Annals oflnternal Medicine
`
`l ORIGINAL RESEARCH
`
`The Epidemiology of Prescriptions Abandoned at the Pharmacy
`William H. Shrank, MD, MSHS; Niteesh K. Choudhry, MD, PhD; Michael A. Fischer, MD, MPH; Jerry Avom, MD; Mark Powell, MA, MEd;
`Sebastian Schneeweiss. MD, ScD; Joshua N. Liberrnan, PhD; Timothy Dollear. MS: Troyen A. Brennan, MD. JD: and M. Alan Brookhart. PhD
`
`Background: Picking up prescriptions is an essential but previously
`unstudied component of adherence for patients who use retail
`pharmacies. Understanding the epidemiology and correlates of pre-
`scripfion abandonment may have an important effect on health
`care quality.
`
`Objective: To evaluate the rates and correlates of prescription
`abandonment.
`
`Design: Cross-sectional cohort study.
`
`Setting: One large retail pharmacy chain and one large pharmacy
`benefits manager (PBM) in the United States.
`
`Measurements: Prescriptions bottled at the retail pharmacy chain
`between 1 July 2008 and 30 September 2008 by patients insured
`by the PBM were identified. Pharmacy data were used to identify
`medications that were bottled and either dispensed or returned to
`stock (RTS) or abandoned. Data from the PBM were used to
`identify previous or subsequent dispensing at any pharmacy. The
`first (index) prescription in a dass for each patient was assigned to
`1 of 3 mutually exclusive outcomes: filled, RTS, or RTS with fill 0n
`the 30 days after abandonment, the patient purchased a pre-
`scription for a medication in the same medication class at any
`pharmacy). Outcome rates were assessed by drug class, and
`generalized estimating equations were used to assess patient,
`
`insurance, and prescription characteristics as-
`neighborhood,
`sociated with abandonment.
`
`Results: 10 349139 index prescriptions were filled by 5 249 380
`patients. Overall, 3.27% of index prescriptions were abandoned;
`1.77% were RTS and 1.50% were RTS with fill. Patients were least
`
`likely to abandon opiate prescriptions. Prescriptions with copay-
`ments of $40 to $50 and prescriptions costing more than $50 were
`3.40 times and 4.68 times more likely,
`respectively, to be aban-
`doned than prescriptions with no copayment (P < 0.001 for both
`comparisons). New users of medications had a 2.74 tim¢5 greater
`probability of abandonment than prevalent users (P < 0.001), and
`prescriptions delivered electronically were 1.64 times more likely to
`be abandoned than those that were not electronic (P < 0.001).
`
`Limitation: The study included mainly insured patient; and ana-
`lyzed data collected during the summer months only.
`
`Conclusion: Although prescription abandonment represents a small
`component of medication nonadherence, the correlates to aban-
`donment highlight
`important opportunities
`to intervene and
`thereby improve medication taking.
`
`Primary Funding Source: CVS Caremark.
`Ann Intern Med. 2010,1531633-640.
`For author affiliations, see end of text.
`
`mannalsnm
`
`Nonadherence to essential long—term medications rep-
`resents a central public health problem (1). Numerous
`studies have demonstrated that patients do not adhere to
`medications as prescribed (2, 3), leading to excess hospital-
`izations, morbidity, mortality, and health care costs (4, 5).
`Improving adherence to essential medications has repeat-
`edly been highlighted as a public health priority (6). How-
`ever, important gaps remain in our understanding of the
`causes of nonadherence and the best ways to intervene to
`support appropriate medication use (7).
`Most adherence research is conditional on a patient
`filling a prescription for a medication, and studies tradi-
`tionally evaluate refill rates (using claims data), patient re—
`ports of subsequent medication use (using self—reported
`data), or rates of administration once a prescription has
`been filled (using electronic pill bottles) (8—10). These ex-
`isting studies of refill
`rates cannot clearly determine
`whether a patient does not adhere to therapy because he or
`she has not followed up with the provider to receive a
`prescription refill, the provider has not written the pre-
`scription, the prescription was written but not delivered to
`the pharmacy, or the prescription was delivered to the
`pharmacy but never picked up (that is, abandoned).
`Prescriptions abandoned at the pharmacy represent a
`potential opportunity to intervene and improve adherence.
`When abandoned, the prescription has been written by the
`physician and called into, faxed to, or electronically deliv—
`
`ered to the pharmacy or hand-delivered by the patient.
