Docstoc

The Impact of Direct-to-Consumer Advertising of Cholesterol

Document Sample
The Impact of Direct-to-Consumer Advertising of Cholesterol Powered By Docstoc
					The Impact of Direct-to-Consumer Advertising of Cholesterol Reducing Drugs on Diagnosis and Treatment of Cholesterol

Hae Kyung Yang* Cornell University Alan Mathios Cornell University Rosemary Avery Cornell University

June 2007

*

Contact author. Department of Policy Analysis and Management, Cornell University, 426 MVR Hall, Ithaca, NY 14853. Tel: (607) 342 7636, Fax: (607) 255 4071. Email: hy75@cornell.edu. This research was funded by an unrestricted educational grant from the Merck Company Foundation.

1. Introduction Cardiovascular disease (CVD) is the leading cause of death in the United States, accounting for about one half of all deaths. About 61 million Americans - almost onefourth of the population - have some form of CVD.1 High blood cholesterol is one of the major risk factors of CVD. For example, clinical studies show that a 1 percent increase in „bad‟ cholesterol level is associated with a 2 percent increase in the risk of CVD (NIH, 2001). More than 50 percent of U.S. adults have elevated blood cholesterol (NIH, 2001). The link between blood cholesterol levels and heart disease was the driving force in the development of new pharmaceuticals that lower LDL cholesterol levels („bad‟ cholesterol) and raise HDL cholesterol levels (these are associated with lower risk of heart disease). Clinical trials of cholesterol reducing pharmacological treatment support the efficacy of drug therapy in preventing CVD in higher risk persons (NIH, 2001). A class of drugs known as statins dominates the cholesterol reducing market (Saftlas, 2005; Sprang et al, 2005). Clinical trials have reported that statins reduce the level of LDL cholesterol by up to 60 percent leading to mortality rate from heart attack or stroke falling by a third.2 Since its introduction in 1987, statins now have annual sales of over $21 billion globally (92.5 percent of the total cholesterol reducing market) and is still growing (Sprang et al, 2005). The market for cholesterol reducing drug is expected to grow with increasing adherence to the updated Adult Treatment Panel III (ATP III) guidelines set by the NIH. ATP III lowered the LDL threshold (from 130 to 100) for triggering a recommendation for drug therapy. This reduction brings an additional 8 million adults into the population for which drug therapy is recommended (NIH, 2001). However,

1 2

CDC report is available at http://www.cdc.gov/nccdphp/bb_heartdisease/index.htm Earlier Clinical Trials such as WHO clofinrate trial or Helsinki Heart Study genfibrozil trial show lipid lowering therapy reduces major coronary events, but do not address the issue of total mortality. Later statin trials WOSCOPS, AFCAPS/TexCaps show that statin significantly reduces the relative risks of major coronary events and WOSCOPS shows a strong trend towards a reduction in total mortality. More recent major Clinical Trials 4S, CARE, and LIPID indicate that lipid lowering therapy especially statins reduces the risks of stroke. Table II.2-1 in ATP III shows CVD outcomes in Clinical Trials of LDL-CholesterolLowering Therapy.

2

studies show that only about one-fifth of all persons who can benefit from cholesterol reducing drugs are taking them (Arnold, 2005). One reason so many consumers who could benefit from the drug are not taking the drug is the difficulty of observing elevated levels of high blood cholesterol. Unless tested for cholesterol, it is years before elevated levels may result in any symptoms. As a result the dissemination of information about the risks of elevated blood cholesterol plays a key role in prevention. There are numerous sources of information from which consumers can learn about the link between cholesterol and CVD and effective prevention strategies. This study focuses on the role of pharmaceutical drug advertising on consumer awareness of their blood cholesterol level and on the demand for drug therapy. Statin drugs are currently only available through prescription, thereby requiring the patient to visit physician‟s office. The ATP III guidelines recommend that before prescribing drugs, physicians should prescribe therapeutic lifestyle changes (TLC). TLC includes improvements in diet and increases in exercise. This suggests that consumers who learn about statin drugs through advertising and see their physician may receive advice about diet and exercise. The second focus of this paper is to examine the empirical relationship between exposure to statin drug advertising and participation in exercise and attempts to change diet. One key unique contribution of this study is the identification strategy used to identify the impact of advertising on awareness of cholesterol, use of statins, diet and exercise. We link survey data on individual knowledge of whether they have high cholesterol, whether they use statin drugs, whether they are attempting to control their diet, and the extent of their exercise behavior with an archive of magazine advertisements for statin drugs. The survey also includes information on which magazines each respondent reads. Thus, we are able to measure the individual exposure to statin drug advertisements based on their individual magazine-reading habits. Because we observe the same information about the consumers that the advertisers observe, we can control for 3

the potential endogeneity of advertising due to firms‟ targeting decisions. Our results demonstrate the important role that controlling for targeting plays in affecting the estimated impacts of advertising. The data employed are extremely rich. The individual-level data comes from a marketing survey, the Simmons National Consumer Survey (NCS), which contain magazine-reading and television-viewing habits. Multiple surveys conducted between the fall of 1995 and the fall of 2004 are pooled to obtain a sample of more than two hundred thousand respondents. Advertising data are obtained from an archive compiled by the authors. This is a collection of all print advertisements for statin drug pharmaceutical products that appeared between 1985 and 2004 in 26 consumer magazines that represent magazines read by individuals with particular demographic characteristics. By combining the advertising archive data with information on magazine reading habits in NCS, we obtain a much richer advertising exposure measure than those used in previous research. Previous studies use market-level data on advertising, which requires the implicit assumption that all individuals in a given market are exposed to the same advertising. Another attractive feature of the research design in this paper is that we can make a case that the advertising exposure measure will not be correlated with unobserved determinants of our dependant variables. Our results provide evidence that, if we do not control for targeting (like other studies that are unable to) DTCA has a statistically significant and positive impact on health behaviors – diagnosis, statins purchase, diet control and regular exercise. However, these empirical results are extremely sensitive to controls for targeting, so much so, that when accounting for which magazines individuals read (and thereby using within magazine variation in advertising exposure to identify our impacts), exposure to advertising no longer has an impact of our behavioral measures. The remainder of this essay is organized as follows. Section 2 provides an overview of the development of the market for pharmaceutical treatment of high 4

cholesterol and reviews the literature. Section 3 discusses a simple theoretical framework for how advertising may influence the demand for statin drugs. Section 4 introduces the data, followed by section 5 which discusses the econometrical specification and our identification strategy. Section 6 and 7 presents the results and discusses implications respectively.

