes in both the first half of the year and in the entire year (as a proxy for no major disruption in the maintenance drug regimen) to ensure that it would have been possible to  maintain synced refill schedules throughout the year.

The primary study group consisted of individuals who had a synchronized refill schedule for all of their maintenance medications for the entire study year—defined as having refills for all medications at every refill date during the year. For example, a patient taking 4 maintenance medications would need to show refills for all 4 medications at each refill date in order to be classified in the synchronized medication refill group.

In sensitivity analyses, we applied alternate eligibility criteria for the study group, representing less restrictive definitions of synchronization. In the first sensitivity analysis, patients were classified in the synchronized refill group if they refilled all maintenance drugs at the same time at the first 2 fills in the year (ie, started as synced, but did not necessarily remain synced). In the second sensitivity analysis, the synchronized refill group consisted of patients who refilled all maintenance drugs at once at any 2 consecutive fill dates during the year, regardless of when in the year this occurred.

The control group, which remained the same for all analyses, included patients who did not have any 2 consecutive fill dates during the year where they refilled all their maintenance drugs at the same time. The study protocol was approved by the University of Pennsylvania Institutional Review Board.

Our primary outcome was medication adherence at the drug class level, as measured by the proportion of days covered (PDC) for CY2012.

 In keeping with CMS’s technical guidance on measuring adherence to Star Rating medications, PDC was calculated as the proportion of days in the observation period (ie, 365 days in our study) with a day’s supply of a medication available from at least 1 prescription in that drug class.

 A continuous PDC measure was calculated for each of the 5 drug classes of interest among patients filling drugs for that class. We also created indicators for whether a PDC measure was ≥0.80, as this is a common cutoff for classifying a patient as adherent and also used by CMS for Part D Plan Star Ratings calculations.

It should be noted that although some mail order pharmacies offer patients the option to receive automatic refills, CMS regulations require pharmacies to obtain consent from Medicare beneficiaries prior to shipping every refill. Thus, patients must take an action (eg, via initiating a refill order or responding to a call) before medication is sent. CMS has noted that this requirement would be expected to increase the value (ie, accuracy) of refill-related adherence measures.

We applied propensity score matching to balance the synchronized refill group members with control members across a variety of characteristics. As a first step, we stratified the sample by 4 key variables likely to affect medication adherence in order to ensure that matched pairs were identical with respect to the following parameters: 1) type of pharmacy where maintenance medications were filled (exclusive use of retail pharmacy vs exclusive use of mail order pharmacy); 2) insurance plan identifier, to account for differences in insurance benefits structure, co-payment requirements, drug formulary and utilization management tools, and other plan-specific features; 3) total number of maintenance medications; and 4) combination of maintenance drug classes being used (eg, a synchronized refill group member taking only antihypertensives and lipid-lowering medications would be matched to a control group member taking only those 2 classes of maintenance drugs).

Only strata with at least 1 synchronized refill group member and 1 potential control were considered in the matching algorithm. Within each stratum, we matched along a wide range of 310 variables, including: 1) sociodemographic variables (eg, age, gender, marital status, Medicare Part D low-income subsidy status, disability status, geographic region, county-level percent on Medicare Advantage); 2) Census tract-level measures for variables including home ownership, education, household income, and home value; 3) clinical variables (ie, medical condition indicators and the CMS Hierarchical Condition Categories (CMS-HCC) score)

; 4) prescription cost variables, including total prescription costs for the year, total prescription costs for maintenance medications only, and total out-of-pocket costs; and 5) consumer meta-data variables used to characterize plan members, including estimated income, interests, hobbies, consumer personas (eg, young urban singles, struggling seniors), occupation (including retirement status), attitudes (eg, positive outlook, control seekers), and tendency to drink alcohol and/or smoke. These variables were obtained from and/or predicted based on demographics, purchasing behavior, public records, Census data, consumer surveys, and other primary sources.

The success of propensity score matching was evaluated by comparing the means of matched synchronized and control group members along the 310 variables used. We calculated Cohen’s d to estimate the size of any differences between the matched groups across all variables; any sample characteristic with a Cohen’s d <0.10 was considered insignificant as per standard practice.

Using the matched sample, we performed paired 

 tests for continuous outcomes and McNemar’s tests for categorical outcomes. We also conducted subgroup analyses by type of pharmacy used to fill all maintenance medications (exclusively retail vs exclusively mail order). We then tested—via regression analyses with interaction terms for the continuous PDC scores and Mantel-Haenszel tests for the dichotomous adherence measure (PDC ≥0.80)&mdash;whether the association between synchronized refills and medication adherence differed for exclusively mail order pharmacy users compared with exclusively retail pharmacy users.

A retrospective claims analysis showed that synchronized refill schedules were associated with better medication adherence among Medicare Advantage patients taking multiple maintenance medications.

Andrea B. Troxel, ScD

Articles

Physician Practice Variation Under Orthopedic Bundled Payment

Patients' Views of a Behavioral Intervention Including Financial Incentives

Synchronized Prescription Refills and Medication Adherence: A Retrospective Claims Analysis

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

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