Context With drug spending rising rapidly for working-aged adults, many employers
and health insurance providers have changed benefits packages to encourage
use of fewer or less expensive drugs. It is unknown how these initiatives
affect drug costs.
Objective To examine how innovations in benefits packages, such as those that
include multitier formularies and mandatory generic substitution, affect total
cost to insurance providers for generic and brand drugs and out-of-pocket
payments to beneficiaries.
Design and Participants Retrospective study from 1997 to 1999 linking claims data of 420 786
primary beneficiaries aged 18 through 64 years who worked at large firms (n
= 25) with health insurance benefits that included outpatient drugs.
Main Outcome Measures Overall drug costs; generic, single-source brand, and multisource brand
costs; and drug expenditures by health insurance providers and out-of-pocket
costs for beneficiaries.
Results For a 1-tier plan with a $5 co-payment for all drugs, the average annual
spending was $725 per member. Doubling co-payments to $10 for all drugs reduced
the annual average drug cost from $725 to $563 per member (22.3%, P<.001). Doubling co-payments in a 2-tier plan from $5 for generics
and $10 for brand drugs to $10 for generics and $20 for brand drugs reduced
costs from $678 to $455 (32.9%, P<.001). Adding
an additional co-payment of $30 for nonpreferred brand drugs to a 2-tier plan
($10 generics; $20 brand) lowered overall drug spending by 4% (P<.001). Requiring mandatory generic substitution in a 2-tier plan
reduced drug spending by 8% (P<.001). Doubling
co-payments in a 2-tier plan increased the fraction beneficiaries' paid out-of-pocket
from 17.6% to 25.6%.
Conclusions Adding an additional level of co-payment, increasing existing co-payments
or coinsurance rates, and requiring mandatory generic substitution all reduced
plan payments and overall drug spending among working-age enrollees with employer-provided
drug coverage. The reduction in drug spending largely benefited health insurance
plans because the percentage of drug expenses beneficiaries paid out-of-pocket
rose significantly.
Spending on outpatient prescription drugs has increased at double-digit
rates for the past decade and is now the third largest component of health
care expenses behind hospital care and physician services.1 The
growth in drug spending can be attributed to many factors, including new drugs
with high prices, an aging US population, legislative mandates, and earlier
diagnosis.2,3 Although drug spending
remains less than 10% of health care expenditures nationally, the share is
much higher (commonly 20% to 25%) for working-aged adults with low rates of
hospitalizations.4
With spending rising so rapidly for working-aged adults, many employers
and health insurance providers have changed their benefits packages to encourage
less and lower-cost pharmaceutical use. Closed or highly restrictive formularies
were one response, in which insurance providers would only cover certain drugs.
However, excluding specific medications or therapeutic classes led to considerable
dissatisfaction among patients and physicians.5 Many
private health insurance plans now offer incentive-based formularies, in which
drugs are placed in different tiers. Under these arrangements, most drugs
are covered, but enrollees pay differing co-payments depending on the tier
to which a drug is assigned. Two-tier plans are commonplace, with a higher
co-payment for brand drugs. However, an increasing number of employers are
offering 3-tier benefits with 3 co-payment levels. These plans typically set
the lowest co-payment for generic drugs, the middle co-payment for formulary
or preferred brands, and the highest co-payment for nonformulary brands. Another
popular benefit design feature to reduce drug spending is mandatory generic
substitution (MGS).
Prior studies suggest that increased patient cost-sharing and formulary
restrictions reduce pharmaceutical use and costs.6-21 However,
most of these studies examined elderly or Medicaid populations,7-10,13,16,17 involved
small changes in co-payments and changes within a single plan,7,11,12,14,15,21 or
preceded the introduction of novel prescription drug benefits such as multitier
formularies and generic substitution.6-12 In
this study, we use data for a wide array of employers and benefit designs
to assess how multitier formularies, increased co-payments, and MGS requirements
affect spending for generic and brand drugs and patients' out-of-pocket costs.
