`
`Effects of Direct-to-Consumer Advertising of
`Hydroxymethylglutaryl Coenzyme A Reductase Inhibitors
`on Attainment of LDL-C Goals
`
`W. David Bradford, PhD’; Andrew N. Kleit, PhD?; Paul J. Nietert, PhD?; and
`Steven Ornstein, MD4
`
`"Center for Health Economic andPolicy Studies, Department ofHealth Administration and Policy, Medical
`University ofSouth Carolina, Charleston, South Carolina; *Department ofMeteorology, Center for Health
`Care Policy, The Pennsylvania State University, State College, Pennsylvania; *DepartmentofBiostatistics,
`Bioinformatics, and Epidemiology, Medical University ofSouth Carolina, Charleston, South Carolina; and
`“Department ofFamily Medicine, Medical University ofSouth Carolina, Charleston, South Carolina
`
`ABSTRACT
`Background: Although highly controversial, direct-
`to-consumer (DTC) television advertising for prescrip-
`tion drugs is an established practice in the US health
`care industry. While the US Food and Drug Adminis-
`tration is currently reexamining its regulatory stance,
`little evidence exists regarding the impact of DTC ad-
`vertising on patient health outcomes.
`Objective: The objective of this research was to study
`the relationship between heavy television promotion
`of 3 major hydroxymethylglutaryl coenzyme A reduc-
`tase inhibitors (“statins”) and the frequency with which
`patients are able to attain low-density lipoprotein cho-
`lesterol (LDL-C) blood-level goals after treatment with
`any statin.
`Methods: We used logistic regression to determine
`achievement of LDL-C goals at 6 monthsafter statin
`treatment, using electronic medical
`record extract
`data from patients from geographically dispersed pri-
`mary care practices in the United States. We identified
`LDL-C blood levels as being at or less than goal, as
`defined by risk-adjusted guidelines published by the
`National Heart, Lung, and Blood Institute from the
`Adult Treatment Panel II] (ATP II) data. A total of
`50,741 patients, identified from 88 practices, were di-
`agnosed with hyperlipidemia and had begun therapy
`with any statin medication during the 1998-2004
`timeperiod. In addition,total dollars spent each month
`on television advertising at the national and local lev-
`els for atorvastatin, pravastatin, and simvastatin were
`obtained. DTC advertising data were merged by local
`media market where the physician practice was locat-
`ed and by the month in which the patient was first
`
`December 2006
`
`prescribed a statin. The models were run for all pa-
`tients who initiated therapy, and also on a subsample
`of patients who continued to receive prescriptions for
`the drugs for at least 6 months. Logistic regressions
`were used to predict the likelihood that each patient
`attained the ATP IM LDL-C blood-level goals as a
`function of DTC advertising and other factors.
`Results: High levels of national DTC advertising
`whentherapy wasinitiated were found to increase the
`likelihood that patients attained LDL-C goals at
`6 months by 6% (P < 0.001)—althoughthe effect was
`concentrated among patients with the least-restrictive
`ATP III LDL-C goals (<160 mg/dL). This result was
`found in both the entire set of patients as well as the
`restricted sample of patients who maintained therapy
`for at least 6 months.
`Conclusions: The results of this study suggest that
`higher levels of DTC television advertising of statin
`treatment were significantly associated with improve-
`ments in the likelihood of attaining cholesterol-
`management goals for at least some patients. While
`this paper does not address the impact of DTC adver-
`tising on the costs of care or on unnecessary switching
`between statin treatments, the results do suggest that
`DTC advertising can have beneficial effects, which
`should be a factor when additional restrictions on DTC
`advertising are considered. This result—that DTC ad-
`
`Accepted for publication October 20, 2006.
`doi:10.101 6/j.clinthera.2006.12.015
`0149-2918/06/$19.00
`
`Printed in the USA. Reproduction in whole or part is not permitted.
`Copyright © 2006 Excerpta Medica,Inc.
