William H. Shrank, MD, MSHS

Oncology management has changed dramatically over the last decade with the approval and rapid adoption of newly targeted oral oncology medications.

 These medications have moved much cancer therapy away from the inpatient setting or outpatient infusion therapy care setting and into the home, offering patients greater convenience and flexibility of timing and location of treatment.

 Oral oncology therapy represents one of the fastest growing expenditures for payers, costing up to $8000 per month per prescription.

The movement to oral oncology medications has altered the balance of risks and adverse effects in the treatment of cancer care. These treatments are associated with a different constellation of adverse effects and drug interactions. Moreover, delivery of these medications at home has raised the possibility of nonadherence to therapy.

 Many studies have indicated that nonadherence is associated with treatment failures and downstream healthcare costs in oncology patients.

 There are also studies that indicate that high patient copayment burden is one of the main reasons for poor medication adherence, with a higher rate of discontinuation and non-adherence as the cost-sharing amount increases.

 Our research will focus on another aspect of benefit design, where designated specialty pharmacies are selected to deliver pharmaceutical care and distribution services to patients taking oral oncology medications.

Specialty pharmacies, in addition to providing basic dispensing and counseling services, use trained nurses and pharmacists to educate patients and more actively manage their care.

 Specialty pharmacies aim to reduce variability in care delivery, improve appropriate medication use and the quality of care, and reduce costs of cancer care.

However, little is known about the effect of specialty pharmacy services in the marketplace.

Accordingly, we compared the effectiveness of a specialty pharmacy program implemented by a large commercial national health plan through specialty pharmacies to improve cancer care compared with services through retail pharmacies in the same insured population.

UnitedHealthcare Pharmacy implemented a s pecialty pharmacy program for specifi c oral oncology medications for its commercial patients in August 2007. The program required the contracted specialty pharmacy to provide clinical expertise and patient education in oncology medications and comorbid conditions, a monthly proactive adherence program including refi ll reminders, adherence screenings, and interventions with the patient and physician if nonadherence was detected. Additionally, an oncology clinical management program of telephonic clinical counseling sessions was required to provide extensive patient education, assessment of disease-specifi c parameters, depression screening, pharmaceutical care interventions, and provider outreach and referral to health resources.

The specialty pharmacy program had requirements for contracted medication reimbursement rates, staff expertise, operational services, and clinical programs that were met by the contracted specialty pharmacies. The interventional adherence program included reminder calls to the patient to coordinate medication refills, with assessment during the call for medication nonadherence in the past 30 days of therapy. If nonadherence was suspected, clinical counseling with specialty-trained pharmacists was provided to address any adherence-related issues through patient education and support strategies, identifi cation of financial assistance opportunities, and/or engagement with the physician.

Consultations were conducted biweekly to monthly for the first 3 months and then approximately every 3 months thereafter while the patient was in the program. Patients were also advised to contact their specialty pharmacist with questions as needed, and this support was available 24 hours a day/7 days a week. When appropriate, the pharmacist engaged in communication with the healthcare provider about their intervention recommendations or immunosuppressive therapy clinical concerns

that were identified in the consultations.

The insurance coverage was offered through employers and consisted of both self-insured and fully insured employers. For those employer groups who enrolled in the specialty pharmacy program, 2 specialty pharmacy vendors meeting the above program requirements were designated as the sole providers of prescriptions for the specific oral oncology medications. Competitive negotiated drug rates incorporated the cost of the additional services and no additional costs were passed on to employers or patients. 

 summarizes the specialty program flow of interactions between the patients and the specialty pharmacies.

The data source was an administrative claims database that included data for approximately 14 million enrollees from United Healthcare with both medical and pharmacy benefits. The claims were de-identifi ed and compliant with the provisions of the Health Insurance Portability and Accountability Act (HIPAA) of 1996.

We included patients that had pharmacy and medical benefits through this health plan and filled 1 or more prescriptions for an oral oncology study drug between August 1, 2007, the program initiation date, and December 31, 2007, providing a 5-month index period to capture 60-day or 90-day fills through either retail or mail order. During the identifi cation period, each patient was assigned an index date (the first prescription fill date) and an oncology index drug (all study drugs listed in 

). Study patients were required to be continuously enrolled for at least 1 year before the index date (baseline period) and 1 year after (follow-up period). The first 2 prescriptions, regardless of where they were filled, were dropped for each patient, due to the existence of a transition period. Each patient was then assigned to the specialty pharmacy network or retail pharmacy network cohort. Patients who subsequently fi lled 80% or more of their oral oncology prescriptions from the 2 designated specialty pharmacies were classifi ed as specialty pharmacy patients; those filling 80% or more of their oral oncology prescriptions from other pharmacies were assigned to the retail pharmacy cohort. Those patients (8% of the study population) who did not meet either criterion were omitted. Specialty pharmacy program participation was also assessed; after applying our assignment, in 98% of employer groups, all patients within a group were assigned to either a specialty pharmacy or retail pharmacy, indicating limited selection bias. The details of sample attrition are shown in 

