Get Connected:


Impact of Patient Cost Sharing on Multiple Sclerosis Treatment

This study evaluates the impact of patient cost sharing on the probability of receiving disease-modifying therapies and treatment compliance among patients with multiple sclerosis.
Published Online: Nov 13,2012
Liisa Palmer, PhD; Safiya Abouzaid, PharmD; Nianwen Shi, PhD; Robert Fowler, MS; Greg Lenhart, MS; Homa Dastani, PhD; and Edward Kim, MD, MBA
Objectives: To evaluate the impact of patient cost sharing on the probability of receiving disease-modifying therapies (DMTs) and treatment compliance among patients with multiple sclerosis (MS).

Study Design: Retrospective claims-based study using Truven Health Analytics MarketScan Commercial and Medicare databases.

Methods: Adult patients with MS with or without DMTs during January 1, 2004, to December 31, 2009, were identifi ed. Patients were assigned to low and high cost-sharing cohorts based on plan-level cost sharing for DMTs. Median cost sharing for DMTs ($29), standardized to 2010 US dollars, was used to determine the threshold between “low” and “high.” The probability of receiving DMTs was assessed. The medication possession ratio (MPR) and risk of treatment discontinuation were evaluated during a 12-month follow-up period. A sensitivity analysis was conducted to evaluate the same outcomes for patients with cost sharing <$100, $101 to $200, and >$200.

Results: A total of 14,497 treated (low cost sharing [n = 6954, 48.0%]; high cost sharing [n = 7543, 52.0%]), and 10,200 untreated patients (low cost-sharing [n = 4957, 48.6%]; high cost sharing [n = 5243, 51.4%]) were identifi ed. Compared with the low cost-sharing cohort, the high cost-sharing cohort had significantly lower odds of receiving DMTs (odds ratio [OR] = 0.79, 95% confi dence interval [CI]: 0.74-0.84, P <.0001), lower odds of treatment adherence (MPR >80%) (OR = 0.87, 95% CI: 0.81-0.94, P =.0006), and higher risk of treatment discontinuation (hazard ratio = 1.08, 95% CI: 1.02-1.15, P =.0060). Sensitivity analysis confi rmed that higher cost-sharing was generally associated with worse outcomes.

Conclusions: Higher cost sharing tends to result in suboptimal adherence or patients being left untreated. Such benefit designs may increase health resource utilization as well as the risk of relapse and progressive disability.

(Am J Pharm Benefits. 2012;4(Special Issue):SP28-SP36)

This study looked at the impact of patient copays and co-insurance on the probability of receiving disease-modifying therapies (DMTs) and on being adherent with treatment in patients with multiple sclerosis. The study results showed that:

  •  Patients are sensitive to cost sharing. High cost sharing results in lower probability of receiving DMTs, lower probability of treatment adherence, and higher risk of treatment discontinuation.

  •  Patients who are not adherent to their DMTs have higher risk of relapse, more emergency department visits and hospitalizations, and higher medical costs.

  •  Reducing cost barriers may improve initiation of DMTs and treatment adherence.
Multiple sclerosis (MS), a chronic, disabling autoimmune disease of the central nervous system, affects approximately 400,000 people—mostly young adults from 20 to 40 years old in the United States.1 Currently the predominant treatment for MS is disease-modifying therapies (DMTs), which slow down disease progression and reduce relapses. Since 1993, 7 DMTs have been approved by the US Food and Drug Administration (FDA) to treat relapsing forms of MS. The 2008 Expert Opinion Paper issued by the National Multiple Sclerosis Society recommends treatment with a DMT once a definite diagnosis of MS is reached. Experts agree that except for certain circumstances (such as lack of benefits, serious side effects, or better treatment options), treatment should continue indefi nitely to avoid increased risk for recurrent disease activity; however, 43% of patients with relapsing-remitting MS (RRMS) are not receiving DMTs.2,3

Suboptimal treatment adherence is common and associated with a higher risk of relapse, emergency department (ED) visits, hospitalizations, and medical costs.4,5 Tan et al5 found that only 59.6% of the study patients had a medication possession ratio (MPR) of 80% or greater for DMTs. In another study examining treatment adherence, average MPR for patients on interferon-beta therapy varied from 72% to 76%, with fewer than 41% of patients with an MPR of 85% or greater.4

