Analyzing the Impact of Different Value-Based Insurance Design Programs

It is widely acknowledged that adherence to long-term pharmacotherapy is poor for the majority of chronic conditions, and that inadequate adherence rates constitute a serious public health concern.¹ Because increasing costs of drugs are implicated, in part, for low adherence levels,² several cost-containment interventions have been implemented over the years.³ These include cost sharing, which is built on the premise that by transferring some costs from payers to patients, plan members will have the incentive to find less expensive options, reduce their out-of-pocket expenses, and help to improve overall cost-effectiveness.⁴ A growing body of evidence, however, suggests that cost sharing further exacerbates already low adherence rates,^5-9 even when copayment increases are modest such as from $12 to $20,⁹ and interferes with preventative care.^10,11 A recent comprehensive literature analysis demonstrated that prescription spending decreased 2% to 6% for every 10% increase in patient cost share, depending on medication and patient health status.³

In addition to increasing patients’ out-of-pocket expenses, cost sharing relies on the dubious expectation that patients will optimally balance likely clinical benefits and costs in their treatment decisions.^4,12,13 To address these issues, Fendrick et al proposed the Value-Based Insurance Design (VBID) model, which links copayments to value (both benefits and costs) of clinical services, versus costs alone as in strict cost-sharing models.¹⁴ In its initial formulation, VBID proposed tailoring cost sharing to individual patients, but administrative challenges and prohibitive costs were immediately apparent. Over time, VBID experimentation, especially by self-insured employers, led to 2 practical approaches. The more common form reduces or waives copayments for clinically valuable products and services irrespective of the health status of the patient. The second, which relies on more advanced data handling capabilities, employs differential copayments based on patient characteristics, and selects specific diagnoses for copayment waiver or reduction.⁴ In essence, VBID has emerged as a clinically sensitive refinement to benefit design that realigns cost share to therapeutic value² so that the greater the clinical value to the patient, the lower the cost share.¹⁵ There are early indications that over the years VBID programs have helped to increase medication adherence and reduce costs.^16-18

One prominent example is the benefit design developed by Pitney Bowes, a large national employer. To address the need for greater adherence to diabetes medication among its employees, the company reclassified all tier 2 or 3 diabetes drugs and devices in its formulary as tier 1. This decision decreased the coinsurance of branded products to 10% from the prior cost sharing of 25% to 50%; the coinsurance on generic products was 10% also. In approximately 2 to 3 years, medication possession ratios (MPRs) increased significantly, mean total pharmacy costs decreased by 7%, and emergency department (ED) visits went down by 26%. Participants’ overall healthcare costs also decreased, with the net per-plan-participant cost in 2003 at $4000 versus the industry benchmark of $6500.^16-18

A handful of employers have made VBID programs available to their employees⁴ in an effort to address the increasing prevalence, medical costs, productivity loss, disabilities, and early mortality associated with high rates of diabetes in their workforces,¹⁶ which largely mirror the national trend. In 2002, more than 1 in every 10 healthcare dollars spent in the United States was linked to diabetes, with the management of diabetes-related chronic complications alone amounting to $24.6 billion that year.¹⁹ The American Diabetes Association estimated that 23.6 million children and adults (7.8% of the US population) had diabetes in 2007, with direct diabetes-related medical costs of $116 billion, and total costs of $174 billion for the year.²⁰ As part of their VBID initiative, a number of employers are taking concerted steps to help improve their workers’ health by providing diabetes treatments at substantially reduced or waived copayments.^16,17 VBID programs, which are often data-driven, seem to be a good fit for diabetes management because both prevalence and treatment adherence rates are readily identifi able via linked medical and pharmacy claims data. A number of VBID programs are accompanied by additional patient support such as instructions on condition management and increased education.

This article provides results of separate analyses for 2 VBID programs, which were implemented in response to high diabetes prevalence rates, inadequate medication adherence, and rising costs. One was conducted in a large national employer and the other was in a northeastern state. We evaluated medication adherence, resource utilization including appropriate preventative care, and healthcare costs for employees with diabetes. For the study conducted within the northeastern state, we evaluated the differences in outcomes for employees who participated in a VBID program that provided waived copayments for participation in a mandatory Telephonic Diabetes Education and Support (TDES) initiative and a matched control group from members in the same state that had neither copayment reductions nor diabetes education support. The second study evaluated the impact of a 50% reduction in copayments, on a prior approval (PA) basis, for diabetes medications for members of a national employer. The comparator group, which included members with diabetes from another national employer in the same industry, had no VBID copayment reductions and served as a quasi-control group.

