Physician Prescription Drug Choice: Investigating the Impact of Multiple Influences

Размещено на http://www.allbest.ru/

Physician Prescription Drug Choice: Investigating the Impact of Multiple Influences

Sanjog Misra

William E. Simon Graduate School of Business Administration

University of Rochester, Rochester NY 14627

Stephen Parente

Carlson School of Management

University of Minnesota, Minneapolis, MN ??

R. Lawrence Van Horn

William E. Simon Graduate School of Business Administration

University of Rochester, Rochester NY 14627



Abstract

This study provides evidence on the determinants of a physician's choice of alternative branded pharmaceuticals in the Statin class of drugs. We develop a physician choice model in which accounts for physician preferences in drug choice as well as the impact of pharmaceutical promotion and countervailing effects of patient cost-sharing. We find that among pharmaceutical promotion channels, physician detailing, and journal advertising have the highest marginal effect. Physician experience with the drug ….

1. Introduction

The demand for prescription drugs in the United States is a topic of much debate and discussion. Prescription drug costs have increased by X and many argue that prescription drugs are one of the primary drivers of increasing health care costs. Understanding the factors which drive demand of prescription drugs is likely to be more important in the future as the US considers the implementation of a Medicare drug benefit.

Demand for prescription drugs is unlike most other consumer products in that they must be prescribed by physicians, and are demanded by patients with limited information who commonly enjoy insurance coverage. To develop a more complete understanding of the demand for prescription drugs it is necessary to develop a model of the prescribing process by physicians.

The factors which determine the final choice of a particular drug is complex. There are three sets of factors that must be considered to model the process. First, one must consider the physicians clinical judgment and their natural preferences towards drugs within a therapeutic class. Second one must consider the role and impact of pharmaceutical promotion on the physician's choice. Finally the consumer's preferences must be modeled. Patient preferences are likely to be affected by both the available information to the patient and promotion to patients in the form of direct to consumer advertising as well as their financial incentives which are affected by the HMOs choice of drug-tiering.

The choice of promotional level and channel by the pharmaceutical industry works to stimulate drug demand while the corresponding choices of patient cost-sharing by the managed care industry are chosen to constrain demand. A tension exists at the patient-doctor interface around which factors will dominate. A complete model of this process is valuable to both industry segments. If one was to know the marginal effectiveness of various promotional strategies in the face of the countervailing forces the pharmaceutical industry would be in a position to better target its promotional dollar. Similarly if managed care firms knew the marginal impact of moving a drug from a preferred to a less preferred tier, they may be able to extract some of that value from the pharmaceutical firms in terms of rebates. Finally policymakers are concerned with the adverse effects of direct-to-consumer (DTC) advertising. A complete treatment of the prescribing process which includes DTC along with detailing to physicians and patient cost-sharing can uncover the true impact of DTC.

To date there does not exist any research which jointly considers all factors simultaneously. Due to our unique patient level pharmaceutical claims data, combined with information on drug detailing and DTC spend this paper address the aforementioned questions. The second main contribution of this paper is to consider the process by which the physician chooses drugs within a therapeutic class, and more specifically models the role of persistence which has long been ignored in prior physician choice models.

Accounts for persistence in the prescription behavior of physicians,

Differentiates between new and existing patients,

Allows for the impact of marketing activity to differ by patient type

Accounts for physician behavior and characteristics and,

Models the influence of HMO's via tiered co-pays.

Our analysis reveals a number of findings. First we find that doctors tend to use idiosyncratic preferences when making prescription decisions for first time patients and then are likely to “stick” to those decisions. We also find marketing activity and managed care influences to be somewhat stronger for new patient decisions than for returning patients. In addition our estimates reveal detailing to have the highest impact of all marketing activity on prescription choice.

The paper is organized as follows: section 2 discusses the nature of promotion in the pharmaceutical industry, section 3 discusses the role of third party payment and the evolution of patient cost sharing, section 4 develops models of physician choice. This is followed by a description of the data in section 5. Section 6 outlines the estimation approach and the results are discussed in detail in section 7. Finally, we conclude with a summary and a discussion of future research possibilities.