Although many factors such as patient medical history, disease stage, physician and patient preference, and access to treatment can infl uence the likelihood of receiving DMTs and achieving treatment adherence, the role of patient out-of-pocket (OOP) costs on medication adherence has garnered interest in recent years. Against the national backdrop of rising healthcare costs, patients are facing increased financial burden for medications because of higher copayments, multitier drug formularies, and the adoption of pharmacy coinsurance. Unlike other chronic maintenance medications, which may have cheaper generic substitutes, DMTs are included under the more expensive specialty drug category; this leaves patients with MS few alternatives. Campbell et al6 examined the annual patient OOP spending on MS medications and found that the median cost increased from $832 in 2002 to $1042 in 2005, with the top 10% of patients incurring a 20% increase from $2598 in 2002 to $3230 in 2005. An evaluation of DMT use among privately insured patients with MS estimated through simulation that increasing drug copayment by 50% would lower the probability of receiving DMTs by 32.9%.7 Recently published studies have shed more light on the impact of cost sharing on treatment abandonment rate, MPR, and persistence with DMTs, finding a negative association between high cost sharing and treatment outcome.8-10 In these studies, patient cost sharing was measured based on either copayment of the first prescription,8 total OOP amount during 6 months preceding the evaluation period,9 or mean copayment per 30 days during the evaluation period.10 Although these approaches refl ect the actual OOP costs, they do not include evaluation of patients who chose not to receive treatment because of cost-sharing burden.

Using a method developed and employed for other conditions and therapeutic agents,11-13 the present study used expected patient cost sharing to characterize the patient’s financial burden with regard to DMTs. Cost sharing for DMTs was measured using a price index calculated at the healthplan level, based on DMT prescriptions filled by all members enrolled in the same plan during the same calendar year. This amount represents the costs that patients would face when deciding on DMT and compliance. This method has the advantage of placing a price tag on DMTs even if the patient declines treatment, thereby allowing analysis of how cost sharing infl uences a patient’s decision on whether to receive DMTs and subsequently adhere to the regimen.

Data Source

This retrospective observational study identifi ed patients from the Truven Health Analytics MarketScan Commercial and Medicare Supplemental and Coordination of Benefits (COB) Databases. These databases contain enrollment details (eligibility, health plan type) as well as medical and pharmacy claims with fi nancial information (provider reimbursement, member copayment, coinsurance, and deductible). Patient data used in this analysis were deidentified in compliance with the Health Insurance Portability and Accountability Act regulations. Institutional Review Board review and approval were not required.

Study Population

Patients diagnosed with MS between January 1, 2004, and December 31, 2009, were identifi ed based on either 1 inpatient or ED MS diagnosis (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] code 340) or 2 or more outpatient MS diagnoses that were at least 30 days apart. Diagnostic claims were not considered for study inclusion. Patients were classified as treated or untreated based on evidence of nonintravenous DMTs (interferon beta-1a, interferon beta-1b, glatiramer acetate) during 2004 to 2009. For treated patients, the fill date for the first DMT was the index date. For untreated patients, the index date was randomly assigned to mimic the distribution of time between the first observed MS diagnosis and the DMTs among the treated cohort. To remain eligible for analysis, patients were required to be 18 years or older at index and have 12 months of continuous enrollment with pharmacy data availability before (preperiod) and after (postperiod) the index date. Patients with any claim for intravenous DMTs or with only medical claims for DMTs during the postperiod were excluded from analysis. Given that plan-level cost sharing is the key independent variable in this study, only patients enrolled in plans with valid (nonmissing, nonzero) cost-sharing data for DMTs were included in the analysis (Figure 1).

Study Outcomes

We hypothesized that patients with MS facing high cost sharing would be less likely to receive DMTs than those facing low cost sharing and that treated patients with high cost sharing would have lower medication adherence and persistence relative to those with low cost sharing. We assessed the probability of receiving DMTs, the probability of treatment adherence, and the risk of treatment discontinuation for patients in the high cost-sharing plans compared with those in the low costsharing plans. Treatment adherence was measured using MPR during the 12-month postperiod, calculated as the total number of days with DMTs on hand divided by 365 days. Treatment adherence was defi ned as an MPR >80%, which was also used in the Tan et al5 study to examine the relationship between MPR and risk of relapse, healthcare utilization, and medical expenditures among MS patients. Treatment persistence was measured by the number of days between treatment initiation and treatment discontinuation. Treatment discontinuation was identifi ed based on a gap in DMT treatment of 30 days or more. Patients were characterized as nonpersistent with treatment based on the first observation of treatment discontinuation.

