Abstract
BACKGROUND:
The National Academy of Medicine has called for value-based drug formularies to address health plan prescription drug spending while maintaining access to high-value medicines. Thirty employer-sponsored plans implemented a “Value-Based Formulary-essentials” (VBF-e) program that uses cost-effectiveness evidence to inform cost-sharing and coverage exclusion.
OBJECTIVE:
To evaluate if the VBF-e was associated with changes in medication use and patient out-of-pocket spending and health plan spending on prescription drugs and other health care.
METHODS:
This was a cohort study using a difference-in-differences design from 2015 through 2019 with 1 year of follow-up after VBF-e implementation at Premera Blue Cross, the largest nonprofit health plan in the Pacific Northwest. The VBF-e exposure group was composed of all individuals aged younger than 65 years and enrolled at least 12 months prior to their employer group’s VBF-e implementation date. The contemporaneous control group was composed of propensity score–matched individuals with the same inclusion criteria but their employer group that did not implement VBF-e. We prespecified the following outcomes: days of medication on hand overall and by VBF-e tier (high-value generic, brand, and specialty drugs were in tiers 1 to 3, respectively, and low-value drugs were in tier 4 or excluded from coverage); prescription drug spending; and other health care use (emergency department visits, hospital days, and outpatient visits).
RESULTS:
Comparing 12,111 exposed (mean age = 36.0; 49.8% female sex) participants with 24,222 control participants (mean age = 34.7; 49.6% female sex), VBF-e reduced use of low-value drugs by 0.3 days per member per month (PMPM) (95% CI = −0.5 to −0.1; 17% decrease) for tier 4 drugs and 0.4 days PMPM (95% CI = −0.5 to −0.4; 83% decrease) for excluded drugs. High-value specialty drug use increased by 0.1 days PMPM (95% CI = 0.0-0.1; 123% increase). Health plan spending decreased by $14 PMPM (95% CI = −26 to −4) and member out-of-pocket spending increased by $1 PMPM (95% CI = 1-2). Other health care use did not change significantly.
CONCLUSIONS:
An exclusion formulary informed by cost-effectiveness evidence reduced low-value drug use, increased high-value specialty drug use, reduced health plan spending, and increased member out-of-pocket spending without increasing acute care use.
Study Registration Number: NCT04904055
Plain language summary
In this study, we evaluate a new kind of prescription drug insurance. This insurance adjusts out-of-pocket costs of each drug according to the benefit and costs that the drug provides. Some drugs with very low benefit and/or very high costs are even excluded from insurance coverage altogether. This insurance increased out-of-pocket spending but reduced health plan spending even more. In the longer term, lower health plan spending should translate to lower health insurance premiums.
Implications for managed care pharmacy
In this cohort study comparing 12,111 exposed participants with 24,222 control participants, a prescription drug exclusion formulary informed by cost-effectiveness was associated with reduced use of low-value drugs—both in the highest cost-sharing tier (17% decrease) and excluded from coverage (83% decrease). Health plan spending decreased ($14 per member per month) and member out-of-pocket spending increased ($1 per member per month).
Prescription drug affordability is a top priority for Americans.1 In the report “Making Medicines Affordable: A National Imperative,” the National Academy of Medicine recommends testing methods to determine the value of drugs and designing drug formularies based on that value, including selectively excluding low-value drugs.2 Cost-effectiveness analysis is a method for determining drug value that has been proposed in formulary design to control health plan spending by limiting low-value drug use while maintaining patient-level access, including high-value specialty drugs for severe conditions such as cancer or hepatitis C.
In 2010, Premera Blue Cross, the largest nonprofit health care payer in the Pacific Northwest, designed an exclusion formulary informed by cost-effectiveness analysis. The pilot, which was implemented among Premera’s own employees, increased the use of drugs with a higher estimated value and decreased pharmacy spending.3 Although promising, that pilot was not adopted by other employer groups and was terminated because of its complexity.
In 2017, Premera streamlined the formulary by reducing the previous 9 cost-effectiveness thresholds to a single threshold. To our knowledge, this new “Value-Based Formulary-essentials” (VBF-e) program is the first formulary in the United States to use a single cost-effectiveness threshold to determine out-of-pocket costs and coverage exclusion. The VBF-e also addresses recent trends important to employer groups, including the increasing costs of specialty drugs and low-value generic medications.4 Unlike the previous pilot formulary, the VBF-e categorizes drugs into the familiar generic, brand, and specialty tiers. Thus, the VBF-e retains the familiarity of typical formulary tiers while applying value-based principles to inform differential access to drugs (including specialty and generic drugs). The VBF-e was adopted by more than 30 employer groups from 2017 to 2019. The objective of this study is to evaluate the impact of the VBF-e on medication use and patient out-of-pocket spending and health plan spending on prescription drugs and other health care.
