Prescription Drug Advertising and Drug Utilization Part 1

Prescription Drug Advertising and Drug Utilization: The Role of Medicare Part D ABSTRACT Pharmaceutical firms currently spend over $4 billion on direct-to-consumer advertising (DTCA) of prescription drugs, a nearly 30-fold increase since 1993 that has led to much debate about its value to patients. We examine how DTCA influences drug utilization along the extensive and intensive margins by exploiting a large and plausibly exogenous shock to DTCA driven by the introduction of Medicare Part D in 2006. Using data on advertising for local media markets from Nielsen, we show that Part D led to large relative increases in DTCA in geographic areas with a high concentration of Medicare beneficiaries compared to areas with a low concentration. We examine the effects of this sudden differential increase in advertising on non-elderly individuals to isolate the effects of advertising on drug utilization from the direct effects of Part D. Using data from pharmacy claims, we find substantial differential increases in drug utilization that mirror the increases in DTCA after Part D. These effects are driven both by increased take-up of treatment and improved drug adherence. Our results imply significant spillovers from Medicare Part D onto the under-65 population and an important role for nonprice factors in influencing prescription drug utilization. 1 1. Introduction Spending on direct-to-consumer advertising (DTCA) of prescription drugs in the U.S. has increased dramatically in the last two decades from $150 million in 1993 to over $4 billion in 2010 (Dave, 2013; Dave and Saffer, 2012). This rise was precipitated by a 1997 FDA policy change that relaxed restrictions on drug advertising. 1 Most drug advertising occurs on television, where pharmaceuticals represented the third highest category of advertising expenditures in 2014 (behind automotive and fast food restaurant advertising). 2 Nielsen estimates that an average of 80 pharmaceutical ads air every hour on American television. 3 Given that adults ages 50+, a population with a high rate of prescription drug use, watch an average of more than 40 hours of live television per week (Nielsen, 2014), the pervasiveness of pharmaceutical advertising could have large effects on prescription drug use. Indeed, Figure 1 shows that the dramatic rise in advertising that occurred over the last two decades has coincided with a striking increase in spending on prescription drugs. Per-capita spending on prescription drugs increased five-fold between 1990 and 2010, following decades of little spending growth (National Health Expenditure Accounts, 2015). While the coincidence of these trends suggests a strong relationship between advertising and spending, the causal channel could go in both directions. The unprecedented number of blockbuster drugs introduced in the 1990s could have induced greater advertising as well as greater drug spending. This, along with other confounding factors, makes it difficult to isolate the independent effect of DTCA. The rise in the level of DTCA has generated much debate about its effects on patient welfare. Most countries (with the exception of the U.S. and New Zealand) ban this type of advertising. On the one hand, DTCA may be informative if it educates patients about available treatments, encourages individuals to seek care for underdiagnosed conditions, and improves communication between patients and physicians. Advertisements may also serve to remind patients to take their existing medications and influence their perception of the benefits of treatment, promoting better drug adherence (Holmer, 2002; Donohue et al., 2004; Wosinska, 2005). On the other hand, the 1 Prior to 1997, ads were required to include essentially all of the information on the product label (which is unlikely to fit in a 30-second television or radio spot), but after 1997 only the major risks and benefits needed to be included. Before 1997, most advertising was in print and it was limited. 2 3 See Nielsen “Tops of 2014: Advertising” Nielsen estimate reported in FiercePharma “Top 10 DTC Pharma Advertisers – H1 2013” 2 persuasive and product differentiation aspects of DTCA may lead to unnecessary treatments and excessive drug spending. There is a lack of consensus on whether DTCA serves primarily to inform or persuade, which matters for assessing its value to patients. This distinction hinges partly on the extent to which DTCA impacts drug utilization and the mechanisms underlying advertising’s impacts, such as whether the effects of DTCA stem from changes in the take-up of therapy versus changes in adherence and whether there are spillovers of advertising on nonadvertised drugs. However, identifying DTCA’s causal effects on utilization has been challenging empirically, since demand factors could influence both the amount of advertising and the timing of advertisements. Some studies have tried to address these endogeneity concerns with instrumental variable strategies, though it is difficult to find appropriate instruments given the close relationship between demand and advertising decisions. We address these challenges by providing one of the first quasi-experimental studies on the effects of DTCA on drug