A retrospective matched cohort study was designed to control for differences in healthcare costs and health services utilization between the 2 groups. The primary outcome measurement was financial: the overall costs (pharmacy and medical). Secondary financial measures include total outpatient costs, total medical costs (inpatient, outpatient hospital and office, and emergency department), and oral oncology pharmacy costs. Secondary clinical resource utilization outcomes based on reimbursement structure include hospitalizations, inpatient and outpatient hospital visits, office visits, emergency department visits, cancer-specific total medical and pharmacy costs, and cancer-specific resource utilization outcomes. Cancer- related medical costs are the costs from physician and facility claims with any of the International 

Classification of Diseases, Ninth Revision, Clinical Modification

) cancer diagnosis codes 140 to 239.99. Oncologyrelated pharmaceutical costs include all oral oncology medications in the analysis as defined by selected Hierarchical Ingredient Code List (HICL) codes (eAppendix A). Additionally, we evaluated medication adherence using 3 methods: 1) the total number of monthly prescriptions filled for all of the study medications, including switches, additions, and deletions; 2) weighted medication possession ratio (MPR) methodology in 

; and 3) number of medication gaps (MGs) per patient, defi ned as a period of at least 60 days without a study medication in the post period.

In order to control for confounding variables, the 2 cohorts were balanced using propensity score matching. The propensity score was derived from a logistic regression, which was then used to construct matched samples from the 2 cohorts.

We used a 1-on-1 greedy matching technique with 3 digits matched at 0.005 to derive the propensity score matched-pair sample.25 The logistic regression used patients’ demographics; baseline costs (medical and pharmaceutical); an indicator for start time of oral oncology agent; baseline comorbidities using the Charlson Comorbidity Index

; 5 cancer drug groups; and baseline cancer complexity score based on methodology adapted from Darkow et al.

 Darkow’s methodology defines complexity by diagnoses that suggest that a cancer patient’s health will be more diffi cult to manage than the typical patient with a particular condition, either because of: a) intrinsic activity of the given condition, or b) the difficulty of managing associated complications, comorbidities, and adverse effects of the treatments used for the management of that condition. Complexity variables in our study were built using primary and secondary diagnosis codes on patients’ medical claims, and patients were classified as having either moderate or high complexity as specified in

. If none of these were present, the patient was considered to have usual complexity.

Additionally, after the matching, the following factors were compared at baseline to assess comparability of cohorts: total costs, medical costs, and pharmacy costs, resource utilization variables such as hospitalization, inpatient and outpatient hospital visits, office visits, and emergency department visits. Post hoc analyses matching on individual cancer diagnoses and cancer drugs were also conducted to validate the cancer groupings utilized (

Statistical tests on the means of outcome measures in the follow-up periods between the 2 cohorts utilizing 

 tests were conducted. Subsequently, sensitivity analyses were conducted on the same outcomes after varying follow-up times to 3, 6, and 9 months to assess whether variations in mortality rates and/or disenrollment from plans and associated cost differences were contributing to results.

The effect of the program on medication adherence, measured as prescriptions filled, weighted MPR, and number of MGs per patient, was compared using 

 <.05 was considered statistically signifi cant. SAS version 9.1 (Cary, North Carolina) was utilized for all statistical analyses. For costs, all outliers were eliminated if over 5 times the standard deviation.

A total of 4165 unique patients were identified (Figure 2) who filled 1 or more oral oncology prescriptions between August 1, 2007, and December 31, 2007. After applying continuous enrollment and prescription filling inclusion criteria, 1588 patients were identified, of whom 502 used specialty pharmacy. Subsequent propensity matching procedures yielded 464 patients in each of the specialty pharmacy and retail pharmacy cohorts with no statistically signifi cant differences at the 5% level for the 2 groups. The patient selection flow chart is shown in Figure 1, and the matching and baseline evaluations are shown in 

The comparison of follow-up costs, healthcare utilization measures, and weighted MPR between the matched specialty pharmacy and retail pharmacy cohorts is presented in 

. For the primary outcome of total healthcare costs (the sum of pharmacy, outpatient, and inpatient medical costs), there were statistically signifi cant differen ces between the 2 groups. The mean total cost per patient per year was 13% lower in the specialty pharmacy group ($84,105 vs $97,196, difference = —$13,092; 

 = .02). Similarly, mean outpatient hospital cost ($16,777 vs $28,629, difference = —$11,852; 

 <.01) was lower by 41% in the specialty pharmacy group. Total medical cost (outpatient and inpatient costs) was 25% lower in the specialty pharmacy group ($45,696 vs $61,137, difference = —$15,440; 

 <.01). The number of oral oncology prescriptions dispensed per patient in the specialty pharmacy group was higher (10.32 vs 7.78, difference = 2.53;

 <.001). The associated overall oral oncology pharmacy costs trended higher in the specialty arm, ($32,545 vs $29,654, difference = $2891; 