`
`ALL 2071
`MYLAN PHARMACEUTICALS V. ALLERGAN
`IPR2016-01128
`
`2105
`
`1
`
`ALL 2071
`MYLAN PHARMACEUTICALS V. ALLERGAN
`IPR2016-01128
`
`
`
`Clinical Therapeutics
`
`vertising might have beneficial effects—should be
`weighed against existing studies that have found that
`patients’ suggestions (conceptually which could be
`induced by DTC advertising) may be associated with
`overprescribing (eg, in the case of the use of antide-
`pressants for adjustment disorder). (Clin Ther. 2006;
`28:2105-2118) Copyright © 2006 Excerpta Medica,
`Inc,
`Key words: LDL-C, statin,
`consumer advertising.
`
`treatment, direct-to-
`
`INTRODUCTION
`Thepractice of advertising directly to consumers (DTC)
`through print and broadcast media has been increas-
`ing since the mid-1990s. The practice was further re-
`inforced in August 1997, when the US Food and Drug
`Administration (FDA)clarified and relaxed apparent
`restrictions on what pharmaceutical companies could
`say in short television and radio advertisements pro-
`moting prescription medications. Despite the ubiquity
`of this practice today, the FDA has begun hearings to
`reevaluate its relatively liberal stance.!
`There are a number of published studies from the
`peer-reviewed literature that have used survey data to
`analyze the impact of DTC advertising on prescribing
`practices.2* The articles can be classified into 2 cate-
`gories: those that examined howpatientsfeel about DTC
`advertising,*»° and those that examined how physicians
`feel about DTC advertising.+° With respect to patients,
`1 article indicated that younger patients, patients with
`chronic health conditions, and parents of children with
`health conditions were positively disposed toward DTC
`advertising.? Older patients, however, appeared more
`likely to ignore DTC advertising and rely more heavily
`on their physicians for prescription advice.> Survey re-
`sults also indicated that physicians quite often pre-
`scribed something other
`than what
`their patients,
`perhaps driven by DTC advertisements, suggested. The
`results with respect to physicians’ attributes were also
`mixed. Onesurveyarticle found that more-experienced
`physicians, physicians with larger caseloads, and
`physicians with more exposure to DTC advertising
`were likely to have more positive attitudes toward
`such advertising.* Another study, however, also found
`that physicians were morelikely to becomefrustrated
`with repeated patient questioning in response to DTC
`advertising than with patients who obtained their in-
`formation from medical publications.*
`
`There are, however,limitations with the use of sur-
`vey data for policy considerations. Surveys cantell us
`what people believe about certain issues, and can give
`us information about their demandfor particular con-
`sumer products. However, they are not effective in re-
`vealing whether DTC advertisements change behavior
`or improve health outcomes. Some studies have exam-
`ined the content of DTC advertisements—primarily
`print advertisements—and found that while many ad-
`vertisements could be classified as “informative,” some
`lacked adequate clinical information.”? However, con-
`tent analysis cannot directly address whether behav-
`iors are changed as a consequence of the advertise-
`ments, and if so, how. One recent study reported the
`results of a randomizedtrial, in which nearly 300 vis-
`its were conducted in family-practice settings using
`standardized patients who presented with 2 prede-
`fined conditions—major depression or adjustmentdis-
`order with depressed mood.® The investigators sug-
`gested that, in the formercase, prescribing an antide-
`pressant at the initial visit would be consistent with
`guideline-based care, although such prescribing in the
`latter case would not. The standardized patients made
`a request for a specific drug, a request for nonspecific
`pharmaceutical treatment, or no specific treatment re-
`quest. The study found mixed results for the possible
`impact of patient suggestions. For major depression, a
`nonspecific request had a larger marginal impact on
`prescribing than a brand-specific request, although in
`both cases more prescriptions were written than if the
`patient made no request. However, patient requests
`for a specific drug or a general request both were as-
`sociated with increased prescribing in patients with
`adjustment disorder (P < 0.001 and P = 0.002, respec-
`tively), in whom a prescription was less clearly war-
`ranted. Thus, DTC advertising might be expected to
`promote someoverprescribing in that case.
