|Home | About | Journals | Submit | Contact Us | Français|
To compare pharmaceutical spending and utilization in a consumer driven health plan (CDHP) with a three-tier pharmacy benefit design, and to examine whether the CDHP creates incentives to reduce pharmaceutical spending and utilization for chronically ill patients, generic or brand name drugs, and mail-order drugs.
Retrospective insurance claims analysis from a large employer that introduced a CDHP in 2001 in addition to a point of service (POS) plan and a preferred provider organization (PPO), both of which used a three-tier pharmacy benefit.
Difference-in-differences regression models were estimated for drug spending and utilization. Control variables included the employee's income, age, and gender, number of covered lives per contract, election of flexible spending account, health status, concurrent health shock, cohort, and time trend.
CDHP pharmaceutical expenditures were lower than those in the POS cohort in 1 year without differences in the use of brand name drugs. We find limited evidence of less drug consumption by CDHP enrollees with chronic illnesses, and some evidence of less generic drug use and more mail-order drug use among CDHP members.
The CDHP is cost-neutral or cost-saving to both the employer and the employee compared with three-tier benefits with no differences in brand name drug use.
For the last 15 years, pharmaceutical expenditures in the United States have increased faster than those in other major medical sectors. Insurers have introduced different strategies to moderate this cost explosion. One mechanism is to increase the financial participation of beneficiaries by adopting multitiered benefits with a lower copayment for generic drugs in the first tier, higher copayment for buyer-preferred brand name drugs in the second tier, and the highest copayment for nonpreferred drugs in the third tier (Schneeweiss et al. 2001). Three-tier benefits have been the dominant pharmaceutical benefit design since 2002 (Curtiss 2002; Kaiser Family Foundation and the Health Research and Education Trust 2004).
A new form of insurance—consumer driven health plans (CDHP)—provides an alternative approach to pharmacy benefits. The consumer is given a fixed amount of dollars in an account that she can spend on medical care or drugs. If that account is exhausted, the consumer pays out-of-pocket until a deductible (usually $3,000–$4,000) is met. This approach has the potential to make the consumer more knowledgeable of the prices of medications and to purchase only medically necessary prescriptions. The CDHP also provides Internet-based tools to help the consumer find lower prices if she is willing to accept generic drugs or agrees to receive her medications by mail order.
This study provides a quasiexperimental evaluation of pharmaceutical use and spending in a CDHP and 2 three-tier benefit plans. We examine one large employer's experience using 1 year of baseline information and 3 years of CDHP and three-tier data.
A growing number of studies have confirmed that increased cost sharing reduces prescription drug utilization (Rector et al. 2003; Gibson, Ozminkowski, and Goetzel 2005). Some studies have found that increased cost sharing is associated with a higher proportion of prescriptions filled by generic drugs (Nair et al. 2003). It has also been reported that increased cost sharing is related to substitution of over-the-counter drugs and increased use of mail-order pharmacy (Gibson et al. 2005).
Though evidence is mixed, cost sharing may reduce the use of essential (health improving) drugs as well as nonessential drugs (Tamblyn et al. 2001; Huskamp, Deverka, and Epstein 2005). Cost sharing may affect consumer behavior in other ways such as reducing the adoption of new therapies, increasing drug discontinuation, and reducing compliance (Gibson et al. 2005).
Little research has focused on prescription drug spending and utilization in a CDHP. In previous work, we examined pharmacy use and expenditures for employees of a large firm that offered a CDHP in 2001 (Parente, Feldman, and Christianson 2004). We found that CDHP enrollees had lower pharmaceutical use and expenditures in 2001 and 2002 compared with those who chose a PPO or POS plan. Our recent study extended that analysis with 2003 claims data (Feldman, Parente, and Christianson 2007). The present study expands this work by examining not only gross measures of pharmacy expenditure and cost, but also utilization by those with chronic illnesses as well as comparisons of brand name and generic drug use in the CDHP and three-tier cohorts. In addition, we examine the impact of the CDHP on mail-order prescriptions. Finally, we significantly improve the robustness of our prior analyses using bootstrapping to provide a more thorough test of statistically significant differences among plan designs.
