Printer Friendly

The effect of direct to consumer television advertising on the timing of treatment.

I. INTRODUCTION

In August 1997, the Food and Drug Administration (FDA) relaxed the rules governing television advertising of prescription pharmaceutical products. Before that time, broadcast ads were permitted only to mention either the name of a drug or a disease against which a drug was effective but not both. After August 1997, pharmaceuticals were allowed to mention both the disease and the drug brand name (as long a a brief list of side effects was mentioned and 1-800 number or World Wide Web site was provided with more detailed information). Spending for direct to consumer advertising (DTCA for prescription drugs went from $985 million in 1996 to approximately $4.2 billion for 2005 (Donohue, Cevasco, and Rosenthal 2007). This has led to a great deal of debate in the medical profession and among health care insurer and managed care organizations and an ongoing review of advertising rules by the FDA However, very little is actually known about the effects of DTCA for the efficient allocation o prescription drugs.

We will examine how DTCA affects physician prescribing patterns and courses of care for patients suffering from a representative chronic condition, osteoarthritis (OA). This condition is of special significance, as one of the major products in this area, Merck's "Vioxx" cyclooxygenase-2 (COX-2) inhibitor was forced to withdraw from the market in October 2004 due to side effects on patients with heart conditions. The side effects of Vioxx have caused considerable criticism of Merck's advertising strategy for Vioxx (see, e.g., editorials from the New England Journal of Medicine; Mukherjee, Nissen, and Topol 2001; Topol 2004).

The primary goal of this paper was to determine what effect local and national television advertising on behalf of the two main COX-2 inhibitors had on the treatment decisions that patients made in collaboration with their physicians. In particular, we will examine the impact of DTCA on the time patients wait after diagnosis with OA and before initiating treatment with a COX-2 inhibitor. The paper will proceed by first reviewing the literature on DTCA in Section II. Section III will present a theoretical model of optimal delay to treatment. Section IV will present details of the data and empirical model we implement. Empirical results are presented in Section V, and Section VI concludes with a discussion about future research.

II. BACKGROUND AND LITERATURE

A. Advertising for COX-2 Inhibitors

Historically, pharmaceutical advertising was done largely through "detailing"--promotion directly from the manufacturer to the physician, either through visits by representatives or contacts by pharmacists or through advertisements in professional journals. Since the mid-1980s, however, drug companies in the United States have increasingly turned their marketing strategies directly toward the consumer. This advertising largely takes place through television media and in newspapers. This change in advertising approach has its share of both critics and advocates.

The pharmaceutical industry in the United States is large--accounting for more than $132 billion in retail sales in 2000 alone (NIHCM 2001). In 2000, Celebrex (celecoxib), the leading COX-2 inhibitor, had sales of approximately $2.6 billion ("Pharmacia has setback for parecoxib" 2001)--while Vioxx (rofecoxib) sold more than $1.2 billion in the first half of 2000 (Knight Ridder News 2001). In support of Vioxx, Merck spent almost $161 million in DTCA in 2000 (Schumann 2001)--which was the most spent on DTCA for any prescription pharmaceutical, making it the 39th most advertised brand of any kind in 2000 ("Top brands in network primetime-2000" 2001). Over the same time periods, Pharmacia and Pfizer jointly spent $78 million in DTCA supporting Celebrex.

We have chosen to examine the role of DTCA in the context of the market for COX-2 inhibitors. During the period spanned by our data (1999-2002), the two available COX-2 inhibitors were Vioxx and Celebrex. These drugs are appropriate subjects of study for a number of reasons. First, they have been heavily advertised. Second, they are significantly more expensive than alternative pharmacologic treatments for chronic pain. Third, data published approximately halfway through our sample period indicated that these products may carry a significant risk of adverse cardiac side effects. Consequently, from mid-2001, many physicians began to question the wisdom in their use (Topol 2004). Additionally, given the widespread suspicion of advertising for prescription drugs, clinicians and policy makers have questioned whether DTCA for COX-2's might be causing fatalities. Our results will speak directly to this issue in that if DTCA had a deleterious effect (in terms of matching high-risk patients to COX-2 use) then we will get one sign pattern on interaction terms between advertising and patient comorbidities. If, however, DTCA is having a positive effect (by matching patients most suited for COX-2 inhibitors with that therapy) then we will get the opposite sign pattern on the advertising-comorbidity interaction terms.

For this study, we have acquired data on television advertising for both Celebrex and Vioxx at the national (network) level and for the top 75 local media markets in the United States. These data are aggregated to the monthly level. Celebrex was approved by the FDA in December 1998 and Vioxx was approved in May 1999. Thus, at least one of the products was available for use over the entire 1999-2002 time period of our analysis. Figure 1 presents the 2000-2002 trend for the spending on national television advertising for Vioxx and Celebrex taken from our advertising database (described below in Section III).

B. Literature on the Impact of DTCA

In economic terms, we would expect advertising for prescription drugs to have three possible effects. First, advertisement for a particular prescription product will provide information regarding the symptoms and regarding the fact that effective treatments are available about the medical condition that the drug treats. This may be labeled a "public good" effect, as the information conveyed by advertisements for one brand may aid sales of all brands selling competing products. Second, advertisements may provide important information regarding side effects, contraindications, and the like that may prompt patients to consult with their physician regarding a treatment modality. This component of the advertising may be labeled as a "matching" effect since it would assist patients and physicians in matching treatment regimes. Third, advertising may simply lead patients to demand a product because of the aesthetic or persuasive characteristics of the ad, or the reputational impact of the ad, rather than the efficacy of the drug. This effect may be labeled as a "brand" effect. Since health care markets, including pharmaceuticals, are often characterized by moral hazard (as patients do not generally pay the full cost of the medications they consume), the welfare implications of this third effect are uncertain.

[FIGURE 1 OMITTED]

The studies on the impact of advertising in the prescription pharmaceutical market that have been published to date have yielded conflicting results. There is an arm of this literature that is generally supportive of advertising in this market, such as work by Telser (1975) and Leffler (1981). Keith (1995) finds that patient suggestions regarding pharmaceuticals (aspirin for cardiovascular disease) are important determinants in prescription decisions and that advertising tends to lead to more appropriate care as a consequence. In this, Keith is advancing an argument made earlier by Masson and Rubin that posits several mechanisms that could lead to positive impacts from advertising on the efficiency of the pharmaceutical market (including that it might encourage people to associate symptoms with a disease and seek care or that it might alert people to treatments they were previously unaware of, which would encourage them to seek care) (Masson and Rubin 1985). For a survey of the more optimistic literature in this area, see Rubin (1991) and Kleit (1998).

Not all economists, however, are so sanguine about the prospects of positive welfare effects from prescription pharmaceutical advertising. Hurwitz and Caves (1988) find that--on net--promotional activities by pharmaceutical firms tend to have the effect of preserving market share for existing products and slowing the penetration of new compounds in the market. King (1996) uses monthly sales data in the ulcer drug market to test the effect that marketing efforts have on the industry. He finds that marketing by a firm causes the demand for the firm's own products to become more inelastic. Similarly, Rizzo (1999) finds that DTCA significantly reduces price elasticity in the market. A reduction in price elasticity would increase opportunities for supracompetitive pricing.

