Value-based insurance design copayment reductions sustained medication adherence 2 years into policy implementation and were most effective in patients with poor adherence before policy implementation.
To determine whether participation in a value-based insurance design (VBID) program was associated with improved medication adherence in 8 drug classes 2 years after implementation and to examine whether adherence changes varied by baseline adherence.
We used a pre-post quasi-experimental study design with a retrospective cohort of 74,748 enrollees using 8 different therapeutic classes of medications to treat diabetes, hypertension, hyperlipidemia, or congestive heart failure.
Brand-name medication copayments were lowered (from tier 3 to tier 2) for all enrollees, while generic copayments were waived only for employers who opted into the VBID program. Medication adherence of VBID program participants and nonparticipants 12 months before and 12 and 24 months after program implementation were estimated on 8 propensity-matched cohorts using generalized estimating equations, as well as on subgroups stratified by baseline adherence. Adherence was measured using the medication possession ratio (MPR) from medication refill records.
VBID was associated with improved medication adherence ranging from 1.4% to 3.2% at 1 year, which increased to 2.1% to 5.2% 2 years following VBID adoption. Adherence changes were most notable among patients who were nonadherent (MPR <.50) before VBID implementation.
Population-based implementation of VBID can improve adherence to medications to treat cardiometabolic conditions, particularly for previously nonadherent patients. VBID guidelines being developed in response to healthcare reform should account for the heterogeneity in patient response to VBID programs.
(Am J Manag Care. 2012;18(5):265-274)We showed significant clinical improvements in medication adherence among patients who experienced a reduction in copayments following the implementation of a valuebased insurance design (VBID) copayment program by a larger private insurer.
Adherence to essential medications is suboptimal for many patients with chronic medical conditions,1,2 resulting in significant morbidity and mortality.3,4 Cost-related nonadherence (CRN) has been well documented as a significant problem in uninsured patients.5-7 Cost-related nonadherence leads patients to forgo medication, skip doses, ration medication by splitting tablets, or forgo other necessities. Insured patients also face CRN due to ever-increasing medication copayments and coinsurance.8,9
One policy innovation that targets CRN in the context of health insurance is value-based insurance design (VBID). Value-based insurance design posits that cost-sharing should be set according to a medication’s clinical value instead of its acquisition cost.10 Following principles of VBID, policy makers may set copayments lower for medications that are more effective or more cost-effective than other medications in the same drug class. An alternative VBID approach would reduce cost sharing for certain populations of patients or subgroups who are most likely to benefit from improved access to treatment.11 The removal of cost barriers for high-value medications under VBID should lead to better medication adherence, better disease management, and lower healthcare spending. Existing VBID evaluations suggest that eliminating copayments for high-value services may result in 1-year adherence improvements ranging from 1.5% to 3.8%, depending on the therapeutic category examined.12,13 However, few studies have examined adherence changes beyond 1 year,14,15 which would help insurers optimize patients’ behavioral response to VBID.
The objective of this study is to determine whether participation in a population-based VBID program was associated with improved adherence in 8 drug classes 2 years after implementation. We also wanted to address the concerns of VBID critics who argue that statistically significant, yet modest, improvements in adherence may not be clinically meaningful by examining adherence changes among populations with varying levels of prepolicy adherence.16,17 Previous research suggested a clinical threshold for adherence at 80%18-20 so we examined whether adherence changes differed between patients who were fully adherent (>80%), somewhat adherent (between 50% and 80%), or nonadherent (<50%) in the year before VBID implementation. The potential heterogeneity in patient response may reflect subgroups of patients for whom CRN is the underlying reason for nonadherence, and who might be exceptionally responsive to VBID. The results from this study extend a 1-year adherence analysis13 and identify which patients would most benefit from VBID programs. This is particularly relevant as the Secretary of Health and Human Services develops guidelines for VBID implementation under healthcare reform as stipulated in section 2713 of the Patient Protection and Affordable Care Act.