Methods
VBF-e DESCRIPTION
Details on the methods for identifying, evaluating, and incorporating cost-effectiveness evidence into the VBF-e and a comparison of the prior cost-based formulary with the VBF-e are in Supplementary Methods 1 and 2 (438.2KB, pdf) (available in online article). The VBF-e applies a value threshold of $150,000 per quality-adjusted life-year (QALY) to all drugs. Drugs below this threshold are generally considered high value,5,6 with decreased out-of-pocket costs in the VBF-e. They are placed in cost-sharing tiers 1 to 3 according to generic, brand, and specialty status (Table 1). Drugs above this threshold are generally considered low value,5,6 with increased out-of-pocket costs in the VBF-e. They are placed in cost-sharing tier 4, regardless of generic, brand, and specialty status. Drugs with no added health benefit but higher cost or drugs with lower health benefit but neutral cost (“dominated” drugs), are excluded from coverage. For example, eszopiclone was moved from tier 1 (generic) in the prior cost-based formulary to tier 4 in the VBF-e, whereas Glumetza was moved from tier 3 (nonpreferred brand) in the prior cost-based formulary to the exclusion list in the VBF-e (Table 1). For simplicity, we refer to drugs above the value threshold (tiers 1 to 3) as “high-value drugs” and drugs below the value threshold (tier 4 and excluded) as “low-value drugs.” Notably, the VBF-e was designed to achieve the greatest decrease in cost-sharing for high-value specialty drugs, with out-of-pocket costs reduced from 30% of the total cost of the drug to a $50 copayment per prescription (Table 1).
TABLE 1.
Comparing the Cost-Sharing Tiers of the Prior Cost-Based Formulary With the Value-Based Formulary-Essentials
| Prior cost-based formulary | ||||
|---|---|---|---|---|
| Tier | Tier description | Percentage of drugsa | Cost shareb | Most common drugs |
| 1 | Generic | 86.2 | $20 | Lisinopril, levothyroxine, atorvastatin, amlodipine, metformin, eszopiclone |
| 2 | Brand | 9.4 | $50 | Proair,c Symbicort,c Novolog,c Zetia,c Vytorinc |
| 3 | Nonpreferred brand | 3.5 | 50% | Viibryd,c Lexapro,c Dexilant,c Oracea,c Glumetzac |
| 4 | Specialty | 0.84 | 30% | Harvoni,c Granix,c Orkambi,c Otezla,c Tecfiderac |
| Value-Based Formulary-essentials | ||||
| Tier | Tier description | Percentage of drugs a | Cost share b | Most common drugs |
| 1 | Generic drugs <$150,000/QALY | 81.4 | $15 | Lisinopril, levothyroxine, atorvastatin, amlodipine, metformin, eszopiclone |
| 2 | Branded drugs <$150,000/QALY | 7.1 | $30 | Proair,c Symbicort,c Novologc |
| 3 | Specialty drugs <$150,000/QALY | 0.3 | $50 | Harvoni,c Granix,c Orkambi,c Otezla,c Tecfiderac |
| 4 | Generic, brand, or specialty drugs >$150,000/QALY | 7.0 | 50% | Zetia,c Vytorin,c Viibryd,c Lexaproc |
| Excluded | Economically dominated | 4.1 | Excluded | Dexilant,c Oracea,c Glumetzac |
a Percentage of drugs is calculated as a percentage of the total days supply for drugs in a particular tier relative to the total days supply for drugs in all tiers. Days supply are based on medication use in the 2 years before Value-Based Formulary-essentials implementation.
b Cost share for a prescription of at least 30 days.
c Indicates branded drug.
QALY = quality-adjusted life-year.
