The Effect of Access Restrictions on the Vintage of Drugs Used by Medicaid Enrollees

January 15, 2005
Frank R. Lichtenberg, PhD

Volume 11, Issue 1 SP

Objective: To examine the extent to which recent Medicaid drug access restrictions, such as preferred drug lists (PDLs), may affect the vintage (or time since Food and Drug Administration approval) of 6 types of drugs used by Medicaid beneficiaries.

Study Design: Retrospective claims database analysis using National Drug Code pharmacy claims data.

Methods: A regression model was developed to analyze the effect that Medicaid access restrictions had on the vintage of medications prescribed in 6 different therapeutic categories. A "difference in differences" approach was used to compare the change in vintage of medications prescribed in Medicaid versus non-Medicaid patients between the January-June 2001 and July-December 2003 study periods.

Results: The results of the regression model showed that PDLs increased the age of Medicaid prescriptions by less than 1 year for drugs in 5 of the 6 therapeutic classes analyzed. In the case of pain management medications, the increase was more than 1.2 years.

Conclusions: The results of the regression model suggest that Medicaid drug access restriction programs (eg, PDLs) have resulted in an increase in the age of drugs prescribed for Medicaid beneficiaries versus non-Medicaid patients. Since previous research has suggested a clinical and economic advantage to utilizing newer versus older drugs, further research should be conducted to explore how these medication restriction policies may unduly affect Medicaid beneficiaries compared with privately insured patients.

(Am J Manag Care. 2005;11:SP7-SP13)

In the last few years, state Medicaid programs have attempted to control spending by restricting access to certain drugs, especially newer, more expensive drugs. One important type of restriction is a requirement for prior authorization. A prior-authorization requirement attempts to restrict or control the utilization of a specific drug or set of drugs by requiring that certain criteria be met and documented before dispensing. Many states have created lists of "preferred" drugs, where the preference is enforced by a mechanism like prior authorization. The intent of these access restrictions is to change physicians' prescribing patterns to minimize the use of certain drugs within a therapeutic category. The decisions guiding what drugs should be included on a preferred drug list (PDL) are generally based on a combination of clinical factors and the willingness of each drug's manufacturer to offer rebates beyond what is already required by federal law. To date, no researcher has attempted to quantify the effect these access-restriction policies have had on the vintage of drugs prescribed for Medicaid beneficiaries. The vintage of a drug is defined as the year in which the Food and Drug Administration (FDA) first approved the drug's active ingredient. For example, the active ingredient in the osteoporosis drug Evista, raloxifene hydrochloride, was approved by the FDA in 1997. A person consuming Evista in 2004 is therefore consuming a 7-year-old drug.

My earlier research, which examined the effects of newer drugs on patient outcomes and overall healthcare costs, suggested that newer drugs produce higher survival rates and lower hospitalization costs when compared with older drugs.1-3 These results were robust in that the benefits of newer drugs persisted despite controlling for numerous factors such as age, sex, race, education, income, and insurance status. Therefore, if Medicaid access restrictions have increased the vintage of drugs used by Medicaid beneficiaries, these policies also may have an indirect effect on the health outcomes and nondrug health expenditures of these beneficiaries.

I used Medicaid and non-Medicaid pharmacy claims data to model the effect that Medicaid access restrictions have had on the vintage of drugs used by Medicaid beneficiaries. If all state Medicaid programs used evidence-based criteria to establish their PDLs, and the drugs requiring prior authorization were those for which there are close therapeutic substitutes that are cheaper, a prior authorization-induced increase in drug vintage would not necessarily be a cause for concern. But as Soumerai4 notes, some states are placing large numbers of medications on prior-authorization lists largely on the basis of price or provision of supplemental rebates. For example, Michigan requires prior authorization for nonpreferred drugs in 40 medication categories. The fact that some states don't use evidence-based criteria to establish their PDLs might lead one to expect that implementation of prior-authorization restrictions sometimes has unintended and possibly adverse consequences. This was the case in 2 of the 3 studies of prior authorization reviewed by Soumerai.4 In 1 study–prior authorization of H2 blockers and nonsteroidal anti-inflammatory drugs (NSAIDs) in the Georgia Medicaid program–prior authorization appears to have reduced drug costs without any unintended side effects. But after the West Virginia Medicaid program required prior authorization of cimetidine, hospitalizations for peptic ulcer disease rose somewhat, and "it is reasonable to assume that the policy reduced quality of care for peptic ulcer disease."4(p138) After Tennessee implemented a prior-authorization policy targeting nongeneric NSAIDs in its Medicaid program, "there was a 26 percent reduction in overall use of NSAIDs, which raises concerns regarding unintended reductions in treatment of pain."4(p139)