utilization. We exploit a large and plausibly exogenous shock to DTCA driven by the introduction of Medicare Part D in 2006. We focus on drugs that treat five chronic conditions that account for a large share of advertising spending—depression, diabetes, hyperlipidemia, hypertension, and osteoporosis 4—and estimate effects along both the intensive and extensive margins to examine the causal pathways through which DTCA influences drug utilization. Our instrumental variable strategy exploits variation across geographic areas in the share of the population that is covered by Medicare (ages 65+) to predict changes in advertising exposure across areas. We show that there was a large relative increase in advertising exposure immediately following the introduction of Part D in geographic areas with a high share of elderly compared to areas with a low elderly share. Prior to Part D, both the levels and trends in advertising exposure across high and low elderly share areas were nearly identical. Since advertising cannot be perfectly targeted to the elderly, we exploit the sudden differential increase in advertising exposure for non-elderly that live in elderly-dominated areas to estimate the effects of advertising on drug utilization. This strategy hinges on the observation that non-elderly individuals are exposed to the increase in DTCA but do not receive Part D insurance coverage, which may independently impact drug utilization. Our focus on the non- 4 Among the 25 most advertised brand-name drugs, these 5 conditions account for half of total advertising expenditures (Kantar Media, 2011). 3 elderly allows us to isolate the effects of advertising on drug utilization from the direct effects of Part D. This paper makes four main contributions. First, we exploit a major policy change to identify the effects of DTCA on drug utilization. The use of policy shocks as natural experiments has been scarce in the existing advertising literature, although it is a promising approach for obtaining variation in advertising that is unrelated to individual demand and health status. Second, we use this empirical framework to estimate the effects of DTCA not only on overall drug utilization but also separately on the extensive and intensive margins—i.e., higher take-up of drug treatments versus better drug adherence by existing patients. We tease apart the contributions of these component effects to the total relationship between advertising and utilization to explore welfare implications of advertising. The prior literature has primarily focused on overall utilization. Third, we use data from two novel sources. We measure pharmaceutical advertising using data on Nielsen ratings in local media markets that we observe separately for the non-elderly (under 65) and the elderly (65+). While most of the DTCA literature uses advertising expenditures or the volume of ads to quantify advertising, ratings are a more direct measure of actual advertising exposure. 5 This measure is more often used outside of the DTCA literature to measure exposure to other types of television programming (e.g., Kearney and Levine, 2014; Kanazawa and Funk, 2001). We obtain measures of drug utilization using administrative pharmacy claims from a database of over 40 large national employers covering about 18 million person-years. Finally, a fourth contribution of our study is to quantify spillover effects of Part D on the non-elderly population. Numerous studies have examined the effects of Part D on the elderly but few have considered the effects on the non-elderly. 6 One important mechanism through which Part D may have an effect on the non-elderly is through advertising, and we find strong evidence of these spillovers. We find that drug utilization is highly responsive to advertising exposure. Following Part D, there was a 6 percent increase in the average number of prescriptions purchased by the nonelderly in areas with high elderly share, relative to areas with low elderly share. Event study results using quarterly utilization data show that this differential effect coincided precisely with 5 To our knowledge, Saffer et al. (2007) – which studies advertising for nicotine replacement therapy – is the only other pharmaceutical advertising study to use Nielsen ratings data. 6 Part D may also have spillover effects on the non-elderly through pharmaceutical R&D investments (BlumeKohout and Sood, 2013) and negotiated drug prices (Duggan and Scott Morton, 2010; Lakdawalla and Yin, 2015). 