Economic theory predicts that when consumers pay the full price of a prescription drug and have information to assess the costs and benefits of the drug, they will consume the optimal amount, given their preferences and incomes. With prescription drug insurance, consumers pay a fraction of the full price and use more than they would have facing the full price (Pauly 1968). To control expenditure and utilization of prescription drugs, health plans are introducing cost sharing. Both the tiered system and CDHP assume that when higher cost sharing is applied, consumers will use fewer drugs or less-expensive substitutes.
Our hypotheses were: (1) three-tiered plans should steer enrollees toward generic drugs; (2) CDHP enrollees with chronic conditions should use more prescription medicines; and (3) CDHP enrollees will use more mail-order prescriptions. The rationale for hypothesis (1) is that three-tier designs explicitly favor generic drugs, unlike the CDHP, which imposes cost sharing equally across different drug tiers. Regarding hypothesis (2), we predict that CDHP enrollees with chronic conditions will have higher drug spending because they are more likely to exceed the CDHP deductible. The rationale for hypothesis (3) is that the CDHP encourages members to buy drugs through mail-order arrangements and provides them with price comparison information.
Because our hypotheses do not predict the potential of the CDHP to reduce overall pharmaceutical utilization and cost compared with three-tiered plans, we asked the following questions: did CDHP enrollees (4) fill fewer prescriptions, and (5) spend less on prescription drugs? Hypotheses (1)–(3) and questions (4) and (5) form the basis for this study.
We analyzed health insurance claims and human resources data from a large self-insured employer that introduced a CDHP with a Health Reimbursement Arrangement in 2001. The employer previously offered a point-of-service (POS) plan and a preferred provider organization (PPO), which it retained when the CDHP was introduced. The POS and PPO both had three-tier pharmacy benefits with the same copayments across tiers ($10 for generic medications, $20 for brand-name formulary medications, and $30 for nonformulary medications), but they used different drug management systems. The POS plan used a local formulary to operate its three-tier benefit while the PPO used a national pharmaceutical benefit management firm. The firm's medical and pharmacy benefits did not change between 2000 and 2003.
Employees were selected for three cohorts using a quasi-experimental pre/post design: (1) enrolled in the POS plan from 2000 to 2003; (2) enrolled in the PPO from 2000 to 2003; and (3) enrolled in the CDHP from 2001 to 2003, after previously enrolling in either the POS or PPO in 2000 (Parente et al. 2004). The cohorts comprise 429 CDHP enrollees, 1,025 in the PPO, and 1,248 in the POS. Using a cohort design restricts the number of eligible employees between 2000 and 2003 by nearly 70 percent. The advantage of this restriction is that we can control for continuous enrollment and exposure to three-tier or CDHP benefits.
We used Ambulatory Diagnostic Groups (ADGs) to account for case-mix differences among the cohorts (Weiner et al. 1991). ADGs are based on ICD9 diagnosis information from all medical claims for each person in the baseline year of 2000. We constructed a case-mix score that was weighted by the expected resource use associated with a given ADG. In addition, we identified ADGs associated with chronic conditions and examined pharmaceutical utilization and expenditures separately for these contracts.
We also used ADGs to develop a measure of catastrophic “health shocks” for the concurrent year. The presence of any ADG associated with an acute major illness, traumatic injury, or cancer is recorded as a categorical variable. The rationale for this variable is that some health care cost differences might be the result of random events, but their impact is so large they need to be accounted for. Baseline case mix would control for chronic illness but would not account for these types of events.
The employer's human resources data identified the employee's age, gender, wage income from the employer, contribution to flexible spending account, and the number of lives in the contract including the employee. We define pharmacy expenditures as the actual amount paid by either the employer or the consumer through cost sharing arrangements. Cost and utilization data reflect all covered contract members. Mail-order prescriptions were identified with as those with >30 days’ supply of the drug.
The distributions of drug utilization and expenditure in populations are skewed since a large proportion of people use no prescriptions. We estimated probit equations for the probability of using any pharmaceutical product (or having any expense) during the year:
where Xi represents a vector of person i variables influencing use, Ci represents health plan choices, T represents each year after the introduction of the CDHP, and ei is a random error. The coefficient α4 represents the CDHP effect on prescription drug use or spending. We estimated separate effects for each year the CDHP was offered.
Second, we estimated regression models for pharmaceutical cost and utilization among those with positive cost or utilization. Using expenditures as an example, these models are
where Xi, Ci, and T are defined as in equation (1). The models account for repeated observations on individuals and tests of statistical significance are based on robust standard errors.