The post-1997 era has presented an opportunity for examination of the new policy regime for DTCA, and much of the literature has been focused on the FDA policy shift. As Zachry and Ginsburg point out, however, there is a paucity of studies that examine the actual impacts of DTCA (Ginsburg 2001).

In one of the few such studies, Dubois (2003) examines the impact of DTCA through the lens of variation in procedure and drug use. He notes previous evidence that there is a wide geographic variety in the use of various medications and suggests that such variations imply underserved populations. Dubois cites several sources that indicate that geographic variations in prescriptions have declined since the relaxation of DTCA regulations, perhaps implying that DTCA is conveying important medication information to previously underserved populations. Calfee, Winston, and Stempski (2002) study whether the August 1997 policy changes at FDA increased the demand for the statin class of drugs using monthly data from IMSHealth and Scott-Levin for a 58-mo period. The authors, however, found that advertising did not have a statistically significant impact on aggregate prescriptions filled. According to the authors, "it may only be possible to detect the effect of DTCA advertising on consumer demand with disaggregated data that link a patient's cholesterol treatment history with the timing of DTCA expenditures."

III. THEORETICAL MODEL

Consider a patient who has been diagnosed with some chronic disease. On diagnosis, the patient faces a choice--either initiate or delay treatment. The benefits of treatment are potential improvements in symptoms of the underlying disease. The costs are monetary (the treatment generally must be purchased at some positive market price) and potentially psychic (people often resist taking medication) as well as the cost of any potential side effects. To decide whether to initiate treatment, the patient will evaluate her utility with treatment and her utility without treatment and pick whichever path yields the highest expected value. If the instantaneous utility associated with therapy is not higher than the utility associated with no therapy then the patient will choose to delay. Each period thereafter, the patient will undertake the same utility calculus, to determine whether to begin therapy or continue to wait.

This model is similar in form to real option theories that have been applied to financial instruments (see, e.g., McDonald and Siegel 1986; Merton 1973), to the timing of land development (see, e.g., Arnott and Lewis 1979; Capozza and Li 1994, 2002; Titman 1985), and even to the timing of initial public offerings (see, e.g., Benninga, Helmantel, and Sarig 2005). An interior optimum for the delay to an action can exist when there is a cost to undertaking some action and when the benefits from the action increase over time.

We will motivate our empirical research by exploring a simple model of the timing of treatment for a stylized chronic disease. Consider a patient who has been diagnosed with a condition, which reduces her health. Furthermore, assume that the impact on health is cumulative--delaying treatment implies that the health state continues to decay. The progression of the disease can be countered by pharmacological therapy. (Assume for the sake of simplicity that there is only one viable therapeutic option.) The value of that therapy is unknown, but the patient has expectations about the treatment effect--expectations that can be affected by information, such as physician advice, testimonials from friends, or DTCA. Thus, the patient's instantaneous health is:

[h.sub.t] = H - [delta]t, if no treatment/H - [delta]t + [theta](a), with treatment,

where H is the base level of health at disease onset, [delta] is the impact of the disease on health during each time period, t represents time, and [theta](*) is the expectations around the pharmacologic treatment effect, which can be affected by advertising, a. Furthermore, [[theta].sub.a] > 0 and [[theta].sub.aa] < 0.

Utility is defined across consumption of some numeraire good, [x.sub.t], and health, [h.sub.t]. Consumers may consume 1 unit of pharmaceutical treatment per period, at price P, or may continue to delay therapy. Once a person chooses to initiate therapy, she will continue to receive treatment until her death at time T. (Given the degenerative nature of the disease assumed in this model, once it becomes optimal to purchase treatment, it will necessarily be optimal in every time period after that.)

Consequently, each patient will maximize the present value of lifetime utility by selecting the optimal delay for treatment onset according to:

(1) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII],

where U(.) is patient utility defined across income (I) and health. The optimal delay to onset, [d.sup.*], is defined by taking the first partial of Equation (1) with respect to d and setting it equal to 0:

(2) [partial derivative]V/[partial derivative]d = [V.sub.d] = U(I, H - [delta]d) -U(I - P, H - [delta]d + [theta](a)) = 0.

The usual convexity assumptions require that the second partial of Equation (1) with respect to d be negative or:

(3) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

Given these assumptions, the optimal time to treatment exists, [d.sup.*] = [d.sup.*](I, H, [delta], r, a). Note, of course, that one condition for an interior solution is that at time = 0, the value of utility without therapy must be greater than the value of utility with therapy. That is, if U(I - P, H - [delta]t + [theta](a)) > U(I, H - [delta]t)[for all]t then the patient will choose to initiate therapy immediately and [d.sup.*] = 0. Similarly, if U(I - P, H - [delta]t + [theta](a)) < U(I, H - [delta]t)[for all]t then the patient will never initiate therapy.

To illustrate the solution, note that:

(4) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

Since diminishing returns in health imply that the higher levels of health resulting from treatment reduce the marginal utility of health, and the complementarity between income and health will imply further reductions in [U.sub.h] with the lower net income resulting from the requirement that some portion of any consumed therapy is paid out of pocket.

Comparative static analysis of the impact of advertising on the optimally selected delay of treatment is straightforward. Inserting the optimal delay time functional, [d.sup.*](.) into Equation (2) and differentiating with respect to the advertising parameter, a, and rearranging yields

(5) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

The sign of this effect depends on the sign of [[theta].sub.a]. If a patient believes himself to be a good candidate for the treatment then greater exposure to advertising, or other positive information about the efficacy of the treatment, will reduce the optimally chosen delay to therapy initiation. However, if the advertising conveys information that leads the patient to believe that he is a poor candidate for the treatment--by emphasizing adverse events or highlighting a contraindication from which the patient suffers--then the expected treatment effect will be reduced, which will lengthen the optimal delay. Thus, whether the information contained in the ad is "positive" or "negative" will depend on patient expectations, other diagnoses, and preferences across the characteristics of the therapy.

There are several implications from the theoretical model for the empirical estimation. First, patients select the optimal delay from diagnosis to therapy. Thus, we will calculate below this delay for each person in the data by estimating a parametric duration model. Second, since pharmacological treatments are better suited to some patients than others, the optimal delay will be a function of the patient's other clinical diagnoses. Finally, the model implies that this optimal delay will be a function of income, health state, opportunity cost of treatment, and advertising exposure--measured at the point of therapy initiation. We will employ either direct measures or proxy measures for each of these factors.

[FIGURE 2 OMITTED]

IV. EMPIRICAL MODEL

A. Empirical Specification

Before specifying an empirical implementation, we note that the decision we will model is one that is joint between the patient and the physician. Thus, our question will be "what impact does DTCA have on the likelihood that the physician/patient interaction will result in a prescription?" We will model the length of the spell between diagnosis and initiation of COX-2 inhibitor therapy. There is a long econometric literature on duration modeling, which presents us with various modeling options. Options range from nonparametric methods that impose few distributional restrictions to parametric methods where the distribution of the "time to failure" (to use the language of much of the literature) must be explicitly specified. All versions of the model are based on estimating the hazard function or the likelihood of a transition (in our case, between nonuse of COX-2 inhibitors and use of COX-2 inhibitors) between states at time [DELTA]t, conditional on the spell (of nonuse) having persisted to time t. One common specification for this hazard rate is to assume it is the product of some baseline population average hazard and an individual specific term (which may or may not depend on covariates).