Explanatory Variables

Patient Cost Sharing

The primary explanatory variable—patient costsharing—was characterized as the cost burden that patients would face if they received DMTs. Cost sharing was estimated for each patient based on the health plan in which they were enrolled. Plan-level cost sharing for DMTs was assessed by calendar year (2004-2009) and payer (commercial and Medicare) for all study plans in the MarketScan Commercial and Medicare Databases. Copayment and coinsurance were standardized to costs per 30-day supply in 2010 dollars. Plan-level cost sharing was determined by: (1) copayment if a plan had a nonzero copayment value, or (2) coinsurance if a plan had a $0 copayment but a nonzero coinsurance value. Study patients received their plan-level cost-sharing value based on the unique plan identifier, the calendar year of their index date, and the payer. Patients who did not receive a cost-sharing value were excluded. The median DMT copayment amount for plans associated with study patients was $29; this value was used as the threshold to classify low versus high cost-sharing plans.

Control Variables
Demographic variables including age, gender, US Census region, urbanicity, insurance plan type, and payer were measured at the index date. Charlson Comorbidity Index (CCI) and medical conditions known to be associated or commonly reported among patients with MS were evaluated during the preperiod as proxies for disease severity. These conditions included: vascular disease, endocrine disease, musculoskeletal disease, urogenital dysfunction, visual loss, neurologic conditions, depression, pain, sleep disorders, and fatigue. Use of systemic corticosteroids, symptomatic therapies for MS (antilymphocyte globulin, methotrexate, and intravenous immune globulin), and MS-related hospitalization and ED visits were also assessed during the preperiod. We created a variable that characterized the patient’s MS and DMT status in the preperiod. This variable classified patients as: previously treated; treatment naïve/previously diagnosed; treatment naïve/newly diagnosed; or treatment naïve/previous diagnosis status unknown. Multivariate Analysis Because study outcomes can be confounded by differences in patients’ demographic and clinical characteristics, multivariate analysis was employed to adjust for imbalances between low and high cost-sharing groups. Specifically, logistic regression was conducted to assess: (1) probability of receiving DMTs in the high cost-sharing cohort compared with the low cost-sharing cohort, and (2) probability of having treatment adherence in the postperiod among treated patients with high cost sharing compared with those with low cost sharing. A Cox proportional hazards model was employed to analyze the relationship between treatment persistence and costsharing level. Demographic and clinical variables were included in these models. Newly diagnosed status was included in the model that analyzed the probability of receiving DMTs; the composite variable that characterized preperiod MS diagnosis and DMTs status was included in treatment adherence and persistence models. The proportional hazards assumption for the model covariates was assessed graphically by plotting the log of the negative log of the estimated survival function versus time. The presence of influential patients on the parameters estimates was verifi ed by plotting the score residuals versus the covariate values.

Patient Demographic and Clinical Characteristics

A total of 110,356 patients were identifi ed as having MS, with 53% characterized as treated and 47% as untreated with DMTs. After applying the study exclusion criteria, 14,497 treated patients and 10,200 untreated patients with a valid nonzero plan-level cost-sharing amount for DMTs were included in the study (Figure 1).

Patients in the high cost-sharing cohorts were generally 3 to 4 years younger than those in the low cost-sharing cohort. The majority of study patients were commercially insured; a larger number of patients with low cost sharing than high cost sharing had indemnity plans. Patients in low cost-sharing plans were more concentrated in the North Central region than those in high cost-sharing plans. The low and high cost-sharing cohorts shared similar clinical profiles, as shown in Table 1.

Compared with treated patients, untreated patients were 6.8 years older and more likely to be Medicare enrollees. A total of 42% of treated patients had DMTs during the preperiod; 39% were newly diagnosed treatment- naïve patients. Only 6% of treated patients were previously diagnosed with MS and did not receive DMTs in the preperiod. By design, all untreated patients were treatment-naïve in the preperiod; 67% of these patients were newly diagnosed.

Cost Sharing for DMTs

The majority of patients—83% and 95% of low and high cost-sharing plans, respectively—had copayment as the only cost-sharing arrangement for DMTs. Approximately 15% of the low cost-sharing cohort and 3% of the high cost-sharing cohort paid coinsurance only. Fewer than 3% of patients had both copayment and coinsurance. Seventy-nine percent of study patients enrolled in plans with monthly cost sharing <$50; 91% were in plans with cost sharing <$100. Only 6% and 2% of patients faced cost-sharing burdens ranging from $100 to $200 and $200 to $300, respectively. And 1% of patients were in plans with cost-sharing arrangements for DMT in excess of $300.