With incentive-based formularies, members pay differential co-payments
or coinsurance rates based on the status of a drug (Table 1). In some plans, members pay a single co-payment for all
drugs, a so-called 1-tier plan. To provide a financial incentive to purchase
generic drugs, some plans charge different co-payments for brand and generics,
with lower co-payments for the generics (2-tier plans). To encourage use of
lower-cost brand medications, some plans further differentiate by adding a
third co-payment for more expensive brand medications. In these 3-tier plans,
generic drugs typically have the lowest co-payment, formulary or preferred
brands have a midrange co-payment, and nonformulary brands have the highest
co-payment. Other variants of the 3-tier plan distinguish brand drugs without
generic substitutes (single-source brands) and brand drugs with generic substitutes
(multisource brands). Savings result from shifting members' drug use to generics
or preferred brands, for which the health plans have negotiated favorable
rates, and by increasing patients' cost-sharing for nonpreferred brands.5
Many health plans have adopted other tools to control drug spending
such as MGS and coinsurance rates rather than fixed dollar co-payments per
prescription. Under MGS, members electing brand drugs over their generic equivalent
generally must pay the generic co-payment plus the full difference in cost
between the brand and generic drugs. Coinsurance rates, or percentage co-payments,
are attractive to plans because they keep pace with rising drug costs. However,
they are more difficult for patients to understand and lead to greater variation
and uncertainty in out-of-pocket expenses.22
We assembled a unique data set linking health care claims to health
plan benefits. Through a health benefits consulting firm, we obtained claims
data from 1997 to 1999 for 35 private employers. The analysis excluded 10
employers with fewer than 1000 employees per health plan or with incomplete
information on drug claims (eg, missing national drug codes). Of the 25 firms
included in the study sample, most offered employees a choice of medical plans,
and some plans changed benefits at the beginning of a calendar year. As a
result, there were 55 unique medical or pharmacy benefit packages (ie, plans)
and 75 plan-years of data since several plans have data for multiple years.
Although employees typically had a choice of medical plans, only 2 firms had
a choice of drug plans, thereby minimizing potential bias from selection of
drug plans based on anticipated use. The study sample consisted of 702 782
person-years of data on 420 786 beneficiaries aged 18 to 64 years who
were continuously enrolled in a plan for 1, 2, or 3 years. We excluded dependents
and employees aged 65 years or older because we could not be sure that their
drug utilization was not covered by other insurance.
Enrollment files included each person's age, sex, ZIP code of residence,
and relationship to employee. Claims files captured all health care claims
and encounters, including prescription drugs, inpatient, emergency, and ambulatory
services. Drug claims included information on the type of drug (drug name,
national drug codes, dosage, supply), place of purchase (retail or mail-order),
and expenditures, including billed charges, negotiated discounts, excluded
expenses, deductibles, co-payments and payments made by the employer, employee,
and other third-party coverage. Data were also available on prescriptions
costing less than the minimum drug co-payment. The medical claims included
the same financial information, as well as the date of service, diagnosis
and procedure codes, type of facility, and provider.
The claims data were linked with information about plan benefits. For
each plan, we obtained photocopies of the summary of benefits provided by
the firms to their employees and abstracted the benefit information. Because
some benefit packages contained more detail than others, we coded information
only when the plan specifically stated that a benefit was covered or excluded.
The few discrepancies were resolved by consensus. The drug benefit design
features we coded included co-payments or coinsurance rates for both retail
and mail-order pharmacies, generic substitution rules, and a list of drugs
or drug classes excluded from coverage. Drugs not covered by the plans consisted
primarily of lifestyle or cosmetic drugs such as sildenafil citrate and other
discretionary medications for hair loss, weight reduction, and smoking cessation.
The medical plan characteristics that we coded included individual plan deductibles,
co-payments or coinsurance rates for physician office visits, and a binary
indicator for enrollment in a managed care plan. No plans had a separate deductible
for prescription drugs.
We estimated 2 sets of regression models: costs by type of drug (generic,
single-source brand, or multisource brand) and drug costs paid by the health
insurance plan and by the patient. Costs reflected total annual payments made
by the enrollee (co-payments, deductibles, excluded expenses) and by all third-party
payers (primary and secondary coverage, net of negotiated discounts) for outpatient
prescription drug claims.
The main independent variables in both models were the drug benefit
design features, including the plan's lowest co-payment at retail pharmacies
and incremental co-payments for second- and third-tier drugs, if applicable.
The model also included binary indicators for coinsurance plans and use of
MGS rules. We did not include cost-sharing arrangements at mail-order pharmacies
because they were highly correlated with retail co-payments and coinsurance
rates. In addition, we did not include indicators for excluded drugs because
there was little variation across plans and the excluded medications comprised
only a small fraction of covered drugs.
