Determinants of HMO Formulary
`Adoption Decisions
`David Dranove, Edward F. X. Hughes, and Mark Shanley
`
`Objective. To identify economic and organizational characteristics that affect the
`likelihood that health maintenance organizations (HMOs) include new drugs on their
`formularies.
`Data Sources. We administered an original survey to directors of pharmacy at 75
`HMOs, of which 41 returned usable responses. We obtained drug-specific data from an
`industry trade journal.
`Study Design. We performed multivariate logistic regression analysis, adjusting for
`fixed-drug effects and random-HMO effects. We used factor analysis to limit the
`number of predictors.
`Data Collection Methods. We held initial focus groups to help with survey design.
`We administered the survey in two waves. We asked respondents to report on seven
`popular new drugs, and to describe a variety of HMO organizational characteristics.
`Principal Findings. Several HMO organizational characteristics, including nonprofit
`status, the incentives facing the director of the pharmacy, size and make-up of the
`pharmacy and therapeutics committee, and relationships with drugs makers, all affect
`formulary adoption.
`Conclusions. There are many organizational factors that may cause HMOs to make
`different formulary adoption decisions for certain prescription drugs.
`Key Words. Formulary, managed care, pharmacoeconomics
`
`A drug formulary is a list of approved drugs.1 The term is historically
`associated with hospitals; a hospital pharmacy stocks those drugs on the
`hospital formulary and some nonformulary items.2 Today, formularies are an
`essential component of managed care.3 Most health maintenance organiza-
`tions (HMOs) provide greater coverage to patients who obtain drugs listed on
`their formularies.4 Some HMOs design their own formularies while others
`rely on third parties for formulary design. Regardless of how HMOs design
`their formularies, there is variation across HMOs in how they make adoption
`decisions for a given set of drugs. Some formularies are relatively ‘‘open’’ and
`include almost all FDA-approved drugs.5 Others are more restrictive, so that
`their HMOs pay for only one or two out of a class of competing drugs. This
`variation in formulary adoption decisions is, needless to say, important to
`
`169
`
`IMMUNOGEN 2284, pg. 1
`Phigenix v. Immunogen
`IPR2014-00676
`
`

`

`170
`
`HSR: Health Services Research 38:1, Part I (February 2003)
`
`pharmaceutical executives who are at a loss to explain why some HMOs
`adopt their drugs while others do not.
`Lyles et al. (1997) examine some of the factors influencing formulary
`adoption decisions by HMOs and other managed care organizations (MCOs).
`The 51 organizations that respond to their survey all report that they use
`formularies to contain costs and assure appropriate drug use. All respondents
`state that they assess the clinical effectiveness and cost effectiveness of drugs,
`while about half also assess the impact of drugs on quality of life. While this
`study sheds light on the goals of formulary makers, it does not identify sources
`of variation in adoption decisions. For example, the study does not address the
`relative importance of cost versus quality goals. Nor does the study assess how
`organizational characteristics affect specific adoption decisions.
`Lyles et al. show that when making formulary adoption decisions,
`MCOs examine a variety of criteria, including cost and quality. While all
`MCOs may weigh similar criteria, they may not reach the same conclusions.
`This can occur for at least three reasons. First, drug adoption decisions may
`vary because MCOs may pursue different objectives (or strategies for value
`creation) and may draw different conclusions about the available evidence on
`drug performance. Variation can also result from the organization-specific
`characteristics of MCOs. These could include incentives, controls, and
`practices that are incomplete or else inconsistent with the economic objectives
`for cost and quality that
`the MCO formally espouses. Organizational
`idiosyncrasies can thus bias adoption decisions away from those outcomes
`that would be most consistent with the MCO’s objectives. Lastly, variation in
`adoption decisions may stem from different relationships between MCOs and
`the drug manufacturers.
`In this study, we report the results of a survey of the directors of
`pharmacy at more than 40 HMOs. We identify economic and organizational
`factors that influence formulary adoption decisions for seven of the largest-
`selling drugs.6 Our intent is to clarify the decision processes that occur at this
`critical
`institutional
`linkage between pharmaceutical firms and HMOs.
`Knowledge of these processes becomes more important as the share of
`pharmaceutical sales that occurs through HMOs continues to grow.
`
`Address correspondence to David Dranove, Ph.D., Professor, Kellogg School of Management,
`Northwestern University, 2001 Sheridan Rd., Evanston, IL. Edward F. X. Hughes, M.D., M.P.H.,
`is a Professor at Northwestern University. Mark Shanley, Ph.D., is an Associate Professor at
`Purdue University in West Lafayette, Indiana.
`
`IMMUNOGEN 2284, pg. 2
`Phigenix v. Immunogen
`IPR2014-00676
`
`

