Objective: To examine the effects of statin cost-sharing (ie,copayments, coinsurance) on adherence to statin medications andthe impact of adherence on healthcare utilization and spending.
Study Design: Retrospective, observational study of statin usersreceiving health benefits or supplemental coverage from employer-sponsoredhealth plans.
Methods: Medical and pharmacy claims were selected from theMedstat MarketScan database for patients who were continuouslyenrolled from 2000 through 2003. Two-stage residual inclusionmodels were estimated. The first stage modeled adherence tostatins, which was derived from the medication possession ratio,and represented the percentage of days on therapy in an 18-month time frame, July 2001 through December 2002. In thesecond stage, generalized linear models were used to estimate2003 utilization and expenditures. Separate estimates were producedfor new statin users (n = 24 113) and continuing statin users(n = 93 253).
Results: Lower statin copayments were associated with higherlevels of statin adherence. In percentage terms, when holding allother variables at their mean value, a $10 increase in copaymentresulted in a 1.8 percentage point reduction in the probability ofadherence for new users and a 3 percentage point reduction in theprobability of adherence for continuing users. For continuing usersadherent to statins, total costs did not change, but fewer negativeevents (emergency department visits, hospitalizations, and coronaryheart disease-related hospitalizations) occurred.
Conclusions: Policy makers and plan managers should considerinterventions that improve adherence to statins, such as lowercopayments.
(Am J Manag Care. 2006;12:SP11-SP19)
The direct medical costs and indirect costs dueto lost productivity from coronary heart disease(CHD) in the United States are estimated toexceed $142 billion in 2006.1 One of the primary riskfactors for CHD is high blood cholesterol, and more than130 million adults in the United States have borderline-highor high total blood cholesterol levels.2 In addition,more than 30% of American adults have high low-densitylipoprotein cholesterol (LDL-C) levels, undesirablecholesterol that is associated with a higher risk of CHD.
The 3-hydroxy-3-methylglutaryl coenzyme A (HMGCoA)reductase inhibitors (statins) are a well-acceptedmedical treatment that reduce LDL blood cholesterollevels. Evidence from clinical trials such as theScandinavian Simvastatin Survival Study,3 the Long-Term Intervention with Pravastatin in IschaemicDisease study,4 the West of Scotland Coronary PreventionStudy,5 and the Air Force/Texas CoronaryAtherosclerosis Prevention Study6 reveal that adherenceto statin medications is associated with lower cholesterollevels and lower levels of CHD morbidity and mortality.3-7These studies also reveal that the clinical benefits ofstatin use increase with the amount of time on statintherapy.3-6 Even so, adherence to statin medications istypically low,8,9 with adherence rates rapidly declininginitially and continuing to decline throughout the firstfew years after initiation of statin therapy.8,10
Adherence to statin medications has been determinedto be related to a variety of factors includingsociodemographic characteristics, CHD risk, andcomorbidities.8,11-15 Recent evidence has demonstratedthat patient financial incentives (ie, copayments,coinsurance) also affect statin adherence.12-15 As statincost-sharing levels increase, adherence to statins falls.Patient cost-sharing also has been demonstrated to be afinancial barrier to the utilization of other classes ofmedication that are typically used to treat chronic disease.16,17 However, the relationship between cost-sharingand adherence to medications has not beenconsistent across all medication classes.16
Prescription drug cost-sharing levels (ie, copayments,coinsurance) for enrollees in employer-based healthplans have increased steadily over time.18 In addition,prescription drug cost-sharing trends show that copaymentsand coinsurance are still rising. In a recent KaiserFamily Foundation/Health Research and EducationalTrust survey of employer-sponsored health benefits,almost half of the firms (45%) responded that they werevery or somewhat likely to increase cost-sharing for prescriptiondrugs in the next year.18
In this study we first analyzed the relationshipbetween statin cost-sharing levels and adherence tostatin medications. Then, we analyzed the relationshipbetween statin adherence and healthcare utilization andexpenditures. This study extends our previous research,which used a panel data framework to examine the relationshipbetween monthly statin cost-sharing andmonthly statin adherence.13 In this study we used apatient-level model linking cost-sharing to adherence.Then, unlike our previous study, we examined the associationbetween adherence and healthcare utilizationand spending. To our knowledge, no study has concurrentlyexamined the relationship between statin cost-sharing,adherence, and healthcare utilization/spending.