The problem with this proportional hazard model is that it requires the population average effect to be constant over time. A more flexible approach for duration models is presented in Kalbfleisch and Prentice (1981) and in Ridder (1990) and known as accelerated failure time (AFT) models. These models permit the baseline hazard to increase, decrease, or remain constant over time. Figure 2 below presents the empirical hazards for the delay to treatment data we explore. The instantaneous failure rates are not constant for our data, as the proportional hazard assumption is strongly rejected for nearly all covariates in our models. Consequently, we will estimate an AFT version of a duration model. In addition, the decreasing hazard we observe is consistent with a Weibull distribution--which is the distribution we will assume.

In addition to nonconstancy in the baseline hazard, we have one other characteristic of the data to accommodate in our model choice. Patients may exit their delay spell in one of two ways: by choosing Vioxx as the treatment to initiate or by choosing Celebrex. These exit strategies are not likely to be random and may consequently affect the length of time that the delay spell itself lasts. Thus, we are faced with the possibility of underling heterogeneity in the duration data, which arises due to unobservable characteristics of the patient. Competing risk versions of the AFT model can account for this data-generating process. For applications of the competing risk models in health care, see Hamilton (1997), Cutler (1995), and Picone, Wilson, and Chou (2003). These models are based on the generalized method of Heckman and Singer (1984), which model heterogeneous transition frailty as a multiplicative term with a gamma distribution.

Thus, we will model the delay to COX-2 initiation as an AFT hazard function with heterogeneous failures in a competing risk model where the instantaneous hazard follows a Weibull distribution, and the unobservable heterogeneity is modeled as a gamma distribution. The hazard rate is then:

(6) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII],

where [x.sub.i] are individual patient and practice characteristics, t is the duration of the individual delay spell, and [[alpha].sub.j] capture the group-level heterogeneity in exit (to Vioxx or Celebrex). For some patients, it will be optimal to delay longer than we observe them, and for some, it will be optimal to delay indefinitely. Given that we only have data spanning the 1999-2002 time period, these two populations will be observationally equivalent--requiring that we take right-hand censoring into account. The model is estimated by maximum likelihood.

Finally, one remaining issue to be addressed is whether any selection effects are at work in the models. It seems plausible that advertising could bring more people to the physician office--which is what we find in prior research (Bradford et al. 2006). While this may lead to more diagnoses of OA, it is not immediately clear that this would bias our estimates on how long individuals wait before adopting therapy. (Even if the shift in patient population did lead to a change in delay, our estimates are reduced form in nature and thus agnostic with regard to what the actual mechanism of a DTCA effect might be.) However, to assess whether such selection issues are driving the results, we reestimated, ex-post, both versions of the model presented below and included proxy variables for the probability that the patient seeks a visit and the probability that the patient receives a prescription during the month when therapy is initiated. These proxy variables are the percent of all the physician's monthly office visits that are to OA patients and the percent of all the physician's monthly visits to OA patients that are associated with a COX-2 inhibitor prescription, respectively. These variables represent naive estimates of the probability of a visit and prescription for each patient during the month in which the patient actually began therapy. Since these are practice average measures, they will be exogenous to the individual patient's characteristics but capture general influences at the practice level on the likelihood that the patient progresses to treatment. The estimated DTCA parameters and the parameters on the DTCA interaction effects were highly robust to inclusion or exclusion of these variables--thus suggesting that selection effects are not playing a large role in generating the results we present below. For the sake of simplicity, therefore, we do not include these models in Table 3; however, they are available on request from the authors.

B. Data

For this study, we have acquired data on television advertising for both Celebrex and Vioxx at the national (network) level and for the top 75 local media markets in the United States. These data are aggregated to the monthly level. Figure 1 presents the 2000-2002 trend for the dollars spent on national television advertising for Vioxx and Celebrex taken from our advertising database.

As Figure 1 indicates, at the national level, monthly advertising exposure for the two brands is roughly comparable over the entire 2000-2002 time period. Interestingly, a very different picture emerges at the local level, where television advertising spots for Celebrex are always greater than that of Vioxx for the entire period. The value of ads purchased at the local level is much lower in dollar terms than at the national level. Since patients do not generally distinguish between the payment source of a television ad (whether the local station or national network receives the revenue), we will include measures of the total spending on television advertising in our empirical models. Consequently, we will have both cross-sectional and across time variation in DTCA exposure.

C. Clinical Data

Practice Partner, Inc. (Seattle, Washington) has marketed a commercial electronic medical record (i.e., Practice Partner) to physician practices for more than a decade. This product is intended to replace paper charts and has been widely adopted--largely by practices that are community based--for clinical reasons and not for research purposes (nor because the practice has any affiliation with a research group or institution). The Medical University of South Carolina collaborates with the vendor to gain access to the record extracts of practices that were willing to have their data used for research purposes. This led to the development of a geographically diverse national research network of ambulatory, mostly primary care practices that use this single electronic medical record system (known as PPRNet). We will examine data on practices from 1999 through 2002. As of 2002 (the end of the study period), 72 practices in 25 states, with 348 physicians, were network members.

Each quarter, participating practices run a computer program, developed and maintained by the electronic medical record vendor, to extract patient activity of the previous quarter from the electronic medical record system. These data are taken from the patient's medical record--similar to chart abstraction. The data capture all diagnoses, medications, patient characteristics (weight, blood pressure, etc.), lab tests ordered, and lab results. Currently, the entire research network database has information on 604,111 patients, including 3.6 million patient contacts, 3.8 million prescription records, 10.1 million vital signs, 12 million laboratory records, and 1.3 million preventive services records. We extract a subset of this data on 22,011 patients who had ever been diagnosed with OA and who had physician visits in the years 1999-2002. (The time period of analysis is dictated by the availability of advertising data and not the availability of clinical data.)

A recent assessment of a subsample of about 500,000 patients from PPRNet practices indicates that the patient population is approximately 57% female, with a mean age ([+ or -] SD) of 43.1 [+ or -] 21.1 yr and has a racial breakdown of approximately 85% Caucasian, 8% African American, and 4.3% Hispanic. PPR-Net practices are located in urban, suburban, and rural areas. Based on a four-tiered classification scheme for the Rural-Urban Commuting Area codes (Morrill, Cromartie, and Hart 1999), PPRNet practices are highly representative of the distribution of the U.S. population. Among the 114 practices active as of December 2005, there are 462 physicians, 51 physician assistants, and 63 nurse practitioners. Eighty-nine (78%) are family medicine practices; four of these are family medicine residency programs. Twenty (18%) are internal medicine practices and five (4%) are combined practices of primary care physicians. Thus, the PPRNet practices appear reasonably representative of other primary care practices in the United States.