`

`Determinants of HMO Formulary Adoption Decisions
`
`171
`
`BACKGROUND ON PHARMACY BENEFITS MANAGEMENT
`
`Pharmacy benefits management emerged during the 1980s when HMOs
`sought to extend the practice of selective contracting to prescription drugs.
`Under selective contracting, HMOs and other managed care organizations
`obtain discounts directly from providers such as hospitals and physicians by
`threatening to exclude high-price providers from their provider network.
`Providers are willing to grant the discounts because they realize that HMOs
`have the ability to move market share.7
`The same principles that enable HMOs to obtain discounts from
`providers apply to the purchase of prescription drugs. The early group and
`staff model HMOs established their own formularies. During the 1980s, IPAs
`outsourced pharmacy benefits management to independent pharmacy benefit
`management firms (PBMs) such as Medco, DPS, PCS, and Value Rx.8
`Through the use of
`formularies, HMOs obtain substantial rebates by
`threatening to exclude high-price drugs. These rebates may be obtained by
`excluding the drug entirely from the formulary or including it, but denying it
`‘‘preferred status’’; that is, requiring enrollees to make a differential copayment.
`The HMO use of formularies has grown rapidly since the 1980s. Today,
`92 percent of HMOs use a formulary.9 All of the HMOs that responded to our
`survey constructed their own formularies. (We excluded from our sample
`those that reported otherwise.) We focus on these HMOs because they make
`their own decisions and would be best able to respond to our survey. But even
`those HMOs that outsource formulary design can have custom-designed
`formularies. While independent PBMs have an ‘‘off-the-shelf’’ formulary that
`they offer to all clients, they will also tailor-make formularies to meet the needs
`of specific clients. Therefore, there is every reason to believe that there is wide
`variation in all HMO formularies, not just those that we study.
`Note that while formularies are important tools for drug use control,
`HMOs also may use other mechanisms such as disease management.10 We
`asked our respondents if they used disease management in the relevant
`therapeutic categories. When we included this measure in our models, our
`results were unchanged.
`
`FACTORS AFFECTING ADOPTION DECISIONS
`
`As a starting point for considering formulary adoption decisions, we would
`expect an HMO to add a drug to its formulary if the expected benefits, which
`
`IMMUNOGEN 2284, pg. 3
`Phigenix v. Immunogen
`IPR2014-00676
`
`