`A limited number of studies of DTC advertising in
`the peer-reviewedliterature have used patient data to
`determine the association between DTC advertising
`and prescribing practices.?-! One of the first exam-
`ined whether the demandfor the statin class of drugs,
`determined using data from national aggregate drug
`sales by class, was increased after the August 1997
`FDA policy change, but did not find any significant
`short-run direct effect.2 A second study used the
`National Ambulatory Medical Care Survey, together
`with national frequencies of advertising of a number
`of drug classes, to determine the relationship between
`
`2106
`
`Volume 28 Number 12
`
`2
`
`
`
`advertising frequencies and prescribing practices for
`the months between 1992 and 1997.!° While that
`study found some significant correlations, the mea-
`sured impacts of DTC advertising were not consistent.
`Morerecently, a study in ~31,000 patients examined
`how likely patients were to use antidepressants when
`they were diagnosed in months with high spending on
`DTC advertising compared with those who were diag-
`nosed during DTC advertising low-spending months.
`That study found that advertising for any brand was
`associated with increased use ofall brands—although
`the magnitude of the effect was small." Finally, a
`2005 study examinedthelikelihood that patients with
`high cholesterol complied with recommendations for
`statin treatment and found that DTC advertisement
`spending on any statin had small positive effects on
`adherence, irrespective of the statin being used.”
`Two other recent studies, although oneis still a
`working paper, are of particular note for this paper.!%+!4
`Those studies used the sameclinical database and ex-
`amined the impact of DTC advertising on the use of
`cyclooxygenase (COX)-2 inhibitors (celecoxib and ro-
`fecoxib). The first of those studies examined the rate
`of prescribing of celecoxib and rofecoxib to patients
`with osteoarthritis at the physician practice level.!3
`That paper found that increases in DTC advertising
`were associated with a greater flow of patients with
`osteoarthritis into the practice to seek care, consis-
`tent with the hypothesis that maintains that DTC ad-
`vertising will encourage patients who are untreated to
`seek care. The second paper examined the delay be-
`tween diagnosis with osteoarthritis and the adoption
`of daily use of a COX-2 inhibitor.!4 Using patient co-
`morbidities, the investigators identified patients who
`had indications for COX-2 inhibitor use, and those
`whohad contraindications for it. The results suggest-
`ed that DTC advertising was associated with in-
`creased adoption among patients with favorable in-
`dications and discouraging adoption among those
`with contraindications.
`Taken together, these results suggest that DTC ad-
`vertising has the effect of increasing the rate at which
`patients seek care and improvingtheclinical matching
`of patients with appropriate therapies. However, no
`study to date has examined whether DTC advertising
`actually leads to improvements or worsening ofclini-
`cal conditions. If DTC advertising can: (1) encourage
`motivated patients to seek care; (2) improve adher-
`ence to therapy; and/or (3) improve matching of thera-
`
`W.D. Bradford et al.
`
`pies, then we would expect that greater exposure to
`DTC advertising might actually have an impact on ob-
`servable aspects of a patient’s health state. We will
`explore the effect of DTC advertising on 1 such
`aspect—reducing elevated blood low-density lipopro-
`tein cholesterol
`(LDL-C)
`levels
`to within clinical
`guidelines.
`We used a unique data set consisting of >600,000 pa-
`tients (including 3.6 million patient-contact records,
`3.8 million prescription records, 10.1 million vital-
`sign measurements, 12 million laboratory records,
`and 1.3 million preventive-services records) extracted
`from the electronic medical records of 88 primary
`care practices in 33 states across the United States. We
`extracted a subset of these data from patients who
`had ever been diagnosed with hypercholesterolemia,
`who had at
`least
`1 physician visit
`in the years
`1998-2004, and who had begun treatment with any
`statin (including, but not limited to, the 3 statins for
`which advertising data were available). These patient-
`level clinical observations were merged with monthly
`television-advertising measures (dollars spent) for both
`national and local metropolitan area media-market
`advertising of 3 brands of statin drugs (atorvastatin, *
`pravastatin, and simvastatint). These data were used
`in a regression framework to measure the association
`between beginning treatment with anystatin drug ina
`location and month that had high DTC advertising and
`the likelihood of a patient being at his/her guideline-
`based LDL-C goal after 6 monthsof therapy.
