Secondary Logo

Journal Logo

Prescription fill rates for acute and chronic medications in claims-EMR linked data

Park, Yoonyoung ScDa; Yang, Hyuna PhDb; Das, Amar K. MD, PhDa; Yuen-Reed, Gigi PhDc,*

Section Editor(s): Osemene., Inyang Nora

doi: 10.1097/MD.0000000000013110
Research Article: Observational Study
Open

Nonadherence to prescribed medications poses a significant public health problem. Prescription data in electronic medical records (EMRs) linked with pharmacy claims data provides an opportunity to examine the prescription fill rates and factors associated with it.

Using a claims-EMR linked data, patients who had a prescription for either an antibiotic, antihypertensive, or antidiabetic in EMR were identified (index prescription). Prescription fill was defined as a pharmacy claim found within the 90 days following the EMR prescription. For each medication group, patient characteristics and fill rates were examined using descriptive statistics. Multivariate logistic regression was used to evaluate the association between fill rates and factors such as age, race, brand vs generic, and prior treatment during 365 days before the index date.

Among 77,996 patients with index antibiotic prescription, 78,462 with index antihypertensive prescription, and 24,013 with index antidiabetic prescription, the prescription fill rate was 73%, 74%, and 76%, respectively. Overall, African American race was negatively associated with fill rates (odds ratio [OR] 0.8 for all 3 groups). Prior treatment history was positively associated with antihypertensives (OR 5.6, 95% confidence interval [CI] 5.4–5.7) or antidiabetics (OR 4.1, CI 3.8–4.4) but negatively with antibiotics (OR 0.6, CI 0.6–0.6). Older age was an additional factor that was negatively associated with first time fill rate among patients without prior treatment.

Significant proportions of patients, especially patients with no prior treatment history, did not fill prescriptions for antibiotics, antihypertensives, or antidiabetics. The association between patient factors and medication fill rates varied across different medication groups.

aIBM Research

bIBM Watson Health, Cambridge, MA

cIBM Watson Health, Tampa, FL.

Correspondence: Gigi Yuen-Reed, IBM Watson Health, 3031N Rocky Point Dr West, Suite 500, Tampa, FL 33607-5878 (e-mail: gigi.yuen@us.ibm.com).

Abbreviations: ACE inhibitors = angiotensin-converting enzyme, BB = beta-blockers, CED = claim EMR data, EMR = electronic medical records, NYISS = The New York State Identification and Intelligence System Phonetic Code, OR = odds ratio.

YP, HY, AD, and GYR are employees of IBM Watson Health, the product owner of the claims-EMR linked data set studied.

The authors have no funding to disclose.

This is an open access article distributed under the terms of the Creative Commons Attribution-Non Commercial-No Derivatives License 4.0 (CCBY-NC-ND), where it is permissible to download and share the work provided it is properly cited. The work cannot be changed in any way or used commercially without permission from the journal. http://creativecommons.org/licenses/by-nc-nd/4.0

Received June 9, 2018

Accepted October 11, 2018

Back to Top | Article Outline

1 Introduction

Medication nonadherence is a major concern in public health. A large body of evidence shows that patients do not adhere to chronic medication such as antihypertensives or statins,[1,2] which can lead to undesirable health outcomes.[3–6] Understanding the extent of nonadherence as well as factors associated with it is therefore important to find effective intervention points for improving adherence.

Administrative claims data have been widely used for adherence studies, since it captures the medication filling events at pharmacy and is considered as more accurate than self-reports.[7] However, due to the nature of claims data where only “filled” events are recorded, it cannot be used to identify patients who never fill a doctor's prescription order. Due to this reason, a majority of prior research examined secondary nonadherence, defined as filling the initial prescription but do not persist to be adherent over a defined period of time. Medication fill rate is related to primary nonadherence in which patients do not fill the very first order for a medication.[8]

With the increasing use of electronic prescribing, electronic medical records (EMRs) linked with claims data are being utilized to investigate the primary nonadherence.[9–15] Since EMR contains information about what was prescribed by providers, subsequently identified pharmacy claim can be used to determine whether a patient has ever filled the prescribed medication. Linked data sets offer this opportunity to study the medication filling behavior of patients. In this study, we explored the fill rates including primary nonadherence to medications for 3 major health conditions using a large, nationwide linked data set.