Several previous studies focused on the relationshipbetween cost-sharing and adherence to statins.12-15,19These studies show that there are at least modestdeclines in statin adherence as statin copayments rise.For example, doubling copayments from $10 to $20 ledto a drop in annual compliance of 6 to 10 percentagepoints for all cholesterol-lowering therapies.14
Two studies examined the relationship between cost-sharingand measures of healthcare utilization oroutcomes. In the first study, Schultz and colleaguesexamined the healthcare experience of 21 229 newstatin users who were enrolled in 23 fee-for-service independentpractice association health plans and who werealso at high risk for CHD.15 Although adherence to statintherapy in the first 90 days after initiation of therapywas positively associated with reaching target levels ofLDL-C, the average statin copayment, which was measuredduring the same time period, was not associatedwith LDL-C goal attainment.
Perhaps the most comprehensive analysis so far wasreported by Goldman and colleagues, who studied thehealthcare experience of 62 744 adults with employer-sponsoredhealth benefits in 88 health plans who initiatedcholesterol-lowering therapy.14 The authors reportedthat higher levels of cost-sharing were associated with areduction in compliance with cholesterol-lowering therapy.The authors also found that compliance with cholesterol-lowering therapy (medication possession ratio[MPR] ≥ 80%) in the past (up to 4 years of experiencewere measured) was associated with a reduction in theannual rate of hospitalization (all cause and circulatory)and in emergency department (ED) visits (all cause andcirculatory). Although Goldman and colleagues estimatedthe financial effects of changes in copaymentlevels based on the changes in utilization, the impact ofvariations in cost-sharing on actual spending was notexamined.
Finally, 1 study analyzed the relationship betweenadherence and utilization and spending, but did notexamine the effects of cost-sharing. In this study of 2981privately insured patients diagnosed with and treatedfor high cholesterol, Sokol and colleagues found that asadherence to cholesterol-lowering medications rose,medical costs declined and the annual risk of hospitalizationdecreased.20 As adherence rose, prescriptiondrug spending also rose and total costs (prescriptiondrug and medical) declined, although neither of theserelationships were statistically significant.
Our article adds to the literature in several ways.First, we studied the relationships among copayments,adherence, and utilization/spending within a singlestudy, which to our knowledge is a more comprehensiveapproach than has been used in the past. Second, wemeasured actual, not imputed, utilization and expendituresto determine the effects on direct medical costsand utilization patterns. Third, separate estimates wereproduced for 2 distinct groups of statin users, new(statin naïve) users and continuing users, whereas otherstudies focused on new users or combined all types ofstatin users. Finally, we used 2-stage residual inclusionmodels to estimate these relationships and produce consistentestimates in the nonlinear utilization and spendingmodels.
This study was based on data from the 2000-2003Medstat MarketScan database, which represents thehealthcare experience of at least 6 million enrollees withemployer-sponsored health insurance each year. Thedatabase also includes Medicare-eligible retirees withemployer-sponsored supplemental medical benefitsand reflects Medicare utilization in addition to theemployer-sponsored portion. Medical and prescriptiondrug claims were linked to patient enrollment informationto create the analytic dataset.
A retrospective study was conducted among statinusers who were aged 18 years or older and did not havean indication of pregnancy (which can affect prescriptiondrug use) during the study time frame. Patients alsowere continuously enrolled from 2000 through 2003.Because statin utilization patterns can differ markedlybetween incident users of statins and prevalent usersof statins,13 patients were categorized into 2 groups: (1)new/incident statin users whose initial statin prescriptionoccurred in the first 6 months of 2001 and who hadno statin prescription fills in the prior year (n =24 113), and (2) continuing/prevalent statin userswho had a first observed prescription fill in 2000 andalso had at least 1 statin fill during the study timeframe (n = 93 253).