In January 2008, BlueCross BlueShield of North Carolina (BCBSNC) instituted a VBID program termed “Medication Dedication” for medications to treat diabetes, hypertension, hyperlipidemia, and congestive heart failure. Generic copayments for these medications were waived for all fully underwritten employers and for a subset of self-funded employers who opted in to this program. All employees and dependents at the employers who participated had the new benefit applied. In addition, brand-name copayments for 8 different therapeutic classes of medication (metformin, HMG-CoA reductase inhibitors [statins], thiazide diuretics, angiotensinconverting enzyme inhibitors [ACEIs], beta-blockers, calcium channel blockers [CCBs], angiotensin receptor blockers [ARBs], and cholesterol absorption inhibitors [CAIs]) were lowered from tier 3 to tier 2 for all enrollees. As a result of this policy, per prescription copayments for VBID participants compared with non-VBID participants declined on average from $15.57 to $2.42 versus $16.23 to $12.91, respectively, for ACEI users; from $15.05 to $2.07 versus $15.63 to $12.74, respectively, for beta-blocker users; from $13.13 to $5.17 versus $14.19 to $14.16, respectively, for metformin users; from $21.93 to $6.14 versus $23.95 to $16.28, respectively, for CCB users; from $24.89 to $19.46 versus $27.15 to $25.66, respectively, for statin users; from $16.91 to $9.14 versus $17.63 to $16.00, respectively, for thiazide users; from $36.31 to $32.28 versus $38.42 versus $32.65, respectively, for ARB users; and from $37.09 to $32.90 versus $40.41 to $33.90, respectively, for CAI users. This policy was population based because it was made available to all employers offering health benefits through BCBSNC in 2008. This study was approved by institutional review boards at both Duke University and the University of North Carolina at Chapel Hill.
Study Design and Sample
This VBID evaluation used a retrospective pre-post quasi-experimental study design with a nonequivalent control group. The 12 months prior to program implementation (January-December 2007) was the pre-period. Administrative claims data were used by BCBSNC to create annual observations for each of the variables described below between 2007 and 2009. Data for the post-period were drawn from the subsequent 24 months (2008 and 2009) to examine adherence changes 1 and 2 years after program implementation. The unit of analysis was the person-year with 3 observations per person.
VBID participants and nonparticipants were included if they were continuously enrolled from January 2007 through December 2009, did not have a change in their VBID enrollment status from 2008 to 2009, were 18 years or older in 2007, and were taking at least 1 of the 8 classes of drugs previously indicated in 2007. The comparison group of nonparticipants was selected from BCBSNC members enrolled in Administrative Services Only benefits. These patients also experienced a reduction in copayments for prescriptions in the 8 therapeutic categories examined from tier 3 to tier 2. However, copayments for generic medications were not eliminated. The same enrollee could be included in analyses for more than 1 class of drugs if using medications from 2 or more of the 8 classes. Given that these medications are all used for chronic health conditions, we used an intention to treat approach whereby patients in the analytical cohort were followed until the end of the study (2009).
After applying these criteria, we identified 5020 participants and 2883 nonparticipants taking metformin; 16,771 participants and 10,204 nonparticipants taking diuretics; 14,978 participants and 8234 nonparticipants taking ACEIs; 12,164 participants and 7298 nonparticipants taking beta-blockers; 21,635 participants and 12,804 nonparticipants taking statins; 8045 participants and 4834 nonparticipants taking CCBs; 3301 participants and 2073 nonparticipants taking CAIs; and 8688 participants and 5705 nonparticipants taking ARBs.
Of the 8 classes, 2 (CAIs and ARBs) did not have any generic options during the period of observation, so VBID participants and nonparticipants both experienced the same copayment reduction from tier 3 to tier 2. We expected similar changes in CAI and ARB adherence between VBID participants and nonparticipants because the copayments were statissimilar for both groups, so CAIs and ARBs were added as nonequivalent dependent variables to strengthen the design of the study.
Medication Adherence Outcome
Medication adherence was assessed using the continuous medication possession ratio (MPR), calculated as days of supply for a specific therapeutic class during each of the 3 annual observation periods. The MPR was calculated as the number of days of supply dispensed per year over the number of days observed in the year (365) and was capped at 1 for patients filling more days of supply than days observed. Adjustments to the days of supply in the MPR were made to account for carryover from previous medication fills, including carryover for medication fills that preceded each observation period including the pre-period (2007). If there were no fills for the drug during the 90 days before the start of the period, then the start date was the date of the member’s first fill for the drug during the period. The MPR accounted for medication switching between different drug therapeutic classes (eg, from a CCB to an ACEI) to avoid undercounting the supply of the drug (eg, a CCB) that members were no longer taking.