D. Variables

The dependent variable for the models is the length of time each patient waits before initiating therapy with one of the two COX-2 inhibitors after being diagnosed with OA. This is calculated as the difference (in days) between the diagnosis date for OA (recorded in the clinical data) and the prescribing date for the first instance of a Vioxx or Celebrex prescription. However, COX-2 inhibitors are used for many complaints in addition to OA; additionally, patients may use a COX-2 for occasional arthritis pain even before a formal diagnosis is made. In these cases, the "delay" would be negative. For such negative delays, the duration between the diagnosis and the prescription is set to 1 d. In addition, most of the patients with a diagnosis of OA do not receive a prescription for Vioxx or Celebrex as long as they are observed in the data. Thus, the duration models we estimate will control for this right censoring.

The independent variables fall into several categories, such as advertising information, patient individual clinical information, and market/practice characteristics.

We obtained national and local advertising information from Competitive Media Reporting, Inc., which collects data on media advertising for all products, including pharmaceuticals, at the market (e.g., city) level. The data are specific to the brand name of the product. Consequently, it is possible to determine which products were advertised, which month they were advertised, how many times they were advertised, and how many dollars were spent on the ads. Patients and physician practices were assigned to the nearest media markets separately by two of the authors. When a practice was close to multiple media markets, they were assigned to the one that was nearest (by driving miles). In addition, we excluded all practices that are unusually far from the nearest media market (more than 100 straight-line miles) to avoid any bias from mismatching of practice and local media market data.

We measure advertising exposure as the total (national and local) dollars spent on ads broadcast for each brand advertised. We add the separate measures for national advertising and local advertising into a single total monthly spending variable. While it is the case that local ads tend to be shown during different times of the day and during different programming, it is unclear that this difference matters empirically. We have estimated versions of our models with the national and local ad spending included separately, and the net results are not meaningfully different. Additionally, the parameters on national advertising are essentially unchanged in magnitude and significance if we exclude local advertising; so, while there may some concern that local advertising could be endogenous (a concern that is ameliorated by the use of individual data, typically from only one practice in each media market), the practical biases appear negligible.

Additionally, since the theoretical model suggests that patient characteristics at the time of the switch are the important factors in initiating therapy, and since we have not captured precisely when in the month therapy begins, we will measure potential advertising exposure as the dollars spent in the month preceding the initiation of therapy.

The patient data contain limited demographic and detailed clinical information. For patient demographic information, we include patient age and an indicator variable for whether the patient is female. We also include variables that capture whether the patient has ever been diagnosed with (or treated for) other relevant comorbidities. These include indicator variables for if the patient has ever been diagnosed with heart disease (coronary disease or hypertension), depression, diabetes, hyperlipidemia, or ever treated for gastrointestinal difficulties with proton-pump inhibitors, H2 blockers, or related products.

Imputations of additional descriptive variables can be made secondary sources. We imputed the price of an intermediate length of physician's office visit with an established patient from the American Chamber of Commerce Research Association's Quarterly Price Reports (http://www.coli.org/). These quarterly reports contain average prices for 50 commodities (including physician office visits) for around 300 metropolitan areas. The linking between average physician visit price and the patient was accomplished by using the average price in the metropolitan area nearest the primary care practice site. Average county per capita income, the percent of the county population covered by Medicare, the percent of the county employed in the labor force, the percent of the county population that is Caucasian and African American, the county population, and the number of physicians per 10,000 population were also merged onto the data from the Area Resource File. Counties were identified as the county in which the practice is located. This information is available on an annual basis.

In addition to the impact of advertising, another source of information that may affect physician prescribing is medical journals. The late 1990s and early 2000s was a period when a significant amount of research was being conducted on the efficacy and side effects of COX-2 inhibitors. We will control for clinical knowledge in two ways. First, over the period of our study (1999-2002), there were more than 900 publications in English-language medical journals about COX-2 inhibitors. Of those, 132 were specifically in the area of OA. To control for the effect of this research on clinical providers, we created a data series, that measures the number of publications in each month that had the keywords: rofecoxib, celecoxib, Vioxx, Celebrex, and osteoarthritis. We further refined the measure by dividing it into three series: the number of publications each month that focused on Celebrex, the number of publications each month that focused on Vioxx, and the number of publications each month that focused on both.

Second, in August 2001, Mukherjee, Nissen, and Topol (2001) published an influential article in a major medical journal where they reviewed data available from a major clinical trial, which indicated serious statistically significant concerns about the cardiovascular risk associated with Vioxx (rofecoxib). To a lesser, and not statistically significant, extent, the paper raised concerns about Celebrex (celecoxib). This was the first publication in a major outlet to raise issues about increased risk of myocardial infarction associated with COX-2 inhibitors in general and Vioxx in particular. These concerns were later to be validated when Merck withdrew Vioxx from the market in October 2004. We will include an indicator variable, which equals 1 after August 2001 and 0 otherwise to test whether the practicing clinical community responded to this new information even in the face of significant DTCA in favor of Vioxx and Celebrex.

E. Possible Impacts of Advertising

There are several variables that are central to understanding the impact of advertising on patient treatment delay decisions. The first set, obviously, are the measures of advertising spending on behalf of Vioxx and Celebrex. In general, one would expect the impact of Vioxx or Celebrex brand television DTCA to shorten the time that patients wait before initiating any COX-2 inhibitor treatment--in which case the parameter on the advertising measures will be negative and significant. It is possible that the only effect of advertising is class level. That is, it may not matter which drug is advertised-any advertising for a COX-2 inhibitor may affect the demand for both approximately equally. To evaluate this, we will estimate two versions of the model. The first will include total dollars spent (as both a linear and a squared term) in the previous month on both Celebrex and Vioxx as one variable. The second will include separate measures (again, both linear and squared terms) for Celebrex and Vioxx ad spending. If television advertising has a pure class effect then advertisement for Vioxx and Celebrex would have roughly equal effects when entered individually on the delay to therapy for Celebrex (Vioxx) in the separate models we run for delay to initiation for each brand individually.

That advertising has an effect, however, says little about the social welfare impact of DTCA. We can, however, learn something about the welfare effects of advertising by examining its effect on delay to treatment for patients that are likely to benefit from, or be poor candidates for, COX-2 inhibitor use. In particular, patients who have required treatment for gastrointestinal problems using protonpump inhibitors, H2 blockers, or other similar treatments are more likely to suffer gastric irritation from nonsteriodal anti-inflammatory drugs (NSAIDs) and so are the good candidates for COX-2 inhibitors. In this case, we identified patients with gastrointestinal comorbidities as those who have ever had a prescription for ranitidine (e.g., Zantac), famotidine (e.g., Pepcid), cimetidine (e.g., Tagamet), omeprazole (e.g., Prilosec), esomeprazole (e.g., Nexium), lansoprazole (e.g., Previcid), rabeprazole (e.g., Aciphex), pantoprazole (e.g., Protonix), sulcralfate (e.g., Carafate), misoprostol (e.g., Cytotec), and Helidac, Prevpac, or metoclopromide (e.g., Reglan). We interacted the advertising measures with the indicator variable for gastrointestinal problems. If advertising improves patient matching then the parameters on those interactions will be negative and significant--indicating shorter delays to initiation.