2. Promotion in the pharmaceutical industry

In an otherwise lackluster economy the pharmaceutical industry has posted phenomenal gains. Sales of prescription drugs in 2002 totaled over $200 billion and have grown at an average of 18% per annum over the past five years. During the same period promotional expenditure relating to prescription drugs have steadily risen, as have prescription drug prices. The cost of promoting a new drug now rivals that of R&D, estimated at $900 million (DeMasi (2001)). The bulk of the promotional activity has traditionally been in the form of visits to physician offices by sales representatives (hereafter detailing.) For instance, pharmaceutical firms spent $5.3 billion on a sales force of 78,840 generating 61.4 million physician visits in 2000, up 10% from 1999. The market for branded pharmaceuticals has, however, changed materially over the last decade. Therapeutic classes have become more crowded with therapeutically equivalent drugs. This competition has not served to reduce prices, much to the contrary, and promotional spend is frequently the target for industry criticism. The industry as a whole has borne the brunt of public criticism for its promotional expenditures (Clinton health care reform), has been accused of excessive profits, and health care cost inflation has been attributed to rising drug expenditures. This has come at a time when the industry has delivered fewer “blockbuster” new therapeutics and an increasing number of “me-too” drugs similar to existing drugs within the therapeutic class. Most recently the industry has been criticized for direct-to-consumer advertising and the potential for inappropriate drug utilization that results.



3. The Role of Third-party payment

The activities of the pharmaceutical industry are not the only force impacting prescription behavior. Insurers and managed care organizations (MCOs) have countered with measures of their own including formulary restrictions - limitations of which drugs will be covered, and now the advent of 3-tier pharmacy benefit programs which provide patients with financial incentives to take a “preferred” drug within a therapeutic class. These tiers are usually cost based with “generic” drugs being the first choice. Pharmacy Benefit Managers (PBMs) additionally exert influence over the final choice of a drug through their rebate programs and preferred wholesale pricing that they may then offer their managed care clients. Last but not the least, the advent of direct to consumer advertising and the free flow of information over the internet has created a mass of informed patients who now play an active role in their own treatment. One might reasonably conjecture that given the multiplicity of these factors the efficacy of detailing to physicians should be reduced. It is therefore even more surprising to note that the industry is adding drug representatives at an increasing rate.

While the influential nature of marketing expenses incurred by drug companies is accepted (see Hurwitz and Caves (1988), Berndt et al. (1994, 2001), Rizzo (1999), Gonul et al. (2001)), its actual effectiveness is hard to assess (see Wazana (2000)). Many of the studies mentioned fail to incorporate the fact that physicians exhibit persistence in their prescription habits (see Hellerstein (1998) and Coscelli (1999)), and that many new prescriptions are simply automatic renewals of users' earlier choices. Accounting for this persistence may affect findings on the effects of promotional activity (Givon and Horsky 1991.)

physician prescription drug choice



4. Choice Models of Physician Behavior

The choice of a prescription drug is unlike most other choice scenarios studied by marketing researcher and economists. It differs from regular choice situations (e.g. brand choice) both in its complexity and in the inter-temporal underpinnings that generate it. Given the discussion earlier, it goes without saying that the institutional dynamics that govern prescription drug choice are complicated. The agents influencing a physician's decision include Insurance Firms, Pharmaceutical Companies and Distributors and the patients themselves. A discussion of these influences is warranted. Pharmaceutical firms and their cronies use detailing, samples, journal advertising, events, gifts and to influence the physician. They also use DTC advertising and promotion to indirectly impact prescription via the patient. A recent trend has been the emergence of tiered copay and its use as a mechanism to mitigate the influence of pharmaceutical firms. Finally, the patient may have his or her own ideas, knowledge base, past experiences etc as would the physician. Given the vast array of these influences it becomes important to specify a framework that captures the nuances of the institution we are studying. However, as a benchmark it is useful to specify a straightforward choice model ala Wosinska (2002) and Gonul et al. (2001).

4.1 Prior Research

There are two papers that use disaggregate choice data and are similar in spirit to our objectives. The first is a paper by Gonul et al. (2001) who use a sample of physicians to model prescription drug choice as a function of detailing, prices and samples. The paper does not however address most of the issues laid out in (a)-(e). The second paper, by Wosinska (2002), is closer to ours in that it uses data very similar to ours (i.e. data obtained from insurance records) and allows for differences in patient types, incorporates copay and interaction effects. The key difference between our framework and Wosinska (2002) is that we specify a structural model of physician behavior she uses a reduced form model to estimate the effects. This allows us to examine and understand physician prescription behavior in more detail and ascertain effects more precisely. Also related to our study is the recent work of Richard and Van Horn (2003) that uses aggregate data to implement a discrete choice specification that accounts for persistence.