There were 3 explanatory variables of interest: (1) an indicator of VBID participation, (2) a time indicator to reflect the pre-period or post-period(s) around VBID implementation, and (3) an interaction of the VBID participation and pre-post indicators. This interaction term indicates whether the pre-post adherence trends differed significantly between program participants and nonparticipants (eg, a differencein- difference analysis). Separate comparisons were made between 2007 and 2008 as well as 2007 and 2009 to understand the potential for VBID to result in both near-term and sustained adherence benefits.
Consistent with past research, each of the models included age in years, male sex, and comorbidity burden (measured as Episode Risk Groups) as covariates. In addition, we controlled for several covariates not accounted for in prior VBID analyses that reduced the extent of unobserved confounding by controlling for the count of unique medications filled, the average generic copayment per 30-day supply, the average brand-name copayment per 30-day supply, whether a patient filled at least 1 prescription with a 90-day supply, and the generic dispensing rate. In addition, patient use of case management or disease management during baseline was included to control for the influence of additional program participation on medication adherence.
Statistical Model Specification
To account for nonnormality in the adherence outcome, we used generalized estimating equations (GEEs) with a gamma distribution, inverse square root function link, robust standard errors to account for repeated measures, and identical covariate specifications for each of the 8 medication classes. Based on GEE results, the impact of the VBID program on adherence was assessed by comparing the difference in program participants’ predicted adherence and their predicted adherence had they not participated in the VBID program. The difference in these 2 predictions (done only for program participants) represents the “treatment effect for the treated.”
To reduce the nonequivalence of the control groups from the imbalance in observed covariates, one-to-one propensity score matching was conducted by iteratively matching program participants to nonparticipants from the eighth to the second digit of the propensity score in the mean adherence models. In the propensity score analysis, we included the covariates described above and 4 interaction terms (case management and disease management, male and case management, male and disease management, and total number of unique medications used and disease management). We present propensity-matched results, because our prior work found concordance between the unmatched adjusted results and matched results.13
>0.80). A GEE with the same specifications as the 2-year adherence analysis was used within each stratum.
Prior to propensity score matching, there appeared to be significant differences between treatment and comparator groups on several variables (). For example, VBID participants taking ACEIs were slightly younger (51.7 vs 52.6 years) and more frequently male (64.6% vs 55.6%); had a higher comorbidity burden (47.8 vs 49.7); used more disease management (20.4% vs 16.9%); and had lower rates of 90-day fills (11.6% vs 21.3%), lower baseline copayment amounts ($10.65 vs $11.30), and lower overall health expenditures at baseline ($7086 vs $7460) than non-VBID participants taking ACEIs. Following propensity score matching, statis
In addition to the primary 2-year adherence analysis, we examined adherence changes following VBID among patients with varying baseline adherence in 2007. Patients were stratified into 1 of 3 adherence categories based on 2007 adherence levels (nonadherent (MPR <0.50), somewhat adherent (0.5 tical differences between groups were eliminated, allowing for meaningful comparisons between groups on adherence differences resulting from the VBID Medication Dedication program.
Unadjusted Adherence Comparisons
Prior to implementation of Medication Dedication in 2007, adherence rates were similar between treatment and comparator patients across each of the 8 therapeutic categories (). Average MPR rates in 2007 ranged from 75% for statin medications to 83% for ACEIs for both program participants and nonparticipants. Following VBID implementation, average adherence declined less for program participants than for nonparticipants in 2008 and 2009. Compared with nonparticipants, 2008 adherence was higher for patients taking ACEIs (0.84 vs 0.82), beta-blockers (0.82 vs 0.80), metformin (0.76 vs 0.73), CCBs (0.84 vs 0.82), thiazide diuretics (0.80 0.78), and statins (0.74 vs 0.73). The adherence differences continued to grow in 2009 for ACEIs (0.83 vs 0.79), betablockers (0.81 vs 0.77), metformin (0.77 vs 0.71), CCBs (0.83 vs 0.80), thiazide diuretics (0.80 vs 0.75), and statins (0.76 vs 0.73). As expected, VBID participants and nonparticipants showed similar adherence changes for ARBs and CAIs.