4 the implementation of Part D in 2006 and persisted through the end of our study period in 2010. The event study also confirms that there were no differential pre-trends in utilization across higher and lower elderly share areas, providing support for the identifying assumption that the trends would have continued to be the same in the absence of Part D. Our results imply that a 10 percent increase in advertising views leads to a 5.4 percent increase in total prescriptions filled for advertised chronic drugs. Our estimate of the elasticity with respect to views exceeds prior estimates of the elasticity with respect to advertising expenditures, which most DTCA studies estimate. This is consistent with prior evidence that an increase in advertising expenditures leads to a less than proportional increase in advertising views (Sethuraman et al., 2011). We find that advertising increases the take-up of drug treatments and improves compliance for existing patients. Expanded take-up of prescription drugs accounts for about 70% of the total effect of advertising, while increased use among existing patients accounts for the remaining 30%. As an important component of this latter effect, we estimate that a 10 percent increase in advertising views increases adherence to a drug therapy by 1 to 2.5 percent. Finally, we assess whether the increase in advertising led simply to substitution from nonadvertised to advertised drugs or whether it generated a net increase in drug utilization for drug classes. We find evidence that advertising also increased the use of non-advertised drugs in the same therapeutic class as an advertised drug. This suggests substantial positive spillover effects on the use of non-advertised drugs within the same drug classes. The utilization results are robust to geographic area-specific trends, sample restrictions, and alternative specifications of the instrument. We also find little evidence in favor of alternate causal channels. First, Part D did not differentially reduce out-of-pocket drug prices in high elderly share areas, ruling out concurrent price effects that could independently impact drug utilization. Second, changes in direct-to-physician advertising after Part D appear to be unrelated to elderly share. Finally, in a placebo test estimating the differential effects of Part D on drug utilization for classes of drugs that do not advertise, we find effects that are very small or null relative to the effects for classes of drugs that do advertise, providing support that the observed changes in utilization are due to advertising and not driven by other potential spillovers of Part D on the non-elderly (e.g. through changes in physician prescribing behavior). While the literature on prescription drug demand has focused heavily on the importance of prices and insurance status in explaining utilization patterns, we generate estimates of the 5 responsiveness of drug demand to a non-monetary factor and find economically important effects. Using the range of price elasticities in the literature (Goldman, Joyce, and Zheng, 2008) combined with our main results, our estimates imply that a 10 percent increase in advertising exposure generates a change in prescription drug utilization equivalent to a 9 to 27 percent reduction in out-of-pocket price. This paper also shows that by increasing insurance coverage for one population, Part D had the effect of generating additional demand for individuals outside of the Medicare program. These demand increases were themselves large and economically important. The rest of the paper proceeds as follows. Section 2 presents background on Part D and the mechanisms for its effect on advertising, as well as a review of the related DTCA literature. Section 3 describes the data sources. Section 4 outlines the empirical framework. Section 5 presents the results. Sections 6 and 7 provide a discussion and conclude. 2. Background and Related Literature 2.1. Why Should Medicare Part D Increase Advertising Exposure? Medicare is a federal program that provides health insurance to the elderly, ages 65 and over, and to qualifying non-elderly disabled individuals. On January 1, 2006, Medicare expanded to include coverage of outpatient prescription drugs through the introduction of Part D. Part D was enacted as a provision of the Medicare Modernization Act (MMA), which was signed into law in December of 2003, and represented one of the largest expansions of the Medicare program since its inception. Part D substantially increased the proportion of elderly with drug insurance and as a result lowered average out-of-pocket drug costs for Medicare beneficiaries. Previous research has shown that this price reduction increased drug utilization among the 65+ population (e.g. Ketcham and Simon, 2008; Yin et al., 2008; Lichtenberg and Sun, 2007). The widespread changes brought about by Part D may have significantly altered pharmaceutical firms’ incentives to advertise. As shown in earlier theoretical work (Lakdawalla, Sood, and Gu, 2013), insurance expansions such as Part D can influence the return to advertising through two mechanisms. First, both theory and prior empirical literature suggest that more profitable markets generate greater returns to capturing new consumers, and in turn stimulate more intense advertising effort. Thus, the returns to advertising are higher when there are more 6 insured consumers in the market, because insured consumers face lower out-of-pocket costs that induce greater use and spending. Second, insurance coverage might alter the responsiveness of consumers to advertising at the margin. If insured consumers are more responsive, firms will face greater incentives to advertise. Indeed, prior empirical studies suggest that better insured consumers are more responsive to advertising (Wosinska, 2002). Intuitively, an undecided consumer might be more likely to try a new drug after being exposed to advertising if the cost of trying the new drug is lower. Taken together, these two effects suggest that insurance