Probability of Receiving DMTs

The unadjusted probability of receiving DMTs was similar between patients with MS in the low and high cost-sharing plans (odds ratio [OR] = 1.03, 95% confi dence interval [CI] 0.98-1.08, P = .3297). Although patients in the low and high cost-sharing plans were similar clinically, there were substantial imbalances in age, region, and health plan type. As shown in Table 2, these variables were significant predictors of receiving DMTs: (1) the probability of receiving treatment decreased with increasing age, (2) the probability of receiving treatment was lower for patients with indemnity health plans relative to those with POS plans, and (3) the probability of receiving treatment was higher for patients in the South relative to those in the West region. After multivariate adjustment for the aforementioned imbalances, the odds of receiving a DMT in the high cost-sharing group were 21% lower than for the low cost-sharing cohort (OR = 0.79, 95% CI: 0.74- 0.84, P <.0001).

Treatment Adherence and Persistence

Compared with treated patients with MS in the high cost-sharing plans, patients in the low cost-sharing plans had a significantly higher unadjusted MPR for DMTs (0.82 vs 0.79; P <.0001), a lower unadjusted discontinuation rate (35% vs 37%; P = .0002), and a higher unadjusted mean number of days on persistent treatment (268 ± 141 vs 256 ± 148; P <.0001). Results of the multivariate analysis, presented in Figure 2, found that the odds of treatment adherence for the high cost-sharing cohort were 13% lower than for the low cost-sharing cohort (OR = 0.87, 95% CI: 0.81-0.94, P = .0006). The risk of treatment discontinuation was 8% higher for patients in the high cost-sharing plans than patients in the low cost-sharing plans (hazard ratio [HR] = 1.08, 95% CI: 1.02-1.15, P = .0060).

Sensitivity Analysis

To better understand the impact of the upper end of the cost-sharing ranges among our study patients, we performed a multivariate sensitivity analysis by creating 3 plan-level cost-sharing categories: <$100, $101 to $200, and >$200. Compared with patients with plan-level costsharing <$100, the odds of receiving DMTs were 37% lower for patients in plans of cost sharing >$200 and 14% lower for patients with cost sharing of $101-$200 (OR = 0.63, 95% CI: 0.54-0.74, P <.0001 and OR = 0.86, 95% CI: 0.76-0.97, P = .0129, respectively) (Figure 3). The odds of treatment adherence were 39% lower among patients with >$200 cost sharing relative to those with <$100 (OR = 0.61, 95% CI: 0.50-0.75, P <.0001). The risk of treatment discontinuation was 27% higher for the >$200 cohort than the <$100 cohort (HR = 1.27, 95% CI: 1.10-1.47, P = .0013). Compared with patients in the <$100 plans, patients in plans of $101 to $200 had less favorable adherence and persistence outcomes, but no statistically significant differences were detected.


Using pharmacy and medical claims data from 2003 to 2010, this study evaluated how patients’ expected financial burden for DMTs would influence the likelihood of receiving treatment and subsequent treatment compliance. We used the median value of plan-level copayment for DMTs to classify 14,497 treated patients and 10,200 untreated patients into plans with low and high cost sharing. Patients in high costsharing plans had lower odds of receiving DMTs than patients in low cost-sharing plans; treatment adherence was lower, and discontinuation was higher among patients in high cost-sharing plans relative to their low cost-sharing counterparts.

Negative associations between high cost sharing and treatment compliance among patients with MS have been documented in several published studies. Despite different methods of defining cost sharing and classifying cost-sharing groups, previous studies consistently suggest that the negative impact of cost sharing on treatment compliance with DMTs is most likely at the higher end of the patient OOP cost spectrum. Express Scripts, a pharmacy benefit manager, evaluated factors predicting patient compliance with specialty drug therapy including DMTs.10 Patients were classified into 5 copayment categories. Results showed that patients with the highest copayment level (>$150) had 19% higher odds of being nonpersistent than those with copayment values <$20. However, the odds were not signifi cantly different between the $51 to $150 copayment groups and the <$20 group. Dor et al9 examined MPR among patients with MS who paid copayment only (n = 987) and coinsurance only (n = 987) for DMTs. Copayment was not significantly associated with MPR because the differences in copayment across alternative adherence thresholds were small, ranging from $31 to $42 per month. However, a 10% increase in cost sharing resulted in a 9% decline in adherence in the coinsurance cohort (patients with a mean OOP of $284). Gleason et al8 analyzed the relationship between OOP for the first filled DMT prescription and the treatment abandonment rate, which was found to be significantly higher for OOP expense >$200 when compared with an OOP expense of <$100. Results from the sensitivity analyses examining the impact of the upper end of the cost sharing on treatment compliance in the present study were consistent with these findings.