Propensity Score—Matched Adherence Comparisons
In adjusted analyses (), percentage point adherence from 2007 to 2008 improved 0.9% (P = .02) for VBID participants taking CCBs, 1.4% (P <.001) for statin medications, 2.2% (P <.001) for beta-blockers, 2.5% (P <.001) for ACEIs, 2.8% (P <.001) for thiazide diuretics, and 3.2% (P <.001) for metformin. Compared with VBID nonparticipants, adherence improvements were sustained into 2009. Percentage point improvements from 2007 to 2009 were higher for each class of medication over this 2 year period than the period from 2007 to 2008. Percentage point improvements from 2007 to 2009 improved 2.2% (P <.001) for CCBs, 2.3% (P <.001) for statin medications, 4.3% (P <.001) for beta-blockers, 4.8% (P <.001) for ACEIs, 4.5% (P <.001) for thiazide diurectics, and 5.0% (P <.001) for VBID participants taking metformin. There were no significant differences in adherence trends for VBID participants and nonparticipants using CAIs or ARBs in 2008 or 2009.
Adherence Changes Greatest Among Those Previously Nonadherent
Consistent with our a priori hypothesis, the greatest adherence improvements were observed for patients with poorer baseline adherence (). Compared with patients who were fully adherent at baseline, percentage point adherence improvements were greatest for previously nonadherent and somewhat adherent VBID participants taking ACEIs (9.7% [nonadherent], 5.4% [somewhat], and 4.4% [fully]); betablockers (6.1%, 5.5%, and 3.7%); metformin (6.6%, 8.8%, and 4.0%); CCBs (5.2%, 4.5%, and 1.9%); statins (3.0%, 3.4%, and 2.2%); and thiazide diuretics (7.3%, 5.2%, and 3.7%).
With the explicit authorization of VBID in section 2713 of the Patient Protection and Affordable Care Act, experimentation with medication cost sharing is likely to increase significantly. Given the escalating interest in cost-sharing reforms, it is essential that these imminent changes be informed with rigorous evidence. In this study, we evaluated the impact of the first population-based implementation of VBID that we know of on adherence to 8 distinct drug classes. Consistent with our earlier work,13 we found that medication adherence improved 1% to 3% for VBID participants 1 year into implementation of the VBID program. We also showed that adherence improvements were sustained and became larger 2 years into the program, consistent with 1 prior study.15 Adherence improvements 2 years into the program ranged from 2 percentage points for CCBs to 5 percentage points for metformin. VBID-related adherence changes may take time to realize, as patients and providers learn which medications are the best high-value treatments to address chronic health conditions.15
We also found that adherence changes varied according to VBID participants’ baseline adherence. Prior adherence analyses of VBID programs have focused on the average patient, which lumps together patients who experience CRN and would be responsive to VBID programs with patients who are nonadherent for other reasons and would be less responsive. In this analysis, VBID participants with the lowest baseline medication adherence (MPR <50% in a year) experienced the greatest adherence increases, ranging from a 3 percentage point adherence increase between 2007 and 2009 for statins to a 9.7 percentage point adherence increase for ACEIs.
Critics of VBID note that a 3% adherence improvement translates to a mere 10 to 12 days of additional medication treatment per year, which may not be clinically meaningful for individual patients.17 Our results are suggestive of a more clinically meaningful improvement in adherence, particularly among certain subpopulations. For example, among patients who were previously nonadherent to metformin before policy implementation, there was an average adherence improvement of 8.8 percentage points, which equates to an additional 32-day supply of medication per year. Prior research has shown that for every decrease of 10 percentage points in adherence there is a 0.14% increase in glycated hemoglobin levels, which suggests a potentially meaningful improvement in disease control from an 8.8% adherence improvement associated with VBID participation.21
In a sensitivity analysis (), we estimate that 4.1% to 11.5% of VBID participants who were nonadherent (<50%) in 2007 became fully adherent (>80%) by 2009. This represents an adherence improvement of at least 30 percentage points, which is likely to be clinically meaningful. These subgroup results suggest that estimates of adherence changes across all patients may mask more clinically meaningful improvements occurring in significant numbers of patients whose nonadherence is driven by CRN.