Our proposed model is similar to the one proposed by Roy (1990) and Resnick and Roy (1990), dubbed the “lightning bolt” model. The framework proposed by these authors was also applied by Roy, Chintagunta and Haldar (1996) and by Chintagunta (1999) in the context of grocery and frequently purchased products using scanner panel data. Because of subtle differences in the data and the data generating process we modify and adapt the framework to suit our needs. In spite of these changes, the proposed model, much like the “lightning bolt” framework allows for a number of possible behaviors on the part of the decision maker. Our empirical findings show that this framework is vastly superior to naive choice models and other models that incorporate some reduced form version of persistence.

4.2 The Benchmark: Naive Choice Model

We start with a simple benchmark, namely the Naive Choice Model. This model assumes that the doctor evaluates each patient visit independently and chooses that drug that maximizes utility. Defining the notion of utility here is difficult, to say the least. However with some reluctance and great benefit in terms of implementation we assume that the physician acts as a perfect agent of the patient. Hence we define the utility (or benefit) that a physician j feels a particular drug k has for patient i at time t as

In this specification are brand/drug specific constants, is copay paid by patient i at time t for drug k, are interactions of physician specific characteristics with the drug and are error terms that are assumed to be distributed i.i.d. extreme value type I. For utility as defined as above we obtain the probability of choosing drug k as

.(1)

Using this probability, and aggregating across drugs, physicians, drugs and patients in the usual manner we can arrive at a likelihood specification. While in most applications of discrete choice this model seems plausible if not appropriate. It is important, however, to understand the implications this structure has for the underlying physician behavior. First this approach does not account for the fact that patients may be new or returning. Consequently, it fails to account for the possibility that a physician may simply renew a prescription without reevaluating the patient. Second it does not account for differences in the impact that detailing and other marketing activity may have on account of there being different patient types. Third it also fails to take into account that doctors may use different decision rules depending on patient type. To account for these behavioral nuances and more we propose a structural model in the next subsection.

4.3 Proposed Model of Physician Behavior

In contrast to the naive model presented above, a more complete model would account for that fact that new patients are different from existing patients and physicians tend to make decisions differently for each. We start by define the following indicator



.

New Patients (): For patients that a physician has not seen before she may engage in an active evaluation of the patient-drug match using information from outside influences or may choose to prescribe a drug without based on simply idiosyncratic preferences. This decision process can be captured by

, (2)

where is the probability that a physician uses the outside information (contained in the vector ) to prescribe the appropriate drug. Correspondingly, with probability the physician may prescribe the drug based on influences that completely idiosyncratic and unconnected to marketing activity of the drug companies, copay or any experience the doctor has had with the drug Past experience may itself be influenced by marketing activity and copay and hence we treat it as an outside influence.. In other words the are as in the benchmark model while the consist of only brand/drug specific constants.

Existing Patients (): If a patient has visited a physician in a previous time period it is possible that the physician simply renews the prescription without any additional evaluation of the match between the drug and the patient. On the other hand it might be the case that the physician re-evaluates the patient just as she would a new patient. The probability that an existing patient is prescribed drug k can be denoted as



. (3)

In this specification is the probability that the physician simply re-prescribes the drug that had been prescribed to the patient on the previous occasion. Combining the two conditional probabilities of we get the unconditional probability of prescribing drug k as

(4)

The model presented in (4) is extremely flexible and allows for a number of interesting behaviors on the part of the physicians. Note that it explicitly accounts for structured heterogeneity on the patient types (existing and new) and on the physician decision process. We outline the main features of our model in the key remarks that follow.

Remark 1: If

then the model reduces to the benchmark model.

Remark 1 shows that our proposed model nests the naive choice model as a special case. In other words a structural interpretation of the naive choice model would imply that at each choice occasion the physician reevaluates every patient independent of her past history and chooses the drug that is best for her.

Remark 2: If



then there is no outside information on choice.

In particular when physicians are completely idiosyncratic in their decision for first time patients and persistence is perfect then marketing activity or experience will have no impact on choice. While this is an extreme case it is important to note that this pattern is permitted by the model.