`
`MATERIALS AND METHODS
`
`Data Collection
`
`Data were obtained from the Practice Partner
`Research Network (PPRNet), which is headquartered
`at
`the Medical University of South Carolina,
`Charleston, South Carolina. PPRNet
`is a practice-
`based learning and research organization among US
`ambulatory primary care practices that use a common
`electronic medical record (Practice Partner, Physician
`Micro Systems, Inc., Seattle, Washington). Practices
`pooled longitudinal data concerning diagnoses,
`laboratory studies, medications, vital signs, and other in-
`formation quarterly for research and quality-improvement
`
`*Trademark:
`
`Lipitor® (Pfizer
`
`Laboratories, Groton,
`
`Connecticut).
`!Trademark: Pravachol® (Bristol-Myers Squibb Company, New
`York, New York).
`+Trademark: Zocor® (Merck & Co., Inc., Rahway, New Jersey).
`
`December 2006
`
`2107
`
`3
`
`
`
`Clinical Therapeutics
`
`activities. Using data from the PPRNet,!’ we extract-
`ed data from all patients who had a diagnosis of
`hypercholesterolemia entered into the diagnosis field
`of the electronic medical record from practices active
`from 1998 through 2004. During this time frame,
`88 community-based primary care practices from
`33 states were represented.
`We obtained national and local advertising infor-
`mation from Competitive Media Reporting, which col-
`lects data on media advertising for all products,
`including pharmaceuticals, at the market
`(eg, city)
`level. The data are specific to the brand nameof the
`product and contain information on which products
`were advertised and how manydollars were spent on
`advertising on both national and local television each
`month. We used thousands of dollars in advertising
`spending by month, summedacross the 3 drug brands,
`as our measure of DTC advertising (thus, we estimat-
`ed only drug-classlevel effects, and did not attempt to
`identify the impact of DTC advertising of the individu-
`al brands separately). The DTC advertising effect is
`measured in terms of dollars per month to capture the
`differential productivity of advertisements in some
`markets in generating visits—which would translate
`into more expensive advertising time per minute.
`Patients and physician practices were assigned to the
`nearest local media market (by mileage to the Metro-
`politan Statistical Area center). We eliminated prac-
`tices that were >100 miles from the geographic center
`of the nearest media market. DTC advertising was
`measured at
`the time (month) at which a patient
`began his/her individual spell of treatment with a
`statin drug. Since the researchcited earlier found simi-
`lar effects between current month DTC advertising
`and measures of lagged month or a stock (eg, several
`months’ advertising added together) of DTC advertis-
`ing, we did not include lagged or stock measures of
`advertising in our models.'3
`Following Donohueet al,!! we created a dichoto-
`mous measure of DTC advertising intensity. We creat-
`ed an indicator variable that equaled 1 if the begin-
`ning of the statin use occurred during a month when
`DTC advertising was in the upper 25th percentile of
`expenditures. For local advertising this indicator vari-
`able corresponded to monthly spending of >$7900 on
`advertising of all 3 statins for which data were avail-
`able. For national advertising, this indicator variable
`corresponded to a monthly spending of >$7,494,900.
`While only 3 statins—atorvastatin, pravastatin, and
`
`simvastatin—had significant DTC advertising during
`the time frame of our study, we analyzed data from all
`patients who received a prescription for any statin for
`the treatment of hypercholesterolemia.
`Theinclusioncriteria for patients in our sample were
`that they must have had an indication in the clinical
`database for hypercholesterolemia and that they must
`have begun therapy with any statin drug. We also es-
`timated a version of the model on a subsample defined
`as all such patients whose prescription duration was
`at least 180 days. In addition, we excluded any patients
`who did not have a cholesterol
`laboratory test on
`record after they began statin therapy. This resulted in
`a sample of 50,741 patients.
`The nature of treatment of dyslipidemias in the
`United States must drive the specific empiric imple-
`mentation of the theoretical framework discussed ear-
`lier. Clinical managementofelevated blood cholesterol
`levels has evolved over the years as evidence has been
`generated from randomized drug trials,
`long panel
`studies in defined populations, and evaluation of ret-
`rospective data sets. The National Cholesterol Edu-
`cation Program (NCEP) periodically conducts expert
`panel assessments of the evidence and makes recom-
`mendations to physicians regarding treatment pro-
`cesses and blood cholesterol
`targets. As mentioned
`earlier, the most recent such guidelines—Third Report
`of the NCEP Expert Panel on Detection, Evaluation,
`and Treatment of High Blood Cholesterol in Adults
`(Adult Treatment Panel [ATP] III) guidelines—were
`published by the National Heart, Lung, and Blood
`Institute in 2001.2° These guidelines set bands for
`what would be considered optimal, borderline, high,
`and very high levels of blood cholesterol—which is
`best measured as the level of LDL-C (mg/dL). These
`guidelines represent thresholds, or targets, such that
`therapies will be adjusted until the target threshold
`is met.