`Some abandoned prescriptions may never be picked up,
`representing a missed opportunity for therapy, whereas
`other prescriptions abandoned may be purchased later at
`the same pharmacy or at another pharmacy, indicating a
`delay in treatment and pharmacy inefficiency.
`Recent studies have used electronic prescribing data to
`assess rates of “primary nonadherence” (rates at which pa-
`tients do not fill prescriptions written by physicians) (11,
`12); however, we are aware of no previous studies evaluat-
`ing the rates and predictors of prescription drug abandon-
`ment at retail pharmacies for commonly prescribed medi-
`cations. To better understand the magnitude of the
`
`See also:
`
`Print
`Editors' Notes ............................. 634
`Editorial comment.......................... 680
`
`Summary for Patients ....................... I-42
`
`Web-Only
`Appendix
`Appendix Figures
`CME quiz
`Conversion of graphics into slides
`
`© 2010 American College of Physicians 633
`
`
`Argentum Pharrn. LLC v. Alcon Research, Ltd.
`Case IPR2017-01053
`
`ALCON 2081
`
`

`

`ORICINAI. RESEARCH l TheEpidemiologyofPrescriptionsAbandoned at the Pharmacy
`
`Context
`
`Failure to retrieve prescription medications at the phar-
`macy is one aspect of nonadherence to therapy.
`
`Contrlbutlon
`
`In this cross-sectional study, the percentage of prescrip-
`tions that were abandoned at the pharmacy was low.
`However, prescriptions for initial therapy, those for expen-
`sive drugs, those that required high copayments, and
`those delivered electronically were significantly more likely
`to be abandoned than others.
`
`Impllcatlon
`
`The increasing use of electronic prescribing may result in
`an increase in the number of prescriptions that patients fail
`to retrieve from the pharmacy. Physicians should be alert
`to factors associated with prescription abandonment.
`
`—The Editors
`
`problem and identify potential strategies to intervene to
`improve medication adherence, we merged a database from
`a large retail pharmacy chain with a database from a large
`pharmacy benefits manager (PBM). This merged data set
`provides a unique opportunity to assess rates and predictors
`of abandonment at a discrete point in the medication fill—
`ing process, as well as subsequent use after abandonment at
`the same or other pharmacies.
`
`data identifying the copayment charged before the prescrip-
`tion was filled (Appendix Figure 1, available at www.annals
`.org). CVS pharmacies generally return prescriptions to stock
`if they are not picked up within 14 days of delivering the
`prescription.
`Electronic pharmacy data from the retail pharmacy
`and PBM were matched on pharmacy store number, pre-
`scription number, fill date, and patient ZIP code. We suc-
`cessfiilly matched 99.93% of retail transactions with PBM
`data. Transactional data from the retail pharmacy was used
`to determine whether a prescription was returned to stock,
`because these data more accurately reflect internal processes
`of the pharmacy than data provided by the PBM.
`Pharmacy benefits manager claims from the baseline
`period, 6 months before the identification period (1 Janu-
`ary 2008 to 30 June 2008), were used to determine
`whether prescriptions filled in the identification period
`were new prescriptions. We defined “new users” as patients
`who had filled no prescriptions in the same class as the
`index prescription in the 6 months before the index. Phar-
`macy benefits manager claims from a 3-month follow-up
`period, 1 October 2008 to 31 December 2008, were used
`to assess whether patients who abandoned prescriptions at
`the pharmacy subsequently filled those prescriptions at the
`same or another pharmacy. We excluded prescriptions at
`all CVS pharmacies that had automatic refill programs
`during the study period because abandonment rates were
`artificially high in these settings.
`
`METHODS
`
`Outcomes
`
`This study was approved by the institutional review
`boards of Partners Healthcare System, Boston, Massachu—
`setts, and Harvard University, Cambridge, Massachusetts.
`Data Sources
`
`Retail pharmacy data were provided by CVS (\Woon—
`socket, Rhode Island), a large national pharmacy chain.
`Pharmacy data contain all prescriptions (regardless of in—
`surer), the mode of transmission (for example, electronic),
`and whether the script was bottled and then returned to
`stock (RTS). Insurance claims data were provided by Care—
`mark (\X/oonsocket, Rhode Island), a large national PBM.
`The PBM data encompass all claims information either
`requested by the pharmacy or reimbursed by Caremark
`and include data from all pharmacies that a patient visited.