before

One possible approach to modeling the effect that Medicaid access restrictions have had on the vintage of drugs used by Medicaid beneficiaries would be to simply regress the mean vintage of Medicaid prescriptions in a state during July-December 2003 on the number of drugs restricted in that state during that period. However, that could result in biased estimates of the effect of prior-authorization restrictions on the vintage of drugs used by Medicaid enrollees, because states that imposed the most restrictions may have tended to use newer (or older) drugs than other states the restrictions were imposed. Therefore, a "difference in differences" approach was used to estimate the relationship across states between the extent of prior-authorization restrictions and the Medicaid versus non-Medicaid difference between a pre-PDL and post- PDL change in the vintage of medications prescribed. In addition, this paper explores how the vintage of drugs prescribed for Medicaid beneficiaries might be affected if every state were to adopt the PDL policies of the most restrictive state.

METHODS

For this analysis, pharmacy claims data were obtained from NDCHealth (http://www.ndchealth.com/). These data contained claims for Medicaid and non- Medicaid beneficiaries from January 2001 to December 2003 for all 50 states and the District of Columbia. The relationship between the extent of prior authorizations and change in vintage of Medicaid prescriptions was examined by comparing an evidently pre-PDL time period to a post-PDL time period. Since the number of states with PDLs grew from 3 in 2000 to 22 in 2003,4(138) it is likely that most Medicaid prior-authorization access restrictions to prescription drugs were put into practice after July 1, 2001. Therefore, the prescription data for the pre-PDL study period included 6 months of NDCHealth prescription claims data from January 1, 2001, to June 30, 2001. Prescription data for the post-PDL study period included 6 months of NDCHealth prescription claims data from July 1, 2003, to December 31, 2003.

all

There may be idiosyncratic state-specific factors that could affect the change in the vintage of medications prescribed (Medicaid and non-Medicaid). These state-specific confounders may be correlated with the extent of prior authorization. To help account for this possibility, a "control group" consisting of non-Medicaid prescription claims was used. That is, by examining the relationship across states between the extent of prior-authorization restrictions and the change in the vintage of non-Medicaid prescriptions, an estimation of within-state factors that may have affected the vintage of all medications could be conducted.

For this analysis, the effect of access restrictions on drug vintage was examined for 6 important therapeutic classes of drugs for which pharmacy claims data were available: antidepressants, antihypertensives, cholesterol-lowering drugs, diabetic drugs, osteoporosis/menopause drugs, and pain management medications. Within each therapeutic class, the most commonly prescribed single-source drugs (jointly accounting for at least 80% of the prescription claim volume for the combined Medicaid and non-Medicaid population) were identified. The brand-name drugs studied in each class are shown in Table 1. The number of brand-name drugs identified in each class ranged from 5 (pain management medications) to 11 (antihypertensives and osteoporosis/menopause drugs). A total of 50 brand-name drugs were identified in the 6 classes.

For all agents in the 6 target therapeutic classes, the vintage of each product was obtained from Mosby's Drug Consult.5 The vintage of each product represented the year in which the new drug application for the product's active ingredient was approved by the FDA.