METHODSStudy Design

For both VBID initiatives, the analyses were designed as retrospective, observational cohort studies and utilized the HealthCore Integrated Research Database (HIRD), which contains linked health plan eligibility, medical, and pharmacy claims data for covered services for 1 in every 9 Americans. The HIRD accesses 14 commercial health insurance plans that operate in the northeastern, southeastern, mid-Atlantic, central, and western regions of the United States.

For the study of the 2 national employers, which operated in the same industry and geographic regions, we performed a cross sectional analysis that used 2008 as the comparison year and 2007 as the baseline year. This design was selected because the employer group with the reduced copayments already had this benefit in place when it initiated coverage with the health plan at the beginning of 2007. Complete data were available on the cohorts from July 1, 2007, through January 1, 2008, and until December 31, 2008. This provided the study with a 6-month baseline phase and a 12-month follow-up period, with January 1, 2008, as the index date. All patients were required to have a prescription for an antidiabetes medication within 3 months prior to January 1, 2008. For the study in the northeastern state, employees who participated in the TDES program between January 1, 2005, and September 30, 2008, were identified, and the date they enrolled in the program was defined as the index date. The control group consisted of employees from the same state who were not participating in the TDES program, as well as other health plan members in the same state who were not offered the program. This analysis complied with the rules covering usage of protected health information (PHI), including privacy requirements embodied in the Health Insurance Portability and Accountability Act (HIPAA) of 1996.

Study Population

For both studies, patients with at least 1 medical claim for diabetes (International Classification of Diseases, 9th Revision, Clinical Modification [ICD-9-CM] 250.xx) or at least 1 pharmacy claim for an anti-diabetes agent (General Product Identifier [GPI] code starting with 27, but excluding 2730) within the study period were included. To be included, patients in the national employer cohort were required to have a minimum of 18 months of continuous eligibility (6 months pre-index and 12 months post-index), while a minimum of 24 months of eligibility (12 months in the pre-index and 12 months in the post-index period) were required for study patients in the TDES group.

Patients were excluded if they were 64 years or older at the index dates because complete medical claims data would cease to be available if Medicare became their primary benefit plan at age 65 years. Also excluded were patients with evidence of high-cost conditions such as cancer, organ transplant, HIV/AIDS, and end-stage renal disease, to avoid disproportionate skewing of the relatively small sample sizes by high-cost outliers.

In the analysis of the single northeastern state cohort, an initial 3:1 exact matching method was used, in which TDES participants were randomly matched with nonparticipating control subjects on the basis of age (±1 yrs), gender, index year, the presence of type 2 diabetes, and oral anti-diabetes drug use in the 12-month pre-index period. Because the plan type and diabetic comorbid conditions showed significant differences between the TDES participants and the exact matched controls, a 1:1 propensity score match was generated via a logistic regression model with predictors of age, gender, plan type, baseline comorbidity index, and the presence of diabetic nephropathy, neuropathy, and retinopathy. The analysis of participating employees and controls at the 2 national employers did not use the matched cohort design.

Study Measures

The main outcome measures for both VBID programs were medication adherence, diabetes resource utilization and costs, use of services for comprehensive diabetes care, and utilization and costs of all healthcare services, regardless of diagnosis. We assessed medication adherence using MPR, which was defined as the sum of the total supply (number of days) of any oral anti-diabetes medication divided by a 1-year period (the duration of the follow-up). Patients with an MPR score >0.8 were defined as adherent. The use of comprehensive diabetes care services for cholesterol, blood sugar (using glycated hemoglobin [A1C]), and kidney function testing, as well as for ophthalmological examinations, was evaluated. Diabetes-related and allcause treatment costs were the sum of plan-paid costs and patient-paid costs. Healthcare utilization was measured by type of service including inpatient hospitalizations, ED visits, physician office visits, other outpatient medical visits, and prescription medications.

Statistical Analyses

The impact of waived or reduced copayments and TDES were assessed by comparing MPR and diabetes-related economic outcomes for employees with diabetes and comparable controls. Chi-squared tests were used for comparisons of corresponding patient cohorts for categorical variables, and t tests for evaluations of statistical differences between cohorts for continuous variables. Multivariate analyses were used to evaluate the effect of the VBID program on medication adherence and diabetes related economic outcomes, adjusting for differences in baseline demographics and disease characteristics. Considering the property of cost and utilization data, the generalized linear model approach was employed using the Poisson family for utilization data and the Gamma family for cost data and the log link function. A conventional alpha of 0.05 was used unless otherwise specified. Statistical analyses were performed using SAS version 9.1 (SAS Institute Inc, Cary, North Carolina).