Despite broader adoption of VBID policies, there remains considerable skepticism surrounding the evidence for VBID programs published in peer-reviewed journals.16,17 Although adherence is an important consideration in examining the benefit of VBID policies, important questions remain. In particular, it is suggested that improvements in adherence resulting from VBID policies should result in better clinical management of health conditions, leading to reductions in the use of high-cost intensive services such as emergency departments and inpatient facilities. The few economic evaluations of VBID to date have suggested that VBID is cost neutral.15,22 However, the need to manage spiraling healthcare costs requires benefit design decisions to be based on more compelling evidence than null results for cost differences. It will be important to establish whether more optimal benefit designs can result in sufficient healthcare cost reductions to offset the increased expense incurred by payers offering VBID products.
Further examination of VBID policies will be necessary to better understand the overall effect of VBID on the care of patients with chronic illness and to inform optimal benefit design as VBID is implemented in healthcare reform. Under section 2713(c) of the Act it is stated, “The Secretary may develop guidelines to permit a group health plan and a health insurance issuer offering group or individual health insurance coverage to utilize value-based insurance designs.” The results of this study should be informative for establishing these guidelines.
This study adopted a number of design features to improve internal validity of our results, including the use of a propensity score—matched nonequivalent comparator group. In addition, the lack of adherence improvement among 2 therapeutic classes (ARBs and CAIs) that experienced the same degree of copayment reduction shifting from tier 3 to tier 2 in both the VBID and comparison groups suggested that the adherence improvements observed in the other therapeutic classes were not simply a design phenomenon. However, as with any observational study, our results should be interpreted in light of potential limitations. Adherence estimates stemming from health claims data assume medication consumption once a prescription is filled, which may not always be true. Although this study controlled for numerous covariates that were thought to influence the relationship of VBID on medication adherence, there remains the possibility for additional unobservable confounders to influence our study results. Finally, caution should be used when extrapolating results to other insured populations or populations of patients using medications for other chronic health conditions that were not examined. Additionally, the results are generalizable to patients previously using medications for chronic health conditions prior to the initiation of the policy and do not account for potential adherence improvements related to initiating new treatments following VBID.
The results from this study suggest that VBID implementation was associated with improvements in medication adherence ranging from 2% to 5% 2 years following VBID adoption. These adherence changes were most notable among patients who were previously nonadherent to treatment prior to VBID. These adherence results adds to a small but growing body of literature suggesting that VBID copayment policies may improve the clinical management of chronic health conditions.Acknowledgments
The views expressed are those of the authors and do not necessarily reflect the views of the Department of Veterans Affairs, Duke University, the University of North Carolina at Chapel Hill, or BlueCross BlueShield of North Carolina.
Author Affiliations: Division of Pharmaceutical Outcomes and Policy (JFF), UNC Eshelman School of Pharmacy, University of North Carolina, Chapel Hill, NC; BlueCross BlueShield of North Carolina (DW, JCP), Durham, NC; Center for Health Services Research in Primary Care (JHL, MLM), Durham Veterans Affairs, Durham, NC; Division of General Internal Medicine (MLM), Department of Medicine, Duke University Medical Center, Durham, NC.
Funding Source: This work was supported by the Robert Wood Johnson Health Care Financing and Organization Initiative (#67461) and BlueCross BlueShield of North Carolina. Dr Maciejewski was also supported by a Research Career Scientist award from the Department of Veterans Affairs (RCS 10-391).
Author Disclosures: Drs Maciejewski and Farley have received consultation funds from Novartis and Takeda Pharmaceuticals, and Dr Maciejewski receives consultation funds from the Research Data Assistance Center (Res- DAC) at the University of Minnesota. Drs Wansink and Parker report employment with BlueCross BlueShield of North Carolina. Ms Lindquist reports no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.
Authorship Information: Concept and design (JFF, DW, MLM); acquisition of data (JFF, JCP, MLM); analysis and interpretation of data (JFF, DW, JHL, JCP, MLM); drafting of the manuscript (JFF, DW, JCP, MLM); critical revision of the manuscript for important intellectual content (JFF, JHL, JCP, MLM); statistical analysis (JFF, JHL, JCP, MLM); provision of study materials or patients (JFF); obtaining funding (JFF, MLM); administrative, technical, or logistic support (JFF, MLM); and supervision (JFF).
Address correspondence to: Joel F. Farley, PhD, UNC Eshelman School of Pharmacy, 2204 Kerr Hall, CB # 7573, Chapel Hill, NC 27599-7360. E-mail: email@example.com. Krueger KP, Berger BA, Felkey B. Medication adherence and persistence: a comprehensive review. Adv Ther. 2005;22(4):313-356.