2. Background

2.1 The Development of the Market for Pharmaceutical Treatment of High Cholesterol

Cholesterol, produced naturally in the body and absorbed from food, is a key component of cell membranes and hormones (NIH, 2001). Cholesterol is transported around the body through a complex called lipoprotein, and each type of lipoprotein plays a different role. Among the key lipoproteins, low density lipoprotein (LDL) is cholesterol rich and is also known as „bad‟ cholesterol. Over time, it contributes to blockage of the blood flow through coronary arteries. Lack of blood flow can lead to chest pain and heart attack (NIH, 2001). Over the years there have been a number of recommendations about maintaining desirable levels of blood cholesterol with specific recommendations focused on limiting LDL cholesterol levels. There are a variety of ways to treat high cholesterol levels including lifestyle changes involving diet (i.e. limitations on intakes of saturated fats) and increases in physical activity. Beginning in the late 1980s, the Food and Drug Administration (FDA) approved the first of the „statin‟ drugs which in clinical trials resulted in marked reductions in LDL cholesterol levels for those treated with the drug. Soon after the first introduction in 1987, a number of other drugs within this class were approved for sale, also as prescription only. The dramatic effectiveness of these drugs soon became a standard way of treating consumers with high levels of blood cholesterol. Table 1 documents the 5

development of the market for statins. The table reports the rapid increase in the number of statin products available on the market and that for the entire period FDA required consumers to obtain a prescription in order to purchase the products. Table 1 also demonstrates that there are other types of cholesterol reducing drugs, some of which are sold over-the-counter (OTC).3 However, these products do not exhibit the same effectiveness and tend to be a small part of the market. For example, in 2004, worldwide statin sales accounted for over $21 billion of the $22.7 billion of sales for all cholesterolreducing drugs (Sprang et al, 2005). Tables 1 and 2 show the types of statins in the market and the time they received FDA approval. Currently there are 5 different types of statins in the U.S.; lovastatin, simvastatin, pravastatin, fluvastatin, and atrovastatin (cerivastatin was withdrawn from the market by the manufacturer in August 2001). Only prescription versions of statins are available in the U.S. However, manufacturers facing patent expiry have applied to the FDA to switch their statin products into OTC.4 So far, three out of five statins have lost patent protection. Mevacor (lovastatin)‟s patent expired in June 2001 while Pravachol (pravastatin) and Zocor (simvastatin) lost their patent protection in April and June 2006 respectively. Since their patent expiry, firms are facing competition from generic manufacturers. While generic versions of statins are available in the market, their market share is still small, though expected to increase significantly over time. For the time period that we consider in our empirical model generic versions are not therefore relevant.

2.2 Direct-to-Consumer Advertising as a Source of Information to Consumers

3 4

OTC niacin, another cholesterol reducing drug, is available as vitamin B-3 complex. By switching the drug to OTC status, manufacturers expect to revive declining sales resulting from competition with generics. Manufacturers get exclusive right to sell their OTC versions for three years.

6

Advertising of prescription drugs falls under the regulatory authority of the FDA. During the last two decades there have been significant regulatory decisions – in 1983, 1985, and 1997 – that have affected the advertising of products that are sold by prescription only. Prior to 1985 there was virtually no Direct-to-Consumer Advertising (DTCA) of prescription drugs. From 1985 until around 1997, the FDA permitted this advertising but subject to significant disclosure requirements. In 1997 the FDA substantially reduced the burden of disclosure for broadcast DTCAs. Many observers believe the 1997 FDA regulatory change led to a sharp increase in DTC broadcast advertisements including statin drug advertising. The use and increase of DTCA led researchers to investigate the impact of this advertising on health outcomes. For this paper we focus on the studies that have focused on DTCA of statin drugs. There have been a number of studies that have focused on the impact of DTCA of statin drugs on health outcomes. Most studies examine the impact of advertising on demand for prescriptions.5 For example, Calfee et al (2002) finds that despite the significant increase in advertising expenditures over time, there is no statistically significant effect on prescriptions. They do find, however, that television advertising of prescription statin drugs is associated with an increase in the proportion of cholesterol patients who had been successfully treated. Wosinska (2005) examines the impact of advertising on compliance and finds evidence that one of the benefits of DTCA is higher compliance among patients under statin drug therapy. In another paper, Wosinska (2002) finds that advertising increases the use of statins if the advertised drug appears on the insurance formulary list. One finding that has appeared in a number of studies is that there is little direct effect of advertising on specific brand choices but that advertising may affect physician

5

For DTCA studies other than statins, see Brekke and Kuhn (2005) for theoretical framework, and for empirical literature see Avery et al, 2007; Donohue and Berndt, 2004; Iizuka and Jin, 2005, 2007; Ling et al, 2002; Mukherji et al, 2004; Rizzo, 1999; Rosenthal et al, 2003.

7

visits. For example, Iizuka and Jin (2005) combine data on DTCA from 1994-2000 with the 1995-2000 National Ambulatory Medical Care Survey (NAMCS) and examine the effect of DTCA on doctor visits. They find that higher DTCA expenditures are associated with increased doctor visits, especially after the FDA relaxed its regulations governing broadcast risk disclosure. Berndt (2005) summarizes the empirical literature in the area and concludes that empirical tests consistently suggest that DTCA induces consumers to visit physicians and increase the size of the market, but prescription drug choice is still primarily determined by the physicians (Berndt, 2005). Berndt adds that DTCA encourages not only appropriate but also inappropriate treatments. One of the weaknesses of these previous studies of DTCA is that they have not been able to measure individual level variation in exposure to statin drug advertising and instead have largely relied on aggregate measures of expenditures either measured at the year level, or at more aggregate levels. Some studies link these aggregate measures of DTCA expenditures with aggregate measures of drug use. Improvements and refinements of these studies focused on obtaining individual level data on the use of statins but these were still combined with aggregate measures of advertising expenditures. We believe that our study is the first to be able to combine micro-level data on both exposure and use.