`

`172
`
`HSR: Health Services Research 38:1, Part I (February 2003)
`
`may be economic, medical, or both, exceed the additional expected costs. To
`the extent that HMOs follow this rational choice model, then adoption
`decisions for specific drugs should be highly correlated across HMOs. In other
`words,
`there should be considerable interdrug variation in adoption
`probabilities, where this variation depends on each drug’s commonly
`perceived benefit/cost ratio. For example, we might expect most HMOs to
`adopt a breakthrough therapeutic drug such as Proscar (for the treatment of
`prostate cancer). At the same time, we expect that fewer HMOs would adopt a
`drug with less obvious benefits and/or substantial side effects, such as Meridia
`(for the treatment of obesity).
`In this study, we do find considerable interdrug variation in adoption
`rates. We study seven drugs, for which the probability of adoption among our
`sample of HMOs ranges from .35 (Amerge and Viagra) to .80 (Allegra). To
`control for this interdrug variation in adoption probabilities, we estimate drug
`fixed-effects (i.e., we include separate indicator variables for each drug). This
`allows us to focus on the factors affecting the adoption of individual drugs
`across HMOs. However, in one specification, we replace fixed-effects with
`limited drug-specific measures of potential benefits and costs, such as the
`extent of competition within the drug’s therapeutic category.
`In addition to interdrug variation in adoption probabilities, there is
`considerable inter-HMO variation (i.e., variation across HMOs for specific
`drugs). Otherwise, the probabilities of adoption would all be zero or one. We
`infer that objective benefit/cost criteria are not the only determinants of drug
`adoption. One reason is that such criteria do not, in general, exist. Each
`organization is therefore free to interpret the available evidence. Organiza-
`tional factors specific to each HMO can also influence adoption by affecting
`(a) how that HMO conducts its benefit/cost analysis and (b) how it weighs the
`benefit/cost ratio against other considerations.
`For example, consider that each HMO with a formulary has a Pharmacy
`and Therapeutics (P&T) committee whose charge is to construct and update
`that formulary. The P&T committee is usually made up of between 10 and 20
`medical and managerial personnel; a few HMOs include one or more
`individuals from outside the organization, such as an independent consultant,
`consumer, or ethicist. A physician specialist or the HMO’s director of
`pharmacy will often chair the committee. Regardless of who chairs the
`committee,
`the pharmacy director has the ultimate responsibility for
`managing the HMO’s drug benefit.
`It stands to reason that adoption decisions are influenced by the
`individuals on the P&T committee and by organizational factors affecting the
`
`IMMUNOGEN 2284, pg. 4
`Phigenix v. Immunogen
`IPR2014-00676
`
`

`

`Determinants of HMO Formulary Adoption Decisions
`
`173
`
`committee as a whole. The individuals on P&T committees vary in their
`training, experience, and perspective.11 The data on which they must base
`decisions are often incomplete and unclear. In addition, the firms seeking to
`have their products adopted provide much of the data, and each committee
`has different experiences to guide their interpretation of this data.12 The P&T
`committee structures, which determine how members of the committee work
`together to make decisions, can vary by HMO, as can the incentives facing
`pharmacy directors to control the pharmacy budget. These variations may be
`only loosely linked to the goals or markets served by the HMOs.13 Some key
`decision makers, including physicians, may have little incentive to consider
`costs at all. This variation in decision makers, decision processes, and
`information may cause considerable inter-HMO variation in adoption
`decisions.
`
`HMO Characteristics and the Evaluation of Benefits and Costs
`
`All HMOs, for-profit and nonprofit alike, will value the pecuniary benefits that
`arise if a generous formulary enhances the overall demand for the HMO,
`especially among relatively healthy individuals. To our knowledge, there are
`no studies that demonstrate a link between formularies and HMO demand.
`However, we conjecture that healthy employees will prefer HMOs with more
`generous formularies, all else equal. One reason is that at the time they select a
`health plan, all individuals have ‘‘option demand’’ for a range of treatments.14
`In other words, they want the security of knowing that their HMO will cover
`whatever medical needs may arise. Another reason is that healthy patients
`might view a generous formulary as a signal of overall HMO quality. Sick
`individuals will likely also prefer a generous formulary, not always to the liking
`of the HMO.
`It is important to note that the benefits of boosting demand may vary
`according to the HMO’s target population and that population’s willingness to
`pay for treatments and drugs. For example, an HMO that serves mostly
`privately insured patients should be able to boost its premium if it offers a
`generous formulary. An HMO that serves mostly Medicare or Medicaid
`patients may be more constrained in its pricing.
`Many drugs provide HMOs with another pecuniary ‘‘benefit’’ by
`substituting for other therapies, sometimes at much lower cost. The anti-ulcer
`medications Zantac and Tagamet, which greatly reduce the need for costly
`ulcer surgery, are good examples. Although all of the drugs in our sample, with
`the exception of Viagra, substitute for other drugs or therapies, there is no
`systematic evidence on the extent to which they reduce costs through this
`
`IMMUNOGEN 2284, pg. 5
`Phigenix v. Immunogen
`IPR2014-00676
`
`