`
`Study Period and Cohort Construction
`All prescriptions filled and either purchased by a patient
`or abandoned (referred to here as “RTS”) at CVS retail phar-
`macies were identified during a 3-month period from 1 July
`2008 to 30 September 2008 (the identification period). The
`CVS consumers who receive pharmacy benefits through Care-
`mark were then identified by matching retail
`transactional
`data to transactional data from the Caremark database.
`
`Among these individuals, all covered and filled prescriptions
`have a paid pharmacy claim in the Caremark database. Fur-
`thermore, all prescriptions that were RTS have transactional
`634 I6 November 2010 Annals ofIntemal Medicine Volume 153'Number 10
`
`For each patient, the first prescription in a class during
`the identification period was considered the index prescription
`and the date on which it was written was considered the index
`
`date. We assigned each such prescription to 1 of 3 mutually
`exclusive outcomes: 1) filled prescription, indicating that the
`patient purchased the prescription; 2) RTS, indicating that
`the patient abandoned the prescription; or 3) RTS with fill,
`indicating that the patient abandoned the prescription and it
`was returned to stock, but the patient purchased a prescription
`for a medication in the same medication class at the same or
`
`another pharmacy. To determine RTS with fill status, we
`identified all RTS prescriptions and evaluated whether the
`patient filled a prescription for any medication in the class,
`determined by the first 4 digits of the Generic Product Index
`code of the abandoned prescription, from any pharmacy in
`the 30 days after the RTS fill date. This time frame was se—
`lected to conservatively estimate the clinical effect of abandon-
`ment. For patients with more than 1 prescription in a given
`class during the identification period, we considered only the
`first of these prescriptions so that we did not assign excessive
`weight to individuals with multiple abandoned prescriptions
`in the same class. For patients whose index RTS occurred in
`the first 2 weeks of the identification period, we also consid-
`ered the prescription an RTS with fill if the patient filled a
`prescription for a medication in the same class in the previous
`m.annals.org
`
`

`

`
`
`The Epidemiology of Prescriptions Abandoned at the Pharmacy ORI GINA I. R ES EAR CH
`
`14 days, because th$e patients were probably not without
`medication at the time of the RTS.
`
`Table 1. Patient Characteristics'
`
`Characteristics of Patients and Prescriptions
`A prescription was considered new if no other pre—
`scriptions in the medication class (determined by Generic
`Product Index codes) had been filled in the 6 months be—
`
`fore the index date. For each index prescription, PBM data
`were used to identify the copayment charged, whether the
`prescription was for a generic or brand—name medication,
`whether the medication was for a chronic or acute condi—
`
`tion, and the source of insurance coverage (Medicare,
`Medicaid, employer sponsored, health plan not through an
`employer, or cash card or other). We also identified
`whether the prescription was transmitted electronically (e-
`prescribed) to the pharmacy. Additional information was
`identified at the patient level: patient age, sex, and the
`number of unique medications filled in the identification
`period (a proxy for comorbidity) (13). The ZIP code of the
`patient’s home residence was identified and linked to 2000
`census tract data to assign the median income in the ZIP
`code of residence of each patient (14). We also used census
`thresholds to determine whether each patient lived in a
`rural or an urban area, on the basis of the population den-
`sity of each ZIP code; rural neighborhood was defined as a
`population density of fewer than 1000 persons per square
`mile (15).
`
`Statistical Analysis
`We used descriptive statistics to summarize the char—
`acteristics of patients in our sample who filled prescrip—
`tions. We then assessed the proportion of prescriptions
`that were filled, RTS, and RTS with fill by medication
`class. Finally, we conducted bivariate and multivariate
`analyses to assess how patient- and prescription-level co-
`variates were associated with RTS rates, by using general-
`ized atimating equations to account for clustering at the
`patient level. Our statistical model was a generalized linear
`model with a log-link function that yielded estimates of
`relative risk. We estimated variable SE5 robustly by using
`the
`empirical variancehcovariance matrix to address
`patient-level clustering. Variables were estimated by using a
`working correlation matrix with an exchangeable structure.
`In our bivariate analyses, we assessed the association
`among medication class, copayment, and brand-name ver-
`sus generic drug on RTS probability and RTS with fill
`probability. Sensitivity analyses were conducted by exclud-
`ing all electronic prescriptions. In the multivariate analysis
`that included all variables, we sought to understand predic-
`tors of true abandonment—that is, patients who did not
`subsequently refill a prescription—and combined all RTS
`with fill prescriptions with filled prescriptions. In this man-
`ner, our dichotomous outcome was RTS versus either a
`
`filled prescription or an RTS prescription with fill. We
`conducted sensitivity analyses that included total medica-
`tion copayment burden as a covariate.