The covariates included a set of variables to describe the medical benefits,
including individual plan deductibles, co-payments or coinsurance rates for
physician office visits, and a binary indicator for enrollment in a managed
care plan. Other covariates were age categories, sex, work status (active
or retired), urban residence, and median household income in the ZIP code
of residence. We controlled for observed differences in comorbid conditions
based on International Classification of Diseases, Ninth
Revision (ICD-9) diagnostic codes from the medical claims files. We
identified individuals who were treated for any of 26 chronic conditions,
such as hypertension, diabetes, congestive heart failure, asthma, and depression,
and included a binary indicator for each condition. Last, the model also included
binary indicators for the calendar year to control for time trends in prescription
drug spending and prices.
Our statistical analyses used a 2-part model.23 The
first part of the model, including the entire study sample, used probit regression
to estimate the probability that a member had at least 1 pharmacy claim. The
second part of the model used a generalized linear model with a logarithmic
link function to estimate the level of drug spending among members with at
least 1 claim, for the outcome of interest. For example, we analyzed use of
generic, multisource, and single-source drugs separately. We chose the generalized
linear model because it predicted component drug expenditures better than
the standard 2-part model that uses linear regression in the second part,
but our conclusions were insensitive to this choice.24,25
We combined the 2 parts of the model to predict average annual drug
spending by drug type and payer status under different plan/co-payment combinations.
Specifically, we used estimates from the first part of the model to predict
the probability of nonzero expenditures for each person under alternative
benefit designs and co-payments. Similarly, we used the second part of the
model to predict expenditures, conditional upon having at least 1 claim for
each person under each plan/co-payment combination. We calculated total expenditures
as the product of the 2 parts of the model and averaged them over all individuals
in the sample for each plan/co-payment combination.23
The simulations used a predetermined set of co-payments that occurred
frequently in our data and were representative of cost-sharing arrangements
in private health insurance plans. Because the coinsurance plans in our sample
lacked sufficient variation in coinsurance rates, we included a binary indicator
for coinsurance in predicting drug expenditures for these plans. We assumed
that 1-tier co-payment plans required MGS, but that other plans did not. We
also compared drug spending with and without MGS in 2-tier co-payment plans.
We adjusted the SEs for clustering of patients within plans. We also used
the bootstrap to derive the SEs of the predictions and compute 95% confidence
intervals (CIs).26 STATA version 7 (STATA Corp,
College Station, Tex) was used for statistical analyses and the 95% CI reflects
.025 in each tail or P≤.05.
The characteristics of the patients in the study were consistent with
a working-age population (Table 2).
About half (46.6%) were aged 45 years or younger, 63.9% were men, and more
than 4 out of 10 were treated for 1 or more chronic health conditions.
We categorized benefits packages into coinsurance plans and 1-, 2- and
3-tier co-payment plans as described in Table 1. The 15 coinsurance plans in our sample had a single coinsurance
rate for prescription drugs of 20% or 30%, with a mean of 27.3% (Table 3). Ten of these plans required MGS.
Among plans with co-payments, the 15 1-tier plans had an average co-payment
of $6.67 (range, $2-$10). Two-tier plans (n = 36) were the most prevalent
benefit design in our sample; the average difference in co-payment between
generic and brand drugs was about $7. Two- and 3-tier plans had similar co-payments
in the first 2 tiers. However, the average co-payment for nonpreferred brands
was $23.56, nearly $12 more than what members would typically pay for preferred
brands. All plans with a single co-payment had MGS programs. In contrast,
only 6 of the 45 multitier plans required MGS.
Unadjusted total drug spending was highest in single co-payment plans
(Table 4). Annual mean spending
was approximately $150 higher per member in 1-tier plans than in coinsurance
or 3-tier plans. The average number of prescriptions dispensed followed a
similar pattern. The fraction of enrollees who filled 1 or more prescriptions
was highest in 1-tier plans and lowest in coinsurance plans. The fraction
of drugs dispensed as generic ranged from 33.2% in 2-tier plans to 38.7% in
1-tier plans.
Table 5 presents predicted
annual drug spending per member within each type of drug plan and co-payment
level. Increasing co-payments within a particular benefit design reduced spending
significantly, controlling for other factors known to affect utilization.