`

`174
`
`HSR: Health Services Research 38:1, Part I (February 2003)
`
`substitution effect. Consequently, we are unable to test directly whether the
`potential for substitution encourages formulary adoption, even though this
`potential often motivates pharmaceutical company arguments for the
`adoption of their products.
`Health maintenance organizations must weigh these benefits of
`including drugs on a formulary against the costs of doing so. An obvious
`cost is the direct cost of paying for the drugs. One factor that may affect this
`direct cost is the number of drugs in the therapeutic category. To the extent
`that the pricing of prescription drugs follows the predictions of oligopoly
`theory, we can expect HMOs to negotiate better discounts when there are
`therapeutically equivalent drugs available. The need to use a formulary to
`control costs may be tempered by the presence of other cost containment
`mechanisms, such as the inclusion of pharmaceuticals within the capitated rate
`paid to primary care physicians. In this case, capitation may substitute for the
`formulary. However, some HMOs may use both formularies and capitation to
`reduce costs. In this case, formularies and capitation are complements. While
`the relationship between formularies and physician capitation is worth
`investigating, the predicted direction is ambiguous.
`Lastly, we might expect for-profit HMOs to take a stricter view of the
`benefit-cost tradeoff than nonprofits. Weisbrod (1988) observes that when it
`comes to quality, nonprofits may offer more ‘‘hard to measure’’ attributes than
`for-profits. For example, Weisbrod argues that nurse training is hard for
`patients to measure, and documents that the nurses in nonprofit nursing
`homes are better trained than those in for-profit homes. From the perspective
`of a managed care enrollee selecting a plan, the exact makeup of the formulary
`may well be hard to measure. Not only is the formulary large and complex, but
`many enrollees will not know their exact drug needs at the time they select
`their health plan, and thus will not make specific inquiries about the formulary.
`This may lead for-profits to be less concerned about how enrollees react to
`decisions to exclude specific drugs.
`
`Organizational Factors Affecting Likelihood of Formulary Adoption
`
`There are many organizational reasons why the individuals who participate in
`formulary adoption decisions may not collectively find the optimal balance of
`economic costs and benefits.15 The P&T decision-makers must gather
`information about each drug, interpret the information, and negotiate with
`other decision makers who may not share the same interpretation of the data
`or the same values.
`
`IMMUNOGEN 2284, pg. 6
`Phigenix v. Immunogen
`IPR2014-00676
`
`