Back to Top | Article Outline

2 Method

2.1 Data source and study cohort

IBM MarketScan Explorys Claims-EMR Data (CED) is a data set obtained through linkage between an EMR database (IBM Explorys Universe database) and a claims database (Truven MarketScan Research Databases). The IBM Explorys Universe database, EMR supplied by more than 300,000 health care providers, contains more than 315 billion clinical and operational data records from approximately 55 million unique patients.[16] The Truven MarketScan Research Databases, claims data supplied by more than 300 contributing employers and 40 contributing health plans,[17] captures more than 25 billion service records from 225 million unique individuals. Both databases have been de-identified, standardized, and normalized using common ontologies for secondary use. Social Security Number along with other demographic information and The New York State Identification and Intelligence System Phonetic Code (NYISS)[18] were used to link the raw data from the two sources, which was then de-identified. There were 4.4 million matched subjects in the linked data set. Ethical approval was not necessary due to the use of de-identified secondary database.

Patients with at least 1 electronic prescription record for any of the study drugs during the study period (index prescription), between January 1, 2014 and June 30, 2015, were included in the study. Patients were required to be continuously enrolled in a health insurance and to have at least 1 clinical activity per year (i.e., at least 1e record of office visit, admission, diagnosis, drug, immunization, observation, problem list, or procedure) between January 1, 2013 and December 31, 2015, which includes the baseline period (365 days before the index prescription) and 180 days after the index prescription.

Back to Top | Article Outline

2.2 Medication and adherence measure

We chose 3 medication groups used for 3 different health conditions to examine the medication fill rate, namely antihypertensive, antidiabetic, and antibiotics. Antihypertensive and antidiabetic were chosen because hypertension and diabetes are two of the most important chronic diseases for which significant degree of nonadherence have been previously reported, and the antibiotic was chosen for its high utilization[11] and the acute nature of use which is in contrast to chronic medications. Therapeutic classes in each group were defined using the RED BOOK (Truven Health Analytic), which classifies National Drug Codes into 31 therapeutic groups and 262 therapeutic classes. Antibiotic classes considered in the study included penicillin, tetracycline, cephalosporin and related agents, erythromycin and macrolides, and antifungals. We excluded the following classes: aminoglycoside due to limited usage in outpatient settings (1368 records), beta-lactam antibiotics (204 records), and miscellaneous (7107 records) due to nonspecific classification. As antihypertensive, we included diuretics (loop, potassium-sparing, and thiazide), angiotensin-converting enzyme (ACE) inhibitors, alpha-beta blockers, beta-blockers (BB), calcium channel blockers, hypotensive agents, and vasodilating agents. As antidiabetics, we included insulins, sulfonylureas, and others which includes the remaining types of medication, as was defined in RED BOOK.

Following the index prescription in the EMR, the first pharmacy claim for drugs within the same class as the index prescription was used to define medication fill. A medication order was considered filled if the first claim was found within the following 90 days.

Back to Top | Article Outline

2.3 Statistical analysis

We examined characteristics of patients in each of the 3 medication groups including demographics (age, gender, and race), specific therapeutic classes used, prior medication history, copay, insurance plan type (fee for service vs partially or fully capitated plans), and the time between electronic prescription order and filled date in claims. Prior use of medication was identified using both electronic prescription records and claims records during the baseline period. For each group, “any prior treatment” was defined as having at least 1 prescription or filling record for any medication in the same therapeutic group during the baseline period (e.g., use of any class of antihypertensive during baseline with an index antihypertensive prescription), and “same class prior treatment” was defined as having at least 1 prescription or filling record for the same therapeutic class medication during the baseline (e.g., use of a penicillin during baseline for an index penicillin prescription). We could not determine the level of copay and insurance plan types for patients not filling their index prescriptions. Thus, we reported the median copay level and proportion of each insurance plan type only among patients who filled medication.

To identify factors associated with the medication fill rate, we used multivariate logistic regression models including age, gender, race, brand vs generic drug use (based on a code identifying products as either original standard product or a generic copy), prior exposure to the same class, and number of distinct therapeutic classes used during baseline within the same therapeutic group (e.g., total number of antihypertensive classes used during the baseline period) as covariates. The models for each medication group and each therapeutic class was fitted separately because the factors affecting the adherence and the magnitude of effects can differ by clinical usage. A separate model was fit among the subgroup of patients without prior exposure to the same therapeutic class medication to examine whether the associated factors differ for first-ever prescriptions (i.e., primary adherence). We reported odds ratios and corresponding 95% of confidence intervals.

Back to Top | Article Outline

3 Results

From the linked data set, we identified 134,434 patients who met our study criteria. There were 77,996 patients who had an index antibiotic prescription, 78,462 patients who had an index antihypertensive prescription, and 24,013 patients who had an index antidiabetic prescription (Table 1). Majority of patients were Caucasians and had a traditional fee-for-service plans in all 3 groups. Patients were more likely to be female in the antibiotic and some of antihypertensive classes but not in the antidiabetic group.