3. Data

To examine the relationships between advertising, knowledge of cholesterol levels, use of statin drug therapy and lifestyle changes requires data on each of these components. Prior to describing our econometric specifications we focus on the various sources of data we utilize to examine the relationships described in the conceptual framework in previous section. The Simmons National Consumer Survey 8

Our individual-level data comes from a marketing survey, the Simmons National Consumer Survey (NCS). The NCS provides detailed information on consumer behavior and magazine-reading and television-viewing habits (NCS 2005). In addition to magazine reading behavior, the data also includes information on whether the respondent reports that they have high cholesterol, whether they use a statin drug, which statin drug they use, whether they are attempting to control their diet and whether they exercise regularly. These variables are the key measures that we need to examine the empirical relationships described in the conceptual framework. As will be described later, the magazine reading questions allow us to create a measure of advertising exposure to statin drugs which can then be linked with the other key variables. The NCS is a repeated cross-sectional survey. The sample for each wave is independently drawn. The NCS employs a multi-stage stratified probability sample. The final sample represents a representative probability sample of all adults living in households in the U.S. (excluding Hawaii and Alaska). In order to minimize respondent fatigue, the data are collected in several phases. In Phase 1 respondents are interviewed (face-to-face). In the first part of Phase 1, interviewers collect demographic data and data on the magazines respondents read. To collect this information, interviewers present respondents with a deck of cards on each of which are printed the logo of one of the 182 magazines in the Simmons sample. If a respondent reports that he has read any portion of a magazine he is asked a set of questions about readership, including whether he read the whole magazine. During the second part of Phase 1 respondents report, by filling out a questionnaire, whether they purchase and use specific products, including specific statin drugs. Survey response rates in the NCS are generally high (approximately 70 percent) and compare well with other widely-used surveys in health economics and health services research. For example: in the 2002 Behavioral Risk Factor Surveillance Survey the 9

median response rate across states was 58 percent (range: 42 percent to 83 percent); for the 2002 National Health Interview Survey (NHIS) the total household response rate was approximately 90 percent and the final response rate for the Adult Sample Person was 74 percent (CDC 2003, 2004). We also examine how the number of statin drug users compares with other national health data sets available. NCS has a comparable amount of statins users compared with other national health data sets available.6 2 percent of total sample in the 1997 NAMCS7 was using statins compared to 3 percent in 2000. 4 percent of total sample used statins in 1997 Medical Expenditure Panel Survey (MEPS)8 increasing to 8 percent in 2000. In NCS, 5 percent of total sample use statins in 1997 increasing to 6 percent in 2000 and 9 percent in 2002. The Pharmaceutical Advertising Data Archive (PHADS) The advertising data are drawn from an archive of all pharmaceutical print advertisements, hereafter PHADS, appearing in the most widely read magazines. The magazine set includes: Better Homes & Gardens, Black Enterprise, Business Week, Cosmopolitan, Ebony, Essence, Family Circle, Glamour, Good Housekeeping, Jet, McCall's, Modern Maturity, Money, National Geographic, Newsweek, People, Playboy, Readers Digest, Rolling Stone, Seventeen, Sports Illustrated, Time, TV Guide, U.S. News &World Report, Vogue, and Women's Day. We selected these magazines to represent magazines most frequently read by individuals with particular demographic characteristics. We selected these magazines to represent magazines most frequently read by

6

NCS sample‟s socio-economic characteristics are very similar to that of NHIS. For comprehensive comparison, see Plassman et al 2005. 7 NAMCS is national survey for the visits to physician office collecting information the provision and use of ambulatory medical care services conducted annually. For more details on NAMCS, see http://www.cdc.gov/nchs/about/major/ahcd/ahcd1.htm 8 MEPS collects information about cost and use of health care and health insurance coverage from individual patients, medical providers and employers. For more details on MEPS see http://www.meps.ahrq.gov/mepsweb/

10

individuals with particular demographic characteristics. We defined 22 groups by race (3 groups), education (5 groups), income (5 groups), age (5 groups), gender (2 groups), and smoking status (2 groups). Using data on magazine reading from the 1998 NCS we chose the ten magazines most frequently read by members of each group. Although we defined 22 groups, members of each group often read the same magazines. Consequently, instead of 220 magazines, the final set includes the above 26 magazines. We were unable to use three magazines because we could not locate all issues in our sample period. In those three cases we substituted the next most widely read magazine for the group in question. Using magazine circulation data from three independent sources, we estimate that the 26 magazines in PHADS account for between 30.0 and 57.5 percent of magazines circulating in the US. The higher figure is estimated from readership data for the 172 magazines included in the NCS. The lower figure is based on total circulation across 580 magazines reported in Audit Bureau of Circulation‟s (2003) Magazine Trend Report. The PHADS magazines represent 5% of the 580 magazines in the Magazine Trend Report data but account for 30.0 percent of circulation. The Frequency of Statin Advertisements in the Print Archive Table 3 shows statin advertisements by magazine. We do not observe any statin ads in teenagers‟, young women‟s or adult only magazines such as, Cosmopolitan, Glamour, Seventeen, Vogue, Playboy, and Rollingstone. The most frequently advertised magazine for statin is Time, followed by Newsweek, US News and World Report. The next highest is Reader’s Digest. It is also interesting to note the increase in the number of print advertisements over time. For most magazines there has been a rise in the number of print advertisements for statin drugs especially in the last couple of years of the data. This increase has occurred despite the relative cost of advertising on television relative to print decreasing in 1997. Prior to 1997 the FDA required pharmaceutical companies to include the full fine print disclosure in broadcast advertising if such an advertisement included a 11

discussion of what the pharmaceutical product did. After 1997, the FDA relaxed this regulation by requiring a shorter disclosure that focused on the main side effects described in lay terms. Thus, despite television advertising becoming relatively more attractive, print advertising continues to be an active form of advertising. It is worthy of note that one way that firms can meet the disclosure requirements in television advertising is to refer consumers to a print advertisement, which still must include the full fine print disclosure. Thus, the regulations may contribute to print and television advertising being complements rather than substitutes. Measure of adverting exposure We base our measure of advertising exposure on the techniques suggested by Avery et al (2007). First, we merge all of the PHADS advertising data to each person in the NCS sample. In the NCS, each respondent is shown copies of the covers of over 200 consumer magazines. For each magazine, the respondent was asked whether he has read or looked at it during the last 6 months, and if so, of the latest 4 issues on average, how many did the respondent see. The variable Readim is the fraction of issues of magazine m read by person i. For each magazine in the archive the NCS respondent reported having read, we multiply the fraction of the four issues he reads time the number of advertisements that appeared in that magazine over the previous twelve months. Note that, in doing so, we assume that an individual‟s reading habits over the four issues reflects his reading habits over the past year. We then sum up all the magazines in the archive. The result is our estimate of advertising to which a person was potentially exposed to by reading the magazines in the archive. For example, the exposure to statin drug advertising of respondent i (AdExpi) is given by:
AdExp i   Adsim * Re ad im
m 1 26