By consulting state Medicaid Web sites, it was determined which of the study drugs required prior authorization in each state and the District of Columbia from July 1 through December 31, 2003. Because the prior-authorization status of agents within a therapeutic class can change over time as more agents are added to a PDL list, a method was devised to quantify the level of restrictiveness attributed to each therapeutic class. This method is described by the equation below:

i

j

ij

where RESTRICTij indicates whether drug was restricted in state during this period. If a drug was restricted during the entire period (ie, it was already restricted at the beginning of the period), RESTRICT=1. If a drug was never restricted during the period (ie, it had not been restricted by the end of the period), RESTRICTij = 0. If a drug was restricted during part of the period, RESTRICTij equals the percentage of the period during which the drug was restricted. For example, if the drug was restricted for the last 4 months of the 6-month period beginning September 1, 2003, RESTRICTij = 4/6 = 0.67.

Using the data described above, models of the following form were estimated for each class of drugs:

b

b

b

change

b

b

b

b

b

b

j

Equation 1 was estimated via weighted least squares, where the weight was the number of prescriptions for the drug. 12 is the estimate of the marginal effect of the number of restrictions on the mean age of Medicaid prescriptions in period 2, and (12 -10) is an estimate of the marginal effect of the number of restrictions on the in the mean age of Medicaid prescriptions; (02 -) is an estimate of the marginal effect of the number of restrictions on the change in the mean age of non-Medicaid prescriptions; and [(12 -10) - (02 - 00)] is a (difference in differences) estimate of the marginal effect of the number of restrictions on the Medicaid versus non-Medicaid difference between the January-June 2001 to July-December 2003 change in the mean age of prescriptions. It is understood that estimates yielded by this procedure of the effects of access restrictions on drug vintage may be conservative, for 2 reasons. First, N_RESTRICTj may be a noisy (error-ridden) measure of the true restrictiveness of state , which would bias the estimates towards zero (ie, it may result in underestimation of the effect of access restrictions on drug vintage). Second, there is evidence that changes in Medicaid drug utilization policy have spillover effects, especially in privately insured patient subpopulations where the caregiving physician treats a high proportion of Medicaid patients.6

RESULTS

Table 2 and Table 3 present summary statistics on the number of drugs requiring prior authorization during the study period. Table 2 shows the distribution of the number of drugs requiring prior authorization for the 6 classes combined, by state. The mean and median numbers of drugs restricted (out of 50 drugs considered) were 8 and 5, respectively. Twelve states did not restrict the use of any of the 50 drugs. Five states restricted the use of more than 47% of the drugs, and one–Vermont–restricted the use of 43 of the 50 drugs.

As Table 3 shows, the extent of restrictions varied across therapeutic classes. Across all states, the mean percentage of drugs with access restrictions was below 8% for 3 classes: osteoporosis/menopause drugs (4%), antidepressants (6%), and diabetic drugs (7%). The mean percentage of drugs with usage restrictions was 12% for cholesterol-lowering drugs and antihypertensives, and 54% for pain management medications. Within individual states, the maximum percentage of drugs in a therapeutic class with access restrictions ranged from 45% for osteoporosis/menopause drugs to 100% for pain management medications.

Summary statistics on the mean age and number of prescriptions, by drug class and payer, during the period 2001-2003 are shown in Table 4. Overall, Medicaid accounted for about 10% of sample prescriptions, but it accounted for 15% of antidepressant prescriptions. In 5 of the 6 drug classes, the mean ages of the Medicaid and non-Medicaid prescriptions differed by less than 4%. Medicaid pain management medications were more than 8% older than non- Medicaid pain management medications.

b

b

b

b

b

b

b

b

b

b

b

b

negative,

less

Estimates of (12-10), (02-00), and their difference, by drug class, are shown in Table 5. For every drug class, the estimate of (12-10) is positive and highly statistically significant. This indicates that the mean age of Medicaid prescriptions increased more, from the first half of 2001 to the second half of 2003, in states with more prior-authorization restrictions by the end of 2003. Moreover, for every drug class, the estimate of (12- 10) is much larger than the estimate of (02-00), and the difference is statistically significant. This indicates that the difference between the mean ages of Medicaid and non- Medicaid prescriptions increased more in states with more prior-authorization restrictions by the end of 2003. For two drug classes–osteoporosis/menopause drugs and cholesterol-lowering drugs–the estimate of (02-00) is indicating that the mean age of non-Medicaid prescriptions increased in states with more prior-authorization restrictions by the end of 2003.