STUDY DESIGN
We used a difference-in-differences design, comparing outcomes for members in employer-sponsored insurance plans that implemented VBF-e (exposure group) to propensity score–matched members in contemporaneous employer-sponsored plans that did not implement VBF-e during the study period (control group), comparing outcomes before and after VBF-e implementation (Supplementary Methods 3 (438.2KB, pdf) has analytic model specification details). The parameter of interest is the impact of the VBF-e on average outcomes per member per month (PMPM) in the exposure group after VBF-e enactment. This parameter combines the coefficients representing the impact of the VBF-e on the level and slope of the outcomes of interest (Supplementary Methods 3 (438.2KB, pdf) ). Our study period was January 2015 through December 2019. Employer plans initiated VBF-e at various times between January 2017 and January 2019. Hence, our analysis included plans administered by Premera Blue Cross 2 years before and 1-3 years after VBF-e implementation. Because most implementations of VBF-e began in January 2019, we have complete data for 1 year after for all plans. Therefore, our primary analysis evaluated outcomes 1 year after VBF-e, with secondary analyses evaluating the second and third years and all 3 years. The exposure and control groups had no other changes in medical or pharmacy benefits during the study period. All employers who initiated the VBF-e kept it during the follow-up period.
We prespecified our analyses and preregistered our study analysis plan on ClinicalTrials.gov.7 This study was approved by the Institutional Review Board at Kaiser Permanente Washington Health Research Institute.
DATA SOURCE AND STUDY POPULATION
We constructed a patient-month–level dataset using administrative claims records from Premera Blue Cross. The records contained information on enrollment, demographic characteristics, and pharmaceutical and other health care utilization and spending. We linked members’ 3-digit zip codes to area-level socioeconomic status measures from the 2015 American Community Survey.8
We included individuals aged younger than 65 years who were enrolled in an employer-sponsored insurance plan with pharmacy and medical benefits administrated by Premera Blue Cross on the index month and the 11 preceding months, with a 1-month allowable gap. For individuals in the exposure group, the index month was defined as the month before VBF-e implementation at the employer group level. For individuals in the control group, the index month was defined as the index month of the matched exposure individual. We excluded employer groups that did not transition all employees to the VBF-e at the same time, thereby limiting individual-level selection into the VBF-e. We censored members aged 65 years or older and members after an enrollment gap of more than 1 month.
PROPENSITY SCORE MATCHING PROCEDURE
To assign index months for control group members, we matched each VBF-e member to 2 members who did not transition to VBF-e during the study period via nearest neighbor matching without replacement (Supplementary Methods 4 (438.2KB, pdf) describes matching details and rationale). We matched on the following covariates: sex, year of birth and year of birth squared, and the following American Community Survey variables and their square: population size, household income in past 12 months, percent aged 25 years or older with a bachelor’s degree, and percent White.
OUTCOME MEASURES
We measured days supply of medications on hand overall, categorized by VBF-e tier, and by change in cost-sharing (higher, lower, or no change in cost-sharing) that resulted from the implementation of the VBF-e. Because these medication use measures included many members who did not use medications, we also described unadjusted changes in medication use among the subset of members with at least 2 prescriptions (overall, in a specific VBF-e tier, and in a specific cost-sharing change category described above) in the year before VBF-e implementation.
We measured total prescription drug spending and prescription drug spending by health plan and patients. We measured other health care use as the number of emergency department (ED) visits, days spent hospitalized, and outpatient office visits made PMPM. We measured total health care spending (drug plus other health care spending). All spending outcomes were inflation-adjusted to 2019 dollars.9,10
Covariate Measures. We adjusted for sex, age at index month, relationship to employee (employee, dependent, spouse/domestic partner), baseline Elixhauser comorbidity score11 in the 11 months before index month plus the index month (0, 1, or ≥2), funding type (fully funded or self-funded employer group), calendar month indicators (January to December), calendar year indicators (2015 to 2019) and employer group-specific indicators. Finally, we included quintile indicators for the following zip code-level measures: population size, percentage of the population that was White, percentage of the population 25 years of age or older with a bachelor’s degree, and median household income.