In contrast, there are a number of conditions that make a person a poor candidate for use of a COX-2 inhibitor. These include a diagnosis of heart disease (hypertension and other coronary diseases). Individuals with these conditions should initiate therapy with a COX-2 inhibitor less frequently, which translates into longer delay times. However, this set of contraindications was not widely discussed in the clinical community until the publication of the Mukherjee, Nissen, and Topol (MNT) article in August 2001. Consequently, we will test for the informational components of the DTCA for heart disease by including a three-way interaction term between the advertising measures, the heart disease indicator variable, and the indicator variable for whether the treatment began after August 2001. If the advertising is conveying clinically useful information then the parameter on this interaction will be positive and significant (indicating that advertisement induces individuals with those comorbidities to wait longer to begin treatment after the contraindication for heart disease was published in the clinical literature).

V. RESULTS

Table 1 presents the means and standard deviations of the variables used in our model, including the average delay to treatment (conditional on initiating therapy). More information on the raw delay measures is presented in Table 2, which shows the average delay time between the diagnosis and the first long-term use of each COX-2 inhibitor as well as the number of people who ultimately adopt each. We find that patients who first adopt Celebrex tend to adopt more rapidly than those who first adopt Vioxx, with a delay of 163 d for the Celebrex users, compared with a delay of 199 d for the Vioxx users. Interestingly, for all the attention paid to the introduction of COX-2 inhibitors, and the large expenditures on promotion on their behalf, 64% of the sample (11,741 patients) did not adopt one of the COX-2 inhibitors.

We estimate two versions of the hazard function presented in Equation (6). Model 1 contains the advertising measures (as combined national and local dollars spent) for Vioxx and Celebrex together, along with the set of independent variables listed above. Model 2 includes separate measures for total Vioxx and Celebrex monthly advertising, along with the other independent variables. Thus, we will estimate the hazard rate in Equation (6) above, where

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

and where TV is either the combined Vioxx and Celebrex national and local advertising dollars or brand-specific national and local advertising dollars, GI is the indicator for a gastrointestinal comorbidity, CVD is the indicator for heart disease comorbidities (cardiovascular disease and hypertension), TNM is the indicator variable for the post-August 2001 period (which corresponds to the publication of the Topol, Nissan, and Mukerjee article that first alerted the medical profession to the increased risk for heart attack associated with the use of some COX-2 inhibitors), and Z represents the remainder of the explanatory variables discussed above. The vector Z also includes individual practice fixed effects. Since in nearly all cases there is only one practice per media market, our practice-level fixed effects correspond largely to media market fixed effects. Finally, we rely on the fixed effects to correct for clustering (repeated observations) at the physician practice level. The coefficient estimates are presented in Table 3.

In the first model, where the COX-2 television advertising dollars for both brands are summed together, we find that television advertising does have a significant effect on the delay in treatment. Total television spending has a positive first derivative and negative second derivative-both significant at better than the 1% level. The parameter estimates in Table 3 are not direct measures of the net effect in terms of days of delay. Table 4 presents these net effects for the variables of primary interest and converts the raw effects to days of delay induced by a 1-unit increase in the variable of interest. When calculated at the mean of the data, a $100,000 increase in all COX-2 advertising during the month of diagnosis has the effect of lengthening the time patients wait to begin COX-2 therapy after being diagnosed with OA by about 30 d (evaluated at the mean of the data). Note that this is in addition to average delay periods of between 160 and 200 d delay (Table 2).

The second model in Table 3 breaks television advertising spending into separate measures for total monthly spending on behalf of Vioxx and total monthly spending on behalf of Celebrex. We find that the results are similar--in that spending on behalf of both brands tends to increase the delay initially and then lead to a decrease--again, with statistical significance in excess of the 1% level. There is, however, a significant difference in terms of where on this quadratic function the two brands are placed.

In the case of Vioxx, the relationship between total spending and delay of therapy becomes negative at $2.4 million per month. During the time period of our data, Merck spent approximately $3.5 million per month on behalf of Vioxx; thus, the marginal dollar spent for Vioxx advertising led to a reduction in the average delay before initiating any COX-2 therapy of approximately 14 d (Table 4). In contrast, the relationship between spending for Celebrex and the delay to therapy adoption does not become negative until Pfizer would spend more than $34.5 million per month. During 1999-2002, Pfizer spent $6.5 million per month on average--and never more than $13.8 million per month. This implies that the marginal dollar spent for Celebrex by Pfizer actually lengthens the delay in initiating any COX-2 treatment by approximately 41 d (Table 4). Thus, the positive effect of joint advertising spending on delay to adoption in the first model is driven completely by Celebrex ads. This curious reverse association between Celebrex television advertising and the use of Celebrex is consistent with other recent research we have conducted that examined the association between television advertising and the rate of prescribing for Vioxx and Celebrex at the physician practice level (Bradford et al. 2006). (l)

Given that we find that television advertising on behalf of the COX-2 inhibitors has some effect on the delay to use, the next question to address is whether this effect is welfare enhancing or not. As discussed above, some patients are better candidates for COX-2 use in that they have exhibited the sorts of gastrointestinal sensitivities for which Vioxx and Celebrex were designed. Other patients have hypertension and coronary disease comorbidities that make them poor candidates for using COX-2 inhibitors because these conditions have been recognized as contraindications for COX-2 use since the publication of the TNM article. As discussed above, we estimate models with interactions between total television advertising (local and national) and the GI indicator variable, the heart disease and pre-TNM indicators, and the heart disease and post-TNM indicators.

Table 3 presents the relevant parameters. We find strong evidence for welfare-enhancing informational effects. The interaction between total COX-2 inhibitor advertising and the indicator variable for gastrointestinal sensitivity is negative and highly significant in both models. In terms of the magnitude of the effect, patients with GI comorbidities reduce their wait time to adopt COX-2 inhibitor treatment by between 2 and 2.4 d for every additional $100,000 in monthly COX-2 inhibitor advertising (Table 4).

The effect of advertising on patients with heart disease comorbidities (cardiovascular disease and hypertension) depends on whether we observe the therapy being initiated prior to the publication of TNM or after. Prior to TNM, patients with cardiovascular comorbidities actually adopted COX-2 inhibitor therapy more rapidly when exposed to increased advertising. Given the information about the potential cardiac dangers associated with COX-2 inhibitor use (especially Vioxx), this is contrary to what would improve social welfare. However, prior to the publication of the TNM paper, this risk was poorly appreciated in the clinical community. However, once the TNM paper was published, we find a strong positive impact on delay to therapy initiation for patients with diagnosed heart disease when exposed to greater levels of COX-2 inhibitor advertising. This is the direction one would expect if the ads provide real information that assists patients and physicians to more optimally match therapies. In fact, the post-TNM interaction effect is larger than the pre-TNM interaction effect by approximately 6-8 days. (2)

Recall that we proposed relatively strong tests of whether DTCA for COX-2 inhibitors leads to welfare-enhancing or-reducing effects. If increasing DTCA is associated with reductions in the time patients who are good candidates for the therapy wait before initiating treatment and is associated with increases in the time patients who are poor candidates wait before initiating treatment then the advertising must be providing useful information to the clinical matching process. In this case, DTCA has at least some welfare-enhancing characteristics. We find exactly this pattern in our interactions between patients with previously treated GI difficulties (good candidates) and with previously diagnosed heart disease (poor candidates). In addition, the expected positive effect from the heart disease only shows up after the first important clinical publication to demonstrate that patients with heart disease are, in fact, at increased risk from COX-2 inhibitor use. Taken together, this evidence strongly supports the neoclassical view of advertising as information-and throws the strong criticism that pharmaceutical DTCA has recently received into question.