`

`Determinants of HMO Formulary Adoption Decisions
`
`175
`
`One way in which these factors can manifest themselves is in the
`relationship between HMO decision makers and the drug manufacturers.
`Drug manufacturers
`invest
`considerable
`resources
`in encouraging
`HMO decision makers to adopt
`their drugs. This includes supplying
`information to decision makers about drug benefits, as well as deve-
`loping personal relationships with decision makers. Thus, the attitudes of
`these decision makers toward drug manufacturers may influence their
`decisions. Drug manufacturers who contact an HMO more frequently and
`otherwise provide more ‘‘service’’ to the HMO may be perceived more
`favorably.
`Adoption decisions may also depend on the characteristics of the P&T
`team.16 For example, up to some point,
`the participation of multiple
`individuals might lead to higher quality decision making, due to the need to
`assess complex information and incorporate multiple perspectives. Past that
`point, however, as a more diverse set of individuals participates in adoption
`decisions, it will be more costly and time-consuming to inform them all with
`appropriate data, reconcile differences in how data are interpreted, and forge a
`consensus out of the different goals and incentives that motivate each
`individual. With the spread of managed care, it is increasingly likely that
`conflicts may arise between management, physicians, and other personnel.
`Differences in goals and values among the involved parties will become more
`difficult to reconcile.
`The size of an HMO may influence adoption decisions. Larger HMOs
`have greater opportunities for achieving scale and scope economies and
`exercising volume purchasing power than will smaller ones. At the same time,
`larger HMOs may have more layers of bureaucracy and face greater pressures
`to formalize their decision processes. Increased bureaucracy could make
`adoption decisions more difficult and impede decision makers responding
`effectively to market pressures.
`Even if decision makers in different HMOs have the same information,
`draw the same conclusions, and face the same organizational pressures, they
`may face different incentives to act on their information. Some pharmacy
`directors report directly to the chief executive officer or chief operating officer,
`whereas others report to the chief medical officer. These reporting relation-
`ships may affect the pharmacy directors’ decisions.
`incentives that
`Pharmacy directors may even face direct financial
`influence their decisions. All pharmacy directors in our sample undergo an
`annual performance evaluation. The performance criteria for these evalua-
`tions, however, vary across HMOs. Some pharmacy directors are evaluated
`
`IMMUNOGEN 2284, pg. 7
`Phigenix v. Immunogen
`IPR2014-00676
`
`

`

`176
`
`HSR: Health Services Research 38:1, Part I (February 2003)
`
`Table 1: Summary of Factors Potentially Affecting Formulary Adoption
`Decisions
`
`Factor
`
`Predicted Direction
`
`Rationale
`
`Nonprofit status
`
`More likely to adopt
`
`Large Medicare/Medicaid
`population
`
`Less likely to adopt
`
`Larger HMO
`
`Ambiguous
`
`Capitated physicians
`
`Ambiguous
`
`Pharmacy director rewarded
`for controlling drug costs
`Pharmacy director rewarded
`for controlling total costs
`Large P&T committee
`
`Less likely to adopt
`
`More likely to adopt
`
`Less likely to adopt
`
`Percentage physicians on
`P&T committee
`Frequent contacts with drug
`company representatives
`
`More likely to adopt
`
`More likely to adopt
`
`Number of drugs in the
`therapeutic category
`
`Ambiguous
`
`Offer more ‘‘hard to observe’’
`attributes of quality
`Public payers less likely to
`boost rates for more
`generous formulary
`Obtains purchasing discounts/
`bureaucratic decision
`making
`Capitation may substitute for
`or complement use of a
`formulary
`Financial incentives
`
`Drugs may substitute for more
`costly interventions
`Harder to reach consensus;
`composition of P&T
`committee may also affect
`adoption
`Favor ‘‘quality’’ over ‘‘cost’’
`
`Receive more favorable
`information and obtain
`better service
`Less need to adopt, but
`potentially lower costs of
`adoption
`
`based on a comparison of the pharmacy budget relative to plan. Others are
`evaluated based on member satisfaction with the pharmacy. These different
`evaluation criteria have important
`implications for formulary adoption
`decisions, since individuals may be biased toward decisions consistent with
`the incentive scheme under which they work.17 For example, a director
`evaluated on pharmacy budget relative to plan may be biased against
`including higher cost drugs on the formulary, even if those drugs offer
`offsetting cost savings that may boost HMO profitability. Conversely, a
`director evaluated on firm profitability or patient satisfaction will be biased
`toward including such drugs.
`Table 1 summarizes the various factors that we have identified that may
`affect formulary decision making. In the data section, we describe the specific
`variables that we use to measure these factors.
`
`IMMUNOGEN 2284, pg. 8
`Phigenix v. Immunogen
`IPR2014-00676
`
`