The number of outpatient hospital visits was significantly lower in the specialty pharmacy cohort (15.75 vs 19.66, difference = —3.90; 

 <.01), but not for inpatient hospital visits or for total days of inpatient stay. The differences in cancer-related expenditures and healthcare utilization show that cancer-related outpatient hospital visits were significantly lower for the specialty pharmacy patients (11.08 vs 13.86, difference = —2.78; 

 = .02), as were the associated outpatient hospital costs ($14,624 vs $23,606, difference = —$8982; 

 <.01). Total cancerrelated medical costs ($41,175 vs $53,554, difference = —$12,369; 

 = .01) were also lower, largely driven by the outpatient hospital costs. Additionally, 36 patients (7.76%) in the specialty group versus 57 patients (12.28%) in the retail group had codes for IV chemotherapy agents (difference = 58%; 

The weighted MPR was higher in the specialty pharmacy patients (0.657 vs 0.580; 

 <.001), indicating greater adherence to oral oncology therapy. Similarly, the mean number of MGs per patient was signifi cantly lower in the specialty pharmacy patients (0.369 vs 0.571, difference 35%; 

The sensitivity analysis using 3-, 6-, and 9-month follow-up periods of outcomes found the results to be consistent with our main findings, as shown in 

As we broadly expand coverage to patients in the United States while attempting to contain costs, it is essential to identify approaches that improve quality and reduce costs. The recent movement of cancer care to the outpatient setting and the new challenges that patients face in appropriately administering complex regimens in their own homes provides a unique opportunity to examine the effect of different models of care delivery on healthcare efficiency.

Our findings indicate that specialty pharmacy programs are associated with improved medication adherence to oncology medications and reduction in costs of select resources. Substantial increases in oral oncology prescriptions filled and several measures of adherence were better in patients who used specialty pharmacies. Specialty pharmacy programs were associated with 13% reductions in overall healthcare costs, largely driven by an over 40% reduction in outpatient hospital costs. These findings suggest that specialty pharmacy services may represent an important opportunity to improve quality and reduce healthcare costs for patients requiring therapy for cancer.

There are a number of reasons why patients in the specialty pharmacy cohort were associated with exhibiting higher medication adherence. The specialty pharmacy program aims to improve education and adherence, to deliver positive reinforcement from a clinician with interventions and follow-up as needed, and to provide directed management of expected adverse effects. Greater adherence in the specialty cohort, though resulting in higher specialty medication costs, was also most likely responsible for reduced healthcare costs. Those who did not adhere more frequently may have required intravenous (IV) chemotherapy. Approximately 58% more patients started IV chemotherapy in the retail setting than in the specialty setting. Increased use of chemotherapy likely represents an important component of the greater outpatient hospital resource utilization that drove increased overall healthcare costs. Unfortunately, due to the retrospective nature of our study, the exact reasons for the decreased use of IV chemotherapy and biological therapy cannot be elucidated. However, it has been shown that improved adherence does lead to decreased need for further add-on or a switch in treatment regimens to IV agents, which increase costs due to the fact they are administered in hospital and/or ambulatory settings.

Reduced rates of outpatient visits were also associated with the specialty pharmacy cohort. Again, this difference could be due to a variety of reasons, including better management from specialty pharmacists of adverse effects of the oral chemotherapy agents. Many of the expected adverse effects of oral oncology agents are vastly different from those of the injectable chemotherapy, and close monitoring and care of these effects lead to better adherence. It is possible that better preparing patients for the expected adverse effects of oral oncology medications and appropriate methods to manage these adverse effects at home would mean less need for additional visits to address these concerns. Other reasons may include better management of comorbid conditions such as management of hypertension. As such, the educational component of specialty pharmacy care and the continuous 24/7  team, may play an important role in improving care effi ciency. Therefore, even in the presence of increased pharmacy cost , the overall total medical costs were statistically lower for the specialty pharmacy cohort.

There are important limitations to this study. Confounding may have resulted from selection bias, as patients or employers may have self-selected into either the specialty or retail pharmacy benefi t programs. Approximately 8% of the possible sample was omitted from the analytic data set because patients did not fi ll 80% of their prescriptions at either retail or specialty pharmacies. There are several reasons why patients may have fi lled in both channels. The most likely reason is that self-insured employers changed their preferred pharmacy outlet during the course of the study period; in this case patient selection would not have infl uenced pharmacy choice. However, some patients may have not used the pharmacyselected by their insurer and would have been omitted if they filled less than 80% from either channel. After omitting that 8% of the sample, we evaluated whether patients within each employer chose the same channel. We found variability in pharmacy choice in only 2% of employers, indicating that the overwhelming majority of benefi ciaries we included in this analysis used the preferred pharmacy channel of their employer, limiting the possibility that selection bias by patients was responsible for the fi ndings here. The selection bias of employers was not explored and remains a limitation of the study.