2. Osterberg L, Blaschke T. Adherence to medication. N Engl J Med. 2005;353(5):487-497.
3. Mazzaglia G, Ambrosioni E, Alacqua M, et al. Adherence to antihypertensive medications and cardiovascular morbidity among newly diagnosed hypertensive patients. Circulation. 2009;120(16):1598-1605.
4. Rasmussen JN, Chong A, Alter DA. Relationship between adherence to evidence-based pharmacotherapy and long-term mortality after acute myocardial infarction. JAMA. 2007;297(2):177-186.
5. Briesacher BA, Gurwitz JH, Soumerai SB. Patients at-risk for costrelated medication nonadherence: a review of the literature. J Gen Intern Med. 2007;22(6):864-871.
6. Madden JM, Graves AJ, Ross-Degnan D, Briesacher BA, Soumerai SB. Cost-related medication nonadherence after implementation of Medicare Part D, 2006-2007. JAMA. 2009;302(16):1755-1756.
7. Soumerai SB, Pierre-Jacques M, Zhang F, et al. Cost-related medication nonadherence among elderly and disabled medicare beneficiaries: a national survey 1 year before the medicare drug benefit. Arch Intern Med. 2006;166(17):1829-1835.
8. Goldman DP, Joyce GF, Escarce JJ, et al. Pharmacy benefits and the use of drugs by the chronically ill. JAMA. 2004;291(19):2344-2350.
9. Goldman DP, Joyce GF, Karaca-Mandic P. Varying pharmacy benefits with clinical status: the case of cholesterol-lowering therapy. Am J Manag Care. 2006;12(1):21-28.
10. Fendrick AM, Smith DG, Chernew ME, Shah SN. A benefit-based copay for prescription drugs: patient contribution based on total benefits, not drug acquisition cost. Am J Manag Care. 2001;7(9):861-867.
11. Chernew ME, Rosen AB, Fendrick AM. Value-based insurance design. Health Aff (Millwood). 2007;26(2):w195-w203.
12. Choudhry NK, Fischer MA, Avorn J, et al. At Pitney Bowes, valuebased insurance design cut copayments and increased drug adherence. Health Aff (Millwood). 2010;29(11):1995-2001.
13. Maciejewski ML, Farley JF, Parker J, Wansink D. Copayment reductions generate greater medication adherence in targeted patients. Health Aff (Millwood). 2010;29(11):2002-2008.
14. Gibson TB, Mahoney J, Ranghell K, Cherney BJ, McElwee N. Valuebased insurance plus disease management increased medication use and produced savings. Health Aff (Millwood). 2011;30(1):100-108.
15. Gibson TB, Wang S, Kelly E, et al. A value-based insurance design program at a large company boosted medication adherence for employees with chronic illnesses. Health Aff (Millwood). 2011;30(1): 109-117.
16. Fairman KA, Curtiss FR. Making the world safe for evidence-based policy: let’s slay the biases in research on value-based insurance design. J Manag Care Pharm. 2008;14(2):198-204.
17. Fairman KA, Curtiss FR. What do we really know about VBID? quality of the evidence and ethical considerations for health plan sponsors. J Manag Care Pharm. 2011;17(2):156-174.
18. Lau DT, Nau DP. Oral antihyperglycemic medication nonadherence and subsequent hospitalization among individuals with type 2 diabetes. Diabetes Care. 2004;27(9):2149-2153.
19. Sokol MC, McGuigan KA, Verbrugge RR, Epstein RS. Impact of medication adherence on hospitalization risk and healthcare cost. Med Care. 2005;43(6):521-530.
20. Hansen RA, Kim MM, Song L, Tu W, Wu J, Murray MD. Comparison of methods to assess medication adherence and classify nonadherence. Ann Pharmacother. 2009;43(3):413-422.
21. Pladevall M, Williams LK, Potts LA, Divine G, Xi H, Lafata J. Clinical outcomes and adherence to medications measured by claims data in patients with diabetes. Diabetes Care. 2004;27(12):2800-2805.
22. Chernew ME, Juster IA, Shah M, et al. Evidence that value-based insurance can be effective. Health Aff (Millwood). 2010;29(3): 530-536.