We analyzed the relationship between statin cost-sharing,adherence, and utilization/spending by firstestimating the relationship between copayments andadherence to statin medications (controlling forcovariates) during an 18-month time frame, which wasJuly 2001 through December 2002. Statin adherencewas measured over the 18-month time period toassess adherence behavior over time. We then estimatedthe relationship between statin adherence andutilization/spending (controlling for covariates) in thesubsequent year, 2003. This framework allowed examinationof the relationship between statin cost-sharingand utilization/spending primarily through the mechanismof adherence to statin medications.
We anticipated that cost-sharing would reduce adherenceto statin medications. We also anticipated thatpatients with higher levels of statin adherence wouldexperience reductions in potentially adverse eventssuch as ED visits, hospital admissions, and CHD hospitalizations.Because patients who are adherent to statinmedications fill a larger number of prescriptions, weanticipated that higher levels of statin adherence wouldbe associated with larger prescription drug expenditures.We also anticipated that higher levels of adherencewould be associated with reductions in medicalspending (due to reductions in adverse events). Weexamined whether the net spending effect, the sum ofmedical and drug spending, was positive or negative. Wealso examined these relationships separately for newand continuing users of statins to determine whetheruse and spending patterns differ for these 2 patientgroups.
Adherence to statin medications is measured as theMPR for statins, which is expressed as a percentage ofdays with statins on hand during the 18-month timeperiod (July 2001 through December 2002). Informationfrom the prescription drug claims, such as the prescriptionfill date and number of days supplied, was usedto calculate the MPR. If a patient refilled the same statinbefore the end of the previous statin prescription fill,then the days supplied for the new prescription wasappended to the end of the previous fill. Because concurrentuse of 2 statins is generally not indicated, if apatient switched statins the remainder of the first statinwas discarded and coverage commenced with the supplyof the new statin.
Patients were classified as adherent to statins if theMPR was greater than or equal to 80%. Clinical benefitsare most likely to occur when this threshold is exceeded.5,7,10,14,15 The MPR was calculated from the utilizationof statin medications and did not include switches toother cholesterol-lowering therapies, which may haveunderestimated the level of adherence to all cholesterol-loweringmedications. Use of other lipid-lowering agentsoccurred infrequently, in 5% to 10% of nonadherentpatients.
Utilization measures for 2003 included physicianoffice visits, ED visits, hospitalizations, and CHD-relatedhospitalizations. Each was coded as an indicator variablewith a value of 1 if the event occurred and with avalue of 0 if the event did not occur. In addition, 2003outpatient prescription drug (mail order and retail pharmacy)spending and medical (inpatient and outpatient)spending were summarized. Expenditures representedthe total amount reimbursed to all providers of carefrom all sources including the health plan, patient out-of-pocket payments, and other insurance (eg, Medicare,Coordination of Benefits). All expenditures were inflatedto 2003 dollars using the Consumer Price Index,Medical Care.
The key explanatory variable was the statin cost-sharingamount (copayment or coinsurance) that thepatient faces when a statin prescription is filled, and wasmeasured as the amount per day supplied. Cost-sharinglargely consisted of copayments. Fewer than 5% ofpatients were subject to coinsurance.
Other explanatory variables associated with adherenceto medications were included in the models.Sociodemographic characteristics consisted of patientsex, age, census region, and urban residence. Medianincome and the percentage of college graduates in thepatient's area of residence (by ZIP code) from the USCensus files also were included. Income was representedcategorically by dividing the distribution of income inthe study sample into thirds: low income (<$25 000),medium income ($25 000-$49 930) and high income(>$49 930).
An indicator for employee status (vs spouse/dependent)was included, along with an array of variables indicatingthe patient's health plan type (eg, HMO, preferredprovider organization [PPO]). An indicator for a visit toa related specialist (cardiologist, endocrinologist, ornephrologist) in the past year was included to accountfor disease severity and possible differences in practicepatterns. CHD risks and secondary prevention wereindicated by the presence of the following CHD diagnosesor procedures in the past year: acute myocardialinfarction, angina, coronary artery bypass graft, chronicischemic heart disease (IHD), coronary atherosclerosis,other IHD, or percutaneous transluminal coronaryangioplasty. Other common comorbidities also wereincluded as covariates (eg, hypertension, depression).
Because the number of concurrent medicationscould influence compliance, the number of unique medicationsfilled in the past year was counted and wasincluded as a covariate. Any mail order use in the pastyear, which can be associated with higher levels ofadherence,13 was also indicated. Finally, outpatientphysician visit cost-sharing has been associated withreductions in the use of prescription drugs21 and wasincluded in the models.