expansions strengthen incentives for advertising and we show that this prediction is borne out in the data. Given this result, we would expect drug advertising to increase more in geographic areas with a higher share of elderly individuals (relative to areas with a low share of elderly), which experienced a greater expansion in insurance coverage. Consistent with this idea, previous research (Lakdawalla, Sood, and Gu, 2013) found that Part D led to a large relative increase in national DTCA spending for drugs differentially used by Medicare beneficiaries. That paper, which focused primarily on the effects of Part D on advertising, also suggested scope for utilization effects of advertising using the Medical Expenditure Panel Survey (MEPS). We build on this previous work, by exploiting a new strategy based on geographic variation in elderly shares. The geographic variation allows us to control for drug-specific shocks, which is of particular importance during our study period given a wave of patent expirations and black box warnings. We also use health insurance claims data to better identify and characterize the causal effects of advertising along both the intensive and extensive margins. 2.2. Previous Literature on Advertising Effects Our paper contributes to a large literature on the impacts of DTCA on drug utilization (see Dave, 2013 for a recent survey). The majority of studies in this literature find positive demand effects of advertising. Although studies consistently find evidence of significant market expansion effects from advertising (e.g. Rosenthal et al., 2003; Iizuka and Jin, 2005; Bradford et al., 2006; Shapiro, 2015), evidence of market stealing—gaining market share from competitors—is mixed. Some studies find no effect, and others find small but statistically significant effects (e.g. Wosinska, 2002; Dave and Saffer, 2012). In general the market expansion effects appear to dominate. 7 Evidence from consumer and physician surveys find that advertising increases the likelihood that a patient initiates a request for a specific drug treatment and the likelihood that physicians fulfill these requests (Hollon, 2005). In one randomized controlled trial, actors are sent to doctors’ offices presenting symptoms of depression. Those who asked for a specific drug treatment or general treatment for depression related to an ad they saw on television were significantly more likely to be prescribed an anti-depressant relative to those who did not request treatment (Kravitz et al., 2005). The magnitude of the effects ranged from 53-76% for those requesting treatment relative to 31% for those not requesting treatment. A persistent challenge for the literature on the impact of advertising on drug utilization has been in identifying a source of variation in advertising that is orthogonal to demand factors. 7 Our study overcomes this problem by using a natural experiment—the introduction of Part D— to study the effects of DTCA on drug utilization among those unaffected by the insurance expansion. To our knowledge, the only other study that provides natural experiment evidence using policy variation is a working paper by Sinkinson and Starc (2015), which exploits changes in advertising due to political election cycles (which temporarily displace DTCA), to examine the effects of own and rival advertising on firm revenue. The estimated elasticities in our study tend to be larger. This may be partially explained by the differences in identification strategies, with Sinkinson and Starc (2015) exploiting temporary reductions in advertising and our study exploiting a permanent increase. Given the long-lasting effects of advertising, we might expect that temporary reductions in advertising intensity would lead to smaller effects on use. In another related working paper, Shapiro (2015) uses variation in advertising expenditures at discrete television market boundaries to estimate the effects of changes in advertising on drug utilization. This strategy assumes that trends in utilization across border-counties are not systematically related to advertising changes and that demand shocks that drive advertising differences are not disproportionately located along one border—assumptions that are difficult to test. 7 Most previous studies of DTCA have had to rely on cross-sectional or time-series variation in advertising expenditures to identify the effect on drug utilization. Studies that attempt to address the endogeneity concern have instrumented for DTCA using variables such as the age of the drug, time until patent expiration, advertising expenditures by the same company in an unrelated drug class, and national advertising costs. 