`

`Determinants of HMO Formulary Adoption Decisions
`
`177
`
`SURVEY METHODS
`
`Prior to administering the survey, we conducted a one-day focus group with
`leading representatives from HMOs, drug manufacturers, and academia.18
`The group made important contributions to both the theory and methods and
`made especially helpful suggestions for designing and administering the
`survey. For example, group participants clearly indicated that directors of
`pharmacy were the decision makers who should be surveyed. The focus group
`also stressed that an instrument that required less than ten minutes to complete
`would enhance the response rate, and that
`it would be necessary to
`compensate survey respondents for their time.19 These suggestions limited
`the scope of our research. Lastly, the group helped us select terminology to use
`in the survey.
`We conducted the survey in two waves, three months apart. We
`initially sent
`the survey to randomly selected directors of pharmacy
`(or the equivalent title) at 30 HMOs that reported they constructed and
`maintained their own formularies.20 We made follow-up calls to non-
`respondents and ultimately received 13 responses. Of
`these, 3 were
`missing responses to more than one key variable,
`leaving us with 10
`usable responses.21 We sent the second wave to directors of pharmacy
`at 45 more HMOs that also constructed their own formularies. After
`follow-ups with nonrespondents, we ultimately received 35 more res-
`ponses. One response was incomplete. One responding HMO serviced
`Veterans Administration patients only. We believed that
`formulary
`decisions at
`this HMO might be driven by different
`factors
`than
`at other HMOs, and deleted it from the sample. Two HMOs serviced
`Medicaid patients only. Consistent with our earlier discussion,
`these
`Medicaid-only HMOs did not include any of our study drugs on their
`formularies. Because of this ‘‘perfect fit,’’ we omitted these HMOs from further
`analysis.22 This left us 31 usable responses from the second wave, for a total of
`41 responses.
`to say to what extent our sample is representative
`It
`is difficult
`of all HMOs. Owing to the brevity of the instrument, we were unable
`to ask detailed questions about many HMO characteristics. Our HMOs
`represent a cross-section of sizes from less than 100,000 enrollees to more
`than 1 million, with a larger percentage of large HMOs than in the general
`population. Our sample also included a roughly equal mix of nonprofits
`and for-profits, compared with 66 percent
`for-profits in the general
`population.23
`
`IMMUNOGEN 2284, pg. 9
`Phigenix v. Immunogen
`IPR2014-00676
`
`

`

`178
`
`HSR: Health Services Research 38:1, Part I (February 2003)
`
`DRUG SELECTION AND MEASUREMENT OF THE
`DEPENDENT VARIABLE
`
`We asked our focus group to identify drugs for our study. We asked them to
`select drugs with the following characteristics:
`
`1. The drugs must be relatively new (approved for marketing in the last
`five years). We restricted our attention to new drugs because these are
`generally of greater interest to drug makers.
`2. The drugs must be in relatively high demand. This criterion again
`reflects the interests of drug makers.
`3. The drugs are not universally included or excluded from
`formularies.24 Given the constraints on survey size, we wanted to
`maximize the inter-HMO variation in adoption decisions.
`
`Note that the third criterion implies that the variation we find in adoption
`decisions across HMOs may exceed the overall variation for all drugs. For
`instance, there may be some drugs that appear on the vast majority of
`formularies. By focusing on those drugs for which there is disagreement,
`however, we can do a better job of explaining the sources of that disagreement.
`Table 2 lists the seven drugs included in our study, along with their
`manufacturers and their principal clinical indications. The list includes some
`drugs commonly prescribed at the time of the survey. For instance, Lipitor,
`
`Table 2: Drugs Studied
`
`Drug
`
`Manufacturer
`
`Indicated fory Year Approved a
`
`Worldwide Sales
`(1998)
`
`Formulary
`Inclusion Rate
`
`Allegra
`Amerge
`Diovan
`Evista
`Lipitor
`
`Rezulin
`
`Viagra
`
`Hoechst AG Allergies
`Glaxo
`Migraine
`Novartis
`Hypertension
`Eli Lilly
`Osteoperosis
`Warner
`Cholesterol
`Lambert
`reduction
`Warner
`Type 2 diabetes
`Lambert
`Pfizer
`
`Erectile
`dysfunction
`
`1996
`1997
`1996
`1997
`1996
`
`1997
`
`1998
`
`$485 million
`$90mn
`$282m
`$144m
`$2,200m
`
`$748m
`
`$1,100mnn
`
`.80
`.25
`.39
`.51
`.68
`
`.66
`
`.39
`
`aFor U.S. Sales
`nNot listed by MedAdNews. Sales estimated
`nnTwelve-month sales estimate extrapolated from nine months sales.
`
`IMMUNOGEN 2284, pg. 10
`Phigenix v. Immunogen
`IPR2014-00676
`
`