(11)

Where subscript m refers to each of the 26 magazines in the archive. This measure of advertising exposure is individual specific. Our measurement approach assumes that two people who read the same number of issues of the same magazines were exposed to the 12

same number of advertisements. This is imperfect because we do not know if both people actually saw the advertisements, but it represents a vast improvement over previous research. Previous studies use market-level data on advertising, which requires the implicit assumption that all individuals in a given market are exposed to the same advertising. Measures of Knowledge of Cholesterol, Use of Statin Drugs, Exercise and Diet We utilize the NCS to generate our primary dependent variables. The goal of the study is to examine the impact of advertising exposure on awareness of high cholesterol and various treatment therapies including pharmaceuticals and lifestyle changes (diet and exercise). To measure a respondent‟s awareness of whether they have high blood cholesterol we utilize the following question. Respondents are asked what ailments they have had in the last 12 months. Respondents can report as many ailments as they choose and one of the choices is High Cholesterol. All respondents who report having this ailment are coded as being aware that they have high cholesterol. The data indicate that 9.46 percent of the survey respondents report having high cholesterol. The NCS data also lists a large number of other ailments and it would be possible to examine the impact of exposure to statin drug advertising on diagnosis of other ailments. Because a blood test is required to assess cholesterol, advertising may be effective at increasing the diagnosis of other ailments that blood tests identify (like diabetes). Moreover, because purchasing statins require a prescription, exposure to advertising may lead to consumers becoming aware of other ailments detected through a physician visit. We will examine these relationships in future work. The NCS also asks each respondent in the sample whether they have used a prescription product in the last 12 months to treat a particular ailment. If a respondent reports using a prescription pharmaceutical product to treat high cholesterol they are coded as using a statin drug. Table 4 presents the number of respondents in each wave of the NCS that uses a statin drug. The data demonstrate the rise in the use of statin drugs 13

over time. In 1995 less than 4 percent of the respondents use the cholesterol reducing drug whereas in 2004 more than 8 percent of respondents were using the drug. A significant portion of respondents, 35 percent, who report that they have high cholesterol are not using a prescription pharmaceutical product to treat this ailment. This could be due to a number of reasons including non-compliance with drug therapy, condition not being severe enough to warrant use of a prescription product, the respondent knows that they have high cholesterol and have not seen a physician (for example, through an application for life insurance), or that the respondent chooses to treat cholesterol through other therapies including dietary changes and exercise. Our measures of whether a respondent is engaged in exercise or is trying to control their diet are derived from the answers to the following NCS questions. For exercise, each respondent is asked if they exercise regularly in last 12 months. With respect to diet, each respondent is asked whether they are controlling their diet. Table 5 shows approximately 43 percent of the sample exercise regularly and approximately 35 percent of the sample are controlling there diet. Other Control Variables We also include in our models standard socio-economic control variables (age, sex, race, education, marital status, family size, presence of children in the household, employment status, family income) as well as controls for overall magazine-reading habits, television-viewing habits, and radio-listening habits. In table 5, over the previous 12 months, the average person in our sample read 6 issues of one or more of the magazines consistently included in the NCS. Note that only 4.9 percent of respondents reported that they read no magazines. The average person in the sample watched 19 hours of television per week and listened to the radio about 2 hours per week. 16 percent of the sample watched no television and 35 percent did not listen to the radio in an average week.

14

4. Empirical Specifications and Identification

As defined above, we use various behavioral outcomes (knowledge of high cholesterol, use of statin drugs, diet and exercise) in the past year as our key dependent variables. We adopt the usual approach by assuming that the discrete outcomes we observe are related to underlying latent variables describing the net benefits of engaging in these behaviors. We assume the underlying latent variables are linear functions of exposure to advertising and characteristics of the person and his household. We specify the general relationship as:
Yi *   0   1 AdExp i   2 Z i   i

Where AdExpi refers to exposure to statin drug advertising, Zi is a vector of demographic control variables and variables that account for readership behavior, and  i is an error term. The Yi* is the latent index and we estimate separate equations for each of the behavioral measures: knowledge of whether they have high cholesterol, use of statin drug, whether they are controlling their diet and whether they exercise regularly. We observe the given behavior (denoted by Yi), only if the continuous latent variable exceeds a critical threshold. Under the standard formulation, Yi = 1 if Yi* >0 and Yi = 0 otherwise. The latent variable Y* describes the net benefits of seeing a physician and thereby finding out about whether they have high cholesterol, need a statin drug etc. We estimate equation (11) by maximum likelihood probit and OLS. We report OLS results since there is little difference in the results and it is easier to interpret the coefficients. Our econometric approach uses variation in exposure to advertisements due to magazine-reading habits to identify the causal effect of advertising on the various behavioral measures we examine. Because the NCS data allow us to include a rich set of control variables in our models, we believe that the identifying variation in advertising