8 Our study offers several other innovations over the prior literature. First, we use data that measure actual exposure to advertising using Nielsen ratings (discussed below) rather than relying on proxies such as advertising spending or number of ads aired. Nearly all DTCA studies use advertising expenditures or volume to quantify advertising. One exception is Avery, Eisenberg, and Simon (2012), which uses survey data from Simmons National Consumer Survey and Kantar/TNS Media Intelligence to construct individual-level exposure to ads for antidepressants. In contrast to that paper, which uses self-reported anti-depressant use in the past 12 months, we have administrative pharmacy claims that enable us to construct comparatively rich measures of utilization such as total prescriptions and days supplied. Second, we estimate the effects of advertising on take-up of treatment and medication adherence, and evaluate how these component effects independently drive the overall relationship between advertising and drug utilization. Since much of the pharmaceutical advertising literature has focused on total utilization effects, little is known about the mechanisms that underlie the relationship between drug utilization and advertising. Specifically, there is little empirical evidence on the effects of advertising on drug adherence and the few existing studies find very small or null effects (Donohue et al. 2006; Wosinska, 2005). Understanding the components of the drug utilization effect is needed to begin to assess whether the increase in use induced by DTCA is welfare enhancing. Finally, we estimate the effects of DTCA for a large number of brand-name drugs across several conditions. Prior studies often focus on a single drug class or a small subset of brandname drugs. Given that FDA policy tends to consider all types of prescription drugs uniformly, our estimates are likely more generalizable for such policy considerations. 3. Data Sources 3.1. Advertising Data The data on viewership of pharmaceutical ads in local media markets come from the Nielsen Ad*Views™ database from 2001-2010. We focus on television advertising, which accounts for more than two-thirds of total DTCA expenditures (Avery, et al., 2012). Nielsen collects data on the universe of television commercials shown in 210 “Designated Market Areas” (DMAs) that span the entire United States. Each DMA is comprised of one or more counties in which the 9 home market television station holds a dominance of total hours viewed. 8 Nielsen viewing stations located in each DMA record all commercials shown and can identify “national” ads shown in all 210 DMAs and “local” ads shown in a subset of these markets. We use data on local ads since there is scope for targeting different amounts of advertising to different markets. Local ads can be shown during network programming (e.g. NBC), syndicated programming, or local television programming (e.g. local news). We obtained local advertising data for the top 100 DMAs (86.5% of TV viewers) and the top 200 advertised brand-name prescription drugs from 2001-2010, which account for more than 96% of advertising spending. Our measure of DTCA exposure is Nielsen rating points. Rating points are derived from data collected on actual viewership of television commercials for a sample of television-owning households in each DMA. Using meters attached to participants’ televisions or paper diaries, Nielsen records who in the household is watching and what they are watching 24 hours a day. “Rating points” are simply the fraction of the sample that watched a particular commercial. In our data, we observe rating points for each brand-name prescription drug, DMA, quarter, and for two age groups (ages 2-64 and ages 65+), which is defined as follows: (1) 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑗𝑗𝑗𝑗𝑗𝑗𝑗𝑗 = # 𝑜𝑜𝑜𝑜 𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣 𝑗𝑗𝑗𝑗𝑗𝑗𝑗𝑗 #𝑜𝑜𝑜𝑜 𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑚𝑚𝑚𝑚𝑚𝑚 x 100 Where 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑗𝑗𝑗𝑗𝑗𝑗𝑗𝑗 are computed as the total number of views of commercials for brand- name drug 𝑗𝑗 in market (DMA) 𝑚𝑚, in age-group 𝑎𝑎, and in quarter 𝑡𝑡 divided by the total number of individuals in the sample in market (DMA) 𝑚𝑚, in age-group 𝑎𝑎, and in quarter 𝑡𝑡, multiplied by 100. We divide rating points by 100 in order to interpret this measure as average views per person. Rating points for a brand can increase if the number of commercials increases, the commercials become better targeted (e.g. primetime vs. late night), and/or more individuals in the market watch television. Nielsen rating points are the industry standard for measuring television viewership and have the advantage of being a more direct measure of advertising exposure than total advertising expenditures or the number of ads, which have been the predominant measures of advertising in the DTCA literature to date. 9 8 For example, the Los Angeles DMA contains 8 counties in the surrounding area which have relatively homogeneous television programming. 9 Nielsen collects very limited data on advertising expenditures at the local level. Expenditure data is not available for local commercials shown during network or syndicated programming, which comprise the majority of local 10 While in recent years, a variety of alternative methods for watching television programming have been introduced—such as time shifted viewing (e.g. DVR) and Internet viewing— traditional live television still remains the dominant medium. In the third quarter of 2014, adults ages 50-64 watched on average 43.2 hours of television programming per week, of which only 3.8 hours were time-shifted and an additional 1.2 hours were spent watching video on the Internet (Nielsen, 2014). Since most of our study period from 2004-2010 precedes the widespread adoption of time-shifted viewing and the introduction of Netflix, YouTube, Hulu, and other Internet streaming services, we expect that the share of our sample that is not watching television live is very small. Moreover, Nielsen does account for most time-shifted viewing in its rating points by including views of all recorded programming watched within seven days of its initial release. 