Main Statistical Analyses. We used 2-part, generalized linear models to estimate associations given the zero-inflated and right-skewed nature of the outcomes.12,13 In the first stage, we estimated the probability of the outcome being greater than zero using a binomial distribution and logistic link. In the second stage, we modeled the nonzero portion of the outcomes. For spending outcomes, we modeled the second stage using a gamma distribution and log link. For utilization outcomes that were counts, we modeled the second stage using a Poisson distribution and log link. We generated SEs using 1,000 bootstrap replicates with replacement and clustering on the individual member.14
Sensitivity Analyses. We evaluated the sensitivity of our drug spending estimates in the first year after VBF-e implementation in a series of analyses. First, we estimated propensity score models that included all the matching variables from our base-case analysis plus funding type using 1:2 matching and 1:1 matching. Second, we conducted separate analyses for individuals enrolled in fully funded or self-funded employer plans. Third, we excluded measures 1 month before and after and 3 months before and after the index month to avoid measuring potential anticipatory or delayed filling of medications. Finally, although the employer group–specific fixed effects may appropriately account for correlations within employer groups in this setting,15,16 as a sensitivity analysis, we included bootstrap resampling that stratified resampling by employer group while clustering by member. We also conducted a falsification test of VBF-e impacts on vision spending (eg, prescription glasses) because we did not expect VBF-e to have an impact on this outcome.
Results
Characteristics of the exposure and possible control group members prior to propensity score matching are in Supplementary Table 1 (438.2KB, pdf) . After matching, the final sample included 12,111 exposed members from 30 employer groups and 24,222 control members from 98 employer groups. Exposure and control group members had similar characteristics except that the exposure group was 1.3 years older and had substantially more members enrolled in self-funded plans than the control group (71.1% vs 28.7%) (Table 2). The sensitivity analysis that included self-funded status as a matching covariate in 1:1 matching produced greater balance in this characteristic (71.1% vs 60.5%) but worse balance in other characteristics (Supplementary Tables 2-4 (438.2KB, pdf) ). The sensitivity analysis that included self-funded status as a matching covariate in 1:2 matching did not produce appreciably greater balance in this characteristic (71.1% vs 35.4%) (Supplementary Table 5 (438.2KB, pdf) ). Pre-period spending in the exposure and control groups exhibited parallel trends (Figure 1). Person-months of observation did not different statistically between exposure and control group members (Supplementary Table 6 (438.2KB, pdf) ).
TABLE 2.
Characteristics of Exposure and Control Group Members in the Year Before VBF-e Implementationa
| Exposure | Control | |
|---|---|---|
| No. of people | 12,111 | 24,222 |
| Individual-level characteristics | ||
| Age, mean (SD), years | 36.0 (18.2) | 34.7 (17.8) |
| Female sex, % | 49.8 | 49.6 |
| Relationship to contract holder, % | ||
| Contract holder | 50.0 | 53.9 |
| Spouse/domestic partner | 32.9 | 29.9 |
| Dependent | 17.0 | 16.1 |
| Funding type, % | ||
| Fully insured | 28.9 | 71.3 |
| Self-insured | 71.1 | 28.7 |
| Elixhauser comorbidity score, % | ||
| 0 | 66.0 | 67.1 |
| 1 | 19.7 | 19.6 |
| 2+ | 14.4 | 13.3 |
| ZIP code–level characteristics | ||
| Population size, mean (SD) | 29,549 (8,024) | 29,703 (7,974) |
| White race, mean (SD), % | 76.1 (11.2) | 76.0 (11.0) |
| Household income in prior 12 months (USD), mean (SD) | 69,508 (14 825) | 70,249 (15 192) |
| Aged ≥25 years with bachelor’s degree, mean (SD), % | 21.3 (6.0) | 21.5 (6.0) |
a For individuals in the control group, the year before VBF-e implementation (ie, index date) was defined as the year before VBF-e implementation of the matched exposure individual
USD = United States dollar; VBF-e = Value-Based Formulary-essentials.
FIGURE 1.
Observed and Modeled Total Drug Spending in the Exposure Group (With and Without VBF-e) and Control Group, 2 Years Before and 1 Year After VBF-e Enactment
MEDICATION USE
Overall, the VBF-e was associated with a decrease in medication use of 1.0-day PMPM (95% CI = −1.8 to −0.2), a 4% reduction (Table 3). This reduction was primarily driven by decreased low-value drug use, with decreases of 0.3 days PMPM (95% CI = −0.5 to −0.1) for tier 4 drugs and 0.4 days PMPM (95% CI = −0.5 to −0.4) for excluded drugs. There was a 123% increase in the use of high-value specialty drugs (VBF-e tier 3) at 0.1 PMPM (95% CI = 0.0-0.1; 95% CI does not cross 0). The use of high-value generic and branded drugs (tier 1 and 2 drugs) showed no changes.
TABLE 3.