A 2-stage model was used to estimate the relationshipbetween statin cost-sharing adherence and utilization/spending.
First Stage: Cost-sharing and Adherence.
A logisticregression model was estimated for statin adherence(MPR ≥ 80%) in the 18-month time period from July2001 through December 2002. Explanatory variablesincluded sociodemographic variables, employee status,health plan type, physician specialist visit, CHD diseaseprevalence, comorbidities, medication variables (numberof medications, mail order use), and prescription drugand physician cost-sharing.
Second Stage: Adherence and Utilization/Expenditures.
In the second stage, logit models were used toestimate 2003 utilization measures (hospitalization andED visits). Generalized linear models with a gammadistribution and a log link were used to estimate 2003expenditures (total, medical, and prescription drug). Inall models, explanatory variables included sociodemographicvariables, employee status, health plan type,physician specialist visit, CHD disease prevalence, andcomorbidities. Actual patient adherence and the residualfrom the first-stage model also were included in thesecond-stage model. The residual was included to produceconsistent estimates in the nonlinear second stageand to correct for any unobservable confounding in thesecond stage.22 Cost-sharing and medication variables(number of medications, mail order use) served asinstruments in the first-stage equation and were notincluded in the second stage.
The characteristics of the study patients are presentedin Table 1. New users of statins were almost equallydistributed among men and women, were 58.8 years ofage on average, and were most likely to live in the South.More than three quarters of patients lived in urbanareas, median income in the patient's area of residenceaveraged about $46 000, and the percentage of collegegraduates in the patient's area of residence averaged justunder 25%. New statin users were much more likely(69.7%) to be an employee than a spouse or dependentand were most likely to be enrolled in a PPO plan.Patients filled prescriptions for about 7 different medicationsin the prior year, and about 40% had used mailorder pharmacy in the past year. CHD prevalence waslow among new users. More than two thirds (72.7%) hadno CHD diagnosis in the past year. On average, thestatin cost-sharing amount was $12.90 for a 30-day supply($0.43/day), and outpatient copayments were about$6 per visit.
Only 28% of new users were adherent during the studytime frame (July 2001-December 2002), a finding that isconsistent with a rapid decline in adherence for newusers.8 In terms of 2003 use and spending, total expenditureswere approximately $5800 per person and almostall new users saw a physician (89.9%). About one fourthof the new users visited the ED, but hospital admissionswere less frequent (11.9%) and CHD-related hospitalizationsoccurred even less frequently (4.2% of patients).
Continuing users of statins were more likely to bemale (55.6%), were 64.1 years of age on average, andwere most likely to live in the South. More than threequarters of patients lived in urban areas, median incomein the patient's area of residence averaged more than$48 000, and the percentage of college graduates in thepatient's area of residence average just over 25%.Continuing statin users were much more likely (71.4%)to be an employee than a spouse or dependent and weremost likely to be enrolled in a capitated point-of-serviceplan. Patients filled prescriptions for about 8.8 differentmedications in the prior year, and almost 60% hadused mail order pharmacy in the past year. Slightlyfewer than two thirds (64.1%) had no CHD diagnosis inthe past year. On average, statin cost-sharing amountswere $12.00 for a 30-day supply ($0.40/day), and outpatientcopayments were about $5 per visit.
Almost two thirds of continuing users (59.1%)were adherent during the study time frame (July 2001-December 2002). Total expenditures were approximately$6600 per person and almost all continuing users sawa physician (93%) in 2003. About one fourth of the continuingusers visited the ED, but hospital admissionswere less frequent (14.2%) and CHD-related hospitalizationsoccurred even less frequently (5.4% of patients) in2003.
Table 2 presents results from the first-stage logit estimationfor statin adherence. Notably, as statin cost-sharingincreases, statin adherence decreases. A $10increase in statin cost-sharingwas associatedwith an 8.9% decrease(< .01) in the odds ofbeing adherent for newusers, and was associatedwith an 11.9%decrease (< .01) inthe odds of beingadherent for continuingusers. Note thatwhile the reduction inadherence was statisticallysignificant fornew and continuingusers, the size of thereduction in adherencewas not statisticallydifferent for newand continuing users.