`wwwannalsmrg
`
`Characteristic
`
`Age, % (n)
`O—17y
`18—34 y
`35—49 y
`50—64 y
`265 y
`
`Sex, % (n)
`Female
`Male
`
`Urban or rural residence, % (n)
`Urban (21000 persons/miz)
`Rural (<1000 persons/miz)
`
`Insurance or payment type, % (n)
`Employer-sponsored
`Cash card/other
`Health plan
`Medicare
`Medicaid
`
`Region, % (n)
`Northeast
`West
`South
`Midwest
`Other territories
`
`Data
`
`11.8(617 041)
`14.7 (770 208)
`23.4 (1 229 463)
`29.3 (1 538 709)
`20.8 (1 092 739)
`
`60.1 (3 134 854)
`39.9 (2 079 784)
`
`68.1 (3 061 167)
`31.9 (1 435 886)
`
`59.0 (3 099 450)
`4.9 (254 336)
`24.9 (1 304 744)
`6.7 (352 018)
`4.6 (238 832)
`
`35.1 (1 830 011)
`6.9 (360 925)
`42.2 (2 198134)
`15.8 (824 603)
`0.0 (730)
`
`Median family Income In ZIP code, 5
`
`61 762.10 (25 349.90)
`
`Mean unlque prescriptions per
`patient (SD), n
`
`2.0 (1.6)
`
`‘ Based on a sample of 5 249 380 persons.
`
`All analyses were performed by using SAS software,
`version 9.2 (SAS Institute, Cary, North Carolina).
`
`Role of the Funding Source
`The work was funded by grants from CVS Caremark
`and a career development award from the National Heart
`Lung and Blood Institute to Dr. Shrank. The authors re-
`tained independent and complete control over the design
`and implementation of the study as well as the analyses and
`writing of the manuscript.
`
`RESULTS
`
`Our cohort consisted of 10 349139 index prescrip-
`tions filled by 5 249 380 patients during the identification
`period. Patients were an average of 47.3 years of age, and
`60.1% were female. They filled 2.0 unique prescriptions
`during the identification period and lived in ZIP codes
`with an average median income of $61762 (Table 1).
`Most patients had employer-sponsored insurance, yet a
`substantial number of patients were insured by Medicare,
`Medicaid, and non—employer-based health plans; approx-
`imately 4% used a cash card to receive discounted medica-
`tions, which probably indicates that they did not have pre-
`scription drug coverage.
`16 November 2010 Annals oflnternal Medicine Volume 153' Number 10 635
`
`

`

`ORICINAI. RESEARCH l TheEpidemiologyofPrescriptionsAbandoned at the Pharmacy
`
`Table 2. Rates of Prescription Flll, RTS, and RTS Wlth FIII, by Drug Class
`
`Drug Class
`
`Prescrlptlon Status
`
`Opiate
`Antihypertensive
`Antidepressant
`Statin
`Proton-pump inhibitor
`Diabetes medication
`Oral
`Insulin
`Antibiotic
`Derrnatologic agent
`Asthma medication or inhaler
`Hormone replacement therapy or oral contraceptive
`Antiepileptic
`Cough, cold, or allergy medication
`Osteoporosis medication
`Antipsychotic
`Antiplatelet or anticoagulant
`Prostate medication
`
`Fllled
`
`RTS
`
`RTS Wlth FIII'
`
`Percentage
`(95% Cl)
`
`98.2 (98.1—98.2)
`97.6 (97.5—97.6)
`97.0 (96.9—97.0)
`97.3 (97.2—97.3)
`95.6 (95.5—95.7)
`
`97.0 (96.9—97.0)
`94.9 (94.8—95.1)
`98.0 (98.0—98.1)
`94.6 (94.5—94.7)
`94.4 (94.3—94.5)
`96.9 (96.9-97.0)
`96.4 (96. 3—96.4)
`95.1 (95.0—95.3)
`96.5 (96.4—96.6)
`95.5 (95.3—95.6)
`97.8 (97.7—97.9)
`97.5 (97.4—97.7)
`
`Number
`
`671 488
`626 631
`443 230
`394 908
`250 969
`