Impact of the VBF-e on Medication Use PMPM 1 Year After VBF-e Implementation
| Medication use (days supply of medication) | With VBF-e (95% CI) | Without VBF-ea (95% CI) | Estimated change (95% CI) |
|---|---|---|---|
| Overall | 26.4 (25.6 to 27.3) | 27.4 (26.3 to 28.6) | −1.0 (−1.8 to −0.2) |
| By value-based tier | |||
| 1 | 22.1 (21.4 to 22.9) | 22.5 (21.5 to 23.4) | −0.3 (−1.0 to 0.3) |
| 2 | 1.9 (1.8 to 2.1) | 1.9 (1.7 to 2.2) | 0.0 (−0.2 to 0.2) |
| 3 | 0.1 (0.1 to 0.2) | 0.1 (0.0 to 0.1) | 0.1 (0.0 to 0.1)c |
| 4 | 1.6 (1.5 to 1.8) | 2.0 (1.8 to 2.2) | −0.3 (−0.5 to −0.1) |
| Excludedb | 0.1 (0.1 to 0.1) | 0.5 (0.4 to 0.6) | −0.4 (−0.5 to −0.4) |
| By change in tier | |||
| Moved into lower tier | 0.3 (0.3 to 0.4) | 0.4 (0.3 to 0.5) | 0.0 (−0.1 to 0.1) |
| No change in tier | 23.9 (23.1 to 24.7) | 24.1 (23.1 to 25.2) | −0.2 (−0.9 to 0.5) |
| Moved into higher tier | 1.7 (1.6 to 1.8) | 2.4 (2.2 to 2.6) | −0.7 (−0.9 to −0.5) |
a The “Without VBF-e” column presents estimates of adjusted mean estimate of the predicted medication use outcome in the exposure group in the post period had VBF-e not been implemented (ie, counterfactual estimate). For example, the 2.4 (2.2 to 2.6) estimate in the “Without VBF-e” column represents the predicted days supply for drugs that were moved into a higher tier in the VBF-e if those same drugs had not actually been moved into a higher tier.
b Members could be granted access to excluded drugs based on an appeals process, therefore use of excluded drugs “with VBF-e” is greater than 0.
c 95% CI crosses 0.
PMPM = per member per month; VBF-e = Value-Based Formulary-essentials.
In the subset of individuals with medication use in the pre-period, we observed unadjusted decreases in medication use in the following categories: overall, VBF-e tier 4 drugs, and excluded drugs (1.8, 3.1, and 5.8 days PMPM, respectively) (Supplementary Table 7 (438.2KB, pdf) ). Use of VBF-e tier 3 drugs showed an unadjusted increase of 3.5 days PMPM.
DRUG SPENDING
In the year after implementation, the VBF-e was associated with a decrease in total (member plus health plan) prescription drug spending of $13 PMPM (95% CI = −25 to −3). This change was driven by a decrease in health plan spending of $14 PMPM (95% CI = −26 to −4), with an increase in member out-of-pocket spending of $1 PMPM (95% CI = 1-2) (Table 4). Combining all 3 years after VBF-e implementation, the VBF-e was not associated with a statistically significant change in total drug spending −$13 PMPM (95% CI = −30 to 0; 95% CI does cross 0) but was associated with a decrease in health plan spending of $15 PMPM (95% CI = −32 to −1) and an increase in member out-of-pocket spending of $2 PMPM (95% CI = 1-2) (Supplementary Table 8 (438.2KB, pdf) ).
TABLE 4.
Impact of the VBF-e on Medication and Total Health Care Spending PMPM 1 Year After VBF-e Implementation
| Spending (USD) | With VBF-e (95% CI) | Without VBF-e (95% CI) | Estimated change in spending (95% CI) |
|---|---|---|---|
| Total drug spending | 115 (104 to 127) | 128 (113 to 144) | −13 (−25 to −3) |
| Health plan drug spending | 103 (92 to 115) | 118 (103 to 134) | −14 (−26 to −4) |
| Member drug spending | 12 (11 to 12) | 10 (10 to 11) | 1 (1 to 2) |
| Total health care spending | 851 (781 to 915) | 886 (779 to 1,000) | −35 (−157 to 72) |
PMPM = per member per month; USD = United States dollar; VBF-e = Value-Based Formulary-essentials.