`

`Determinants of HMO Formulary Adoption Decisions
`
`179
`
`Viagra, and Allegra were among the top one hundred selling drugs in the
`world. The table also includes the year in which each drug was approved for
`marketing in the United States and 1998 worldwide drug sales, as reported in
`MedAdNews, an industry trade journal.
`We asked respondents to state whether each drug was included on their
`formulary. All respondents gave unambiguous ‘‘yes/no’’ responses for every
`drug. Based on the responses, we constructed a dichotomous dependent
`variable that equaled one if the HMO included the drug on the formulary and
`zero if they did not. The final column of Table 1 reports the fraction of HMOs
`that reported including these drugs on their formularies. In our empirical
`analysis, we used logistic regression to estimate models of
`formulary
`inclusion.25
`As mentioned earlier, an HMO may approve several drugs in a
`therapeutic category but encourage physicians and patients to use a subset of
`them, perhaps by use of differential copayments. To capture this possibility,
`we asked respondents to state whether drugs received preferential status.
`Preferential status was granted in about 20 percent of
`the cases. We
`recomputed the dependent variable by creating ordered categories of
`formulary inclusion; for example, we scored ‘‘on formulary-preferred’’ higher
`than ‘‘on formulary-not preferred.’’ We then reestimated our models of
`formulary inclusion using ordered probit, but obtained qualitatively similar
`findings to the logistic regression estimates. Specifically, the predictors that are
`statistically significant in the logistic regression remain statistically significant
`in the ordered probit regression, and the magnitudes of their effects on the
`probability of adoption are largely unchanged.
`There is another meaningful distinction between drugs on the formulary
`that are exclusive within their therapeutic class and those that share formulary
`status (even preferred status) with substitutes. Many of the factors that we
`consider might affect decisions about whether to grant exclusivity to a drug
`and which drug to select. We did not collect information about exclusivity,
`however, and leave this issue to future research.
`
`PREDICTOR VARIABLES
`
`We asked the pharmacy directors to provide general descriptive information
`about their HMOs, as well as specific information about various organiza-
`tional characteristics that might influence formulary adoption decisions. We
`also obtained information about each drug from MedAd News.26 From this
`
`IMMUNOGEN 2284, pg. 11
`Phigenix v. Immunogen
`IPR2014-00676
`
`