`198 272
`62 814
`933 701
`477 415
`339 009
`326 478
`200 772
`132 529
`77 134
`71 666
`59 782
`51 659
`
`Percentage
`(95% CI)
`1.0 (1.0-1.0)
`1.1(1.1—1.1)
`1.4 (1.4—1.5)
`1.4 (1.4—1.4)
`2.6 (2.5—2.7)
`
`1.3 (1.2—1.3)
`2.2 (2.1—2.4)
`1.3 (1.3—1.3)
`3.0 (2.9—3.0)
`3.5 (3.4—3.6)
`1.3 (1 .3—13)
`1.7 (1.7—1.8)
`3.6 (3.5—3.7)
`1.7 (1.6-1.8)
`2.3 (2.2—2.5)
`1.0 (0.9—1.1)
`1.3 (1.2—1.4)
`
`Number
`
`6850
`7160
`6591
`5654
`6817
`
`2614
`1482
`12131
`15 011
`12 595
`4368
`3630
`4942
`1346
`1754
`611
`679
`
`Percentage
`(95% Cl)
`0.9 (08—09)
`1.3 (1.3—1.4)
`1.6 (1.5-1.6)
`1.3 (1.3—1.4)
`1.8 (1.8—1.9)
`
`1.8 (1.7—1.8)
`2.9 (2.7—3.0)
`0.7 (0.7—0.7)
`2.4 (2.4—2.5)
`2.1 (2.1—2.2)
`1.8 (1.8-1.8)
`1.9 (1.8—2.0)
`1.3 (1.3—1.4)
`1.8 (1.7—1.9)
`2.2 (2.1—2.3)
`12 (1.1—1.3)
`1 2 (1.1-1.3)
`
`Number
`
`5800
`8585
`7176
`5432
`4817
`
`3586
`1884
`6719
`12 249
`7551
`6032
`3956
`1824
`1446
`1665
`738
`635
`
`RTS = returned to stock.
`1' Prescription was RTS but was subsequently filled at the same or another pharmacy.
`
`index prescriptions (0.34
`Approximately 3.27% of all
`million prescriptions) were abandoned; 1.77% of those pre-
`scriptions were RTS, and no prescription was filled by the
`same patient for a medication in the same class in the
`
`Figun. Blvarlate relatlons between prescrlptlon cost or
`brand-name or generic status and rates of abandonment.
`
`i}
`
`I
`
`1}
`
`'
`
`1]-
`
`-I
`
`0—9.99
`
`10-1939
`
`‘0
`
`§ 20-2939
`E.
`3. 30—3999
`0
`U
`
`0-49 99
`.
`
`4
`
`250.00
`
`4}
`
`-I-
`
`_'_
`
`-l-
`
`l]-
`
`i
`
`fl—
`
`_D-
`
`8. Brand-name
`.3
`so
`5 Generic
`
`11
`
`i
`l—l—l—l—l—l—l—l—l—l—l
`0
`0.5
`1
`1.5
`2
`2.5
`3
`3.5
`4
`4.5
`5
`
`Prescriptions, 96
`
`El RTS
`
`I RTS with fill
`
`We controlled for clustering at the patient level by using generalized
`estimating equations. Bars indicate 95% C15. RTS = returned to stock.
`636 I6 November 2010 Annals ofIntetnal Medicine Volume 153'Number 10
`
`subsequent 30 days at any pharmacy (RTS), whereas
`1.50% were filled at some pharmacy in that time frame
`(RTS with fill).
`
`Abandonment rates varied by medication class (Table
`2). Opiates and antiplatelet medications were least likely to
`be RTS prescriptions (1.0% and 0.9%, respectively) or
`RTS with fill (0.8% and 1.1%). Antihypertensives, oral
`diabetic medications, and statins also had comparatively
`low abandonment rates. Among daily-use therapies, higher
`rates of RTS were seen for proton-pump inhibitors (2.6%),
`asthma medications (3.5%), and insulin (2.2%). Medica-
`
`tions used on an as-needed basis, such as dermatologic
`agents (RTS rate, 2.9%) and cough and cold medications
`(RTS rate, 3.6%) were also abandoned more commonly.
`In bivariate analyses,
`the copayment charged was
`strongly associated with rates of abandonment. Prescrip-
`tions with copayments of less than $10 were abandoned
`1.4% of the time, and abandonment rates increased con-
`
`sistently to 4.5% for copayments greater than $50 (Fig-
`ure). Similarly, abandonment rates were greater for brand—
`name medications than for generic medications (Figure).