OTHER HEALTH CARE USE AND TOTAL HEALTH CARE SPENDING
The VBF-e was not associated with changes in office visits, ED visits, days in the hospital, or total health care spending (drug plus other health care spending) (Supplementary Table 9 (438.2KB, pdf) ; Table 4).
SENSITIVITY ANALYSES OF DRUG SPENDING OUTCOMES
Results from all sensitivity analyses were similar to results from base-case analyses (Table 4 and Supplementary Table 10 (438.2KB, pdf) ), except the sensitivity analysis that excluded 3 months around the index month and the sensitivity analysis that only included fully funded plans both yielded point estimates for member out-of-pocket spending ($0) that were below the 95% CI of the base-case result, suggesting no impact of the VBF-e on member out-of-pocket spending in these scenarios. As expected, the falsification test showed no significant association between the VBF-e and vision care spending (Supplementary Table 11 (438.2KB, pdf) ).
Discussion
In the first year after implementation of a value-based drug formulary, we found a decrease in overall medication use of 4%, driven primarily by the reduced use of low-value drugs. This may be because of the observed increase in member out-of-pocket spending of $1 PMPM. Increases in out-of-pocket costs are known to lead to lower medication use. Such reductions may lead to poor patient and provider experience and, to the degree that users derive benefit from these medications, clinical harm. We also found that VBF-e implementation resulted in a large relative increase (123%) in high-value specialty drug use. Among individuals with high-value specialty drug use in the pre-period, this corresponds to an unadjusted increase of 3.5 days PMPM, equating to 1.4 additional prescriptions per year. This increase has promising patient-centered implications, given the importance of medication adherence for conditions treated by specialty drugs (eg, cancer and multiple sclerosis) and the typical constraints on access to these drugs in traditional formularies. These implications should be evaluated in future studies of value-based formularies.
We also note that drugs with decreases in cost-sharing did not experience increases in use (Table 3). This is similar to the results of a randomized trial that reduced cost-sharing for high-value blood pressure–lowering drugs did not result in an increase in use.17 The authors pointed out that cost-sharing increases are targeting nonadherent patients. These patients have less of a chance of noticing and acting on the cost-sharing change because they are not regularly filling the prescription. Further, cost-sharing decreases, which are a gain to patients, are more likely to be felt less strongly than cost-sharing increases. However, decreasing cost-sharing can increase medication use or adherence in other settings.
We also observed that health plan costs decreased by $14 PMPM. The health plan savings appear to be largely from reduced use of low-value drugs (assigned to tier 4 or excluded). Such reduction in low-value drug use did not increase the number of office visits, ED visits, days spent in the hospital, or total health care spending. The $14 PMPM decrease represents a 12% reduction in health plan drug spending. For comparison, this amount of savings equals spending for all antidiabetic and cardiovascular medications (antihyperlipidemics, antihypertensives, and anticoagulants) for a typical commercially insured plan.18 Such health plan savings may reduce future premiums, which is particularly salient because premiums for employer-sponsored plans have increased by 47% over the past decade, outpacing both general inflation and workers’ earnings.19
With drug spending becoming a key concern for employers, value-based insurance designs are increasingly setting health plan goals around drug savings or at least cost neutrality.20 Our study contributes to a body of evidence that shows that value-based insurance designs can incentivize the use of higher-value drugs.21 However, most prior designs focused on only high-value drugs, neglecting low-value drugs, which are arguably harder to identify. By including low-value drugs, we uncover possible health plan savings. It may be that the combination of potential drug savings, focus on specialty drugs, and the familiarity of generic, brand, and specialty tiers would be needed to incentivize uptake of cost-effectiveness analysis in formulary design. By applying cost-effectiveness estimates to identify low-value drugs, health plans can reduce use of these drugs and generate savings that could be used to lower future premiums and expand coverage of high-value services.22
LIMITATIONS
The VBF-e was implemented along with an exclusion tier. Exclusion tiers themselves are expected to reduce plan spending.23 Therefore, it is difficult to disentangle the effect of tier adjustment informed by cost-effectiveness from the introduction of the exclusion tier.
A limitation to the generalizability of this study was that the cost-sharing levels in the pre-period (Table 1) were higher than typical employer plans of the time. According to a 2019 survey of 2012 employers, the average cost-sharing for drugs in tiers 1-3 or the specialty tier were $11, $33, 34%, or 24%, respectively.24 A VBF-e implemented among employer plans that had more generous pre-period benefits may result in greater savings but also further reductions in medication use.