Remark 3: If

then physicians use outside information only to make prescription decisions for first time (established) patients.

It is possible that doctors evaluate new patients on the basis of all available information and then “stick” to their choices. Similarly for

we would have a scenario where doctors choose drugs for new patients based on their idiosyncratic preferences but later may reevaluate them using all information available, including marketing input.

Remark 4: If then doctors are more likely to prescribe drugs for new patients without relying on outside information.

For the case of new patients it is important to understand whether most doctors are affected by outside influences such as marketing activity, copay and/or their past experiences. Clearly if we have a case where, it is more likely that, doctors use some form of their idiosyncratic preferences to make the drug choice decisions for new patients. From a medical ethics standpoint if one finds that the parameter is small it implies that doctors are fairly independent of outside influences.

Remark 5: If then doctors are more likely to re-evaluate matches based on all available information including marketing input.

The parameter captures the degree of persistence there is in prescription behavior. However, it also acts as a weight for the possibility that the doctor reevaluates existing patients based on outside influences.

Remark 6: If

()

then impact of marketing mix elements on prescription choice is smaller (larger) for established (new) patients than for new (established) patients.

Understanding where outside influences have a larger impact is important to both managed care and the pharmaceutical industry not to mention policy makers. The condition outlined in the remark suggests that if persistence is high and doctors are not very idiosyncratic the main impact of marketing activity and copay will be on the prescription choices of first time patients. On the other hand should the reverse be true then reevaluations of established patients would be more likely to be based on outside influences.

While the above discussion highlights some of the possible behavior patterns on the part of physicians, many are extreme cases used to illustrate the flexibility of the framework and in all likelihood would not map on to reality. Our conjecture is that in most cases moderate values of the structural parameters will be seen. We expect that there will be high (but not perfect) persistence and that doctors will use a mix of idiosyncratic preferences and marketing information to make decisions for first time patients. In particular we also expect that doctors will tend to be fairly patient specific and hence idiosyncratic in their choices for first timers. More formally, we expect

and .

5. Data

Pharmaceutical and medical data data were provided by a large health insurer serving six states located along the Eastern Coast of the United States. The study period examined was claims data for dates of service in calendar years 1999 and 2000. A three-tier pharmacy pricing system was introduced in 1999. However, this pricing system was later `reset' by the selection of another pharmaceutical benefit management (PBM) firm at the start of 2000. This change forced the managed care plan to reassess all of their preferred and nonpreferred prescriptions and consequently led to many changes in both the selection of drugs for the preferred tier, but also changes in prices. In addition, we were able to better identify price change effects.

The pharmaceutical claims data included detailed information on the type of prescription drug received by the patient, the copayments paid, the reimbursement to the pharmacy, the date of prescription, the prescribing physician and the therapeutic class of the drug. Most importantly, we were able use the prescription data to identify employer-specific heterogeneity in copayments within the three tiers. This permitted the identification of the specific three-tier pricing contracts developed and agreed on between the health insurer, the employer, and the PBM. This data was critical to completing this analysis because it allowed us to identify the pricing changes due explicitly to the three-tier system.

Medical claims data for physician, outpatient, inpatient and laboratory services was also identified. This was used to identify the total expenditure and utilization of a patient for medical care. In addition, the diagnosis information contained in these data were employed in a case-mix algorithm developed by Weiner et al (1991) to produce a vector of 34 patient-level non-exclusive categorical variables known as Ambulatory Diagnostic Groups (ADGs) representing the major acute and chronic care factors driving differences in medical care utilization and expenditure.

Operationalization of Variables

Details of IMS Data (Larry)

How copay was computed (Larry)

How Experience was computed (Larry)

Why Sampling was omitted

Describe Table 1



6. Estimation and Results

6.1 Estimation

To estimate our models we require a particular specification for the utilities used in equations (1) and (4). In particular we specify

,(5)

And

.(6)

The are coefficients of relevant variables while and are drug specific constants. For identification purposes we set the constants for the generic drug Gemfibrozil to be equal to zero. Given these specifications the estimation follows the standard maximum likelihood approach. Recall that is an indicator which is 1 if drug k was prescribed by physician j for patient i at time t and zero otherwise. We can then write the overall sample likelihood as

.(7)

Maximizing (7) gives us estimates of the parameters of interest.