Table 1

Table 1

The overall medication fill rate was 73% for the antibiotic, 74% for the antihypertensive, and 76% for the antidiabetic group (Table 2). The adherence rate was comparable in antibiotic group between patients who had prior treatment records with any antibiotic compared to patients who did not have any prior antibiotic treatment during baseline (74% vs 71%). However, a slightly lower adherence rate was observed for patients with index antibiotic prescription who had prior treatment with the same class antibiotic compared to those without same class prior treatment (69% vs 74%). Unlike antibiotics, adherence rate was higher among patients with any prior treatment compared to patients without prior treatment in the past year, in both antihypertensive (78% vs 41%, respectively) and antidiabetic groups (81% vs 48%, respectively). Similar pattern was observed with regard to the same class prior treatment. The mean time to fill was 1.7 days for antibiotic prescriptions and 13.5 days for both antihypertensives and antidiabetics (Table 2). As expected from the acute nature of use, antibiotics had much shorter time from prescription to filling compared to the chronic medications. Across the 3 medication groups, prior treatment history was associated with longer duration from the prescription to the filling.

Table 2

Table 2

In multivariate logistic regression models, different factors were associated with the medication fill rate in varying directions and degrees, even within the same therapeutic group depending on the therapeutic classes (Table 3). Overall, African American race was negatively associated with the fill rates, adjusting for other factors in the model. The age effect varied, with negative association observed in antibiotic group and no association observed in other medication groups. The number of medication classes used during baseline in the same therapeutic group was positively associated with the fill rates in all 3 medication groups. Same class prior treatment had the strongest association among the factors we examined. However, it was positively associated with the antihypertensive and the antidiabetic groups, whereas it was negatively associated with the antibiotics group.

Table 3

Table 3

The factors associated with primary adherence rate among patients who never received a same class treatment during the baseline were similar to those affecting medication fill rate in the entire study population (Table 4). One notable difference was the effect of older age (>65), which was negatively associated with first-ever fill rate more than the overall fill rate.

Table 4

Table 4

Back to Top | Article Outline

4 Discussion

In a large linked data set, we observed that 24% to 27% of patients who were prescribed either an antibiotic, antihypertensive, or antidiabetic medication did not fill their prescriptions in the following 90 days. Notably, prior treatment history was associated with higher fill rates, suggesting that patients adhere better once the treatment begins, but adherence to the first-ever treatment is poorer.

The overall rate of prescription fill is comparable to what was reported earlier using a similar data set.[10,11] Previously reported rates of primary nonadherence for antimicrobial medications are around 23%, and for chronic medications the rates range from 3% to 4% to greater than 40%. Most of the previous studies were either done outside of the United States,[19,20] restricted to patients in a specific integrated managed care,[9,12,14,21] or restricted to a specific pharmacy benefit manager or insurance plan.[10,11,13] Lower nonadherence was seen in integrated managed systems, reflecting that better integration of care may lead to improved adherence in patients with chronic diseases.

Higher fill rates for medications that the patient had previously used was also observed in the previous studies.[11] An interesting exception to this observation was prior treatment with the same class antibiotic, which was associated with poorer adherence to subsequent antibiotic prescription. It may partly reflect the use of leftover antibiotics reported in a previous study.[22] This observation suggests that medication fill behaviors for acute and chronic medications can be associated with different factors, and methods to improve overall fill rates or primary adherence may need to differ depending on target drug.

One of the strengths of this study is the generalizability of the result to a larger population, because the study data set has contributions from multiple payers including large employers, managed care organizations, as well as Medicare and Medicaid. In addition, we were able to examine the difference in fill rates and factors associated with the fill rates at therapeutic class levels rather than at an aggregated level. However, this study is not without limitations. It was previously reported that adherence is higher for drugs on formulary[11] but we could not account for formulary in this analysis. In the presence of sample use obtained from physicians’ offices, the fill rate would have been underestimated. But the effect of sample use is expected to be small since most samples do not last for extended period of time and we used 90-day period to capture the medication fill.

In conclusion, we observed that a significant proportion of patients did not fill their prescription for antibiotics, antihypertensives, or antidiabetics, and medication fill rate is strongly associated with the prior treatment history. The implication of the fact that a quarter of patients are not filling their prescriptions is significant with respect to both public health and policy point of view. Further research is needed to identify causal factors for nonadherence and targets for intervention to improve medication fill rates.

Back to Top | Article Outline

Author contributions

Conceptualization: Yoonyoung Park, Hyuna Yang, Amar K Das, Gigi Yuen-Reed.

Data curation: Hyuna Yang.

Formal analysis: Hyuna Yang.

Methodology: Yoonyoung Park, Hyuna Yang.

Supervision: Gigi Yuen-Reed.

Writing – original draft: Yoonyoung Park, Hyuna Yang.