Availability and quality cost-effectiveness data are sometimes limited. We discuss how the VBF-e handles such limitations in Supplementary Methods 1 (438.2KB, pdf) . Further, tier assignments were based on population average cost-effectiveness estimates. However, the value of a drug may differ depending on the subgroup it is used for. We therefore find it reassuring that despite these limitations, the VBF-e still reduced average health plan spending. However, we lacked all-cause mortality and other clinical endpoint data in this claims dataset, precluding a full evaluation of the VBF-e on health outcomes.
Residual confounding is possible with any observational study. Individuals in the exposure group had 30% higher drug spending levels in the pre-period than individuals in the control group. However, the pre-period spending trends were similar. The difference-in-differences approach assumed no unobserved confounding factors coinciding with the VBF-e. This assumption was supported by the initiation of VBF-e among 30 different employer groups at different times across 3 years. A confounding factor would need to coincide with the varying initiation dates for the different employers. Furthermore, our falsification test suggested that for bias to occur, unobserved confounding must not have simultaneously affected vision care. Moreover, our results were robust to multiple alternative model specifications.
Cost-effectiveness thresholds delineating good value in the United States typically range between $100,000/QALY and $200,000/QALY.5,6 Payers who insure populations with different preferences or budget constraints may choose to apply different thresholds, which would produce different results. For example, we would expect that decreasing the $150,000/QALY threshold used in the VBF-e would further increase health plan savings but also increase member out-of-pocket spending.
Conclusions
We find that a drug formulary informed by cost-effectiveness evidence can be accepted by employers and potentially achieve a better balance between population-level prescription drug affordability and patient-level access to high-value specialty drugs. Surrogate measures of patient health (acute care utilization measures) did not suggest any decrement in health. However, further research should be performed to determine the VBF-e impact on health outcomes. Considering both the original pilot and the VBF-e, Premera has now implemented a formulary informed by cost-effectiveness for over a decade, with substantial health plan savings and no impeded access to high-value treatments. The VBF-e has also been accepted by a large number of employers. Other payers may consider this approach to balance prescription drug affordability and access.
REFERENCES
- 1.Kirzinger A, Kearney A, Stokes M, Brodie M. KFF Health Tracking Poll - May 2021: Prescription drug prices top public’s health care priorities. Kaiser Family Foundation. Accessed March 9, 2023. https://www.kff.org/report-section/kff-health-tracking-poll-may-2021-findings/ [Google Scholar]
- 2.National Academies of Sciences Engineering and Medicine. Making Medicines Affordable: A National Imperative. The National Academies Press; 2018. [PubMed] [Google Scholar]
- 3.Yeung K, Basu A, Hansen RN, Watkins JB, Sullivan SD. Impact of a value-based formulary on medication utilization, health services utilization, and expenditures. Med Care. 2017;55(2):191-8. doi:10.1097/MLR.0000000000000630 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Joyce G, Henkhaus LE, Gascue L, Zissimopoulos J. Generic drug price hikes and out-of-pocket spending for Medicare beneficiaries. Health Aff (Millwood). 2018;37(10):1578-86. doi:10.1377/hlthaff.2018.0628 [DOI] [PubMed] [Google Scholar]
- 5.Anderson JL, Heidenreich PA, Barnett PG, et al. ; ACC/AHA Task Force on Performance Measures; ACC/AHA Task Force on Practice Guidelines . ACC/AHA statement on cost/value methodology in clinical practice guidelines and performance measures: a report of the American College of Cardiology/American Heart Association Task Force on Performance Measures and Task Force on Practice Guidelines. Circulation. 2014;129(22):2329-45. doi:10.1161/CIR.0000000000000042 [DOI] [PubMed] [Google Scholar]