6.2 Results

Naive Model

Table 2 presents the results obtained via estimation of the benchmark model. A cursory glance reveals that the results are consistent with expectation and the extant literature. We find that Detailing emerges as the most important determinant of prescription behavior with Journal Advertising as a close second. The difference between the two coefficients is, however, not statistically significant. Comparing the coefficients does not portray a complete picture since the elasticities may be very different. In Table 5a we have presented elasticities for the three main drugs (Lipitor, Pravachol and Zocor) which account for over 90% of the prescriptions in this class of drugs. The others drugs were ignored primarily because they have zero (or fairly small) expenditures for certain marketing activity which gives zero (or negligible) estimates of elasticity. For example brands such as the generic Gemfibrozil and Mevachor have virtually zero detailing and hence the zero detailing elasticities.

The computed detailing elasticities are in the range [0.18, 0.21]. On the other hand journal advertising elasticity ranges between [0.02, 0.034]. To get an idea of that these elasticities imply we note that for Lipitor the market leader the required expenditure on detailing to gain one percentage point of market share would be $59 million whereas it would require $63 million to get the same gain using journal advertising. The estimates suggest that detailing has a much larger impact per dollar that journal advertising. The effect of direct to consumer advertising was found to be smaller than detailing with elasticities in the [0.03, 0.11] range. The fact that DTC has larger elasticities than journal advertising is surprising but occurs primarily because of (a) the large DTC advertising expenditures incurred by these three drugs (b) the nature of elasticities of the logit model. In the next section we show that accounting our proposed model results in elasticities that are significantly different from those obtained here.

The model also shows that doctors seem to be sensitive to the co-pay faced by the patients. The co-pay elasticity lies in the range [-0.07,-0.34] and though it seems small, it implies a rather strong effect since copay changes are usually in large amounts. For example, Lipitor's copay elasticity under the benchmark model is estimated to be -0.07 and on average Lipitor is in the preferred tier with a copay of about $11. If Lipitor were to be moved to a lower tier with a copay of $20 the copay elasticity effect would imply that Lipitor would drop from the current 60.6% market share to 54.5%!

The basic choice model also reveals that experience is a significant determinant of drug choice. Overall, the model seems to point towards the importance of detailing as a determinant of prescription choice.

Results from the Proposed Model

The estimates from our proposed framework are presented in Table [3 or 4]. The first noticeable aspect of the result is the improvement in the log-likelihood and the information criteria. While the BIC was 22535 in the naive choice model it now stands at 19056. Since the proposed model nests the naive choice model we conduct a likelihood ratio test and no unexpectedly find that our proposed model fits the data better. As a robustness check we also estimated a model with only persistence and found that while it outperformed the benchmark it faired poorly compared to the full specification.

The estimates of and some interesting aspects of physician behavior. These are outlined below.

Finding 1:

,

i.e. doctors are more likely to prescribe drugs to new patients without relying on outside influences.

The estimate of turns out to be equal to 0.36 and is significantly less that 0.5. The finding implies that in more cases than not doctors tend to base decisions for new patients on idiosyncratic factors and don't rely on marketing information, managed care influences or past experience with the drug. This is not to say that these influences do not play a role but that their influence is limited.

Finding 2:

Prescription behavior is persistent.

We also found strong evidence of persistence in prescriptions. Out estimate of is statistically significant and stands at 0.76 which is consistent with those found by Van Horn, Suponsic and Richard (2003) for the same class of drugs. This finding is significant in that it points out that outside influences have a limited role to play in repeat prescriptions.

Finding 3:

,

Outside information has more of an impact on first time patient decisions.

Our estimates reveal that

while .

In our choice framework it implies that the impact of outside influences is larger for first time patient decisions. While this in itself is not surprising it is important to note that the estimate of the difference

while significantly different from zero is only at approximately 0.11. The fact that this difference is small implies that when using outside influences to make decisions doctors may not be differentiating between patient types. Clearly our estimates are only for one particular therapeutic class and these effects may be different for other drug classes.

In addition to helping understand physician behavior our framework allows us to compute the response elasticities both for new and existing patients. Table 5b presents these elasticities. A quick look at this table shows that the impact of all variables is smaller for established patients than it is for new patients. This is directly related to the estimates of persistence and the degree of idiosyncratic behavior . A comparison to table 5a also reveals that the naive model overestimates the impact of copay and DTC advertising and underestimates detailing elasticities.