In percentage terms,when holding all othervariables at their meanvalue, a $10 increasein cost-sharing resultedin a 1.8 percentagepoint reduction in theprobability of adherencefor new users anda 3 percentage pointreduction in the probabilityof adherencefor continuing users.Figure 1 and Figure 2show the relationshipbetween statin cost-sharingamounts rangingfrom $0 to $50 for a30-day supply and theprobability of adherencefor new (Figure 1)and continuing (Figure2) users.
Consistent withprevious studies, theeffects of selected explanatoryvariables onadherence were in theexpected direction. Forexample, female sexwas associated withlower levels of adherenceto statins. As ageincreased toward 65 years, adherence alsoincreased, but as age exceeded 65 years, adherencedeclined. Patients residing in middle-and highincomeareas and areas with higher college graduationrates were more adherent to statins. Thepresence of a CHD diagnosis or procedure in thepast year was typically associated with higherstatin adherence, except for the diagnosis of otherIHD, which was not associated with adherence. Asthe number of different medications increased,adherence declined slightly, but if patients hadused mail order service in the past year for anyprescription, then they were much more likely tobe adherent to statins. Plan type had an effect onadherence, with the highest adherence levels forpatients enrolled in capitated point-of-serviceplans and the lowest adherence levels for patientsenrolled in noncapitated point-of-service plans.
The relationship between adherence and expendituresis detailed in Table 3. For continuing users,prescription drug spending was higher for adherentpatients. There was a nonsignificant offset in medicalspending associated with adherence, and theeffect on total spending was not statistically significant.For new users, as adherence improved, prescriptiondrug spending increased, medicalspending decreased, and the total effect (totalexpenditures) also was negative, although none ofthese relationships were statistically significant.
Continuing users who were adherent to statins had alower likelihood of the 3 types of potentially adverseevents: ED visits, hospitalizations, and CHD hospitalizations(Table 4). However, the likelihood of a physicianvisit was not associated with adherence to statins in thecontinuing user group.
New users who were adherent to statins had a lowerlikelihood of CHD hospitalizations in 2003; however,unlike continuing users, adherence was not associatedwith the likelihood of an ED visit or a hospitalization. Incontrast to continuing users, adherent new users had amuch higher likelihood of a physician visit.
Examination of the coefficients of the first-stageresiduals (Tables 3 and 4) revealed that in about half ofthe models, the coefficient of the residual term was notstatistically significant, indicating that the estimateswould have been consistent without use of the 2-stageresidual inclusion method. In the other half of themodels, the estimated coefficient was significant,and it needed to be included in the models to produceconsistent estimates. In these cases, the residual termcorrected for unobservable confounding in the second-stagemodel.
Figure 3 and Figure 4 display the simulated relationshipbetween statin adherence and the predicted probabilityof an ED visit, a hospitalization, or a CHD-relatedhospitalization. As predicted adherence improves, thelikelihood of a potentially adverse event declines.
DISCUSSION AND CONCLUSION
In this large cohort of statin users enrolled inemployer-sponsored plans, patient copayments forstatins were a financial barrier to statin adherence. Forcontinuing/prevalent users of statins, statin adherencewas related to higher prescription drug expendituresand a nonsignificant offset in medical expenditures. Fornew/incident users of statins, statin adherence was notassociated with changes in medical or prescription drugexpenditures.
Although there was no net effect on direct medicalpayments, reductions in utilization were clearly evidentfor adherent continuing users in terms of ED visits,inpatient hospitalizations, and CHD-related hospitalizations. Because these events can be an indication ofadverse events, improved adherence can lead to betteroutcomes. If the indirect costs associated with lowerlevels of adverse events were included in the financialcalculations (eg, absence and short-term disabilitycosts), then the net financial effects might have resultedin savings.
Adherent continuing users had significantly higherprescription drug expenditures, which were undoubtedlyassociated with higher expenditures for statin medication.Perhaps, in addition, patients who are adherent to1 medication for a chronic illness tend to be more adherentto other medications, which could result in evenhigher prescription drug spending. Even so, higher levelsof adherence did not result in higher net direct costs.