We were unable to identify dates of death using administrative claims, and had to choose a specific time frame in which to follow patients. Our base case included only patients who were consistently enrolled in the health benefit plan for 1 year after initiation of therapy. By excluding those who did not survive the year or disenrolled within the post—follow-up period, we enriched the sample with healthier, consistently treated patients and excluded the costs associated with death for those who did not survive and/or switch providers and plans.

We would not expect higher mortality rates in the specialty pharmacy cohort, and expect that this would lead to conservative estimates of healthcare cost reductions that resulted from specialty pharmacy care. However, we addressed this limitation by conducting sensitivity analysis requiring only 3, 6, and 9 months of continuous enrollment. The results were qualitatively unchanged in each of the time periods, suggesting that our fi ndings were not influenced by censoring due to death and/or disenrollment. Furthermore, we were not able to match on duration of therapy with the oral oncology agent prior to the study period, leading to potential residual confounding. However, after the match, there was no statistical difference in the number of new starts within 3, 6, and 9 months of the baseline period, mitigating the impact of this variable. Follow-up period beyond 1 year was not within the scope of this study and death and/or disenrollment post 1 year could not infl uence results and is a limitation of the study period we chose.

There may have been confounding related to stage of disease and cancer type. Based on previous studies, some of these limitations are inherent to retrospective claims studies.

We attempted to address this issue by matching on multiple variables that served as a proxy for disease complexity, including type of cancer drug, comorbidity score, and a calculated cancer complexity disease score. Additionally, the post-hoc analysis matching on cancer diagnoses yielded similar results. We have used a comprehensive medical claims—based proxy variable for cancer complexity developed by Darkow et al in our econometric models.

 However, the key factor for complexity is the diffi culty of managing the patient and, as such, is a clinical construct. Note that complexity is not health status, nor severity, nor disease stage; instead, complexity is shorthand for the clinical challenges posed by an individual patient with a particular disease state. However, we admit that the different stage and trajectory of disease progression for different cancers can impact costs as well as outcomes, and remains a limitation of the study.

Additionally, we were not able to capture quality of life measures in this study setting, which would have provided information from a patient perspective and may have significant policy implications.

Articles

Our research focuses on the clinical and economic outcomes of patients receiving care from a payer-designated specialty pharmacy program compared with regular retail pharmacies.

Outcomes of a Specialty Pharmacy Program for Oral Oncology Medications

In our highly fragmented healthcare system, patients deal with untold complexity when managing their chronic conditions. Patients frequently take multiple medications at several different times during the day. Prescriptions for these drugs often are written by many different prescribers, and patients may fill these prescriptions at different pharmacies, often on different days.

 This complexity of care contributes to nonadherence to prescribed therapy

 and to the inefficiency and high costs of care, and may meaningfully affect health outcomes. As a result, efforts to improve the quality of care and reduce healthcare costs must account for complexity, and we must strive to create simpler systems to streamline care and to make it easier for patients to adhere to therapy. 

But first we need to understand the problem. In this issue of 

The American Journal of Pharmacy Benefits

, Nordstrom and colleagues compared the levels of synchronicity of care for 2 different preparations of erythropoiesis-stimulating agents (ESAs) and the chemotherapeutic agents that cause the anemia that these therapies ameliorate.

 The authors found that most ESAs are administered on the same visit as chemotherapy, but that one agent (darbepoetin alfa) was more likely to be synchronized with chemotherapy than the other (epoetin alfa). The authors suggest that this synchronicity of therapy offers potential benefits in terms of patient convenience and healthcare costs. 

Nordstrom and colleagues’ study is of particular interest because it is, in essence, a novel type of comparative effectiveness analysis.  Comparative effectiveness research is intended to provide a better understanding of how different treatment regimens affect outcomes by comparing their impacts on health benefits (eg, length or quality of life) and safety. However, comparative effectiveness studies also should evaluate different models of care because how care is delivered may be just as important as what is delivered. This factor is especially important because of substantial deficits in healthcare quality and the massive variability in care that patients receive in the United States. As a result we must look to comparative effectiveness evaluations to compare how different healthcare systems, policies, and structures influence outcomes. 

In the specific case of cancer, we need to better understand how to coordinate complex care, because we must treat the underlying disease while also managing a myriad of symptoms and side effects. We also need to move toward a shared understanding about reasonable goals, expectations for therapy, and futility, so that we do not needlessly administer expensive care that can do little to improve outcomes. Comparative effectiveness studies that compare different approaches to cancer management will unquestionably move us in  this direction.

But comparative effectiveness studies, regardless of whether they are evaluating specific therapies or their mode of delivery, are of value only when they provide truly important guidance about outcomes. In theory, improving the synchronization of complex treatment regimens sounds appealing, especially given the existing literature documenting  the relationship between synchronization and improved medication adherence. But in the case of cancer and ESA, do we know whether patients whose care was more synchronized are any more likely to adhere to the therapy that was advised by their physician? More importantly, do we know whether greater adherence to ESAs and synchronization with chemotherapy administration lead to improvements in clinical outcomes? 