The qualitative aspects of the results, however, are similar to those found in the naive model. Detailing continues to be the most important factor. However DTC doesn't seem to be as important as in the naive choice model. The detailing elasticities are all in the range [0.20, 0.35] which is consistent with the findings of Van Horn, Suponsic and Richard (2003) who find that the estimate is on average 0.28 for the same drug class. Chintagunta and Desiraju (2003) also report (for the SSRI Selective Serotonin Reuptake Inhibitor therapeutic class) that based on their US sample detailing elasticities in the range [0.20, 0.30]. The finding that detailing elasticities are virtually ten times that of journal advertising or DTC advertising is also consistent with past findings.

Our empirical findings have implications for all players in the industry, these are discussed next.

7. Discussion and Implications (Larry & Steve)

There are three key findings of this paper. First, among the pharmaceutical promotion channels, physician detailing, and journal advertising have the highest marginal impact. Direct to consumer advertising appears to be less important. Second, the persistence of an individual physician's prescribing behavior overshadows the impact of marketing information, managed care influences or past experience with the drug. As was found in our proposed model, the effect of persistence is strong for both new and existing patients. Third, we find that marketing information has a greater impact on physician behavior for first time patients rather than returning patients.

We discuss the implications of these findings with respect to different actors involved in physician prescribing behavior including: insurers, pharmaceutical manufacturers, patients and finally, physicians themselves. For insurers and possibly self-insured employers administering their own health plans, these results suggest tiered pricing of pharmaceutical are having a significant impact after controlling for individual physician effects. If we were to have found that persistence crowds out the impact of other factors to the point of making them random noise, insurers as well as their agents, pharmaceutical benefit managers (PBMs) should reconsider the use of three tier structures. However, the effects are present and as such, signal the strategy as viable in at least its early stage of development. The degree of physician persistence moderating the impact of tier-pricing may be of concern, but some level of physician practice style is always present and the degree of success insurers and employers associate with strategy needs to be traded off against the disutility of using a more restrictive formulary that might discourage enrollment.

Pharmaceutical firms also have successfully affected the behavior of physicians. Our results have greatest pertinence for this group since they report on the success of their marketing channels to affect prescribing behavior. For a given class of drugs, the marginal impact of detailing, DTC and journals could be compared to relative share of a firm's marketing budget to develop a crude cost-effectiveness estimate. Since we are not privy to such budgets, we can't comment further on which strategy makes the most sense in terms of resource allocation. However, if budget constraints are not a tantamount concern for marketing, clearly detailing and journal advertisement are having a significant effect and should not be diminished unless they are prohibitively costly.

The impact of our results with respect to patients is difficult to gauge without full information of the consequences of persistence and marketing strategies of their consumption of pharmaceuticals. If a patient taking Lipitor has better health outcomes than if she had taken because of a one or several marketing strategies, than the impact of than the impact of the pharmaceutical promotion would be welfare improving for this consumer. If on the other hand, side-effects caused from taking Lipitor and would not have been caused by taking Xanax, the strategy would harm the patient. A situation that would be welfare improving for a the patient is where the marketing strategies challenge a physician's existing persistent prescription pattern and get him to consider better alternatives than the physician's default behavior. Without a fuller accounting of patient outcomes, we can not comment further.

For physician's the results demonstrate that they retain control over their prescribing behavior and are comprehensively affected by marketing strategies. At best, these strategies alert physicians to ways of treating new and existing patients that they may have not previously considered, and as a consequence, improve the patient's health status in ways that their habitual practice would have not. At worse, they serve as a nuisance to the physician and they willfully ignore the advice and communication regarding a new therapy because of their unwillingness to separate marketing from a clinical communication. Our results clearly demonstrate the effect of persistence and that it remains the driving factor affect prescribing behavior. Although persistence can be thought of as a good quality where the physician is precise in their treatment decisions, persistence can hurt a patient too. If they insist against clinical evidence of a possible better outcome for their patient solely on the basis of habit, these strategies are ultimately welfare improving for society as a whole because they moderate the effect of `deleterious' persistence.