- 6.Neumann PJ, Cohen JT, Weinstein MC. Updating cost-effectiveness--the curious resilience of the $50,000-per-QALY threshold. N Engl J Med. 2014;371(9):796-7. doi:10.1056/NEJMp1405158 [DOI] [PubMed] [Google Scholar]
- 7.ClinicalTrials.gov. Evaluating the impact of a Value-based Prescription Drug Formulary (VBF-e). U.S. National Library of Medicine. Updated May 27, 2021. Accessed March 9, 2023. https://classic.clinicaltrials.gov/ct2/show/NCT04904055 [Google Scholar]
- 8.U.S. Census Bureau. American Community Survey data. Updated November 23, 2021. Accessed March 9, 2023. https://www.census.gov/programs-surveys/acs/data.html
- 9.U.S. Bureau of Economic Analysis. Personal consumption expenditures: Health care Index. FRED, Federal Reserve Bank of St. Louis. Accessed March 9, 2023. https://fred.stlouisfed.org/series/DHLCRC1Q027SBEA [Google Scholar]
- 10.Dunn A, Grosse SD, Zuvekas SH. Adjusting health expenditures for inflation: A review of measures for health services research in the United States. Health Serv Res. 2018;53(1):75-96. doi:10.1111/1475-6773.12612 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Agency for Healthcare Research and Quality. Elixhauser comorbidity software refined for ICD-10-CM Healthcare Cost and Utilization Project (HCUP). Accessed March 9, 2023. https://hcup-us.ahrq.gov/toolssoftware/comorbidityicd10/comorbidity_icd10.jsp [Google Scholar]
- 12.Nelder JA, Wedderburn RW. Generalized linear models. J R Stat Soc [Ser A]. 1972;135(3):370-84. doi:10.2307/2344614 [Google Scholar]
- 13.Belotti F, Deb P, Manning WG, Norton EC. twopm: Two-part models. Stata J. 2015;15(1):3-20. doi:10.1177/1536867X1501500102 [Google Scholar]
- 14.Cameron AC, Gelbach JB, Miller DL. Bootstrap-based improvements for inference with clustered errors. Rev Econ Stat. 2008;90(3):414-27. doi:10.1162/rest.90.3.414 [Google Scholar]
- 15.Wooldridge JM. Econometric Analysis of Cross Section and Panel Data. 2nd ed. MIT Press; 2010. [Google Scholar]
- 16.Yano I, Beal SL, Sheiner LB. The need for mixed-effects modeling with population dichotomous data. J Pharmacokinet Pharmacodyn. 2001;28(4):389-412. doi:10.1023/A:1011586814601 [DOI] [PubMed] [Google Scholar]
- 17.Volpp KG, Troxel AB, Long JA, et al. A randomized controlled trial of co-payment elimination: the CHORD trial. Am J Manag Care. 2015;21(8):e455-64. [PMC free article] [PubMed] [Google Scholar]
- 18.Evernorth. What is drug trend and how to manage it. 2018:10. Accessed March 9, 2023. https://www.evernorth.com/articles/specialty-drug-trends-and-utilization [Google Scholar]
- 19.Palosky C. Average family premiums rose 4% this year to top $22,000; employers boost mental health and telemedicine amid COVID-19 pandemic, benchmark KFF Survey Finds. Kaiser Family Foundation. 2021. Accessed March 9, 2023. https://www.kff.org/health-costs/press-release/average-family-premiums-rose-4-this-year-to-top-22000/ [Google Scholar]
- 20.Courtnay S. In evaluating rheumatoid arthritis treatments, involving patients is key. Am J Manag Care. 2021. Accessed March 9, 2023. https://www.ajmc.com/view/in-evaluating-rheumatoid-arthritis-treatments-involving-patients-is-key [Google Scholar]
- 21.Agarwal R, Gupta A, Fendrick AM. Value-based insurance design improves medication adherence without an increase in total health care spending. Health Aff (Millwood). 2018;37(7):1057-64. doi:10.1377/hlthaff.2017.1633 [DOI] [PubMed] [Google Scholar]
- 22.Kim DD, Griauzde DH, Richardson CR, Fendrick AM. Clinically driven payment and benefit design to improve health equity: the case of obesity prevention and treatment. Health Affairs Blog. 2021. September 9, 2021. Accessed March 9, 2023. https://www.healthaffairs.org/content/forefront/clinically-driven-payment-and-benefit-design-improve-health-equity-case-obesity [Google Scholar]
- 23.Chambers JD, Rane PB, Neumann PJ. The impact of formulary drug exclusion policies on patients and healthcare costs. Am J Manag Care. 2016;22(8):524-31. [PubMed] [Google Scholar]
- 24.Claxton G, Rae M, Damico A, Young G, McDermott D, Whitmore H. 2019. Employer Health Benefits Survey: Summary of Findings. Kaiser Family Foundation. Accessed March 9, 2023. https://www.kff.org/health-costs/report/2019-employer-health-benefits-survey/ [Google Scholar]