Our sensitivity analyses further confirmed that higher cost sharing signifi cantly lowered the likelihood of receiving DMTs. As noted, the average age for the untreated patients was nearly 7 years older than the treated patients. In part, these large imbalances are because DMTs are only indicated for RRMS, which is more prevalent among younger patients. In addition, MS is a highly heterogeneous disease, and patients with milder disease may be less likely to receive DMT treatment, given its cost and inconvenience. Because diagnosis codes do not specify MS type, we were unable to directly assess treatment availability with respect to disease progression. However, because older people are more likely to have progressive forms of MS than young people, age as a covariate in the multivariate models served as a proxy for disease status.

Previous research indicates that poor DMT adherence is associated with a higher risk of MS-related hospitalization, ED visits, relapse, and MS-related medical costs.4,5 The consequences of delayed DMT initiation, low adherence, and poor persistence because of patient cost sharing were not assessed in this study, but it is possible that any revenue obtained through cost-sharing arrangements may be partially offset by increases in medical services and costs.

Results of this analysis should be interpreted in light of several limitations. Manufacturers offer patient assistance and copayment programs to reduce patient fi nancial burden, but these subsidized payments were not distinguishable from the actual patient OOP costs on the prescription claims. Although patient assistance programs are typically offered to patients with low income and/or those without insurance, copayment programs are offered to a broader patient population. Patients in the current analysis have commercial insurance, which often makes them ineligible for patient assistance programs. Very few study patients (0.1%) were enrolled in a high-deductible health plan, which is often considered a proxy for low income. Nonetheless, the study may underestimate the effects of benefit design on the outcomes, because the impact of copayment program enrollment cannot be assessed with available data. A multivariate technique was employed to adjust for imbalances in patient demographic and clinical characteristics using available data elements from the MarketScan databases. However, there are unobservable factors that may impact patients’ enrollment in high versus low cost-sharing plans, the choice to initiate treatment with DMTs, or treatment adherence. Some of the factors that could not be assessed include patient income, type/stage of MS, physician preferences, and distance to providers. Study results should be interpreted accordingly. This study relied on prescription claims and assumed that dispensing was equivalent to utilization of the prescription. Previous research suggests that claims-based evaluation of medication compliance is a reliable assessment method.14 The choice of using 80% MPR to define treatment adherence was somewhat arbitrary. We evaluated the relationship between cost sharing and adherence using a 70% and a 90% threshold. The relationship between copay amount and adherence was modestly sensitive to the way adherence was defi ned; however, the study conclusions were unaffected. For example, the odds ratio for adherence, comparing copay amount greater than the median versus less than the median, fluctuated from 0.835 to 0.874 to 0.905 when adherence was defi ned as MPR >70%, MPR >80%, and MPR >90%, respectively. All odds ratios were highly statistically signifi cant. A large percent of patients were excluded from the study because they did not meet the continuous enrollment criteria. The magnitude of this criterion was larger for the untreated cohort than for the treated cohort. Disenrollment from MarketScan databases is typically related to termination in employment or data contribution thresholds of MarketScan contributors. Consequently, we believe that patients excluded because of continuous enrollment were not markedly different, from a clinical perspective, from those who remained eligible for analysis.

Finally, our findings were limited to patients with commercial insurance and employer-sponsored Medicare supplemental coverage and cannot be generalized to the US population.


Our findings suggest that patients with MS are sensitive to the financial costs associated with DMTs and may forgo treatment or end treatment prematurely when facing a high cost-sharing burden. Benefit plan designs that negatively impact treatment initiation and/or adherence may increase the risk of relapse, health resource use, and disease progression. Future studies are desired to quantify clinical and economic consequence of undertreatment among patients facing high cost sharing.