For example, increasing single fixed co-payments for all drugs from $5 to
$10 reduced annual average drug spending from $725 to $563 per member (22.3%
reduction, P<.001). Similarly, doubling co-payments
in multitier plans reduced annual average drug spending by about one third
(32.9% in 2-tier plans, 34.5% in 3-tier plans; P<.001).
Higher patient cost-sharing led to less spending on both generic and brand
name drugs.
Adding co-payments also significantly reduced average drug spending.
Changing from a single co-payment of $5 to a 2-tier plan with co-payments
of $5 for generic and $10 for brand drugs reduced average drug spending from
$725 to $678 (6% reduction, P<.001). Similarly,
changing from a single co-payment of $10 to a 2-tier plan with co-payments
of $10 for generic and $20 for brand drugs reduced average drug costs from
$563 to $455 (19% reduction, P<.001). Adding another
co-payment for nonpreferred brands reduced spending further, albeit more modestly.
For example, adding a co-payment of $5 or $10 for nonpreferred brand drugs
lowered overall drug spending an additional 2% to 4%, respectively (P = .004 and P<.001). Spending
on brand drugs declined, while spending on generic drugs increased with the
addition of a third tier. For instance, annual expenditures on generic drugs
increased from $71 to $81 per member with the addition of an incremental co-payment
of $5 for nonpreferred brand drugs.
Higher co-payments for physician office visits had no effect on drug
spending (data not shown). Also, total drug spending was similar in managed
care and nonmanaged care plans, although use of brand drugs was modestly lower
in managed care settings.
We also examined the share of drug spending borne by patients and all
third-party payers under different cost-sharing arrangements (Table 6). Patient out-of-pocket spending did not change substantially
within a specific benefit design, because the reduction in overall drug use
due to higher patient cost-sharing largely offset the effects of higher co-payments
per prescription. However, the fraction of drug costs borne by patients rose
considerably. Doubling co-payments in 2- and 3-tier plans increased the fraction
of drug expenses beneficiaries paid out-of-pocket from 17.6% to 25.6% and
20.1% to 32.3%, respectively. Overall, patient out-of-pocket payments were
highest in coinsurance plans and 3-tier plans with higher co-payments.
Mandatory generic substitution also lowered drug costs significantly
(Table 7). Specifically, adding
MGS in 2-tier plans reduced drug spending by $36 to $52 per person (8% reduction, P<.001), depending on the level of co-payments. Requiring
MGS reduced expenditures on multisource and single-source brands, but had
no appreciable effect on generic drug spending. However, separate analyses
examining the number of prescriptions dispensed rather than drug spending
found a modest increase in generic prescriptions and little change in total
prescriptions with the addition of MGS.
The desire to control health care costs has led to considerable variation
in how employers and health insurance providers structure formularies, design
benefits, and provide incentives to both physicians and patients. We found
that many of the tools used to influence pharmaceutical use were effective
in reducing drug expenditures for working-age enrollees with employer-provided
drug coverage. Adding an additional level of co-payment, increasing existing
co-payments or coinsurance rates, and requiring MGS all reduced health insurance
plan payments significantly. Doubling patient co-payments lowered average
drug spending by as much as one third (Table 5), reducing both the likelihood of having a claim and the
level of spending conditional upon use. The reduction in drug spending largely
benefited employers, as the fraction of drug costs borne by patients increased
significantly. We also found evidence that requiring MGS was an alternative
to adding an additional level of co-payment. All 1-tier plans in our sample
required MGS; in contrast, less than 1 in 7 multitier plans imposed it.
There is optimism among some health insurance plans and providers that
requiring larger co-payments on nonpreferred brands will dampen the rapid
growth in drug spending. We found that this benefit had modest effects. Adding
a third tier with incremental co-payments of $10 reduced drug expenditures
by only 4%. This effect is smaller than reported elsewhere for a single preferred
provider organization that changed from a 2-tier benefit with co-payments
of $7 and $12 to a 3-tier plan with co-payments of $8, $15, and $25.21 While some of the discrepancy between studies can
be explained by the dollar difference in co-payments, the potential savings
from a 3-tier benefit depend on many other factors, including the formulary
structure (how drugs are classified) and utilization patterns within a plan.5
The differences in drug spending are likely to be driven by patient,
not provider, behavior. Physicians are generally not familiar with the costs
of the medications they prescribe.27 Moreover,
the patients in our study had generous drug coverage with modest differences
in co-payments for alternative medications. Modest differences in out-of-pocket
cost for a subset of a provider's patients are unlikely to induce significant
changes in physician-prescribing patterns.