Following the estimation, we computed the predicted probability of use/expenditure and the conditional values for each contract as if they had been enrolled in each health plan. The conditional values were obtained from a smearing retransformation. We then multiplied the results to arrive at unconditional predictions. Standard errors of the predictions were obtained by bootstrapping 500 iterations of the model. With the additional bootstrapping component, this approach is identical to our previous analysis of CDHP cost and utilization (Parente et al. 2004).
Table 1 presents descriptive statistics for the baseline year 2000. The three cohorts had similar ages but the CDHP had the highest proportion of males (64 percent). One of the largest differences was employee income: the CDHP had the greatest share (38 percent) of employees with income above the firm's 75th percentile compared with the POS (19 percent) and the PPO (29 percent). The CDHP had the lowest baseline illness burden at 6.55 per contract compared with the PPO (7.13) and POS (6.83). POS enrollees used more mail-order prescriptions in the base year than the other cohorts.
CDHP and PPO impacts on pharmacy expenditure and prescriptions are reported in Table 2. The results are based on a simple OLS regression where subscribers with zero expenditure or cost are coded as “1” and included. Although CDHP subscribers experienced less pharmacy expenditure in 2001 and 2002 than POS, these findings are not statistically significant. The only significant CDHP differences from the POS were less utilization of any scripts and generic prescriptions in 2002. The PPO had less brand and generic drug utilization than the POS in later years. The CDHP population had considerably more mail-order prescription use than the other cohorts.
Table 3 presents the results of our two-part model with tests for the significance of differences in relation to the POS plan. In contrast to Table 2, the CDHP cohort has lower costs (−30.4 percent in 2001 to −19.5 percent in 2003). Of these, the −30.4 percent difference for 2001 is statistically significant. In addition, brand name drug use is consistently higher in the PPO population compared with the POS. Generic drug use is lower in both the CDHP and PPO compared with the POS plan with the greatest differences in 2002. Prescription drug use by chronically ill patients in the CDHP and PPO is lower and statistically different from the POS cohort in 2002. Mail-order prescription use by CDHP enrollees was larger, but not statistically different from the POS cohort from 2001 to 2003. However, compared to the PPO cohort, CDHP enrollee mail order use was larger and a statistically significant difference. In contrast, the PPO cohort's mail-order consumption was significantly less than the POS cohort from 2000 to 2003.
We suspect that the variance in pharmaceutical expenditures is positively related to the level of expenditures due to substantial differences in prescription drug treatment plans. For example, a patient treated for seizure disorders could take medications that cost from $5 to $500 per month depending on the compound and availability of a generic substitute. A physician may work with the patient to find the appropriate dosing and chemical compounds that could lead to a high degree of variation. A Breusch–Pagan/Cook–Weisberg test (using STATA 9.0) rejected the assumption of homoskedasticity. To gauge the impact of heteroskedastic errors we estimated a full generalized least squares regression model, forcing homoskedasticity and zero autocorrelation across the three panels. We found little change in the standard errors of the coefficients.
We also were concerned about omitted variable bias. However, a Ramsey RESET test did not reject the hypothesis that there were no omitted variables in the pharmaceutical expenditure equation.
Finally, our difference-in-differences approach may not be appropriate if the CDHP and comparison cohorts already were experiencing different trends in pharmacy use/spending prior to offering the CDHP. The largest observed difference in employee demographics among the cohorts was the contract-holder's wage income. Therefore, a plausible explanation for different trends might be that higher-income employees already had different trends of drug use/spending. To investigate this possibility, we interacted income with the time trend. We found no significant differences in the trend of pharmacy use/spending for employees with different wage incomes. This test does not rule out different trends based on other interactions.
Interpreting the findings with respect to our study hypotheses, we find some support for our first hypothesis that three-tiered plans steer enrollees toward generic drugs. This result suggests that CDHP cost sharing does not favor generic drugs to the extent found in three-tier benefits, which provide a substantial price reduction for generic drugs. However, we note differences among the three-tier designs where the PPO cohort used more brand name drugs and fewer generic drugs than the POS cohort. Interestingly, the results suggest that CDHP enrollees continued to use brand name drugs and decreased their consumption of generic drugs in the second year. These reductions did not persist into the third year suggesting that CDHP enrollees subsequently updated their prescription use patterns.