15

exposure is econometrically exogenous, i.e., AdExpi is uncorrelated with the error term in equation. In addition to including standard controls for individual socio-economic characteristics, we take several steps to further strengthen our identification strategy. First, our models include controls for general magazine-reading habits and exposure to other media. In this way, we control for the possibilities, say, that heavy magazinereaders or television-watchers somehow differ in their unobserved propensity to see physicians, exercise regularly, use statins or control their diet. Second, and probably more important, our models include a set of dummies reflecting whether an individual read a particular type or category of magazine. We determine these categories by conducting a factor analysis of readership characteristics of different magazines. The factor loadings suggest five particular groups of magazines and we put in variables to control for these groups. We have labeled these groupings as Home: (Better Homes and Gardens, Family Circle, Good Housekeeping, McCalls/Rosie, Reader’s Digest, TV Guide, and Woman’s Day); Young Adults (Cosmopolitan, Glamour, Playboy, Rolling Stone, Seventeen, Vogue); African-American (Black Enterprise, Ebony, Essence, Jet); Business (Business Week, Money); Retirees: (Modern Maturity); and General Interest (National Geographic, Newsweek, People, Sports Illustrated, Time, U.S. News and World Report). With these controls, our identification is based on within-category variation in exposure to advertisements in the archived magazines. For example our strategy will identify the impacts based on respondents being exposed to a different number of advertisements because they read Time versus Newsweek. Third, we include a dummy variable for every magazine (whether the respondent reads that magazine) in the NCS. With this included, our specification now captures the remaining unobserved factors that are possibly related with firms‟ targeting or individual‟s magazine readership. This previously could not have been captured through the use only of demographic characteristics and magazines‟ categories. Here identification is based on respondents being exposed to different amounts of advertising because they are reading the same 16

magazine at different times. nometric specification approximates the model firms use to target their advertisements. Our identification strategy should also minimize any bias stemming from the fact that data limitations force us to omit from our empirical models statin drug prices and the price of a physician visit. Because firms simultaneously choose advertising levels and product prices, in some empirical contexts it would be important to control for the product prices consumers face. However, we exploit individual level variation in advertising exposure that arises because individuals read different nationally distributed magazines. In this empirical context, the variation in advertising exposure can be expected to be uncorrelated with state and local variation in product prices, physician visit prices, the prices of healthy foods and or the price of exercise clubs or exercise equipment.

5. Results

5.1 Summary Statistics

Table 5 provides descriptive statistics on all the variables used in the analysis. Since statins are almost exclusively used by older respondents, our sample is limited to those who are aged over 45. The demographic makeup of the sample as shown in Table 6 is representative of people aged over 45 in the United States.9 As reported in the crosstabulations in our 10 years of NCS data in Table 4, 16.2 percent of respondents report high cholesterol as an ailment they have had in the last 12 months and were diagnosed,

9

The demographic characteristics - such as age, family income, education, ethnicity and health insurance of NCS respondents are very similar to that of the other national health data sets such as NHIS. For details see Plassman et al (2005).

17

11.2 percent purchase statin drugs, 34.7 percent control diet, and 42.9 percent exercise regularly. Around half (50.1 percent) are covered by private insurance, while 17.4 percent do not have health insurance. The average number of hours of television watching in a week equals 18.8, with 15.9 percent of the sample reporting that they do not watch television at all. Only 4.9 percent of the sample reports not reading any magazines.

5.2 Cross Tabulations

To get a first glimpse of the relationship between the key variables of interest and exposure to advertising, Table 6 presents how these variables are distributed across different quartiles of the distribution of exposure to advertisements for statin drugs. Table 6 shows a clear relationship between the number of statin drug advertisements a respondent sees and reporting that they have high cholesterol. For example, while 13.1 percent of respondents in the first quartile of the exposure distribution reported cholesterol, 19.5 percent reported having high cholesterol if they were in the highest quartile of the exposure distribution. Similarly, whereas 8.6 percent of respondents in the first quartile of the exposure distribution reported using a statin drug, 13.8 percent reported using a statin drug if they were in the highest quartile of the exposure distribution. Interestingly, there is a monotonic relationship between levels of exposure and use of statins. For the second quartile of exposure, 9.8 percent of respondents use a statin drug and 11.2 percent of those in the third quartile use a statin drug. In general, these cross-tabulations show a strong correlation between exposure to advertising, knowledge of high cholesterol and use of statins. The results for diet and exercise show similar relationships. 33.9 percent of respondents in the first quartile of the exposure distribution reported exercising regularly, whereas 49.8 percent reported exercising if they were in the highest quartile of the 18

exposure distribution. The relationship between exercise and exposure also appears to be monotonic. 38.8 percent of the second quartile and 45.3 percent of the third quartile is exercising regularly. Finally, with respect to controlling one‟s diet, 26.9 percent of respondents in the first quartile of the exposure distribution reported controlling diet, while 37.0 percent reported controlling diet if they were in the highest quartile of the exposure distribution. These results appear to be less monotonic in that the results for 3rd and 4th quartiles are very similar. As with cholesterol knowledge and use of statins, the cross-tabulation results show strong correlations between advertising exposure and our therapeutic lifestyle change variables. The challenge we face is to understand whether there are causal relationships between advertising and exposure. It could very well be that advertising exposure is targeted to those who have high cholesterol, need statin drugs and are interested in improving their health. This could drive these correlations and say nothing about whether the advertising itself drives these behavioral outcomes. The specifications we outline below directly address the „causality‟ issue.

5.3 The Effect of DTCA Exposure on Behavioral Outcomes

We begin our multivariate analysis of the impact of advertising exposure on diagnosis, drug use and life style changes. Table 7 presents estimates of the effect of DTCA advertising exposure on each of the behavioral outcomes.10 All specifications include the demographic variables discussed above as well as our controls for television watching and magazine readership behavior. In particular, we include the number of PHADS magazines that each individual reads in order to control for unobserved heterogeneity

10

We also tried probit estimation for robustness check purposes. As expected, the qualitative results from probit estimations were identical to OLS estimations.

19

between statins use and readership behavior. This is important because our measure of advertising exposure is based, in part, on the readership behavior of the respondents. We use alternative specifications to test the sensitivity of our estimated impact of advertising exposure. Model 1 is the baseline specification which does not control for magazine types or magazine fixed effects, model 2 includes magazine categories, while model 3 replaces magazine categories with a set of dummy variables on whether an individual has read the magazines found in NCS. We first focus on the models where the dependent variable is whether the respondent reports having high blood cholesterol. The results in model 1 and model 2 show that the coefficients on advertising exposure reveal a positive and statistically significant effect on exposure on being diagnosed high blood cholesterol. The coefficient on advertising exposure in model 1, which controls for magazine reading habit, is positive and statistically significant. The magnitude of the effect of advertising exposure is reduced significantly when controlling for magazine categories suggesting that some of the estimated impact is driven by what category of magazines individuals read rather than the advertising in these magazines. Moreover, when we replace the magazine categories with the NCS magazine dummy variables, the coefficient of the advertising exposure is not statistically significant. Throughout the specifications, when we include NCS magazine fixed effects, the effect of DTCA exposure on behavior is eliminated. We next focus on the impact of advertising exposure on the use of statin drugs. The results for models 1 and 2 in Table 7 show that the coefficients on advertising exposure reveal a positive and statistically significant impact of exposure on the purchase of statin drugs. As we have seen previously in the results for diagnosis, the coefficient of DTCA exposure in model 2 is reduced from model 1. The coefficient of DTCA exposure in model 3 is positive, but not statistically significant. Our next focus is on the impact of advertising exposure on diet control and exercise. The results for these models are presented in table 7. Overall, the results suggest 20