`Thus, while one might be tempted to model the im-
`pact of statin treatment (and the derived effect of
`DTC advertising on the outcome) in terms of changes
`in measured LDL-C, the resulting estimator would be
`biased. To see why, consider the treatment process
`using statins for the treatment of high cholesterol.
`One characteristic of these drugs is that the effect, in
`terms of LDL-C reductions, largely depends on the
`dose of the drug used (and so it is limited by the pa-
`tient’s tolerance for adverse effects). In general, clini-
`cians prescribe the lowest starting dose that
`they
`
`2108
`
`Volume 28 Number 12
`
`4
`
`
`
`believe can achieve the goals, and retest the patient.If
`the goal is not met on retest, then the dose is increased
`until
`the target LDL-C level
`is met. Now, consider
`2 hypothetical patients. Assume the first patient pre-
`sents with 2 risk factors for ischemic heart disease, but
`has not yet been diagnosed with the condition, and has
`an LDL-C level of 150 mg/dL. The ATP III guidelines
`call for a target LDL-C level of <130 mg/dL—so the
`patient begins statin therapy and achieves the goal after
`a 20-mg/dL decrease in the LDL-C level. A second pa-
`tient presents with an LDL-C level of 200 mg/dL. This
`patient’s dosage is
`titrated until she achieves goal
`(<130 mg/dL). Although both patients have achieved
`the recommendedtreatment goal, one has doneso after
`achieving a 20-mg/dL decrease in blood LDL-C level,
`while the other has doneso after achieving a 70-mg/dL
`decrease. Both the 20-mg/dL decrease and the 70-mg/dL
`decrease in LDL-C values achieved—in one meaning-
`ful sense—the desired clinical outcome.
`Howthen should one model this process? In essence,
`there are 2 separate questions: “Whateffect does DTC
`advertising have on reducing blood LDL-C levels, ir-
`respective of whether clinical targets are met?” and
`“What effect does DTC advertising have on helping
`patients achieve LDL-C clinical goals?” While both
`are important (since any significant reduction in LDL-C
`levels is thought to have clinical benefit), it is the latter
`question that most directly drives the clinical decision-
`making, and so drives the process that generates the
`data we observe. Consequently,
`for
`this research,
`we focused on whether the ATP III treatment thresh-
`old goals are met. Evidence-based LDL-C goals are
`defined in the ATP III guidelines.!® We extracted all
`relevant clinical information from the PPRNet data
`(with the exception of smoking status and family his-
`tory of premature cardiovascular disease [CVD], since
`these 2 factors were not available in our data). Table I
`summarizes our adaptation of the ATP III guidelines
`to determine LDL-C goals. Then for each patient, we
`extracted the LDL-C laboratory result that was mea-
`sured closest to the date 6 monthsafter thefirst statin-
`prescription date. Patients were defined as being at
`goal if their follow-up LDL-C level was at or below
`those levels listed in Table I for that patient.
`
`Statistical Analysis
`Data consisted of observations on 50,741 individu-
`al patients from 88 different physician practices. Un-
`observable physician or practice characteristics may
`
`W.D. Bradford et al.
`
`Table |. Defining low-density lipoprotein cholesterol
`(LDL-C) goals.'?
`
`Risk Factors
`
`Hypertension,
`HDL-C <40 mg/dL,
`Age >44 (Men)
`or >54 (Women),
`Diagnosed
`with COPD
`(Smoking Proxy)
`
`O or]
`
`2or3
`
`Any
`
`LDL-C Goal,
`me/dL
`
`<160
`
`<130
`
`<100
`
`Diagnosed
`CVD
`
`None
`
`None
`
`All
`
`HDL-C = high-density lipoprotein cholesterol; COPD =
`chronic obstructive pulmonary disease; CVD = cardiovascular
`disease.