Debate over the effects of cost-containment strategies often fail to
distinguish between the level of drug spending and growth in spending. The
rapid increase in pharmaceutical spending from 1987 to 1993 was due both to
rising drug prices and to higher per capita utilization. However, since 1994,
the growth in spending has been largely due to increased utilization.4 If this trend continues, increased patient cost-sharing
will play a larger role in reducing the level of drug spending than slowing
the growth in expenditures.
Our analysis has several limitations. First, we examined a working-age
population with employer-provided drug coverage. Thus, our findings may not
generalize to lower-income groups or the elderly population. However, our
findings reflect behavioral responses of more than 400 000 enrollees
to a wide range of drug and medical benefits.
Second, some plans imposed higher co-payments or coinsurance rates for
drugs dispensed at out-of-network pharmacies. We did not control for this
feature in our analysis because this information was not consistently reported
in the benefits package. In addition, we had no information on use of over-the-counter
medications, which could potentially mitigate the effects of increased patient
cost-sharing.
Third, we could not assess the full impact of extremely high co-payments.
In our sample, the mean difference in co-payments was $6.50 between generic
and brand drugs and $12 between preferred and nonpreferred brands, with a
maximum of $15. Therefore, we could not reliably predict the effect of a plan
with co-payments in excess of $30. Although such high co-payments were unusual
in our data, they are increasingly becoming more common as costly new drugs
and biotech agents enter the market.28,29
Finally, we could not control for selection of health insurance plans
because we did not know the full range of choices offered to employees. Most
of the firms in our sample offered employees a choice of medical plans, which
typically included a managed care and indemnity option. However, in all but
2 firms there was no choice of drug plan, which minimizes any potential bias
from employees selecting benefit package designs that suit their particular
needs or preferences. Furthermore, reestimating the models without these plans
did not change our results.
A large fraction of the increase in drug spending in 2000 was due to
higher expenditures on a small number of drugs and drug categories.3 Where drugs are placed in the formulary will substantially
affect utilization patterns and costs. Currently, drug classification is often
a function of ingredient cost and manufacturer rebates rather than clinical
outcomes.5 As a result, pharmacy benefit managers
and their sponsors may be designing prescription benefit packages that reduce
the costs of pharmaceuticals but increase overall medical costs.
There is little evidence about whether lower pharmaceutical use resulting
from higher patient cost-sharing adversely affects clinical outcomes. Several
studies have found that spending caps and formulary restrictions reduced use
of both essential and nonessential medications among low-income and elderly
populations.8-10,16,19,20 However,
few studies have found a consistent link between higher co-payments and patients'
health, particularly among persons with employer-provided coverage whose drug
spending comprises a much smaller percentage of their income. A recent study
of reference pricing for angiotensin-converting enzyme inhibitors, in which
insurance covers the cost up to the reference price and patients pay the extra
cost of more expensive medications in a class, found little evidence that
patients stopped treatment for hypertension or that health care costs increased.30 Future research should examine the impact of benefit
design on a wide range of therapeutic classes and whether changes in drug
spending affect medical care utilization and overall health care costs of
different patient populations.
1.Levit K, Cowan C, Lazenby H.
et al. Health spending in 1998: signals of changes.
Health Aff (Millwood).2000;19:124-132.Google Scholar 2.Dubois RW, Chawla AJ, Neslusan CA.
et al. Explaining drug trends: does perception match reality?
Health Aff (Millwood).2000;19:231-239.Google Scholar 3. Prescription Drug Expenditures in 2000: The Upward Trend Continues. Washington, DC: National Institute for Health Care Management Foundation;
2001.
4.Berndt ER. The US pharmaceutical industry: why major growth in times of cost containment?
Health Aff (Millwood).2001;20:100-114.Google Scholar 5.Penna P. Three-tier copay systems and consumer-centric care.
J Managed Care Pharmacy.2000;6:351-354.Google Scholar 6.Leibowitz A, Manning WG, Newhouse JP. The demand for prescription drugs as a function of cost-sharing.