3.2. Drug Utilization Data We construct measures of drug utilization using a database of insurance claims from more than 40 large national employers, including many Fortune 500 companies, for 2004-2010. 10 These data were compiled by a prominent health benefits consulting company and cover approximately 18 million person-years during the study period. The claims dataset is described in more detail in several previous studies (e.g. Goldman et al., 2004; Goldman and Joyce, 2007; Joyce et al., 2007). The pharmacy claims include detailed information about all outpatient prescription drug purchases including the drug name, National Drug Code (NDC), days supplied (e.g. 30 days, 60 days, etc.), and payments. We link the claims data by NDC with data from IMS Health to obtain consistently defined drug names and therapeutic drug classes for each prescription. Limited demographic information is provided on the claims, including gender, age, marital status, and the three-digit ZIP code of residence. We restrict our analysis to individuals with full-year insurance coverage and aged 40-60. 11 This group is closer in age to Medicare eligibility and thus more likely to be using similar types commercials. Only commercials shown during “local television” programming (e.g. local news) have expenditure data. For this reason, we do not use expenditure data in this study. 10 Data from this company prior to 2004 is not defined in a consistent way with data from 2004 onwards, thus we cannot use it in our analysis. 11 We exclude ages 61-64 out of concern that individuals close in age to Medicare eligibility may change their drug utilization behavior in anticipation of future Part D coverage (Alpert, 2015). 11 of prescription drugs as Medicare beneficiaries. We only include individuals who live in the top 100 Nielsen DMAs, which represents about 95 percent of pharmacy claims. We use the three-digit ZIP code to match the claims data with the Nielsen advertising data. Each person in the claims data is assigned to a local advertising market (DMA) based on their ZIP code of residence to determine their potential advertising exposure in each quarter. One limitation of our data is that Nielsen DMAs are defined in terms of five-digit ZIP codes, while we only observe individuals’ three-digit ZIP codes in the claims. Some three-digit ZIP codes overlap multiple DMAs, so it is not possible to assign these individuals to a single DMA with certainty. Instead we assign these individuals the population-weighted 12 average of advertising exposure (rating points) across all of the possible DMAs where they could reside. About 30 percent of the individuals in the claims data receive this probabilistic measure of advertising exposure. Consequently, we use the three-digit ZIP code as the effective advertising market rather than the DMA, since advertising exposure is constant within three-digit ZIP codes for all individuals in the sample. As we will show below, effects are similar if we restrict the data to the subsample with a single DMA match. We initially focus on two variables: total number of prescriptions purchased and total days supplied. We aggregate these measures to quarterly totals for each person by drug. For most analyses, we focus on drugs that treat five chronic conditions: depression, diabetes, hyperlipidemia, hypertension, and osteoporosis. There are 50 drugs for these conditions that were advertised during our study period and are contained in the Nielsen top 200 (see list of drugs in Appendix Table B.1). We collapse the data to the three-digit ZIP code level by condition, to conduct our analyses at the level of variation in advertising exposure. Since Part D affected advertising incentives for all drugs and due to the possibility of spillovers across drugs treating the same condition, we do not conduct a drug-level analysis and instead perform our analysis at the condition-level. This results in 107,345 ZIP code-by-condition-by-quarter observations. In computing mean prescriptions purchased and mean days supplied within a ZIP code and condition, zeros are included for individuals who were enrolled in a health insurance 12 Population weights for the 5-digit ZIP code level come from the 2000 Decennial Census. In the Census, population estimates are aggregated by ZIP Code Tabulation Areas (ZCTAs) and we use a 5-digit ZIP code to ZCTA crosswalk to obtain the 5-digit ZIP code population estimates. 12 plan but did not purchase any drugs for the condition. 13 Separate means are computed for advertised and non-advertised drugs 14 within each of the five chronic conditions. We focus primarily on advertised drugs and use non-advertised drugs in select specifications. We also construct a measure of drug adherence for individuals who filled at least one prescription for any of the five chronic conditions. We measure adherence between the quarter of the individual’s first drug claim through the quarter of their last drug claim for that condition. 