Author Affiliations: From Truven Health Analytics, Washington, DC (LP, RF), Cambridge, MA (NS, GL); Novartis Pharmaceuticals Corporation (SA, HD, EK), East Hanover, NJ.

Funding Source: This study was funded by Novartis Pharmaceuticals Corporation.

Author Disclosures: Drs Palmer and Shi and Mr Fowler and Mr Lenhart received consultancies from Novartis Pharmaceuticals Corporation for conducting the study analysis and for preparing the manuscript. Dr Kim reports employment and stock ownership with Novartis Pharmaceuticals Corporation. Drs Abouzaid and Dastani report former employment with Novartis Pharmaceuticals Corporation.

Authorship Information:
Concept and design (LP, SA, NS, GL, HD, EK); acquisition of data (RF); analysis and interpretation of data (LP, SA, NS, RF, GL, HD, EK); drafting of the manuscript (LP, NS, EK); critical revision of the manuscript for important intellectual content (SA, RF, HD, EK); statistical analysis (LP, NS, RF, GL); obtaining funding (EK); administrative, technical, or logistic support (LP, NS); and supervision (LP, EK).

Address correspondence to:
Liisa Palmer, PhD, Director, Outcomes Research, Thomson Reuters, 4301 Connecticut Ave NW, Ste 330, Washington, DC 20008. E-mail:
1. National Multiple Sclerosis Society. sclerosis/what-we-know-about-ms/who-gets-ms/index.aspx. National Multiple Sclerosis Society website. Published 2011. Accessed October 21, 2011.

2. National Multiple Sclerosis Society. Expert Opinion Paper-Disease Management Consensus Statement. professionals/publications/expert-opinion-papers/index.aspx. Published 2008. Accessed October 20, 2011.

3. Margolis JM, Fowler R, Johnson BH, Kassed CA, Kahler K. Disease-modifying drug initiation patterns in commercially insured multiple sclerosis patients: a retrospective cohort study. BMC Neurol. 2011;11:122.

4. Steinberg SC, Faris RJ, Chang CF, Chan A, Tankersley MA. Impact of adherence to interferons in the treatment of multiple sclerosis: a non-experimental, retrospective, cohort study. Clin Drug Investig. 2010;30(2):89-100.

5. Tan H, Cai Q, Agarwal S, Stephenson JJ, Kamat S. Impact of adherence to disease-modifying therapies on clinical and economic outcomes among patients with multiple sclerosis. Adv Ther. 2011;28(1):51-61.

6. Campbell RS, Patel V, Gleeson M, Calimlim B, Zagari M, Curkendall SM. Outof-pocket health care expenditures and expenditure caps in an insured population with chronic conditions. Am J Pharm Ben. 2011;3(2):98-105.

7. Ozminkowski RJ, Marder WD, Hawkins K, et al. The use of disease-modifying new drugs for multiple sclerosis treatment in private-sector health plans. Clinical Therapeutics. 2004;26(8):1341-1354.

8. Gleason PP, Starner CI, Gunderson BW, Schafer JA, Sarran HS. Association of prescription abandonment with cost share for high-cost specialty pharmacy medications. J Manag Care Pharm. 2009;15(8):648-658.

9. Dor A, Lage MJ, Tarrants ML, Castelli-Haley J. Cost sharing, benefi t design, and adherence: the case of multiple sclerosis. Adv Health Econ Health Serv Res. 2010;22:175-193.

10. Express Scripts, Inc. Factors predicting patient compliance with specialty drug therapy for multiple sclerosis, infl ammatory conditions and hepatitis C. /pharmacyservices/.Published October 2006. Accessed October 2, 2011.

11. Goldman DP, Joyce GF, Escarce JJ, et al. Pharmacy benefi ts and the use of drugs by the chronically ill. JAMA. 2004;291(19):2344-2350.

12. Solomon MD, Goldman DP, Joyce GF, Escarce JJ. Cost sharing and the initiation of drug therapy for the chronically ill. Arch Intern Med. 2009;169(8):740-748; discussion 748-749.

13. Gibson TB, Jing Y, Kim E, et al. Cost-sharing effects on adherence and persistence for second-generation antipsychotics in commercially insured patients. Manag Care. 2010;19(8):40-47.

14. Steiner JF, Prochazka AV. The assessment of refi ll compliance using pharmacy records: methods, validity, and applications. J Clin Epidemiol. 1997;50(1):105-116.