As the authors note, during the period that this study was conducted, evidence surfaced indicating that ESAs may cause harm when administered to patients with hemoglobin levels greater than 10 g/dL. As a result, increased synchronization may well have led to the overuse of ESAs in patients for whom the medication results in more harms than benefits. Thus, comparative effectiveness research can help us establish whether it is preferable to carefully follow a patient’s blood counts and only administer ESAs when hemoglobin falls to a specified level or to administer them on a regular basis.

Studies that compare ways in which we deliver care are every bit as essential as studies that compare the effects of different medications. For comparative effectiveness studies comparing medication effects, we have a limited set of potential acceptable outcomes. Similarly, efforts to compare strategies for healthcare delivery must evaluateeffectiveness in terms of the  outcomes that we care about most. We must not blindly assume that greater synchronization of care will benefi t patients and save the healthcare system money. Such an assumption would be no different from assumptions made by pay-for-performance programs when they create incentives for changes in surrogate markers that turn out to be  associated with worse, not better, healthcare outcomes.

We must be very thoughtful about the outcomes on which we focus in comparative effectiveness research and be sure that these outcomes correspond to real benefits for our patients. Nordstrom and colleagues have described patterns in the complexity of cancer management. This study should be interpreted as just the beginning of a line of inquiry to better understand the effects of simplification on patient outcomes and to improve the systems we use to deliver that care.

Efforts to compare strategies for healthcare delivery must evaluate effectiveness in terms of the outcomes that we care about most.

Comparative Effectiveness Research: Is "Synchronicity" a Valid Outcome?

	The greater use of generic prescription medications has been widely advocated as a policy solution to rising healthcare costs.1,2 Although generics accounted for 63% of prescriptions dispensed in the United States in 2006 and the rate of generic drug use is increasing,3 there still are substantial opportunities for further improvement. By simply switching prescriptions from branded medications to molecularly identical generics, overall drug spending in the United States could be reduced by 11%.4 Moreover, the greater use of generics need not negatively impact quality and in some cases may improve it. For example, prescribing in accordance with established guidelines for the treatment of hypertension could lead to more generic drug use and substantial prescription drug cost savings (approximately 25% of total drug costs for hypertension medications) while providing higher-quality, evidence-based care.5 More extensive generic medication use also could reduce patient costs and promote medication adherence.6 Tiered formularies, in which patients pay higher copayments for brand-name medications, are widely used to create incentives for patients to use generics; 86% of Medicare beneficiaries who receive drug benefits through the Part D program and 91% of workers with employer-sponsored coverage are enrolled in tiered plans.7,8 Although implementing a tiered benefit structure leads to greater generic drug use,9,10 as do larger differences between the copayments charged for branded and generic drugs,11,12 a better understanding of patients' price sensitivity for choosing a generic over a branded medication could assist with the design of formularies that maximize generic medication use. One explanation for the underuse of generic medications may be that the current structure of tiered formularies, in which there is an average of a $13 difference in copayments between brand and generic medications,3 is insufficient to motivate some patients to use generics. Accordingly, we conducted a national mailed survey of a random sample of commercially insured adults to estimate how much patients would need to save in order to choose a generic over an equivalent brand-name medication.

	We surveyed a random sample of 2500 of the approximately 50 million Medicare and private health plan beneficiaries of a large, national pharmacy benefits manager between February and April 2007. Our survey sample included patients in any state or the District of Columbia who were between 18 and 94 years of age, and who had at least 1 pharmacy claim through their pharmacy benefits manager in the previous year. Beneficiaries were excluded if during the same 1-year period they had at least 1 pharmacy claim for a medication to treat cognitive impairment based on Generic Product Identifier (GPI) codes beginning with 6205 (eg, donepezil, rivastigmine, tacrine). We also excluded patients insured by Medicaid and patients with any pharmacy claims for antiretrovirals (GPI codes beginning with 1210) during the prior year because they may have obtained overlapping or additional prescription drug coverage. We selected our sample size to estimate descriptive statistics after conservatively predicting our response rate at 40%, with narrow confidence intervals (95% confidence intervals with a width of fewer than 5 percentage points). We mailed each beneficiary in our random sample an introductory postcard followed by a survey including a $1 cash incentive. The survey cover page identified the affiliations of the investigators and stated that the survey was to be used for research purposes. Beneficiaries who did not respond to the first survey were mailed up to 2 more surveys, and the second survey also included a $1 cash incentive. This study was approved by the Brigham and Women's Institutional Review Board. 

	We provided survey subjects with hypothetical scenarios in which they were prescribed medications to treat high cholesterol, back pain, or depression and asked about their willingness to purchase generic medications. These diagnoses were chosen because perceptions may differ based on the acuity and nature of the conditions being treated (ie, acute symptomatic, mental health, and chronic asymptomatic). For each condition, we asked respondents (1) whether they would only buy the generic medication if it were less expensive for them than the branded drug (true/false); (2) among patients who responded true to the first question, how much they would have to save per month to choose the generic over the branded drug (10 categories, ranging from $1 to >$50); (3) whether they would take the generic drug if it were free (true/false); and (4) whether they would never take the generic drug (true/false). We also collected information about patient sociodemographic characteristics (sex, ethnicity, income, education) and self-reported health to assess factors associated with patients' perceptions about generic drugs. The survey instrument was developed iteratively by the investigators and pilot-tested extensively to improve face validity. 