that DTCA exposure to statin drugs has statistically significant effect on exercise only. Even after being diagnosed, the effect of DTCA exposure is still positive and statistically significant on exercise throughout the different specifications. Except for model 1, we do not find any statistically significant effect of DTCA exposure on diet control for people aged over 45. Conditional on being diagnosed, DTCA exposure does not significantly affect diet control. We also describe the coefficients on a subset of the other variables included in the model. The coefficients on other variables in the model are generally what we would expect. The coefficients on the age variables show a positive and significant relationship between age and diagnosis of high cholesterol, purchase of statins, and dietary control. Not surprisingly, the relationship between age and regular exercise is negative and statistically significant. Females are less likely to be diagnosed with high cholesterol and are less likely to purchase statins. Interestingly, even after conditioning the sample on those with high cholesterol, females are less likely to be prescribed statin drugs. Females are also more likely to engage in dietary control and in regular exercise. Compared with Whites, African Americans are less likely to be diagnosed with high cholesterol and less likely to purchase statins. As with females, even after conditioning the sample on those who have high cholesterol, they are less likely to be prescribed statin drugs. African Americans are less likely to engage in diet control but are more likely to engage in regular exercise. The coefficients on Hispanic are very similar to the coefficients described for African Americans. There appears to be some education gradient with respect to diet and exercise. Compared with high school graduates, dropouts are less likely to control diet and exercise than college graduates. One possible explanation for this result is that the relatively more educated respondents are more likely to engage in regular exercise and dietary control. This is true even when conditioned on having high cholesterol. It is possible that the

21

more educated prefer to invest in therapeutic lifestyle changes compared with drug therapy. The coefficients on magazine reading and television watching habits are positive and significant in the diagnosis and purchase models. One possible explanation for this is that watch more hours of television are exposed to more statins advertisements through this medium. However, for television watching, because we do not measure the number of advertisements they are exposed to, this is speculative. The uninsured are less likely to be diagnosed, purchase statins, control their diet or engage in regular exercise. This is also true conditional on diagnosis. That is, even for those diagnosed with high cholesterol the uninsured are less likely to purchase statins. Whereas Medicare recipients are more likely to be diagnosed, purchase statins and engage in dietary control. These results are as expected. Finally, the coefficients on the survey wave variables are consistent with what we would expect. Diagnosis is increasing over time as is statins use. This is as expected because the ATP III has lowered the threshold level of high cholesterol recently and size of the statins market has been increased.

7. Discussion

Using a unique set of advertising data that allows us to measure individual level advertising exposure together with a rich set of individual level control variables, this paper studies the effect of DTCA exposure on patient‟s getting diagnosis, purchasing statins, controlling diet and exercising regularly. Our results show that there is a positive and significant effect of DTCA exposure on getting diagnosed, purchasing statins and exercising regularly, but the effect of DTCA exposure largely goes away with exception on exercise when we control more intensely on firms‟ targeting. Some of our results are opposite to previous findings that DTCA increases the likelihood of visiting physician‟s 22

office or decrease exercising. But in most studies of the impact of DTCA it is difficult, if not impossible, with the data that is utilized, to adequately address the targeting issue. Adequately accounting for targeting is one of the most significant challenges to research on the impact of advertising. Careful examination of how other identification strategies used in other research account for targeting is warranted. For example, some studies use instrumental variables to identify the causal impact of advertising. In these studies advertising of alternative products are used to instrument the product under examination. However, it is likely that all DTC Rx advertising is likely to be targeted toward those with health insurance coverage thus creating a need to deal with controls for individual characteristics that account for this targeting. Additionally, if there are common factors that determine the targeting of prescription drug advertising then instrumental variables using alternative products is not likely to solve this problem.

23

References American Heart Association. 2005. Heart Disease and Stroke Statistics – 2005 update. http://www.americanheart.org/presenter.jhtml?identifier=1200026 Arnold, C.J. 2005. “A deep dive into the cholesterol debate,” Credit Suisse Equity Research Report.

Audit Bureau of Circulation. 2003. Magazine Trend Report, Schaumburg, IL.

Avery, R.J., Kenkel, D.S., Lillard, D.R., Mathios, A.D.2006. “Regulating Advertisements: The Case of Smoking Cessation Products,” NBER Working Paper No. 12001. ___________________________________________ 2007. “Private Profits and Public Health: Does Advertising Smoking Cessation Products Encourage Smokers to Quit?,” Mimeo. Berndt, E.R. 2005. “The United States‟ Experience with Direct-to-Consumer Advertising of Prescription Drugs: What Have We Learned?” Paper presented at the International Conference on Pharmaceutical Innovation, Taipai, Taiwan. Brekke, K.R., Kuhn, M. 2005. “Direct to Consumer Advertising in Pharmaceutical Markets,” Journal of Health Economics, Vol. 25, No. 1, 102-130. Calfee, J., Winston, C., Stempksi, R. 2002. “Direct-to-Consumer Advertising and the Demand for Cholesterol-reducing Drugs,” Journal of Law and Economics, Vol. 45 (October), pp. 672-690.

24

Donohue, J.M., Berndt, E.R. 2004. “Effects of Direct-to-Consumer Advertising on Medication Choice: The Case of Antidepressants,” Journal of Public Policy & Marketing, 23 (2) Fall, 115-27. Grossman, M. 1972. “The Demand for Health: A theoretical and empirical investigation,” National Bureau of Economic Research Occasional Paper, New York: Columbia University Press. Iizuka, T., Jin, G. 2005. “The Effect of Prescription Drug Advertising on Doctor Visits,” Journal of Economics & Management Strategy, Vol. 14, No. 3, Fall. 701-27. _____________ 2007. “Drug Advertising and Health Habits,” mimeo. Ling, D., Berndt, E., Kyle, M. 2002. “Deregulating Direct to Consumer Marketing of Drugs: Effects on Prescription and Over-the-Counter Product Sales,” The Journal of Law and Economics, Vol. 45, No. 2. pp. 691-723. Mukherji, P., Dutta, S., Rajiv, S. 2004. “Estimating the Effect of Direct-to-Consumer Advertising for prescription drugs: A Natural Experiment” http://ssrn.com/abstract=502863.