Finally, Table 3 presents parameters which test whether patients and clinicians respond to information from the clinical literature. As discussed above, we included measures of gross publication rates in the month preceding initiation of COX-2 use for papers that discuss COX-2 inhibitors (either generically or focusing on a specific brand) in the context of care for patients with OA. While the measures of clinical publication rates are uniformly statistically significant, there is little consistency in the parameter values (except for publications involving only Celebrex--which tended to always increase delay times). We take this as evidence that clinical publications are measurably important for the prescribing patterns of primary care clinicians in community practice; we cannot characterize how the effect is felt. This is likely because some publications are favorable toward the effectiveness (and cost-effectiveness) of COX-2 inhibitors and some are pessimistic. However, our data do not currently characterize the tenor of the publication. More research on how clinical information and DTCA information interact appears warranted.

VI. CONCLUSIONS

The increased use of television advertising by manufacturers of prescription pharmaceutical has been a controversial development over the past 5-10 yr in the United States. While the use of such advertisement has grown dramatically since the early 1990s, to date, there have been few studies which have empirically examined the effect of these ads on patient care. The primary goal of this paper was to determine what effect local and national television advertising on behalf of the two main COX-2 inhibitors had on the treatment decisions that patients made in collaboration with their physicians. In particular, the treatment decision we studied is the time patients choose to wait before initiating treatment with either Vioxx or Celebrex. Using data on 18,235 patients from a set of geographically dispersed community-based primary care practices, we have measured the determinants of the delay between diagnosis for OA and onset of COX-2 inhibitor therapy. To accomplish these goals, we estimated a series of competing risk duration regressions using an AFT model.

Despite the importance of our study, there are limitations. First, without monthly data for pharmaceutical detailing, we were unable to account for the impact and interaction of detailing directly. Such physician-based marketing remains a larger component of pharmaceutical marketing efforts than DTCA advertising. It is possible that DTCA and detailing efforts were coordinated; if so, the DTCA effects measured here might have included some detailing effect. Personal communication with pharmaceutical representatives, however, suggested that since both Vioxx and Celebrex were such important products to the manufacturers, the representatives discussed these products at every opportunity. This implies that there would be little correlation between the levels of detailing and local or national variation in advertising, which in turn would minimize the potential for bias in the results. Second, practices also generally have a supply of pharmaceutical samples on hand to give patients when they write a prescription. The availability of samples may influence which product is prescribed. Again, personal communication with physicians and pharmaceutical representatives suggested that most physicians would typically have had a stock of samples of both on hand, implying that the potential for omitted variables bias is limited. Additionally, the practice-level fixed effects included in the models capture any general tendency to favor one drug over another.

In summary, we find that increases in television advertising for Vioxx is associated with shorter wait times between diagnosis and use. On the other hand, Celebrex television advertising is associated with longer delays to COX-2 adoption. Finally, we also present evidence that the effect of DTCA may tend to improve economic efficiency in that advertising tends to shorten the delays to adoption for patients who are better candidates for COX-2 use and lengthens the delay to adoption for patients who are worse candidates for the use of COX-2 inhibitors.

ABBREVIATIONS

AFT: Accelerated Failure Time

COX-2: Cyclooxygenase-2

DTCA: Direct to Consumer Advertising

FDA: Food and Drug Administration

OA: Osteoarthritis

doi: 10.1111/j.1465-7295.2009.00215.x

REFERENCES

Arnott, R., and F. Lewis. "The Transition of Land to Urban Use." Journal of Political Economy, 87, 1979, 161-69.

Benninga, S., M. Helmantel, and O. Sarig. "The Timing of Initial Public Offerings." Journal of Financial Economics, 75, 2005, 115-32.

Bradford, W. D., A. Kleit, P. Neitert, T. Mcilwain, and S. Ornstein. "The Effect of Direct to Consumer Television Advertising on Physician Prescribing Behavior for the Treatment of Osteoarthritis." Health Affairs, 25(5), 2006, 1371-77.

Calfee, J., C. Winston, and R. Stempski. "Direct-to-Consumer Advertising and the Demand for Cholesterol-Reducing Drugs." Journal of Law and Economics, 45(2, pt 2), 2002, 673-690.

Capozza, D., and Y. Li. "The Intensity and Timing of Investment: The Case of Land." American Economic Review, 84, 1994, 889-904.

Capozza, D., and Y. Li. "Optimal Land Development Decisions." Journal of Urban Economics, 51, 2002, 123-42.

Cutler, D. "The Incidence of Adverse Medical Outcomes Under Prospective Payment." Econometrica, 63, 1995, 29-50.

Donohue, J., M. Cevasco, and M. Rosenthal. "A Decade of Direct to Consumer Advertising of Prescription Drugs." New England Journal of Medicine, 357, 2007, 673-80.

Dubois, R. "Pharmaceutical Promotion: Don't Throw the Baby Out with the Bath Water." Health Affairs Web Exclusive, W3, 2003, 96-103.

Hamilton, B. H., and V. H. Ho. "Estimating Surgical Volume-Outcome Relationships Applying Survival Models: Accounting for Frailty and Hospital Fixed Effects." Health Economics, 6, 1997, 383-95.

Heckman, J., and B. Singer. "A Method for Minimizing the Impact of Distributional Assumptions in Econometric Models for Duration Data." Econometrica, 52, 1984, 271-320.

Hurwitz, M., and R. Caves. "Persuasion or Information? Promotion and the Shares of Brand Name and Generic Pharmaceuticals." Journal of Law and Economics, 31, 1988, 299-320.

Kalbfleisch, J., and R. Prentice. "Estimation of the Average Hazard Ratio." Biometrika, 68, 1981, 105-112.

Keith, A. "Regulating Information About Aspirin and the Prevention of Heart Attack." American Economic Review, 85, 1995, 96-99.

King, C. "Marketing, Product Differentiation and Competition in the Pharmaceutical Industry." Working Paper, MIT Department of Economics, 1996.

Kleit, A. N. "Using Advertising to Generate Information and Signals for Product Quality: Lessons for Biotechnology in Canada from Pharmaceutical Markets in the United States," in Biotechnology and the Consumer, edited by K. A. Mathios. Boston: Kluwer, 1998, 257-75.

Knight Ridder News. "Drug Maker Merck reports 5% Increase in Profits." Tribune Business News, July 21, 2001.

Leffler, K. "Persuasion or Information? The Economics of Prescription Drug Advertising." Journal of Law and Economics, 24, 1981, 45-74.

Masson, A., and P. Rubin. "Matching Prescription Drugs and Consumers: The Benefits of Direct Advertising." New England Journal of Medicine, 313, 1985, 513-15.

McDonald, R., and D. Siegel. "The Value of Waiting to Invest." Quarterly Journal of Economics, 101, 1986, 707-28.

Merton, R. C. "An Intertemporal Capital Asset Pricing Model." Econometrica, 41, 1973, 867-87.