15 We consider adherence as receiving continuous treatment for a condition, 16 rather than for a specific brand since this is the more relevant dimension from a policy and welfare perspective. Adherence is measured quarterly and separately for each condition as the medication possession ratio (MPR), which is a widely used method for measuring medication compliance with claims data (Peterson, et al., 2007; Hess, et al., 2006). The MPR is calculated as the number of days with drug on-hand (i.e. days supplied) divided by the number of days in the quarter. We adjust the numerator of the MPR to account for claims with overlapping days supplied. For example, it is typical to refill a prescription before finishing the days supplied for the initial prescription. The extra days supplied during the overlap period may not indicate better drug adherence. If overlapping claims have the same active ingredient, 17 we assume that the person finishes the days supplied in the first claim before starting the days supplied in the second claim. A refill of an existing medication or a switch from a brand name to generic product would fall under this category. For overlapping claims with different active ingredient names (for the same condition), we assume that patients start using the days supplied for the second claim on the fill date and throw away the remaining days supplied for the first claim. This case likely represents a drug switch that resulted from an unsatisfactory response to the initial medication. Since 13 In other words, in the individual-level data, each person gets 5 observations for each of the conditions in each quarter whether or not they were diagnosed with the condition. If they do not use a drug for that condition, the observation is zero. We do not condition on having a diagnosis as advertising might affect the rate of diagnosis. 14 Non-advertised drugs (typically off-patent brands or generics) are drugs that belong to the same therapeutic drug classes as the 50 advertised chronic brand-name drugs, but do not advertise during our study period. 15 In some specifications, we account for discontinuation by computing an alternative measure of adherence where we assume that a person is non-adherent after their last observed drug claim. 16 Specifically, we combine utilization for advertised drugs with the non-advertised drugs in the same therapeutic classes as the advertised drugs. 17 Combination drugs are viewed as a unique combination of two or more active ingredients. 13 advertising may lead to more drug switching, it is especially important to account for this case to avoid overstating the effect of advertising on adherence. 18 After constructing the quarterly MPR for each individual by condition, we also create a binary indicator for individuals who had MPR ≥80%, which is considered high adherence and is the threshold most commonly reported in the pharmaceutical literature (Andrade, et al., 2006). As before, we collapse the data by three-digit ZIP code, condition, and quarter, computing the mean MPR and the proportion of individuals with MPR ≥80% in each cell. 3.3. Population Data We compute the share of the population that is 65 and over (i.e. eligible for Medicare) in each local advertising market (DMA) using the 2000 Census. From the advertiser’s perspective, the DMA is the relevant market. For individuals who cannot be matched to a single DMA, they are assigned the population-weighted average of the elderly share across all possible DMAs where they could reside. There is substantial heterogeneity in the share of the population that is 65+ across markets, ranging from 8% in the Houston DMA to 26% in Fort Myers-Naples DMA (see Table 1). The elderly share is used to construct the instrument. This share is held constant at the DMA’s 2000 value so that no identification originates from changes in the elderly share. 3.4. Descriptive Statistics In Table 2, we present sample means for the advertising variables by elderly share for 2005 and 2007, before and after Part D. For each year, we split the 100 DMAs into above-median elderly share markets and below-median elderly share markets. Average views per person of pharmaceutical ads are much greater for elderly viewers (ages 65+) relative to non-elderly viewers (ages 2-64). For example, in low elderly share markets in 2005, elderly viewers saw on average 1,184 pharmaceutical ads per year compared to 387 ads for non-elderly viewers. This difference is likely driven by the targeting of pharmaceutical ads to programming that elderly viewers watch, as well as the fact that older viewers watch more hours of television per year. Comparing the change in views per person from 2005 to 2007 across market types, we observe a 18 Days in the hospital were assumed to be fully compliant and patients resumed their prescriptions after they were discharged. 14 three to five times larger increase in ads viewed in high elderly share markets for elderly and non-elderly viewers following Part D. Table 2 also reveals differences in the average population size across high and low elderly share markets. High elderly share markets tend to be less populated than low elderly share markets. This may be partially due to retiree preferences for less urban areas. This result suggests that it will be important to include market fixed effects in all of our analyses. 