b

b

b

b

The implications of these estimates are considered in Table 6. (Similar implications were obtained when an alternative measure of restrictions–the share of period 0 prescriptions that were for drugs that were restricted by the end of period 2–was used.) The second column shows the difference-in-differences estimate [(12 - 10) - ( 02 - 00)] of the marginal effect of the number of restricted drugs on the mean age of Medicaid prescriptions. The third column shows the marginal effect multiplied by the mean number of products restricted [mean(N_RESTRICTj)]. This indicates how much the restrictions that were adopted increased the difference between the mean ages of Medicaid and non-Medicaid prescriptions. For 5 of the 6 therapeutic categories, these effects were fairly small, increasing the age difference by less than one sixth of a year. In the case of pain management medications, the effect was much larger–more than 1.2 years.

The last column shows the marginal effect multiplied by the maximum observed number of products restricted [max(N_RESTRICTj)]. This indicates how much the difference between the mean ages of Medicaid and non-Medicaid prescriptions would have increased if all states had been as restrictive as the most restrictive state (or how much the difference will increase if they become that restrictive in the future). Under this scenario, the Medicaid versus non-Medicaid difference in the mean age of antihypertensive medications would have increased by almost a year, the differences in the mean age of antidepressants and cholesterol-lowering drugs would have increased by just over a year, and the differences in the mean age of pain management medications would have increased by 2.26 years.

DISCUSSION

This study suggests that Medicaid PDLs increase the vintage of medications prescribed to Medicaid beneficiaries compared with non-Medicaid enrollees. These differences varied by state and therapeutic area, with the increase in vintage ranging from less than a year to 1.2 years. Since, as Table 7 indicates, the fraction of the Medicaid population that is black or Hispanic (48%) is almost twice as great as the fraction of the total US population that is black or Hispanic (25%), Medicaid drug restrictions have increased the difference between the average age of drugs used by blacks and Hispanics and the average age of drugs used by other Americans. Although the prior-authorization restrictions may not have exacerbated socioeconomic health disparities, in 2 of the 3 studies of prior authorization reviewed by Somerai, implementation of prior-authorization restrictions may have had unintended and possibly adverse consequences.4

This study was subject to several methodologic limitations. First, since the post-PDL time period was only 6 months, the full impact of a PDL on prescription utilization may not have manifested, since in some cases refills are "grandfathered," or there may have been certain exemptions for refills on existing prescriptions. Therefore, this study may underestimate the effect of Medicaid access restrictions on drug vintage by producing conservative estimates of this effect. Second, this analysis did not consider the strictness of the prior-authorization criteria across states. That is, some states may have a very simple prior-authorization process while others may use more complex processes, involving documented failure of multiple generic or preferred agents before a nonpreferred agent may be dispensed. These types of differences in prior-authorization complexity can occur across states or across classes within a state. Differences in prior-authorization complexity can bias statewide aggregate estimates of policy effects on drug vintage either upward or downward, depending on the prior-authorization structure across classes within a state.

As noted above, previous research indicates that, in general, use of newer drugs increases longevity and productivity, and reduces use of other medical services, especially hospitals. If the drugs that have been restricted are typical new drugs in terms of their effect on longevity, productivity, and hospitalization, then these restrictions are likely to have reduced the rate of growth in longevity and productivity and the rate of decline in hospitalization in the Medicaid population, and to have widened racial health disparities. Proponents of these restrictions would probably argue that the drugs subject to restriction offer fewer therapeutic benefits than other new drugs; therefore, there have been few, if any, adverse consequences of restricting access. Unfortunately, the data required to assess directly the consequences of Medicaid drug restrictions on mortality, morbidity, and medical expenditure are not yet available. This topic merits future research.