Soc Sci Med.1985;21:1063-1069.Google Scholar 7.Reeder CE, Nelson AA. The differential impact of co-payment on drug use in a Medicaid population.
Inquiry.1985;22:396-403.Google Scholar 8.Soumerai SB, Avorn J, Ross-Degnan D, Gortmaker S. Payment restrictions for prescription drugs under Medicaid: effects
on therapy, cost and equity.
N Engl J Med.1987;317:550-556.Google Scholar 9.Soumerai SB, Ross-Degnan D, Avorn J.
et al. Effects of Medicaid drug-payment limits on admission to hospitals and
nursing homes.
N Engl J Med.1991;325:1072-1077.Google Scholar 10.Soumerai SB, McLaughlin TJ, Ross-Degnan D.
et al. Effects of a limit on Medicaid drug-reimbursement benefits on the use
of psychotropic agents and acute mental health services by patients with schizophrenia.
N Engl J Med.1994;331:650-655.Google Scholar 11.Harris BL, Stergachis A, Ried DL. The effect of drug co-payments on utilization and cost of pharmaceuticals
in a health maintenance organization.
Med Care.1990;28:907-917.Google Scholar 12.Smith DG. The effects of co-payments and generic substitution on the use and
costs of prescription drugs.
Inquiry.1993;30:189-198.Google Scholar 13.Walser BL, Ross-Degnan D, Soumerai SB. Do open formularies increase access to clinically useful drugs?
Health Aff (Millwood).1996;15:95-109.Google Scholar 14.Johnson RE, Goodman MJ, Hornbrook MC, Eldredge MB. The impact of increasing patient prescription drug cost sharing on
therapeutic classes of drugs received and on the health status of elderly
HMO members.
Health Serv Res.1997;32:103-122.Google Scholar 15.Johnson RE, Goodman MJ, Hornbrook MC, Eldredge MB. The effect of increased prescription drug cost-sharing on medical care
utilization and expenses on elderly health maintenance organization members.
Med Care.1997;35:1119-1131.Google Scholar 16.Horn SD, Sharkey PD, Phillips-Harris C. Formulary limitations and the elderly: results from the Managed Care
Outcomes Project.
Am J Manag Care.1998;4:1105-1113.Google Scholar 17.Stuart B, Zacker C. Who bears the burden of Medicaid drug co-payment policies?
Health Aff (Millwood).1999;18:201-212.Google Scholar 18.Hillman AL, Pauly MV, Escarce JJ.
et al. Financial incentives and drug spending in managed care.
Health Aff (Millwood).1999;18:188-200.Google Scholar 19.Motheral BR, Henderson R. The effect of a closed formulary on prescription drug use and costs.
Inquiry.1999;36:481-491.Google Scholar 20.Tamblyn R, Laprise R, Hanley JA.
et al. Adverse events associated with prescription drug cost-sharing among
poor and elderly persons.
JAMA.2001;285:421-429.Google Scholar 21.Motheral B, Fairman KA. Effect of a three-tier prescription copay on pharmaceutical and other
medical utilization.
Med Care.2001;39:1293-1304.Google Scholar 23.Duan N, Manning WG, Morris CN, Newhouse JP. A comparison of alternative models for the demand for medical care.
Journal Business Economic Statistics.1983;1:115-126.Google Scholar 24.Mullahy J. Much ado about two: reconsidering retransformation and the two-part
model in health econometrics.
J Health Econ.1998;17:247-281.Google Scholar 25.Manning WG, Mullahy J. Estimating log models: to transform or not to transform?
J Health Econ.2001;20:461-494.Google Scholar 26.Goldman D, Leibowitz A, Buchanan J. Cost containment and adverse selection in Medicaid HMOs.
J Am Stat Assoc.1998;93:54-62.Google Scholar 27.Ernst ME, Kelly MW, Hoehns JD.
et al. Prescription medication costs.
Arch Fam Med.2000;9:1002-1007.Google Scholar 29.Reissman D. Issues in drug benefit management: have we pushed member co-payments
too far?
Drug Benefit Trends.2001;13:24.Google Scholar 30.Schneeweiss S, Walker AM, Glynn RJ.
et al. Outcomes of reference pricing for angiotensin-converting-enzyme inhibitors.
N Engl J Med.2002;346:822-829.Google Scholar