`These relationships were confirmed in our multivariate
`models that included all variables being studied. When we
`compared the associations between prescription- and
`patient-level variables and true prescription abandonment
`(RTS with no subsequent fills), medication copayment was
`most strongly associated with abandonment rates. Com-
`pared with prescriptions with no copayment, prescriptions
`with copayments of $40.01 to $50.00 had a 3.40 times
`greater probability of being abandoned, and prescriptions
`costing more than $50.01 had a 4.68 times greater proba-
`bility of being abandoned (P < 0.001 for all pairwise com-
`www.mnalsnrg
`
`

`

`
`
`The Epidemiology of Prescriptions Abandoned at the Pharmacy ORI GINA I. R ES EAR CH
`
`parisons with the referent category) (Table 3). Similarly,
`median income in ZIP code of residence was significantly
`associated with abandonment rates; patients living in ZIP
`codes in the highest income quintile were 21% less likely
`to abandon prescriptions (P < 0.001). Medicaid beneficia-
`ries were 8% more likely to abandon prescriptions than
`were persons with employer-sponsored health insurance.
`Young adults aged 18 to 34 years were most likely to
`abandon prescriptions. Seniors were 45% less likely to
`abandon prescriptions than young adults (P < 0.001). Pa-
`tients with more comorbid conditions were more likely to
`abandon prescriptions; each additional unique prescription
`medication filled was associated with a 4% increase in the
`
`probability of abandonment (P < 0.001).
`New users of medications had more than 2.74 times
`
`greater probability of abandonment than prevalent users,
`and maintenance medications had slightly less probability
`of being abandoned (P < 0.001 for both comparisons). Of
`note, prescriptions delivered electronically to the pharmacy
`had a 64% increase in the probability of being abandoned
`compared with those that were not electronically delivered
`(P < 0.001).
`
`DISCUSSION
`
`To our knowledge, ours is the first study to compre-
`hensively evaluate the phenomenon of prescriptions aban-
`doned at the pharmacy. We found that 3.27% of prescrip-
`tions bottled at the pharmacy were abandoned and RTS,
`and on more than half of those occasions, the patient did
`not fill an alternate prescription for the same medication at
`any pharmacy. This represents a relatively small proportion
`of all prescriptions that are filled at pharmacies, comprising
`a small component of overall medication nonadherence or
`failure to appropriately use long-term medications. How-
`ever, the total number of abandoned prescriptions in the
`population is great, and every essential prescription aban-
`doned could represent an important clinical concern if the
`patient does not subsequently restart the medication or
`identify a substitute. Moreover, the likelihood of abandon-
`ment for patients who fill multiple medications can be
`substantial and clinically important.
`By evaluating prescription abandonment rates, we as-
`sess a discrete event in the continuum of the prescription
`drug delivery process that may represent an opportunity to
`intervene and support better medication adherence. Physi-
`cians or pharmacists should be aware of the patient and
`prescription characteristics associated with higher rates of
`abandonment to assist patients to improve medication use.
`We have created a simple prediction rule with 4 covariates
`that providers can use to rapidly assess risk and to best
`identify who may benefit most from additional counseling
`or the selection of a less expensive medication (Appendix
`and Appendix Figure 2, available at www.annals.org).
`Copayments charged to patients were the strongest
`predictors of abandonment, suggesting that patients expe-
`wwwannalsmrg
`
`Table 3. Multivariate-Adjusted Associations Between
`Patlent and Prescription Characteristics and Rates of
`Abandonment
`
`Characterlstlc
`
`Age
`18—34 y
`0—17 y
`35—49 y
`50—64 y
`265 y
`Sex
`Female
`Male
`
`Urban or rural resldence
`Urban (21000
`persons/mil)
`Rural (<1000
`persons/m?)