Morrill, R., J. Cromartie, and G. Hart. "Metropolitan, Urban, and Rural Commuting Areas: Toward a Better Descriptions of the United States Settlement System." Urban Geography, 20, 1999, 727-48.

Mukherjee, D., S. Nissen, and E. Topol. "'Risk of Cardiovascular Events Associated with Selective COX-2 Inhibitors." Journal of American Medical Association, 286, 2001, 954-59.

National Institute for Health Care Management. "'Prescription Drugs and Mass Media Advertising, 2000." Washington, DC: NIHCM Foundation, November 2001.

"Pharmacia Has Setback for Parecoxib.'" Chemical Market Reporter, July 2001, 8.

Picone, G., R. Wilson, and S. Chou. "Analysis of Hospital Length of Stay and Discharge Destination Using Hazard Functions with Unmeasured Heterogeneity." Health Economics, 12, 2003, 1021-34.

Ridder, G. "The Non-Parametric Identification of Generalized Accelerated Failure-Time Models." Review of Economic Studies, 57, 1990, 167-81.

Rizzo, J. "'Advertising and Competition in the Ethical Pharmaceutical Industry: The Case of Hypertensive Drugs." Journal of Law and Economics, 42, 1999, 89-116.

Rubin, P. "Economics of Prescription Drug Advertising." Journal of Research in Pharmaceutical Economics, 3(1), 1991, 29-39.

Schumann, M. "Top 100 Megabrands: Chevrolet Leads the Race Along Megabrand Road, but Marketers Hit 2001 Potholes." Advertising Age, 72, 2001, I.

Telser, L. "The Theory of Supply with Application to tile Ethical Pharmaceutical Industry." Journal of Law and Economics, 18, 1975, 449-78.

Titman, S. "Urban Land Prices Under Uncertainty." American Economic Review, 75, 1985, 505-14.

"Top Brands in Network Primetime-2000." Brandweek, April 23, 2001, 30.

Topol, E. "'Failing the Public Health-Rofecoxib, Merck and the FDA." New England Journal of Medicine, 351. 2004, 1707-09.

Zachry, W. M., S. J., Coons, and Ginsburg, D. B. "Patient Autonomy and the Regulation of Direct-to-Consumer Advertising." Clinical Therapy, 23, 2001, 2024 37.

(1.) We also estimated versions of the duration of delay for Vioxx and Celebrex separately, including spending on both brands in each equation. In those models, we found that the general effects demonstrated in the second column of Table 3 hold: Vioxx spending generally has a net effect of reducing the delay to therapy adoption for both Vioxx and Celebrex, while Celebrex has a net effect of lengthening the delay for both brands. Those results are available on request.

(2.) Note that while the absolute magnitudes of the pre-and post-TNM interaction effects are quite different between the two models, the net effects are quite similar.

W. DAVID BRADFORD, ANDREW N. KLEIT, PAUL J. NIETERT and STEVEN ORNSTEIN *

* The authors would like to thank Jack Calfee, John Cawley, and Terry Steyer for their comments and assistance on this paper as well as participants at the Sloan Program Conference on Consumers, Information and the Evolving Healthcare Market Place, the 2005 Annual Health Economics Conference (State College, Pennsylvania), and the 2005 International Health Economics Biannual Congress (Barcelona). This paper was funded by grants from the Agency for Healthcare Research and Quality (1 R01 HS011326-01A2) and from the National Heart, Lung and Blood Institute (1 R01 HL077841-01).

Bradford: Busbee Chair in Public Policy, 201C Baldwin Hall, Department of Public Administration and Policy, University of Georgia, Athens, GA 30602. Phone 706-542-2731, Fax 706-583-0610, E-Mail bradfowd@uga.edu

Kleit: Professor of Energy and Environmental Economics, MICASU Faculty Fellow, Department of Meteorology, Department of Energy and Mineral Engineering, Pennsylvania State University, 507 Walker Building, University Park, PA 16802-5013

Nietert: Associate Professor, Department of Medicine, Medical University of South Carolina, 135 Cannon St., Room 403J, Charleston, SC 29425. Phone (843) 876-1204, E-mail nieterpj@musc.edu

Ornstein: Professor, Department of Family Medicine, Medical University of South Carolina, Charleston, SC 29425. Phone (843) 876-1213, E-mail ornstesm@musc.edu
TABLE 1
Data Description (N = 18, 235)

 Standard
Variable Mean Deviation

Average delay between diagnosis with 178.422 304.496
 OA and first use of COX-2 inhibitor
 (for 6,494 uncensored observations)
Total dollars in COX-2 advertising 99.767 40.053
 squared, month preceding therapy
 (in $100,000s)
Total dollars in COX-2 advertising 11,557.730 6,317.694
 squared, month preceding therapy
 (in $100,000s)
Total dollars in Vioxx advertising, 35.009 17.176
 month preceding therapy
 (in $100,000s)
Total dollars in Vioxx advertising 1,520.648 1,413.734
 squared, month preceding therapy
 (in $100,000s)
Total dollars in Celebrex advertising, 64.758 31.272
 month preceding therapy
 (in $100,000s)
Total dollars in Celebrex advertising 5,171.554 3,203.501
 squared, month preceding therapy
 (in $100,000x)
Patient has received gastrointestinal 0.381 0.486
 treatment
Patient has been diagnosed with heart 0.696 0.460
 disease
Interaction between gastrointestinal 35.715 53.269
 treatment and total COX-2 inhibitor
 advertising dollars
Interaction between heart disease and 56.575 57.886
 total COX-2 inhibitor advertising
 dollars during pre-TNM period
Interaction between heart disease and 53.578 58.177
 total COX-2 inhibitor and advertising
 dollars during post-TNM period
Number of journal publications 0.344 0.996
 discussing COX-2 inhibitors in
 month preceding therapy
Number of journal publications 0.977 0.760
 discussing Vioxx in month preceding
 therapy
Number of journal publications 2.223 1.175
 discussing Celebrex in month
 preceding therapy
Therapy initiated in 1999 0.071 0.257
Therapy initiated in 2000 0.512 0.500
Therapy initiated in 2001 0.061 0.240
Patient age at therapy initiation 65.531 13.864
Patient is female 0.678 0.467
Patient has been diagnosed with 0.181 0.385
 depression
Patient has been diagnosed with 0.179 0.383
 diabetes
Patient has been diagnosed with 0.354 0.478
 hyperlipidemia
County population 523,489 1,358,298
Number of physicians per 1,000 147.098 52.533
 population in county
County per capita income 2,5845.46 5,223.66
Number of Medicare enrollees in county 2.277 2.279
Percent of county residents who are 47.738 6.685
 employed
Percent of county population that is 80.914 8.026
 Caucasian
Percent of county population that is 11.896 7.963
 African American
Average price for intermediate length 56.574 7.255
 physician visit

TABLE 2
Delay between OA Diagnosis and First COX-2 Use (Number of Patients)

 Average Delay
 for Patients Average Delay
 Who Do Not for Patients
 Adopt Vioxx Who Adopt Vioxx

Average delay for patients 776.9 (11,741) 199.1 (2,760)
 who do not adopt Celebrex
Average delay for patients 162.7 (3,734) --
 who adopt Celebrex
Total patients 18,235