	We used descriptive statistics to examine characteristics of the respondents and to summarize our overall results. Multiple imputation was used to impute missing independent variables,13 and no variables were missing more than 10% of the time. Generalized estimating equations with contrasts were used to evaluate whether the proportions of patients willing to purchase generics ever, if they were free, or if they were less expensive than brand-name drugs differed by medication type. The dependent variable for each of these models was the true/false response to each question, and the only independent variable was medication type. This analysis was performed with the GENMOD procedure (SAS 9.1 software; Cary, NC) using a binominal distribution, a logit link function, and an exchangeable variance-covariance structure. Among patients who reported for all 3 classes of medications that they would only use a generic if it were less expensive than a brand-name drug (ie, among those patients who might use a generic but who were least likely to do so), we calculated each patient's price threshold for each medication based on the midpoint of the range they selected (see eAppendix available at www.ajpblive.com). For example, patients who indicated that they would need to save "$11-$15" per month in order to use a generic cholesterol medication were considered to have a price threshold of $13 for that medication. For patients who indicated that they would need to save ">$50" per month, we defined their threshold as $50. We calculated an average price threshold for all 3 medication classes and estimated the median of the average price savings that these patients would need to receive for them to use generic medications. We then evaluated whether a patient's price threshold was influenced by his or her beliefs about generics. Specifically, we assessed the average price threshold for patients who did and did not agree that (1) brand-name drugs are more effective than generic drugs and (2) generic drugs cause more side effects than brand-name drugs. Finally, we developed a linear regression model to identify whether health status, education, race/ethnicity, sex, income, and age independently predicted a patient's price savings threshold.

	Of the 2500 beneficiaries who were mailed surveys, 1054 responded (see Figure 1). Six responses were duplicates (patients who had responded, were mailed another survey, and responded to the next survey as well). We only included the first response for each respondent, with an overall response rate of 42%. An additional 298 addresses were identified as incorrect because the mailed surveys were returned to the sender. After removing incorrect addresses from the denominator, our response rate among correct addresses was 48% (1047 respondents from 2202 correct addresses surveyed). The demographic characteristics of our respondents are shown in Table 1.

	Few respondents would never buy a generic medication (Figure 2). Responses varied by the type of medication prescribed; 13.1% of respondents indicated that they would not buy a generic antidepressant at any price, 5.7% would not buy a generic cholesterol medication, and 5.9% would not buy a generic back pain medication (P <.001). Similarly, although the majority of respondents would buy a generic medication if it were less expensive than the branded drug (cholesterol medication 72%, back pain medication 68%, antidepressant 65%) (Figure 2), the proportion that would do so was smallest for antidepressants (P <.001). More than 80% of respondents would take the generic rather than a brand-name drug, regardless of indication, if the generic were free (Figure 2).

	Among patients who reported that they would be willing to use a generic medication if it were less expensive than a brand-name medication for all 3 medication classes (n = 235), the median amount they would have to save each month to choose the generic was $25.50 (interquartile range $18-$50). The distribution of per-patient average price thresholds is shown in Figure 3. Patients who agree that generics cause more side effects than brand-name medications had a slightly higher price threshold ($27.22) than patients who did not agree ($25.50), although this difference was not statistically significant (P = .32). Patients' beliefs about the effectiveness of generics did not influence their price threshold. On regression analysis, race significantly predicted a patient's savings threshold for choosing a generic. White patients needed to save $15.15 less per month than patients of other races in order to choose a generic (P <.001) (Table 2). In contrast, age, sex, education, income, or health status were not significantly associated with a patient's price threshold.