Effect sizes in this study were calculated at thepatient level. However, if the effects are evaluated at theemployer or the plan level, across hundreds or thousandsof enrollees, then improvements in adherencethat may seem small at the patient level would be multipliedacross the membership, and the cumulativeeffects could be significant.
This study has several limitations. First, it was basedon administrative data, so we were unable to determinea patient's actual statin consumption patterns. Usingadministrative claims to measure adherence, weassumed that prescription filling behavior was correlatedwith prescription drug consumption patterns. Also,clinical data—including information on side effects, diseaseseverity, or the degree of cholesterol elevation—were not included. In the somewhat unlikely event thatpatients would discontinue statin use after returning tonormal cholesterol levels and manage their conditionvia diet and exercise during the study time frame, thenthese patients would be classified as nonadherent. Ifthese patients were actually adherent and adherencetypically produced better utilization andspending patterns, then our results are biasedtoward zero.
Other explanatory variables such as racealso were not available. In addition, our studyfocused on a well-insured population ofpatients with employer-sponsored healthcarebenefits and employer-sponsored supplementalMedicare insurance. In patient populationswhere the percentage of income spent onmedical care and prescription drugs is higher,the effects of cost-sharing on adherence arelikely to be larger than those presented herein.
Also, many clinical studies show that thebenefits of statins are most pronounced afterat least 2 years of therapy. In this case, wemeasured adherence in an 18-month timeframe and use and spending in the subsequentyear. If these time periods were extended,then the benefits (both financial andclinical) might have become more apparent inboth the new and continuing-user groups.
There was some concern that selectionbias might occur in the adherence and theutilization/spending models. First, in theadherence models, if more adherent patientsselected plans with lower prescription drugcost-sharing levels, then the cost-sharingeffects would be biased upward. However,examination of the benefit structures of theemployers indicated that although plan structures mightdiffer markedly for different classes of employees, manyemployees were presented with plans that had littleto no variation in prescription drug cost-sharingacross the available medical plans. Second, in the utilization/spending models, sicker patients might havebeen less adherent to medications, which would bias theadherence effects upward. However, measuring adherencebehavior before utilization and spending helped toaddress these concerns. More importantly, use of the 2-stage residual inclusion model helped to alleviate theseconcerns by revealing that in about half of the modelsthere was no evidence of unobservable confounding,and in the other half the correction (inclusion of theresidual term) aided in mitigating these effects and producedconsistent estimates.
The results suggest that interventions to improvestatin adherence, including reductions in patient cost-sharing,23 will improve outcomes among statin users,especially those who are continuing on statin therapy.The clinical literature reveals that long-term statintreatment reaps the greatest benefits to patients. Plans,policy makers, and employers should consider implementationof interventions targeted to improve statinadherence levels and to maintain high statin adherencelevels as health promotion and health improvementmeasures.
From Thomson Medstat, Ann Arbor, Mich (TBG, OB), and Washington, DC (TLM); andPfizer Global Pharmaceuticals, New York, NY (KA, DAR, KAM).
Funding for this manuscript was provided by Pfizer Global Pharmaceuticals. Three ofthe authors (KA, DAR, KAM) are Pfizer researchers.
Previous versions of these results were presented at the American College of Cardiology55th Annual Scientific Session, Atlanta, Ga, March 11-14, 2006; and the InternationalSociety for Pharmacoeconomics and Outcomes Research 11th Annual InternationalMeeting, Philadelphia, Pa, May 20-24, 2006.
Address correspondence to: Teresa B. Gibson, PhD, Thomson Medstat, 777 E EisenhowerPkwy, Ann Arbor, MI 48108. E-mail: firstname.lastname@example.org.
1. Thom T, Haase N, Rosamond W, et al. Heart disease and stroke statistics—2006update. 2006;113:e85-e151.
National Health and Nutrition Examination
2. National Center for Health Statistics. Hyattsville, MD: National Center for Health Statistics; 1999-2002.
3. Scandinavian Simvastatin Survival Study Group. Randomised trial of cholesterollowering in 4444 patients with coronary heart disease: the Scandinavian SimvastatinSurvival Study. 1994;344:1383-1389.
N Engl J Med.