Writing – review & editing: Yoonyoung Park, Hyuna Yang, Amar K Das, Gigi Yuen-Reed.

Yoonyoung Park orcid: 0000-0002-5990-1220.

Back to Top | Article Outline

References

[1]. Abegaz TM, Shehab A, Gebreyohannes EA, et al. Nonadherence to antihypertensive drugs: a systematic review and meta-analysis. Medicine (Baltimore) 2017;96:e5641.
[2]. Bates TR, Connaughton VM, Watts GF. Non-adherence to statin therapy: a major challenge for preventive cardiology. Expert Opin Pharmacother 2009;10:2973–85.
[3]. Ho PM, Rumsfeld JS, Masoudi FA, et al. Effect of medication nonadherence on hospitalization and mortality among patients with diabetes mellitus. Arch Intern Med 2006;166:1836–41.
[4]. Ho PM, Spertus JA, Masoudi FA, et al. Impact of medication therapy discontinuation on mortality after myocardial infarction. Arch Intern Med 2006;166:1842–7.
[5]. Sokol MC, McGuigan KA, Verbrugge RR, et al. Impact of medication adherence on hospitalization risk and healthcare cost. Medical Care 2005;43:521–30.
[6]. De Vera MA, Bhole V, Burns LC, et al. Impact of statin adherence on cardiovascular disease and mortality outcomes: a systematic review. Br J Clin Pharmacol 2014;78:684–98.
[7]. Hess LM, Raebel MA, Conner DA, et al. Measurement of adherence in pharmacy administrative databases: a proposal for standard definitions and preferred measures. Ann Pharmacother 2006;40:1280–8.
[8]. Raebel MA, Schmittdiel J, Karter AJ, et al. Standardizing terminology and definitions of medication adherence and persistence in research employing electronic databases. Med Care 2013;51:S11–21.
[9]. Karter AJ, Parker MM, Moffet HH, et al. New prescription medication gaps: a comprehensive measure of adherence to new prescriptions. Health Serv Res 2009;44:1640–61.
[10]. Fischer MA, Stedman MR, Lii J, et al. Primary medication non-adherence: analysis of 195,930 electronic prescriptions. J Gen Intern Med 2010;25:284–90.
[11]. Fischer MA, Choudhry NK, Brill G, et al. Trouble getting started: predictors of primary medication nonadherence. Am J Med 2011;124:1081.e9– 22.
[12]. Shah NR, Hirsch AG, Zacker C, et al. Predictors of first-fill adherence for patients with hypertension. Am J Hypertens 2009;22:392–6.
[13]. Shrank WH, Choudhry NK, Fischer MA, et al. The epidemiology of prescriptions abandoned at the pharmacy. Ann Intern Med 2010;153:633–40.
[14]. Raebel MA, Ellis JL, Carroll NM, et al. Characteristics of patients with primary non-adherence to medications for hypertension, diabetes, and lipid disorders. J Gen Intern Med 2012;27:57–64.
[15]. Johnson BH, Henriques C, Bonafede MM. Primary nonadherence: do differences exist between life sustaining and quality of life medications? Value Health 2017;20:A700.
[16]. Kaelber DC, Foster W, Gilder J, et al. Patient characteristics associated with venous thromboembolic events: a cohort study using pooled electronic health record data. J Am Med Inform Assoc 2012;19:965–72.
[17]. Thomson Healthcare, Adamson D, Chang S, Hansen LG. Health Research Data for the Real World: The MarketScan Databases. 2008;1–32.
[18]. Taft RL. NAME SEARCH TECHNIQUES - SPECIAL REPORT NO. 1. New York State Identification and Intelligence System. 1970. Available at: https://www.ncjrs.gov/App/Publications/abstract.aspx?ID=1743. Accessed December 1, 2017.
[19]. Storm A, Andersen SE, Benfeldt E, et al. One in 3 prescriptions are never redeemed: primary nonadherence in an outpatient clinic. J Am Acad Dermatol 2008;59:27–33.
[20]. Tamblyn R, Eguale T, Huang A, et al. The incidence and determinants of primary nonadherence with prescribed medication in primary care: a cohort study. Ann Intern Med 2014;160:441–50.
[21]. Shin J, McCombs JS, Sanchez RJ, et al. Primary nonadherence to medications in an integrated healthcare setting. Am J Manag Care 2012;18:426–34.
[22]. Richman PB, Garra G, Eskin B, et al. Oral antibiotic use without consulting a physician: a survey of ED patients. Am J Emerg Med 2001;19:57–60.
Keywords:

drug utilization; electronic medical record-claims data; medication adherence

Copyright © 2018 The Authors. Published by Wolters Kluwer Health, Inc. All rights reserved.