`
`affect the degree to which patients adopt or adhere to
`statin therapy. Thus, we corrected for clustering (re-
`peat observations by physician practice) in the data
`for all regression models presented below. The data
`were analyzed using STATA 9.0 (STATA Corporation,
`College Station, Texas).
`We modeled the effect of DTC advertising by ex-
`amining the impact of high television advertisement
`spending on the likelihood that patients whoinitiated
`any statin therapy achieve their ATP IIT LDL-C goals
`within 6 months. In this case, high DTC advertising
`was defined as advertising that occurred during a
`month (national) or location/month (local) that corre-
`sponded to the upper 25th percentile of DTC spend-
`ing in our data. Postinitiation LDL-C laboratory values
`were measured as the laboratory values taken nearest
`the date of initiation plus 6 months, with the excep-
`tions that: (1) the LDL-C level must have been mea-
`sured at least 45 days after beginning therapy; and
`(2) the LDL-C test must have occurred no more than
`1 year after beginning therapy. Patients who did not
`have a posttreatment LDL-C measurement conform-
`ing to these restrictions were excluded from the analy-
`sis. To evaluate the effect of DTC advertising on goal
`attainment, we estimated 2 logistic models to predict
`the dichotomous outcome variable (at goal = 1 if
`blood LDL-C level was below the ATP III goals, and
`goal = 0 otherwise). The first model included a con-
`
`December 2006
`
`2109
`
`5
`
`
`
`Clinical Therapeutics
`
`stant term, clinical risk-adjusters, and the indicator
`variables for high DTC local and national television
`advertising during the month in which the patient ini-
`tiated therapy. The second model included those vari-
`ables as well as physician practice fixed effects (and
`excluded 21 patients whose physician practices had
`too few patient observations to support the fixed-effect
`estimation). We also estimated the 2 models separately
`across the subset of 33,047 patients who maintained
`statin therapy for at least 6 months. All models are ad-
`justed for clustering at the practice level, and are esti-
`mated with Huber/White heteroskedasticity-corrected
`errors using the “robust” option in STATA.
`One concern that we addressed prior to estimating
`the models was how to represent the effect of time
`in the process of achieving LDL-C blood level goals.
`Certainly, the medical profession has paid increasing
`attention to the need to control LDL-C levels as evi-
`dence has mounted aboutthe risks associated with ele-
`vated blood LDL-C levels. In addition, clinical guide-
`lines support evidence that statin use is associated
`with a range of protective effects, such that clinicians
`have become increasingly careful
`to encourage pa-
`tients to adopt daily statin therapy.?° This increased
`attention to LDL-C control raised 2 questions.
`First, we needed to determine whether we were ob-
`serving a different type of patient population for ele-
`
`vated LDL-C with statins as time progressed. The sta-
`tistical problem that this raises is that if clinicians
`were persuading patients with relatively borderline
`LDL-C levels to begin using statins, then the likeli-
`hood of a successful outcome (blood LDL-C levels
`below those in the ATP III guidelines) could have been
`increased due simply to the fact that the average pa-
`tient had less far to go to reach his or her goals. If
`DTCalso generally increased over time, then this se-
`lection effect would lead to spurious correlation.
`Figure 1 graphs the mean LDL-C blood level prior
`to initiation of statin therapy from our sample over
`time. (Recall that all patients in the sample had begun
`therapy.) There was some apparent downward trend
`in starting LDL-C levels, which suggested that selec-
`tion effects may play a role in the process. We con-
`trolled for this by including starting blood LDL-C
`levels as a regressor. (Pretreatment LDL-C levels are
`missing from some observations since practices did
`not retroactively enter data from the paper charts
`when they adopted the electronic medical records. We
`imputed missing LDL-C levels using a multivariate re-
`gression and also included an indicator variable in the
`estimated models that equaled 1 when pretreatment
`LDL-C was imputed, and equaled 0 otherwise. The
`parameter estimates for this nuisance indicator vari-
`able are not shownin the tables.)