National Consumer Survey (NCS). 2005. http://www.directionsmag.com/companies/Simmons_Market_Research _Bureau/

National Institute of Health. 2001. Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel III). Third Report of the National Cholesterol Education Program Expert Panel. National Heart, Lung, and Blood Institute, National Institute of Health. NIH Publication No. 01-3670. Norton, E.C., Wang, H., Al, C. 2004. “Computing interaction effects and standard errors in logit and probit models,” The Stata Journal, Vol. 4, No. 2, 154-167.

25

Plassmann, V., Kenkel, D., Lillard, D.R., Mathios, A. 2005. “Smoking Cessation: An Analysis of Quit Methods,” Working Paper, Department of Policy Analysis and Management, Cornell University. Rizzo, J. 1999. “Advertising and Competition in the Ethical Pharmaceutical Industry: The Case of Antihypertensive Drugs,” Journal of Law and Economics, 42:1, pp.89-116. Rosenthal, M.B., Berndt, E., Donohue, J., Epstein, A., Frank, R. 2003. “Demand Effects of Recent Changes in Prescription Drug Promotion,” Kaiser Family Foundation. http://www.kff.org/content/2003/6085/Demand_Effect_revised_61803.pdf Saftlas, H. 2005. “Industry Surveys – Healthcare: Pharmaceuticals,” Standard & Poors Industry Surveys. Sprang, H., Purcell, M., Ryan, M. 2005. “Pharmaceuticals for Beginners – Third Edition,” Deutsche Bank Industry Research Report. World Health Organization, 2002. “The World Health Report 2002 –Reducing Risks, Promoting Healthy Life.” http://www.who.int/whr/2005/whr2005_en.pdf Wosinska, M. 2002. “Just What the Patient Ordered? Direct-to-Consumer Advertising and the Demand for Pharmaceutical Products,” Harvard Business School Marketing Research Papers, No. 02-04. ___________ 2005. “Direct-to-Consumer Advertising and Drug Therapy Compliance,” Journal of Marketing Research, August. 42(3), 323-32.

.

26

Table 1: The Development of statins market

Year 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001

FDA approval dates Mevacor (Lovastatin) - 8/1987

Number of products per year Brand Generic Total 1 0 1 1 0 1 1 0 1 1 0 1 3 3 4 4 4 5 6 6 6 7 7 0 0 0 0 0 0 0 0 1 1 5 3 3 4 4 4 5 6 6 7 8 12

Pravachol (Pravastatin) - 10/1991 Zocor (Simvastatin) - 12/1991 Lescol (Fluvastatin) - 12/1993

Lipitor (Atrovastatin) - 12/1996 Baycol (Cerivastatin) - 6/1997 Lovastatin - Teva 9/1999 Lescol XL (Fluvastatin) - 12/2000 withdrawal of Baycol - 8/2001 Advicor (Lovastatin) - 12/2001 Lovastatin - Purepac Pharm, Gen Pharm, Mylan, Sandoz, 12/2001 Altoprev4) (Lovastatin) - 6/2002 Lovastatin - Carlsbad 6/2002 Pravigard pac (Aspirin/Pravastatin) 6/2003 Crestor (Rosuvastatin) - 8/2003 Vytorin (Extimibe/Simvastatin) - 7/2004

2002

8

6

14

2003 2004

10 11

6 6

16 17

Source: Drugs@FDA Drug information Pathfinder http://www.fda.gov/cder/Offices/DDI/pathfinder.htm, Medline Plus (National Library of Medicine) Note: 1) Based on FDA's approval date. 2) Product is counted if approved by September of given year. 3) Other cholesterol reducers, Zetia, Niaspan (niacin) and Welchol are not included as they are not statins. 4) Altoprev was previously Altocor but changed the name in 2004 to avoid possible confusion with Advicor

27

Table 2: Brand Profile

Brand Lipitor Zocor Pravachol Zetia/Vytorin Crestor Mevacor Advicor Lescol

Substance atrovastatin simvastatin pravastatin extimibe/simvastatin rosuvastatin lovastatin lovastatin fluvastatin

Manufacturer Pfizer Merck BMS Schering-Plough/Merck AstraZeneca Merck Kos Pharmaceutical Norvatis

Max dose 80mg 80mg 80mg 80mg 40mg 80mg 40mg 80mg

Total cholesterol -45% -31% -27% -43% -46% -12% NA -27%

LDL HDL -60% 5% -36% 16% -37% 3% -60% 6% -63% 10% -17% 12% -45% 41% -36% 6%

Source: Company data; Facts and Comparisons; SG Cowen

28

Table 3: Number of statins Advertisements by Magazine

Magazines Better homes & gardens Black Enterprise Business Week Ebony Essence Family Circle Good Housekeeping McCall's Money Modern Maturity National Geographic Newsweek People Reader's Digest Sports Illustrated Time TV Guide U.S. News & World Report Women's Day Total

Lipitor 2 0 0 0 0 0 0 1 0 2 0 1 2 4 0 2 1 8 0 23

Mevacor 0 5 0 0 0 0 4 0 3 0 0 3 0 5 0 6 0 7 5 38

1995 - 1999 Pravachol Zocor 3 3 1 0 13 0 0 1 0 0 5 3 5 5 1 0 5 0 0 1 4 0 18 17 6 9 8 15 0 1 18 15 3 1 16 13 3 9 109 93

Subtotal 8 6 13 1 0 8 14 2 8 3 4 39 17 32 1 41 5 44 17 263

Crestor 2 0 8 0 0 0 5 0 4 0 2 13 7 2 12 11 4 12 0 82

Lipitor 15 0 0 4 0 4 8 0 10 8 14 12 0 19 0 16 12 10 4 136

Pravachol 5 0 10 0 0 0 17 0 0 6 0 14 2 15 0 20 6 12 10 117

2000 - 2004 Vytorin Zocor 0 8 0 12 3 0 0 17 0 8 0 7 0 3 0 17 0 4 0 1 0 5 3 29 5 0 0 24 4 11 4 23 3 19 6 28 0 13 28 229