RESULTSPatient Characteristics at Baseline

In the cross-sectional analyses of the 2 national employers, there were 715 employees with diabetes in the PA program, of which 53% were female, and 497 in the control cohort, of which 50% were female (

Table 1

). Patients in the PA cohort were older, with a mean age of 50 (±10) years compared with 48 (±12) years for the control group. The vast majority of subjects, 88% in the PA program and 87% in the control group, were on oral antidiabetes medication; 21% of the subjects in the PA cohort were on insulin compared with 20% in the control group. The median copayment for diabetes medications was $4.00 in the PA program and $12.40 in the control group (P <.0001). Baseline Deyo-Charlson Comorbidity Index (DCI) scores were significantly higher for the PA group versus control group (1.31 vs 1.06, P <.0001). There were also significant between-group differences in hypertension and diabetic nephropathy; however, no significant differences were reported in all-cause or diabetes-related medical utilization at baseline.

A total of 474 patients were included in the analysis of the northeastern state employees (237 in the TDES program versus 237 in the matched control). In both groups more than 50% of the employees were female (Table 1). The mean age of the patients at baseline was 54 (±7.5) years, and baseline demographic and clinical characteristics were similar for the 2 groups, except that participants in the TDES program had a higher percentage of diabetes-related medication use (84.8% vs 75.9%, P = .02). In the TDES group, 69.2% of the patients were on oral diabetes medications versus 67.1% in the control group (P = .62), and 33.8% and 21.5% were on insulin in the respective groups (P <.01). Adherence rates were higher for both case and control groups (0.76 vs 0.74) during the baseline period. There were no significant differences in baseline diabetes-attributable costs (P = .64).

Medication Adherence

Figure 1

shows the proportion of patients who were adherent to medication at 1-year follow-up. Medication possession ratio scores in the follow-up period showed that in both the PA and control groups less than half of all patients were adherent, although there was a significantly greater proportion of adherent patients (43%) in the PA group versus 36% in the control group (P = .028). The mean MPR was 0.68 in the PA group versus 0.64 in the control group, which was not significant, after adjusting for baseline MPR, DCI, and comorbidities. In the TDES versus control analysis, although there were more patients with MPR scores >0.8 in the TDES group, there was not a significant difference in the percentage of patients (68% TDES vs 65% control, P = .60) adherent with diabetes medications. The estimated mean MPR was not significant (0.82 for TDES vs 0.77 for control, P = .06) after controlling for baseline anti-diabetes agent use.

Diabetes Costs

In the PA group versus control, there were no differences in diabetes comprehensive care services such as A1C, low-density lipoprotein cholesterol (LDL-C) tests, and eye examinations. Despite the lower medication copayment, the group with the PA benefit had significantly more inpatient admissions, ED visits, physician office visits, and ancillary outpatient services, resulting in significantly higher total diabetes-related healthcare costs. As

Table 2

shows, patients in the PA case cohort had mean total all-cause costs of $10,872 compared with $8819 in the control group. For diabetes-attributable costs, the PA case group had significantly higher unadjusted costs compared with the control group, $4836 versus $3452 (P <.05), respectively. Similar differences were observed in the adjusted costs, with mean estimated diabetes costs significantly greater for the PA case group ($3600) versus the control group ($2945; P <.05).

Also shown in Table 2, at 1-year follow-up, there were no significant cost differences between the TDES case and control groups. After controlling for baseline anti-diabetes medication use and baseline diabetes-related costs, adjusted post—12-month diabetes-related costs for all participants in the TDES program were estimated at $5380, and all-cause costs were $13,296, compared with $4482 and $13,375, respectively, for the control group. For participants using diabetes medications during the follow-up period (n = 314), compared with the control group, the estimated diabetes-related costs were about $837 higher, whereas the all-cause costs were about $762 lower for the case group.

Diabetes Resource Utilization

In the TDES and matched control study, there were no significant differences in inpatient admissions, ED visits, and physician-office visits; however, participants in the TDES program used significantly more outpatient services (P <.01). Reviewing these claims more closely, it becomes evident that participants in the TDES program had significantly better comprehensive diabetes care, as measured by a significantly higher percentage of patients receiving A1C tests (85.7% vs 69.2%), LDL-C tests (65.8% vs 56.5%), kidney tests (57.8% vs 43.5%), and eye examinations (46.6% vs 35.4%), as shown in

Figure 2.