4. The Long-Term Intervention with Pravastatin in Ischaemic Disease (LIPID)Study Group. Prevention of cardiovascular events and death with pravastatin inpatients with coronary heart disease and a broad range of initial cholesterol levels.1998;339:1349-1357.
N Engl J Med.
5. Shepherd J, Cobbe SM, Ford I, et al. Prevention of coronary heart disease withpravastatin in men with hypercholesterolemia. 1995;333:1301-1307.
6. Downs J, Clearfield M, Weis S, et al. Primary prevention of acute coronaryevents with lovastatin in men and women with average cholesterol levels: results ofAFCAPS/TexCAPS. Air Force/Texas Coronary Atherosclerosis Prevention Study.1998;279:1615-1622.
7. Wei L, Wang J, Thompson P, Wong S, Struthers A, MacDonald T. Adherence tostatin treatment and readmission of patients after myocardial infarction: a six yearfollow up study. 2002;88:229-233.
8. Benner JS, Glynn RJ, Mogun H, Neumann PJ, Weinstein MC, Avorn J. Longtermpersistence in use of statin therapy in elderly patients. . 2002;288:455-461.
9. Perreault R, Blais L, Lamarre D, et al. Persistence and determinants of statintherapy among middle-aged patients for primary and secondary prevention. 2005;59:564-573.
10. Benner JS, Tierce JC, Ballantyne CM, et al. Follow-up lipid tests and physicianvisits are associated with improved adherence to statin therapy.2004;22(suppl 3):13-23.
Arch Intern Med.
11. Grant RW, O'Leary KM, Weilburg JB, Singer DE, Meigs JB. Impact of concurrentmedication use on statin adherence and refill persistence. 2004;164:2343-2348.
Manag Care Interface.
12. Coombs JH, Cornish L, Hiller P, Smith DG. Compliance and refill patternbehavior with HMG-CoA reductase inhibitors after acute myocardial infarction.2002;15:54-58.
Am J Manag Care.
13. Gibson TB, Mark TL, McGuigan KA, Axelsen K, Wang S. The effects of prescriptiondrug copayments on statin adherence. 2006;12:509-517.
Am J Manag Care.
14. Goldman DP, Joyce GE, Karaca-Mandic P. Varying pharmacy benefits with clinicalstatus: the case of cholesterol-lowering therapy. 2006;12:21-28.
Am J Manag Care.
15. Schultz JS, O'Donnell JC, McDonough KL, Sasane R, Meyer J. Determinants ofcompliance with statin therapy and low-density lipoprotein cholesterol goal attainmentin a managed care population. 2005;11:306-312.
Am J Manag Care.
16. Gibson TB, Ozminkowski RJ, Goetzel RZ. The effects of prescription drug cost-sharing:a review of the evidence. 2005;11:730-740.
17. Goldman DP, Joyce GE, Escarce JJ, et al. Pharmacy benefits and the use ofdrugs by the chronically ill. 2004;291:2344-2350.
2005 Annual Employer Health Benefits Survey.
18. Kaiser Family Foundation and the Health Research and Educational Trust.Washington, DC: Kaiser FamilyFoundation; September 14, 2005.
J Gen Intern Med.
19. Ellis JJ, Erickson SR, Stevenson JG, Bernstein SJ, Stiles RA, Fendrick AM.Suboptimal statin adherence and discontinuation in primary and secondary preventionpopulations: should we target patients with the most to gain? 2004;19:638-645.
20. Sokol M, McGuigan K, Verbrugge R, Epstein R. Impact of medication adherenceon hospitalization risk and healthcare cost. 2005;43:521-530.
21. Joyce GE, Escarce JJ, Solomon MD, Goldman DP. Employer drug benefit plansand spending on prescription drugs. . 2002;288:1733-1739.
Endogeneity in Nonlinear Parametric Models: A Guide for Applied
Researchers in Health Economics.
22. Terza JV. Gainesville, Fla: Department of Epidemiologyand Health Policy Research, University of Florida; 2006. Working paper.
Am J Manag Care.
23. Fendrick AM, Smith DG, Chernew ME, Shah S. A benefit-based copay for prescriptiondrugs: patient contribution based on total benefits, not drug acquisitioncost. 2001;7:861-867.