4. Empirical Strategy To understand the impact of DTCA on drug utilization, we exploit quasi-experimental variation in advertising exposure after the introduction of Part D for non-elderly adults living in high elderly share areas. Specifically, we capture the differential change in DTCA exposure across high and low elderly share areas by estimating the following difference-in-difference equation, our first-stage relationship: (2) 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝑚𝑚𝑚𝑚𝑚𝑚,<65 = 𝛽𝛽(𝑆𝑆ℎ𝑎𝑎𝑎𝑎𝑎𝑎65𝑚𝑚 × 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑡𝑡 ) + 𝛾𝛾𝑡𝑡 + 𝜇𝜇𝑚𝑚 + 𝛼𝛼𝑐𝑐 + 𝜖𝜖 𝑚𝑚𝑚𝑚𝑚𝑚 Where 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝑚𝑚𝑚𝑚𝑚𝑚,<65 is average views per person (rating points) for non-elderly individuals ages 2-64 in market 𝑚𝑚 in quarter 𝑡𝑡 for ads related to condition 𝑐𝑐. The market 𝑚𝑚 is the three-digit ZIP code. 19 𝑆𝑆ℎ𝑎𝑎𝑎𝑎𝑎𝑎65𝑚𝑚 is the share of population 65+ in market 𝑚𝑚 in 2000 20, and 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑡𝑡 is an indicator that equals 1 in the post-Part D period (2006-2010). Thus 𝑆𝑆ℎ𝑎𝑎𝑎𝑎𝑎𝑎65𝑚𝑚 × 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑡𝑡 is our instrument for DTCA exposure for the non-elderly. In some specifications we use an alternative form of this instrument, 𝐻𝐻𝐻𝐻𝐻𝐻ℎ𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸ℎ𝑎𝑎𝑎𝑎𝑎𝑎𝑚𝑚 × 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑡𝑡 , where 𝐻𝐻𝐻𝐻𝐻𝐻ℎ𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸ℎ𝑎𝑎𝑎𝑎𝑎𝑎𝑚𝑚 is an indicator that equals 1 if the market 𝑚𝑚 has an above-median elderly share. This instrument does not restrict the impact of elderly share to have a linear relationship with advertising but, instead, considers the possibility of non-linear effects. We use the share of the population 65+ rather 19 As discussed in Section 3.2, we collapse the data to the 3-digit ZIP code level rather than the DMA-level, since a subset of the sample resides in 3-digit ZIP codes that cross multiple DMAs and cannot be matched to a single DMA. We assign these individuals the population-weighted average advertising exposure across all the DMAs where they might reside. Thus, our constructed advertising exposure measure is constant within 3-digit ZIP codes, but not always within DMAs. 20 The 65+ share is also computed at the DMA level since this is the relevant market from the advertiser’s perspective. For individuals whose 3-digit ZIP code cannot be matched to a single DMA, they receive the population weighted average of the 65+ share across all the DMAs where they might reside. 15 than the level 65+ as our instrument to reflect advertising incentives. The cost of advertising is based on the total number of viewers in a market, so the cost per elderly viewer is lower in markets with a high elderly share. Thus, we would expect that firms targeting elderly viewers would advertise more in high elderly share markets. Since geographic areas with a high concentration of elderly are potentially different from areas with a low concentration, all of our analyses condition on market fixed effects, which account non-parametrically for cross-sectional differences by elderly share. We also include time fixed effects to account for secular time trends and condition fixed effects to account for differences in utilization and returns to advertising across conditions. Standard errors are clustered at the three-digit ZIP code level (market 𝑚𝑚) to account for serial correlation within areas as well as correlation across conditions within areas. Next, we estimate a reduced form equation comparing changes in drug utilization for the non-elderly across areas with a high elderly share relative to a low elderly share. By focusing on the non-elderly population, we can isolate the effects of advertising on drug utilization from the direct effects of Part D on utilization. The equation for the reduced form is as follows: (3) 𝑌𝑌𝑚𝑚𝑚𝑚𝑚𝑚,<65 = 𝜃𝜃(𝑆𝑆ℎ𝑎𝑎𝑎𝑎𝑎𝑎65𝑚𝑚 × 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑡𝑡 ) + 𝜌𝜌𝑡𝑡 + 𝜎𝜎𝑚𝑚 + 𝜏𝜏𝑐𝑐 + 𝜖𝜖 𝑚𝑚𝑚𝑚𝑚𝑚 Where 𝑌𝑌𝑚𝑚𝑚𝑚𝑚𝑚,<65 is mean total prescriptions or days supply for non-elderly individuals in market 𝑚𝑚 in quarter 𝑡𝑡 for advertised drugs that treat condition 𝑐𝑐. We also use this reduced form model to estimate effects on other measures of drug utilization including prescriptions purchased (or days supplied) conditional on use, the probability of any drug use, and drug adherence. We scale the reduced form effect by the first stage to obtain the 2SLS estimate: the effect of advertising on prescriptions purchased. 21 21 It should be noted that the Nielsen data on advertising exposure is available for the 2-64 age group, while we select on ages 40-60 (individuals who are likely to be using similar drugs as Medicare beneficiaries) in the utilization data. Our advertising elasticity estimates will not be biased by this age-group selection if the proportional change in advertising exposure after Part D is the same across ages. As shown below, individuals 2-64 and 65+ have the same proportional change in advertising exposure after Part D, which supports this assumption. 16