TABLE 3
AFT Duration Models of Time between Diagnosis and Treatment for OA
Patients Coefficients of Model

 Model 1 Combined
 Vioxx and Celebrex
Variable Total Advertising

Total dollars in COX-2 advertising, month 0.0730 (17.300)
 preceding therapy (in $100,000s)
Total dollars in COX-2 advertising -0.0002 (-11.91)
 squared, month preceding therapy
 (in $100,000s)
Total dollars in Vioxx advertising, month
 preceding therapy (in $100,000s)
Total dollars in Vioxx advertising
 squared, month preceding therapy
 ($100,000s)
Total dollars in Celebrex advertising,
 month preceding therapy (in $100,000s)
Total dollars in Celebrex advertising
 squared, month preceding therapy
 (in $100,000s)
Patient has received gastrointestinal -0.0182 (-0.16)
 treatment
Patient has been diagnosed with heart -0.3110 (-2.51)
 disease
Interaction between gastrointestinal -0.0029 (-2.17)
 treatment and total COX-2
inhibitor advertising dollars
Interaction between heart disease and -0.0451 (-22.87)
 total COX-2 inhibitor advertising
 dollars during pre-TNM period
Interaction between heart disease and 0.0554 (32.82)
 total COX-2 inhibitors advertising
 dollars during post-TNM period
Number of journal publications discussing 0.1709 (5.39)
 COX-2 inhibitors in month preceding
 therapy
Number of journal publications discussing -0.0542 (-1.77)
 Vioxx in month preceding therapy
Number of journal publications discussing 1.2403 (26.41)
 Celebrex in month preceding therapy
Therapy initiated in 1999 1.9057 (5.42)
Therapy initiated in 2000 2.5651 (8.18)
Therapy initiated in 2001 -2.9144 (-9.41)
Patient age at therapy initiation 0.0157 (5.87)
Patient is female -0.0118 (-0.16)
Patient has been diagnosed with depression -0.0762 (-0.90)
Patient has been diagnosed with diabetes 0.1694 (1.85)
Patient has been diagnosed with 0.1950 (2.51)
 hyperlipidemia
County population (in 1,000s) -0.00003 (-5.99)
Number of physicians per 1,000 population 0.0021 (1.18)
 in county
County per capita income 0.00003 (1.52)
Number of Medicare enrollees in county -0.1345 (-1.39)
Percent of county residents who are -0.0871 (-5.30)
 employed
Percent of county population that is -0.0159 (-1.30)
 Caucasian
Percent of county population that is -0.2625 (-4.44)
 African American
Average price for intermediate length 0.0686 (6.35)
 physician visit
Constant 9.8063 (4.40)
Likelihood ratio test of overall 11, 957.97 (<0.0001)
 significance p value

 Model 2 Separated
 Vioxx and Celebrex
Variable Total Advertising

Total dollars in COX-2 advertising, month
 preceding therapy (in $100,000s)
Total dollars in COX-2 advertising
 squared, month preceding therapy
 (in $100,000s)
Total dollars in Vioxx advertising, month 0.0482 (4.17)
 preceding therapy (in $100,000s)
Total dollars in Vioxx advertising -0.0010 (-9.12)
 squared, month preceding therapy
 ($100,000s)
Total dollars in Celebrex advertising, 0.0687 (14.79)
 month preceding therapy (in $100,000s)
Total dollars in Celebrex advertising -0.0001 (-2.39)
 squared, month preceding therapy
 (in $100,000s)
Patient has received gastrointestinal 0.0970 (0.87)
 treatment
Patient has been diagnosed with heart -0.3271 (-2.63)
 disease
Interaction between gastrointestinal -0.0037 (-2.69)
 treatment and total COX-2
inhibitor advertising dollars
Interaction between heart disease and -0.0044 (-2.16)
 total COX-2 inhibitor advertising
 dollars during pre-TNM period
Interaction between heart disease and 0.0132 (7.03)
 total COX-2 inhibitors advertising
 dollars during post-TNM period
Number of journal publications discussing -0.1664 (-4.79)
 COX-2 inhibitors in month preceding
 therapy
Number of journal publications discussing 0.1468 (4.31)
 Vioxx in month preceding therapy
Number of journal publications discussing 0.9796 (22.66)
 Celebrex in month preceding therapy
Therapy initiated in 1999 -0.3832 (-1.03)
Therapy initiated in 2000 1.6536 (5.44)
Therapy initiated in 2001 -2.4908 (-8.10)
Patient age at therapy initiation 0.0163 (6.23)
Patient is female -0.0283 (-0.39)
Patient has been diagnosed with depression -0.0919 (-1.11)
Patient has been diagnosed with diabetes 0.1244 (1.38)
Patient has been diagnosed with 0.2036 (2.66)
 hyperlipidemia
County population (in 1,000s) -0.00003 (-6.15)
Number of physicians per 1,000 population 0.0034 (1.95)
 in county
County per capita income 0.00001 (0.48)
Number of Medicare enrollees in county -0.3372 (-3.47)
Percent of county residents who are -0.0838 (-5.30)
 employed
Percent of county population that is -0.0(08 (-0.93)
 Caucasian
Percent of county population that is -0.2604 (-4.53)
 African American
Average price for intermediate length 0.0517 (4.85)
 physician visit
Constant 12.7530 (5.89)
Likelihood ratio test of overall 13, 164.55 (<0.0001)
 significance p value

Note: t statistics are given in parentheses.

TABLE 4
AFT Duration Models of Time between Diagnosis and Treatment for OA
Patients. Net Marginal Effects: Change in Days of Delay to Treatment
for a 1-Unit Change in the Explanatory Variable

 Model 1 Model 2
 Combined Separated
 Vioxx and Vioxx and
 Celerex Celebrex
 Total Total
Variable Advertising Advertising

Total dollars in COX-2 advertising, 30.70 --
 month preceding therapy (in $100,000s)
Total dollars in Vioxx advertising, -- -13.7
 month preceding therapy (in $100,000s)
Total dollars in Celebrex advertising, -- 40.80
 month preceding therapy (in $100,000s)
Interaction between gastrointestinal -2 -2.4
 treatment and total COX-2 inhibitor
 advertising dollars
Interaction between heart disease and -31 -2.8
 total COX-2 inhibitor advertising
 dollars during pre-TNM period
Interaction between heart disease and 38.10 8.60
 total COX-2 inhibitor advertising
 dollars during post-TNM period

TABLE 5
Rural/Urban Distribution of PPRNet Practices
and Patients

 PPRNet PPRNet U.S.
 Practices Patients Population
 (%) (%) (%)

Urban core areas 64 66 71
Small town/rural area 17 15 10
Suburban area 10 12 9
Large town area 9 7 10
COPYRIGHT 2010 Western Economic Association International
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2010 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Bradford, W. David; Kleit, Andrew N.; Nietert, Paul J.; Ornstein, Steven
Publication:Economic Inquiry
Article Type:Report
Date:Apr 1, 2010
Words:10384
Previous Article:Centralized or decentralized information: which is better for providing incentives?
Next Article:Do sunk costs matter?
Topics:

Terms of use | Privacy policy | Copyright © 2020 Farlex, Inc. | Feedback | For webmasters