`
`1507
`
`140-7
`
`130
`
`120
`
`1104
`
`
`
`ProportionatLDL-CGoal
`
`Mean prestatin LDL-Clevel
`SS Mean 6-month poststatin LDL-Clevel
`
`
`
`rN
`INTE
`Vv
`
`OY
`\
`
`ot
`
`A/
`\
`SUN A Ko
`Voy vy
`
`wiry
`
`
`100—|
`01/1998
`01/2000
`01/2002
`01/2004
`
`Figure 1. Mean low-density lipoprotein cholesterol (LDL-C)levels.
`
`Date
`
`2110
`
`Volume 28 Number 12
`
`6
`
`
`
`W.D. Bradford et al.
`
`Figure 1 also sheds light on the second question
`that was raised regarding the effect of including time
`in our models. The lower line in Figure 1 graphs the
`mean LDL-C level measured posttreatment. Again,
`these measurements were the lab measurements clos-
`est in time to 6 months after initiation of statin treat-
`ment. A downward trend in posttreatment LDL-C
`levels was apparent. The implications were clearer in
`Figure 2, which graphs the percentages of patients
`whowere at goal 6 monthsafter initiating statin thera-
`py, as well as the average total (summed local and na-
`tional) DTC advertisement spending. Clearly, rates of
`LDL-C goal attainment were rising over the entire
`range of the data. In addition, the trend appears to
`have been relatively linear. Consequently, we needed
`to control for time in the logistic models. We did so by
`including “0/1” indicator variables for the year that
`statin therapy began (with 2004 being the excluded
`categorical variable).
`Finally, we needed to accommodate the fact that
`DTC advertising may have affected patients different-
`ly. Healthier patients—with the less restrictive LDL-C
`goals of <160 and <130 mg/dl—may have been more
`responsive to health messages of all types, including
`DTC advertising. If so, the impact of DTC advertising
`
`on matching therapy or adherence would have dif-
`fered across patients with different LDL-C goals. To
`test for this we included interactions between the indi-
`cator variables for initiating therapy during a high
`DTC advertising month and indicator variables for
`having LDL-C goals of <160 and <130 mg/dL.
`
`RESULTS
`Table TI lists the relevant characteristics of our sample.
`Table III presents means of LDL-C goals and goal
`attainmentfor patients in the entire sample, and those
`patients who began therapy during a high overall
`DTC advertising month (defined as being a month in
`the 75th percentile or higher of total DTC advertis-
`ment spending) and a low overall DTC advertising
`month (defined as being in the 25th percentile or lower
`of total DTC advertisement spending). Approximate-
`ly 17.4% of the sample had an LDL-C goal of
`<100 mg/dL, 39.2% had a goal of <130 mg/dL, and
`43.4% had a goal of <160 mg/dL. The groups were
`different in ways other than their LDL-C goals, as
`shown in Table II: in the group with an LDL-C goal
`of 100 mg/dL, the mean age was 66.4 years, 37.8%
`were women, and the average pretreatment LDL-C
`levels were 117 mg/dL; in the group with the LDL-C
`
`0.95
`
`0.85
`
`«CO.754
`
`6
`UO
`ES
`QO
`Zz
`8
`c
`°
`5
`
`°
`a
`
`
`
`
`
`
`Proportion at goal after 6 months
`-—----- Average monthly total ad spending
`
`> 20,000
`
`<
`L15,000 & §
`S <
`= 3
`ya &
`SS
`10,000 os
`=> ao
`26
`PR ¢.
`
`- 5000
`
`ga
`
`01/1998
`
`01/2000
`
`01/2002
`
`01/2004
`
`Date
`
`Figure 2. Percentages of patients achieving low-density lipoprotein goals and mean total direct-to-consumer
`advertisement spending.
`
`December 2006
`
`2111
`
`7
`
`
`
`Clinical Therapeutics
`
`Table Il. Characteristics of the study sample.