Our findings do not support the second hypothesis that CDHP enrollees with chronic conditions would use more prescription medicines than the other plan designs. This is surprising since most of the chronically ill CDHP enrollees were over the deductible and faced no cost sharing restrictions. One possibility is that the prospect of cost sharing led those with chronic illnesses to limit their drug purchases before reaching the deductible. In other work, we have found this behavior among all CDHP subscribers (Feldman and Parente 2007).
Our third hypothesis that CDHP enrollees would use more mail-order prescriptions than other cohorts was supported in all 3 years, but the results were not statistically different from the POS cohort. We interpret this finding as different from the PPO population. We interpret this finding as a suggestive response by consumers to reduce pharmaceutical expenditures. Since brand name drugs are available by mail order, this may have been a way to conserve costs on low-volume, high-cost brand name prescriptions.
With respect to our research questions, CDHP enrollees spent somewhat less for prescription drugs than the POS cohort with three-tier benefits, even though the POS plan used a very restrictive formulary. This result was statically significant in the first year the CDHP was offered (2001). A significant decrease in CDHP prescriptions also occurred in 2002. One explanation may be the way the CDHP “balance billed” the insured in the early part of the second year for pharmaceuticals consumed during the first year. These were not paid as part of the deductible when the service occurred. In effect, the employer gave the insured an interest-free loan on pharmaceutical expenditures.
Beyond our research questions, we found that cost and utilization of PPO enrollees was not significantly different from the POS plan. This is surprising since the POS plan used a more restrictive formulary.
The study has two limitations. First, it is based on one employer and is not generalizable to other CDHPs. However, the use of one employer controls for variation in employer-specific benefit designs as well as price variations that are not commonly available in multi-employer research.
Second, some of the results for the POS and PPO plans might be due to changes in drug benefit management that we could not measure. For example, annual negotiations with PBMs could lead the employer to place different drugs on the preferred list. This would affect the prices faced by POS and PPO enrollees, but not CDHP enrollees. We addressed this limitation by using a multi-year cohort design, where the effects of changes in benefits are controlled by interactions of time indicators with cohort membership. We also were informed by the employer that almost no changes in benefit design occurred between 2001 and 2004.
Drug spending in the CDHP is cost-neutral to cost-saving compared with three-tier benefits with no major differences in brand name drug use. We find limited evidence of less drug consumption by CDHP enrollees with chronic illnesses, and some evidence of less generic drug use and more mail-order drug use among CDHP members. As the CDHP market evolves, the empirical evidence “engaging the consumer” in prescription drug coverage appears to be plausible, but not yet overwhelming.
This paper was directly funded by grants provided by the Robert Wood Johnson Foundation Health Care Financing and Organization initiative and Pfizer. The data were obtained using a data use agreement signed by the University of Minnesota and an anonymous large employer. A trusted third party firm, T3Health LLC, was retained to collect the human resources and claims data from the principal employer and the three health plans involved in the study. Primary ownership of the data is with the employer due to their self-insurance contracts with the three health plans involved in the study. T3 Health LLC signed a Business Associate agreement with the employer to serve as their agent for collecting confidential data from all sources, developing a nontraceable study identification number, deleting any personal health information from the files and then providing limited-use data to the University of Minnesota researchers. T3 Health LLC's services were paid for by the Robert Wood Johnson Foundation and Pfizer as a subcontractor to the University's master contract. Finally, three sets of 4-way confidentiality statements were signed by the University's principal investigator, the employer, the health plan providing claims data and the trusted third party.
This research was supported by the Robert Wood Johnson Foundation's Health Care Financing and Organization Initiative and Pfizer Inc. Essential trusted third party data security services were provided by Lisa Schoff of T3 Health LLC. Critical project coordination and institutional review board preparation was facilitated by Ruth Taylor at the University of Minnesota. The paper was sufficiently enhanced by comments from two anonymous reviewers.
Disclaimer: The research presented in this article is the product of the researchers only and does not represent the views of the associated employer, health plans, and financing organizations.
The following material is available for this article online:
Appendix SA1: Author matrix.
This material is available as part of the online article from
http://www.blackwell-synergy.com/doi/abs/10.1111/j.1475-6773.2008.00857.x (this link will take you to the article abstract).
Please note: Blackwell Publishing is not responsible for the content or functionality of any supplementary materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.