`

`180
`
`HSR: Health Services Research 38:1, Part I (February 2003)
`
`Table 3: Factors Affecting Likelihood of Formulary Adoption
`
`E2
`
`Economic and Organizational Factors
`E1
`Worldwide drug sales in 1998 (drug specific). Worldwide sales typically exceed U.S.
`HMO sales by a factor of four or more, so may be treated as exogenous to the
`decisions of U.S. HMOs
`Number of direct competitors in 1998 (drug specific). These are drugs within the
`same therapeutic category and class as the drug in question
`Is the HMO nonprofit? (HMO specific)
`Percentage of HMO revenue derived from Medicare and Medicaid (HMO specific)
`Percentage of primary care practitioners at financial risk for pharmacy cost (HMO
`specific)
`
`E3
`E4
`E5
`
`A2a
`A2b
`A2c
`A2d
`A3
`
`Administrative Factors
`A1
`To whom does the director of pharmacy report? (HMO specific)
`A2
`Importance of the following elements of director of pharmacy’s performance
`evaluation (HMO specific)
`Pharmacy budget relative to plan
`Overall profitability of HMO
`Member satisfaction with pharmacy
`NCQA accreditation
`Number and composition of the Pharmacy and Therapeutics committeen (HMO
`specific)
`Extent to which P&T committee obtains information from the following sources
`(HMO specific)
`HMO management
`Other MDs in HMO
`Published literature
`Consulting firms
`Specialty societies
`Competing health plans
`
`A4
`
`A4a
`A4b
`A4c
`A4d
`A4e
`A4f
`
`Relationship with Pharmaceutical Companies
`R1
`Number of times company representative visited HMO in previous year (HMO/
`drug specific)
`Overall satisfaction with company (1 5 not satisfied; 5 5 very satisfied) (HMO/drug
`specific)
`
`R2
`
`Other characteristics Not Specified
`O1
`HMO size (1 5 smallest size class; 4 5 largest size class) (HMO specific)
`O2
`Per member, per month pharmacy costs, and rate of PMPM increase (HMO
`specific)
`
`nFive HMOs did not report this information. We set their P&T committee size and makeup equal
`to the average, so as to preserve degrees of freedom.
`
`information, we constructed variables that correspond to the various factors
`that potentially drive formulary adoption decisions. Table 3 lists and describes
`these variables, and indicates whether they are HMO-specific, drug-specific,
`or HMO/drug-specific.
`
`IMMUNOGEN 2284, pg. 12
`Phigenix v. Immunogen
`IPR2014-00676
`
`

`

`Determinants of HMO Formulary Adoption Decisions
`
`181
`
`Several variables may be more meaningful when expressed in relation to
`others. For example, variable A2a, the ‘‘importance of pharmacy budget
`relative to plan,’’ is reported on a scale of 1–5. But this may be more
`meaningful when measured in comparison with the importance of other
`elements of performance evaluation, A2b–A2d. Thus, we measured ‘‘relative
`importance of pharmacy budget’’ as equaling A2a divided by the average of
`A2a–A2d. Similarly, we computed relative importance of opinions for
`questions A4a–A4d. Lastly, we measured R1 and R2 (relationships with drug
`manufacturers) for each HMO as follows. We asked the HMO to report the
`number of visits by representatives from each drug maker. We then computed
`the nominal difference between the number of visits by each drug maker and
`the average across all drug makers. Table 4 reports summary statistics for key
`predictor variables.
`
`Empirical Issues and Data Reduction
`
`As described above, 41 of the 75 surveyed HMOs returned usable responses.
`Although we nominally have 287 observations (7 drugs and 41 HMOs), we
`effectively have far fewer degrees of freedom, because an HMO’s adoption
`decision for one drug is likely to be correlated with that HMO’s decision for
`other drugs. In other words, our observations are not necessarily independent
`within HMOs. We adjust our standard errors to account for this intra-HMO
`correlation in adoption decisions.27
`Correcting for intra-HMO correlation reduces the effective degrees of
`freedom, thereby limiting the number of predictor variables that we may
`include in our model. To maximize the information contained in predictor
`variables while minimizing the number of predictors, we performed a factor
`analysis of two sets of predictors, thereby reducing them to two composite
`scores. Specifically, we computed factor scores from a varimax factor analysis
`of the variables measuring the composition of the P&T committee and the
`extent to which the P&T committee relied on different sources of information
`inside and outside the firm. We selected three factors with eigenvalues greater
`than one. These factors were heavily loaded (loadings greater than 0.5) on the
`following underlying variables:
`
`Fac