`
`Insurance or payment type
`Employer-sponsored
`Cash card/other
`Health plan
`Medicare
`Medicaid
`
`Income
`50—541 094
`541 095—$51 393
`551 394—563 972
`563 973—580 330
`$80 331—5200 001
`
`Number of unlque
`prescrlptlons per
`patient
`
`Copayment
`$0
`5001—51000
`51001-52000
`52001-33000
`53001-54000
`54001-35000
`2550.01
`
`Prescription dellvery
`method
`Not electronic
`Electronic
`
`New user
`No
`Yes
`
`Maintenance drug
`No
`Yes
`
`Prescrlptlons, Unadjusted Relative Rlsk
`n'
`Frequency
`(95% cm
`of RTS, %
`
`1 222 000
`954 000
`2 270 000
`3 278 000
`2 622 000
`
`6 183 000
`4 101 000
`
`5 919 000
`
`2 917 000
`
`6 000 000
`499 000
`2 424 000
`945 000
`480000
`
`1 876 000
`1 851 000
`1 790 000
`1 698 000
`1 610 000
`
`—
`
`824 000
`5 759 000
`1 435 000
`1 028 000
`239 000
`247 000
`527 000
`
`9 928 000
`421 000
`
`4 190 000
`6 159 000
`
`3 980 000
`6 369 000
`
`2.4
`2.4
`2.0
`1.5
`1.4
`
`1.8
`1.6
`
`1.7
`
`1.7
`
`1.8
`2.7
`1.6
`1.2
`2.3
`
`1.9
`1.8
`1.7
`1.7
`1.6
`
`—
`
`1.5
`1.3
`1.6
`2.0
`2.6
`3.4
`4.5
`
`1.7
`2.3
`
`0.9
`2.4
`
`1.9
`1.7
`
`1.00 (reference)
`0.98 (0.96—1.00)
`0.87 (0.86-0.89)
`0.65 (0.64—0.66)
`0.55 (0.54—0.56)
`
`1.00 (reference)
`0.88 (0.87—0.89)
`
`1.00 (reference)
`
`0.95 (0.94—0.96)
`
`1.00 (reference)
`1.01 (0.98—1.03)
`0.83 (0.81—0.84)
`0.96 (0.94-0.99)
`1.08(1.04—1.12)
`
`1.00 (reference)
`0.93 (0.92—0.95)
`0.90 (0.88-0.91)
`0.86 (0.84—0.87)
`0.79 (0.77—0.80)
`
`1.04(1.04—1.05)
`
`1.00 (reference)
`1.21 (1.17—1.25)
`1.58 (1.53—1.63)
`2.05 (1 .98—2.12)
`2.60 (2.51—2.69)
`3.40 (3.27—3.54)
`4.68 (4.53-4.84)
`
`1.00 (reference)
`1.64 (1.60—1.67)
`
`1.00 (reference)
`2.74 (2.70—2.78)
`
`1.00 (reference)
`0.97 (0.96—0.98)
`
`RTS = returned to stock.
`* Includes prescriptions that were filled, RTS, and RTS with fill.
`1' Results from a multivariate modcl that includes all variables listed in the tablc.
`The 95% CIS are based on robust SE5 that account for clustering at the patient
`level; the dichotomous outcome is RTS vs. filled prescription or RTS with fill.
`
`rience “sticker-shock” at the pharmacy and choose not to
`fill
`those prescriptions. Improved physician awareness of
`patient cost-sharing requirements, and communication
`16 November 2010 Annals ofInternal Medicine Volume 153' Number 10 637
`
`

`

`
`
`O RIG INA L R ES EA RC H The Epidemiology of Prescriptions Abandoned at the Pharmacy
`
`about those costs with patients before arrival at the phar-
`macy, may reduce abandonment rates (16). However, phy-
`sicians are often unaware of their patients’ out-of-pocket
`costs at the time of prescribing and often believe that phar-
`macists should play a central role in these discussions (17,
`18). The advent of electronic health records may facilitate
`greater awareness of and communication about medication
`costs. Moreover, if pharmacies proactively seek to commu-
`nicate with physicians to identify less costly therapies and
`thus reduce cost-sharing requirements before the patient
`attempts to purchase the medication, some benefit may
`result. Similarly, benefit designs that reduce cost-sharing
`for the most effective medications, known as “value-based
`
`insurance designs,” may reduce abandonment rates (19, 20).
`New prescriptions are almost 3 times more likely to be
`abandoned than previously filled prescriptions; particular
`care could be directed to communicating with patients
`who are filling new prescriptions to support therapy initi-
`ation. Patients with more oomorbid conditions abandoned
`
`prescriptions at higher rates than those with fewer oomor-
`bid conditions, even after patient age was controlled for. In
`sensitivity analyses controlling for total oopayment burden,
`these relationships were qualitatively unchanged. Our find-
`ings highlight the effects of increasing the complexity of a
`patient’s medication regimen and the importance of sim-
`plifying therapy when possible (21).