`
`Characteristic
`
`Age, mean, y
`
`Sex, no. (%)
`Male
`Female
`
`LDL-C Target, mg/dL
`
`<100
`(n = 8281)
`
`<130
`{n = 18,638)
`
`<160
`(n= 20,655)
`
`66.4
`
`63.4
`
`55.5
`
`All Patients
`(N = 47,574)
`
`60.6
`
`(SD, 12.33)
`
`5089 (62.2)
`3093 (37.8)
`
`9580 (51.4)
`9058 (48.6)
`
`9006 (43.6)
`11,649 (56.4)
`
`22,956 (48.3)
`22,902 (48.1)
`
`Baseline LDL-C level, mean, mg/dL
`
`117
`
`131
`
`143
`
`133.4 (SD, 39.41)
`
`Comorbidities, no. (%)
`Hypertension
`Diabetes
`Coronary artery disease
`COPD
`
`No. (%) of patients using HMG-CoA drug
`Pravastatin
`Atorvastatin
`Simvastatin
`Other*
`
`DTC television advertising
`Treatmentinitiated at high local and
`metropolitan DTC advertising, no. (%)
`Treatmentinitiated at high national
`DTC advertising, no. (%)
`Year treatment wasstarted
`1998
`1999
`2000
`2001
`2002
`2003
`2004
`
`-
`-
`-
`-
`
`-
`-
`-
`-
`
`-
`-
`-
`-
`
`24,689 (51.9)
`12,392 (26.0)
`7040 (14.8)
`1878 (3.9)
`
`982 (12.0)
`3666 (44.8)
`2635 (32.2)
`908 (11.1)
`
`4529 (24.3)
`2665 (14.3)
`6001 (32.2)
`2069 (11.1)
`
`2396 (11.6)
`10,782 (52.2)
`4771 (23.1)
`2706 (13.1)
`
`7907 (16.6)
`17,113 (36.10)
`13,407 (28.2)
`5683 (11.9)
`
`-
`=
`7
`-
`-
`-
`-
`
`13,655 (28.7)
`
`12,868 (27.0)
`
`2023 (4.3)
`2257 (4.7)
`3066 (6.4)
`6738 (14.2)
`11,102 (23.3)
`14,638 (30.8)
`7750 (16.3)
`
`LDL-C = low-density lipoprotein cholesterol; COPD = chronic obstructive pulmonary disease; HMG-CoA = 3-hydroxy-3-methyl-
`glutaryl coenzyme A reductase inhibitor (“statin”); DTC = direct-to-consumer.
`*Included cerivastatin, fluvastatin, and lovastatin.
`
`goal of <130 mg/dL, the mean age was 63.4 years,
`48.6% were women, and the average pretreatment
`LDL-C levels were 131 mg/dL; and in the group with an
`LDL-C goal <160 mg/dL, the mean age was 55.5 years,
`56.4% were women, and the average pretreatment
`LDL-C levels were 143 mg/dL. (Note that in the least-
`restrictive LDL-C grouping, the average patient had a
`pretreatment LDL-C level below goal—sothatthere-
`
`gression parameters reported subsequently must be in-
`terpreted in part as associations between the explana-
`tory variables and maintaining LDL-C goals in atleast
`some of the patients.) Finally, the relative usage of
`each statin wasrelatively constant across the 3 groups,
`with a slightly higher use of simvastatin among the
`most-restrictive target group (<100 mg/dL) than the
`2 less-restrictive groups. As may be expected, patients
`
`2112
`
`Volume 28 Number 12
`
`8
`
`
`
`W.D. Bradford et al.
`
`Table Ill. Rates of low-density lipoprotein cholesterol (LDL-C) target achievement within 12 weeks
`of treatment with a 3-hydroxy-3-methylglutaryl coenzyme A reductase inhibitor
`(“statin”), by LDL-C goal.
`
`LDL Target, mg/dL
`
`<100
`
`<130
`
`<160
`
`Characteristic
`
`(n = 8281)
`
`(n = 18,638)
`
`(n = 20,655)
`
`Achieved target, no. (%)
`High DTC advertising exposure
`no. at target
`Low DTC advertising exposure
`no. at target
`
`DTC = direct-to-consumer.
`
`4771 (57.6)
`
`15,181 (81.5)
`
`18,953 (91.8)
`
`1214 (14.7)
`
`3707 (19.9)
`
`4692 (22.7)
`
`859 (10.4)
`
`2233 (12.0)
`
`2487 (12.0)
`
`whose goals were higher (ie, easier to attain) were
`more likely to achieve those goals.
`We found preliminary evidence that DTC advertis-
`ing had an