	Our survey of randomly selected adults with commercial insurance found that most patients would be willing to choose a generic medication over a comparable brandname medication if the generic medication cost less and that, on average, the patients most resistant to using generics would need to save $25.50 per month for them to choose the generic alternative. Insurers have widely implemented tiered formularies with reduced costs for generic use with the goal of increasing patient demand for generic products. These benefit structures have been shown to effectively stimulate generic use. For example, Huskamp et al found that the introduction of a 3-tier formulary by 2 large employers resulted in 49% of statin users switching to lower-tier drugs.9 In addition, the amount of copayment difference between generics and preferred brand-name drugs predicts the proportion of generics that patients use. For example, Mager and Cox showed that plan sponsors with a $21 and higher copayment differential had a generic fill rate that was 5.2% higher than those plan sponsors with a $0 to $5 copayment differential.14 Yet at present, the average difference between copayments that US patients pay for generic and preferred brand-name drugs is $13,3 which is substantially lower than the monthly savings patients in our survey reported as needing to save in order to preferentially choose a generic. Accordingly, our results suggest that efforts to increase the proportion of generic medication used may be aided by increasing the copayments associated with preferred brand-name drugs or decreasing them for generics. Because we found that more than 80% of patients would choose a generic if it were free, eliminating cost-sharing entirely for generics, which currently averages $11,15 may be the more effective of these 2 strategies. We expect that eliminating copayments for generic medications would lead to improvements in adherence. Such observations were made by Pitney Bowes16 and Chernew and colleagues17 in conjunction with policies that eliminated or reduced copayments for medications used to treat chronic diseases such as diabetes, hypertension, and coronary artery disease. Further, we suggest that the savings from averted clinical events associated with greater medication adherence may more than offset the increased costs that insurers and employers may face by offering generics for free. Unfortunately, this hypothesis remains inadequately evaluated. In our survey, white patients required a smaller price differential than nonwhites for purchasing generics. Minority patients have been shown to trust the healthcare system less than white patients,18 and their resistance to choosing a generic may be a reflection of greater distrust of generic medications. Paradoxically, elderly black patients also are more likely to report cost-related drug underuse.19,20 These factors combined may contribute to the well-documented disparities in prescription drug use. Of course, our sample was predominantly white, and additional qualitative study of the relationship between patient ethnicity and preferences for generics would be informative. There are several limitations to our analysis. First, we surveyed only commercially insured patients and cannot generalize our findings to patients who are uninsured or who are covered by Medicaid or the Veterans Affairs healthcare system. Second, fewer than 50% of subjects responded to our survey. Respondents may have differed systematically from nonrespondents, and we did not test for systematic differences between respondents and nonrespondents. Therefore, our results may not be generalizable to all Medicare and private health plan beneficiaries. Third, we assessed patients' willingness to purchase generics on the basis of hypothetical scenarios in which they have been prescribed medications to treat high cholesterol, back pain, and depression. As such, it is possible that their responses may differ from their actual medicationtaking behavior. Additionally, we randomly sampled medication users to increase generalizability and did not enrich our sample with patients using medications to treat the conditions we studied. As a result, a small proportion of respondents were taking medications to treat these conditions, and we could not assess the proportion of patients who responded to each hypothetical scenario who actually used medications for those conditions. Actual use may have altered some patients' perceptions about generics. Finally, because our intention was to assess patients' willingness to accept therapeutically equivalent generic medications, we did not specify whether respondents were to assume that the brand and generic medications they were choosing between were necessarily molecularly identical. This distinction may have influenced the results, as patients may be more likely to accept molecularly identical substitutes. That said, therapeutic interchange is a widely used strategy. Thus our results are, at the least, conservative estimates of patients' willingness to use generics under these circumstances. Despite these limitations, we believe this is the first study to explicitly survey patient price sensitivity for generic medications. Our findings indicate that current pharmacy benefit designs may not require sufficient cost-sharing differentials between generic and branded medications to stimulate some patients to select generic medications. If we believe that generics are equally effective and offer greater value than molecularly identical branded medications, increasing that difference may effectively change behavior. Reducing cost-sharing requirements for generic medications may offer the most sensible solution; such a strategy may reduce overall prescription drug costs, improve adherence to essential chronic medications, and improve the management of chronic disease.

The greater use of generic prescription medications has been widely advocated as a policy solution to rising healthcare costs.

Setting Prices for Generic Medications: A Survey of Patient's Perceptions

, Ganguli and Hong contributed a provocative piece evaluating the relationship between initiating newly approved antihypertensive medications and outcomes.

 As expected, newer medications were associated with increased medication costs. More interestingly, use of new medications also was associated with greater overall healthcare costs. Although some important limitations in this research must be acknowledged, these results contribute to a growing body of literature that warns us of both the safety risks and cost implications of adopting new medication therapies for diseases that already have highly effective options. The main policy question we must grapple with now is this—how should insurers develop and implement formulary restrictions for new medications that will promote cost-effective and safe prescription drug use without restricting effective and necessary care?

Ganguli and Hong evaluated medication-related and non—medication-related healthcare costs captured in the Medical Expenditure Panel Survey in patients with hypertension who received newly approved antihypertension medications and medications that were not newly approved. Only 63 patients (5.8% of the sample) were prescribed newly approved antihypertensives. After controlling for sociodemographic characteristics as well as comorbidity in multivariate analyses, the total healthcare expenditures of new drug adopters were found to be 1.63 (P = .019) times and the nondrug healthcare expenditures 1.79 (P = .013) times higher than the expenditures of non–new drug adopters.

As the authors acknowledge, they used a very blunt measure of comorbidity—the number of unique medications used&mdash;and were unable to fully account for the severity of patients’ hypertension, their response to other medications, or their previous healthcare spending. As a result, some of the differences they found may have been due to unmeasured confounding. However, the results are consistent with a multitude of studies that show that patients frequently are not prescribed guideline-recommended hypertension therapies. These patients often receive more expensive treatments, many of which have been shown to be less effective in head-to-head trials, leading to unnecessary healthcare spending without clinical benefit.