Subtotal 30 12 21 21 8 11 33 17 18 15 21 71 14 60 27 74 44 68 27 592

Total 38 18 34 22 8 19 47 19 26 18 25 110 31 92 28 115 49 112 44 855

29

Table 4: Number of Prescription Cholesterol Reducing Drug Users in NCS by Year

Year 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 Total

Total 15,038 8,516 26,758 29,826 33,054 31,575 21,452 20,024 18,173 22,410 226,826

All age Users 594 346 1,187 1,483 1,915 1,870 1,593 1,766 1,692 1,890 13,092

Percentage 3.95 4.06 4.44 4.97 5.79 5.92 7.43 8.82 9.31 8.43 5.77

Year 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 Total

Age over 45 Total Users 6,980 500 3,883 289 13,162 1,056 14,830 1,328 16,692 1,749 15,980 1,722 11,970 1,455 11,447 1,641 10,359 1,544 11,066 1,696 116,369 12,980

Percentage 7.16 7.44 8.02 8.95 10.48 10.78 12.16 14.34 14.90 15.33 11.15

30

Table 5: Descriptive Statistics of NCS 1995 – 2004 Age Over 45 (N=116,369)
Variable % Purchase % Diagnosed % Exercise % Diet control Exposure to statins ad Age Family size Magazine readership % Read no magazines Avg. weekly hours of TV watching % Watch no TV in avg. week Avg. hours of radio listening % Listen to no radio in avg. week % Female % Children in the household % Employed Family income % Race White Black Hispanic Other race % Education Less than high school Some college High school graduate College graduate Mean 11.15 16.21 42.90 34.65 23.97 59.47 2.34 6.06 4.86 18.79 15.87 2.10 34.63 54.99 47.86 76.85 63589.21 88.19 5.87 6.94 2.07 12.53 25.93 31.03 30.51 St.d Min Max Variable % Marital status Never married Divorced/separated/widowed % Health insurance Private insurance Medicaid Medicare Uninsured % Magazine type Home Young Adults African American General Interest Business Retiree % Year 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 Mean 5.21 24.06 50.06 2.74 20.53 17.42 68.26 21.51 6.87 65.75 14.63 15.55 6.00 3.34 11.31 12.74 14.34 13.73 10.29 9.84 8.90 9.51 St.d Min Max

29.61 9.97 1.33 6.00 18.47 3.94

0.00 47.00 1.00 0.00 0.00 0.00

292.25 77.00 14.00 131.00 166.80 72.00

60275.62

2500.00

341984.60

31

Table 6: Percentage of Diagnosed, Purchase, Exercise and Diet Control by Quartile of Distribution of statins Advertisements (Age over 45)
Quartile Q! (0 to 0.5 advertisements) Q2 (0.5 to 9 advertisements) Q3 (9 to 28 advertisements) Q4 (28 or more advertisements) Total Quartile Q! (0 to 0.5 advertisements) Q2 (0.5 to 9 advertisements) Q3 (9 to 28 advertisements) Q4 (28 or more advertisements) Total Quartile Q! (0 to 0.5 advertisements) Q2 (0.5 to 9 advertisements) Q3 (9 to 28 advertisements) Q4 (28 or more advertisements) Total Quartile Q! (0 to 0.5 advertisements) Q2 (0.5 to 9 advertisements) Q3 (9 to 28 advertisements) Q4 (28 or more advertisements) Total Not diagnosed 21,404 21,745 24,882 29,479 97,510 Not purchased 22,505 22,895 26,393 31,596 103,389 Not controled diet 16,494 14,786 17,786 21,784 70,850 Not exercised 14,924 14,089 14,988 17,022 61,023 Diagnosed 3,208 3,642 4,849 7,160 18,859 Purchased 2,107 2,492 3,338 5,043 12,980 Controlled diet 6,073 8,642 10,050 12,803 37,568 Exercised 7,646 8,916 12,402 16,873 45,837 Total 24,612 25,387 29,731 36,639 116,369 Total 24,612 25,387 29,731 36,639 116,369 Total 22,567 23,428 27,836 34,587 108,418 Total 22,570 23,005 27,390 33,895 106,860 % Diagnosed 13.03 14.35 16.31 19.54 % Purchased 8.56 9.82 11.23 13.76 % Controlled Diet 26.91 36.89 36.10 37.02 % Exercised 33.88 38.76 45.28 49.78

32

Table 7: OLS Estimates of Behavioral Outcomes for NCS 1995 – 2004 Adults Age Over 45
Diagnosed Model 2 0.0157 *** (0.0051) Yes Yes NA 116369 0.04 Controlled diet Model 2 *** 0.0008 (0.0067) Yes Purchased Model 2 0.0162 *** (0.0044) Yes Yes NA 116369 0.04 Exercised Model 2 0.0278 *** (0.0071) Yes

Variable Ad exposure/100 Wave fixed effects Magazine category fixed effects NCS magazine fixed effects Observations R-squared Variable Ad exposure/100

Model 1 0.0266 (0.0048) Yes No No 116369 0.04 Model 1 0.0237 (0.0063) Yes

***

Model 3 -0.0038 (0.0065) Yes NA Yes 116369 0.05 Model 3 0.0088 (0.0085) Yes

Model 1 0.0255 (0.0041) Yes No No 116369 0.04 Model 1 0.0382 (0.0067) Yes

***

Model 3 0.0041 (0.0056) Yes NA Yes 116369 0.04 Model 3 0.0148 (0.0090) Yes NA Yes 106860 0.09

***

*

Wave fixed effects Magazine category fixed effects No Yes NA No Yes NCS magazine fixed effects No NA Yes No NA Observations 108418 108418 108418 106860 106860 R-squared 0.09 0.09 0.10 0.07 0.07 Note: Robust standard error in parentheses; * significant at 10%; ** significant at 5%; *** significant at 1%

33

Note: Robust standard error in parentheses; * significant at 10%; ** significant at 5%; *** significant at 1%

34

35


				
DOCUMENT INFO