DISCUSSION

The findings in this analysis confirm suboptimal medication adherence rates even for chronic conditions.¹ There is wide acceptance that commonly employed cost-effectiveness measures such as copayment and coinsurance may be worsening these already poor treatment adherence rates.^5-7 Employers faced with escalating costs such as those of the northeastern state in this study, which paid about $20 million a year for its 2000 employees with diabetes, are seeking to implement initiatives that will improve treatment outcomes while keeping costs under control. Both the TDES and PA programs were implemented as VBID initiatives intended to advance those goals.

In the analysis of the large national employers, we found a significantly higher percentage of individuals with the PA benefit to be adherent with medication. This difference could be attributable to the significant difference in the copayment amounts; $8.00 between employees with the PA benefi t and the control group. However, because we do not know what the adherence rates were in this population prior to the copayment reduction, it would be speculative to attribute this difference to the copayment disparity, especially since the adjusted mean MPR was not significantly different. Additional rationale for not attributing this outcome entirely to copayment differences may be found in the mixed results from prior work, which suggest that the consequences, especially long term, of benefit changes such as financial incentives were still largely uncertain.^3,4 Targeted educational and disease management support, along with patient coaching, are increasingly being linked to better adherence, with or without financial incentives. In this analysis, however, the majority of the patients were receiving generic anti-diabetes medication, and reducing the lowest copayment tier even further may not have offered enough of a cost differential to change their response. The evidence from the Pitney Bowes initiative suggests that if branded medications were reduced to lower tier costs for the 2 groups in this study, it would have been reasonable to expect more pronounced differences in adherence rates.

Although the TDES program had a strong educational component, we did not find that educational support in addition to waived copayments had a significant impact on medication adherence. However, the TDES enrollees were associated with significantly more utilization of preventative services, which is reflective of a higher level of patient awareness and overall care, suggesting a benefit from the educational component. This indicates that patients who are more compliant with their medication regimen may be more active in the pursuit of other healthcare services, a trend that was observed among the participants in the PA program. The participants who were adherent with diabetes medications had significantly greater levels of comprehensive diabetes care such as A1C tests, LDL-C tests, and eye examinations. The more adherent members of this cohort also had significantly less ED and inpatient hospital services than participants who were not compliant. In addition, they had higher rates of services, levels generally associated with preventative care, including larger numbers of physician office visits, more laboratory testing, and higher pharmacy utilization and costs. Greater adherence is associated with reduced total costs over the medium or long term. In the first 1 or 2 years with patients with diabetes, as we demonstrated in this study, the increased use of medication, diabetes supplies, and laboratory and other services are expected to increase patient costs. However, we did not find a significant reduction in high-cost services, including inpatient hospitalizations and ED visits, which would offset the greater use of services for preventative care.

The implementation of VBID programs involves a number of challenges, including administrative, cost, data management, legal, privacy, and appropriateness of diagnosis and medication choice issues.⁴ While it appears that patient education is an essential element for successful implementation and improved health and cost outcomes, additional studies on different aspects of VBID initiatives are still required. A recent study to evaluate the effectiveness of TDES among high-risk and high-cost patients with diabetes and coronary artery disease showed that patients who received telephone-based management had significantly lower all-cause hospital admissions, diabetes-related admissions, and all-cause medical costs compared with patients who did not.²¹ While we observed a trend toward seeking better care among the TDES patients in this analysis, it is possible that the impact of the intervention was less pronounced because all of the patients were not in the high-risk category, and had lower risk of short-term events. These results point to the need for additional studies over longer time frames to further evaluate the impact of VBID programs as well as their components such as TDES.

Limitations

Observational claims data were used throughout this analysis. While this is an excellent starting point in observational research, analyses using claims data may be subject to being incomplete and to unobservable factors, which can influence outcomes. This also means that a number of clinical metrics, such as patient glucose control, were not factored into the analysis. Even though we excluded high-cost conditions from this analysis, cost results are sensitive to outliers; that is, a few extreme values capable of skewing the impact of the intervention, particularly when sample sizes are small. This analysis did not control for different income levels among the participants; income levels have been known to influence long-term medication adherence. The copayment reduction was not substantial between the PA and control patients employed by the 2 large national employers. In fact, the lack of significance and small effect sizes observed between these groups may be an indication that the size of the copayment reduction in this study constitutes a weak intervention by itself.

CONCLUSIONS

While VBID initiatives are still in the fledgling stage, and may not always result in cost savings, as in the 1-year follow-up studies discussed here, they do offer a way to further refine such blunt cost-containment instruments as cost sharing. Accumulating evidence seems to suggest that while copayment reductions or waivers are a vital part of VBID initiatives, they may prove to be insufficient for the attainment of the goals of improved long-term health outcomes and cost savings unless some efforts are focused on patient education.