There are three limitations to this paper. First, we only examine one class of drugs and draw inferences from it with respect to entire industries. While focusing on only statins limits the generalizability of our results, it also providers an excellent controlled setting to compare and contrast the results of the naive, persistence and proposed complete models. Future studies can use this work as an estimation template focused on other classes of drugs. A second limitation is that we only looked at one insurer. However, while we can't use this insurer to generalize to the nation, we know from earlier work (Parente, et al, 2004), that the insurer is not extraordinary compared to a national sample of insurers in their physician prescribing behavior. We chose to use this insurer's data because of accessibility, but are confident the analysis can be repeated by other insurers. Finally, we don't explicitly account for consumer search behavior. Ideally, we had information on the consumer preferences and their sources of information regarding the utility of a drug we could have developed a more complete empirical specification. However, it was unavailable to us and we would not expect it be commonly available to any insurer or researcher at this time. At the very least, we assume the effect of consumer preferences regarding pharmaceutical consumption to be random and don't expect it to bias our estimates. Similarly, we assume patient-physician pairing is also random in that one group of consumer does not systematically know more than another to inform their choice of physician.



8. Conclusion and Directions for Future Research (Larry & Steve)

We extended a model the model and if nothing else it show persistence is still king, but promotion does have an impact other applications for other Pharma. Other classes. Interesting to get patient behavior explicitly involved for prescribing behavior. Could do this with new research from consumer directed health plans where the insured bahvior to examine pricing differentials is tracked and can be formally added into the empirics. To do this correctly requires access to employer's full data to see consumer counter-factuals for other employer insurance offerings.

Another possible effect is senior discount card use to disentangle effect of consumers BUT begin data requirements to do this right. Still the card and it's use may serve as good consumer identifying information separate from physician behavior.



References

Chintagunta P. and R. Desiraju (2003) “Strategic Pricing and Detailing Behavior in International Markets” forthcoming Marketing Science.

Chintagunta, P. (1999) “Variety Seeking, Purchase Timing and the “Lightning Bolt” Brand Choice Model”, Management Science, vol. 45, No.4 April. pp 486-498

Harris B.L, Stergachis A, Ried L.D. The effect of drug co-payments on utilization and cost of pharmaceuticals in a health maintenance organization. Med Care 1990; 28: 907-917.

Hillman A.L, Pauly M.V, Escarce J.J, et al. Financial incentives and drug spending in managed care. Health Aff (Milwood) 1999; 18: 189-200.

Johnson R.E, Goodman M.J, Hornbrook M.C, et al. The effect of increased prescription drug cost-sharing on medical care utilization and expenses of elderly health maintenance organization members. Med Care 1997; 35: 1119-1131.

Johnson T.E, Goodman M.J, Hornbrook M.C, et al. 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.

Motheral B, Faiman K.A. Effect of a three-tier prescription copay on pharmaceutical and other medical utilization. Med Care 2001; 39: 1293-1304.

Motheral B.R, Henderson R. The effect of a copay increase on pharmaceutical utilization, expenditures, and treatment continuation. Am J Manag Care 1999; 5: 1383-1394.

Nelson A.A, Reeder C.E, Dickson W.M. The effect of a Medicaid drug copayment on the utilization and cost of prescription services. Med Care 1984; 22: 724-736.

Rector T.S. Finch M.D. Danzon P.M. Pauly M.V. Manda BS. Effect of tiered prescription copayments on the use of preferred brand medications. Medical Care. 41(3):398-406, 2003 Mar.

Reeder C.E, Nelson A.A. The differential impact of copayment on drug use in a Medicaid population. Inquiry 1985; 22: 396-403.

Resnick, S. and R. Roy (1990) “Multivariate Extremal Processes, Leader Processes and Discrete Choice Models” Advances in Applied Probability, v. 22 pp. 309-331.

Roy, R. (1990) “Two essays on consumer choice” Unpublished Dissertation Thesis, Cornell University, Ithaca, N.Y.

Roy, R., P. Chintagunta and S. Haldar, (1996) “A Framework for Investigating Habits, ``The Hand of the Past'', and Heterogeneity in Dynamic Brand Choice”, Marketing Science, Vol. 15, No. 3, pp. 280-299.

Weiner J.P. Starfield B.H. Steinwachs D.M. Mumford L.M. Development and application of a population-oriented measure of ambulatory care case-mix. Medical Care. 29(5):452-72, 1991 May.

Размещено на Allbest.ru