Not only are our healthcare dollars at risk. A perusal of the lay press reveals repeated evidence of medications that were newly approved to great fanfare and broadly adopted, only to find later that they caused unexpected harm (eg, rofecoxib, rosiglitazone) or had disappointing efficacy (eg, ezetimibe). The US Food and Drug Administration (FDA) approval requires demonstration of efficacy compared only with a placebo control rather than the standard of care, and efficacy may be demonstrated by using surrogate outcomes. The data used to gain FDA approval are generally uninterpretable for those who hope to compare the efficacy of newly approved therapies with that of medications that have been proven to work. Moreover, the trials needed for approval are not of sufficient size or scope to evaluate unexpected risk. Such data usually become evident only after widespread marketing of the drug.

Many countries have implemented policies that require consideration of cost, safety, and comparative efficacy before making drugs available to their citizens.

 In the United States, approval to market a drug is based only on efficacy, and we rely on the marketplace to encourage use of the most effective and cost-effective medications. For the majority of patients who have prescription drug insurance, this responsibility lies squarely on the shoulders of those who create formularies. The costs and safety risks inherent in covering newly approved medications are well known to all pharmacy and therapeutics (P&T) committees that determine coverage. In the case of hypertension, virtually all insurers have implemented some financial incentive to initiate therapy with a generic thiazide, angiotensin-converting enzyme inhibitor, or beta-blocker.

 Many have implemented administrative barriers such as prior authorization or step therapy to reduce unnecessary use of angiotensin II receptor blockers or other brand-only options.

 Some insurers educate prescribers about evidence-based care.

 However, without more formidable coverage restrictions, unnecessarily costly and potentially unsafe medications are commonly prescribed despite the availability of safe, effective, and less costly options.

Ganguli and Hong call for more pharmacoeconomic studies to support coverage decisions. We all can agree that this would be helpful. But these data are rarely available when a drug first comes to market, because approval is based on ungeneralizable and imperfect data.

What can P&T committees do in the absence of better outcome and economic data? They must lead—and restrict coverage in the absence of convincing data that indicate a new medication offers meaningful benefits over existing options. They must require manufacturers to provide the necessary data to make thoughtful decisions about efficacy and safety before covering medications on any tier. They must abandon the myth that educated prescribers and consumers will consistently make informed decisions about the most effective, safe, or cost-effective medications. They must make these decisions for their beneficiaries to protect them from harm, to reduce their costs and the costs to the healthcare system, and to enhance the likelihood that patients will be able to afford and adhere to their medications.

 This approach must include an education component for prescribers and consumers (preferably developed in collaboration with prescribers) to instill trust in the mechanisms and metrics used in such evaluations, and to quell any misunderstanding that this approach is solely a financially motivated one.

Making the argument for greater formulary restrictions coherently and clearly can be challenging, especially in the face of compelling stories from sympathetic patients who believe they have been denied access to needed treatments. More strict guidelines for formulary development could be generated by consensus, under the auspices of the Academy of Managed Care Pharmacy or another convening body, and adopted broadly. Insurers should identify clear principles that will be adhered to in formulary development. The recipe is not a new one. Medications that have been shown to be superior to existing therapies, offer a meaningful additive benefit to existing therapy, or are more safe for even a select patient population will be granted formulary coverage. New drugs must be compared with active comparators when other effective options exist. Surrogate outcomes are not sufficient for broad formulary coverage. Active postmarketing surveillance data are essential and must be updated regularly to maximize safety. Pharmacoeconomic analyses are needed to support placement on formularies and levels of coverage once medications have been demonstrated safe and effective. None of these concepts are novel—yet they often are not required by P&T committees.

As the health reform debate evolves, it seems increasingly likely that market forces and private insurers will be relied on to stem the rising costs of healthcare. Doing so will require hard decisions and forming a relationship of trust with prescribers and patients. Limiting coverage of new medications will be unpopular to a vocal group of patients and providers who will argue that limitations are based on profit motives and disregard for patient health. Insurers must take control of this argument publicly and educate the skeptics to demonstrate that limiting coverage for new medications would raise the bar for the quality of trials and evidence that manufacturers would need to bring a medication to market. These actions would help protect patients from potential harm and limit unnecessary costs to patients and to the health system. 

The path of least resistance will lead to continued broad formulary coverage, use of steeper financial incentives to influence medication choice, and continued prayer that those financial incentives will be stronger than the marketing acumen of the manufacturers selling their newest medications—a strategy that has led to our current predicament. The most effective fix will require leadership from P&T committees in the form of more rigorous formulary restrictions that stimulate manufacturers to meet a higher standard when bringing new therapies to market, especially when effective alternatives exist. Such an approach would meaningfully reduce costs and improve safety, resulting in health and financial benefits to both individual patients and the healthcare system as a whole.

P&T committees should implement more rigorous formulary restrictions that stimulate manufacturers to meet a higher standard when bringing new therapies to market.