`Variations in abandonment rates by drug class were
`also informative. High rates of abandonment of insulin
`may suggest inefficiencies in the delivery system for these
`medications, which are not as easily packaged into monthly
`supplies as other medications and may be a source of sub-
`stantial morbidity if patients who require insulin therapy
`experience any gaps in therapy. Higher rates of abandon-
`ment of proton-pump inhibitors, an often overutilized
`medication (22), or as-needed medications for symptom-
`atic conditions may not signify as much clinical risk and may
`result from the availability of alternate, over-the-counter op-
`tions that may be more cost-effective for patients. The ex-
`tremely low rate of opiate abandonment probably indicates
`less cost-sensitivity and a grater demand for the medication
`than for other classes. Many states do not allow electronic
`prescribing of opiates; sensitivity analyses that excluded elec-
`tronic prescriptions continued to demonstrate lower rates of
`opiate abandonment than of other classes. Patient addiction
`or plans to divert opiates may also contribute to reduced rates
`of abandonment.
`
`Prescriptions delivered electronically to the pharmacy
`were almost 65% more likely to be abandoned than those
`delivered by other means. This finding is not surprising,
`because some patients who receive electronic prescriptions
`do not have to hand-deliver the prescription to the phar-
`macy or otherwise initiate the fill request themselves. Be-
`cause they lack a patient-initiated step, electronic prescrip-
`tions may be more likely to be delivered to the pharmacy
`for patients who never intended to fill the prescription.
`However, this finding highlights unintended consequences
`638 16 November 2010 Annals ofInterml Medicine Volume 153 ' Number 10
`
`of electronic prescribing. One feature of electronic pre-
`scribing that has been widely promoted is the improved
`ability to document medication lists. If these lists represent
`what was ordered and not what was filled, they may serve
`to obfiiscate rather
`than clarify therapeutic regimens.
`Moreover,
`if electronic prescriptions are contributing to
`pharmacy ineflficiency rather
`than simplification, addi-
`tional costs may result from electronic prescribing. In ur-
`ban areas, electronic prescribing may be even more com-
`plex as prescribers must identify the correct pharmacy from
`many competitors. Inaccurate electronic prescriptions, de-
`livered to the wrong pharmacy, may also lead to greater
`rates of abandonment and may explain greater abandon-
`ment rates in urban than rural regions. Pharmacies can
`choose to be more selective about filling electronic pre-
`scriptions before the patient’s arrival in order to improve
`efliciency. Physicians may choose to print a reminder for
`patients when prescriptions are electronically delivered to
`the pharmacy to help them remember that a prescription is
`waiting.
`This study also highlights an important source of un-
`necessary oost and inefficiency in the delivery of prescrip-
`tion drugs. Every prescription that is bottled and then re-
`turned to stock has an associated cost to the pharmacy,
`estimated at more than $10 per prescription (23). In total,
`more than 3.6 billion prescriptions were filled at pharma-
`cies in the United States in 2008 (24). Assuming a national
`abandonment rate similar to these findings, more than 110
`million prescriptions are abandoned at US pharmacies
`annually. Conservatively estimating the cost to the phar-
`macy of each abandoned prescription to be $5, abandoned
`prescriptions probably cost pharmacies more than half a
`billion dollars annually in the United States. These costs
`could increase substantially as more prescriptions are deliv-
`ered electronically.
`By requiring patients to have data in the PBM system,
`our study probably underrepresents patients who were un-
`insured, which would lead to conservative estimates of rates
`
`of abandonment. We included a large sample of Medicaid
`beneficiaries, who have low incomes, as well as patients
`who purchased medications with a cash card, who most
`likely did not have comprehensive coverage. However,
`those who signed up for a cash card may diflfer somehow
`from other uninsured patients, limiting generalizability to
`that population. Patients who use CVS pharmacies are
`probably representative of patients who use other large
`pharmacy chains, but they may differ from patients who
`purchase prescriptions at grocery stores or independent
`pharmacies (25).
`In addition, we cannot comment on
`abandonment at mail-order pharmacies, where rates may
`be substantially lower.
`We dichotomized prescription delivery into “elec-
`tronic prescriptions” and “all others.” Among non—
`electronically transmitted prescriptions, we did not have
`systematic information about whether prescriptions were
`phoned or faxed in by the physician. These prescriptions
`www.mnalurg
`
`

`

`The Epidemiology of Prescriptions Abandoned at the Pharmacy
`
`Original Research
`
`may have abandonment rates that