Assessing medication adherence using Stata

Abstract

In this article, I introduce the medadhere command, which computes medication adherence rates for two commonly used measures in research and practice: the medication possession ratio and the proportion of days covered. medadhere computes adherence rates for a single medication or multiple medications, and its options provide great flexibility to support the specific needs of the user.

Keywords

st0578 medadhere medication adherence medication compliance medication possession ratio proportion of days covered pharmacy claims

1 Introduction

Although most patients leave the doctor’s office with a medication prescription, many fail to take their medication as prescribed. According to the World Health Organization, medication adherence (also referred to as compliance) rates in developed countries average only about 50% (Sabaté 2003), and even patients in closely monitored clinical trials have been reported to maintain adherence rates of only 43% to 78% (Osterberg and Blaschke 2005). Given its association with increased morbidity, mortality, health services use, and cost (Psaty et al. 1990; Rasmussen, Chong, and Alter 2007; Kulkarni et al. 2008; Linden and Adler-Milstein 2008; Sokol et al. 2005; Dragomir et al. 2010; Roebuck et al. 2011), medication nonadherence is a major concern to practitioners, payers, and policy makers alike. Accordingly, measuring adherence is an essential component of any strategy to improve clinical outcomes using pharmacotherapy.

Over the past two decades, administrative pharmacy claims have increasingly been used as a data source for measuring medication adherence under the assumption that filling (and refilling) a prescription is consistent with taking the medication as directed. Adherence measures derived from administrative data are appealing because they are convenient, objective, noninvasive, and cost effective for use in large populations (Steiner and Prochazka 1997; Osterberg and Blaschke 2005; Hess et al. 2006). Two of the most widely used adherence measures reported in the literature are the medication possession ratio (MPR) and the proportion of days covered (PDC) (see below for descriptions) (Andrade et al. 2006; Cramer et al. 2008; McMahon et al. 2011; Sattler, Lee, and Perri 2013; Lam and Fresco 2015). The Centers for Medicare & Medicaid Services currently reports PDC rates for diabetes, hypertension, and cholesterol drugs as part of the annual star rating system that measures the quality of care received by beneficiaries enrolled in Medicare Advantage Plans (managed care) and Prescription Drug Plans (Centers for Medicare & Medicaid Services 2018).

In this article, I introduce the new medadhere command, which computes medication adherence rates using MPR and PDC. medadhere can compute rates for a single medication and for multiple medications (for each individual drug separately), and its options provide great flexibility to support the specific needs of the user.

2 Methods

The computation of any pharmacy claims–based adherence measure requires patientlevel records captured over a particular time interval. Patient-level information includes the dates on which the prescriptions were filled and the number of days of supply for the medication (for example, 60 pills to be consumed twice per day for 30 days). These variables are used to calculate the amount of medication patients have on hand over the course of a given study period. The two most common methods investigators use to set the study period are a fixed date range (for example, January 1, 2017 to December 31, 2017) or a variable study period, in which each patient’s follow-up period is allowed to vary within a set time interval (for example, 180 days from each patient’s first prescription fill). As described in the following sections, MPR and PDC use these date elements somewhat differently for computing adherence rates.

2.1 Medication Possession Ratio

The MPR is generally defined as a ratio of the total days of available supply (of a medication) to the number of days in a specified study period (Steiner and Prochazka 1997). As an example, table 1 presents pharmacy data for a patient who filled her prescription seven times between December 30, 2012 and June 27, 2013. To illustrate the computation of MPR for a fixed study period, let us assume the study seeks to compute her MPR between January 1, 2013 and June 29, 2013 (180 days, inclusive).

Table 1.

Example pharmacy fill data for a single patient

	User provided		Derived
Fill no.	Fill date	Days supply	Fill end date	Supply
1	30Dec2012	30	28Jan2013	28
2	23Jan2013	30	21Feb2013	30
3	27Feb2013	30	28Mar2013	30
4	03Apr2013	30	02May2013	30
5	30Apr2013	30	29May2013	30
6	29May2013	30	27Jun2013	30
7	27Jun2013	30	26Jul2013	3

As shown, only 28 days of supply (out of 30) are counted from the first fill because 2 days of supply are from before the study start date (January 1, 2013). Similarly, only 3 days of supply are counted in the seventh fill because the remaining days of supply are beyond the study end date (June 29, 2013). Thus, the MPR is computed as the sum of supply (181) divided by the number of days in the study period (180), which in this case equals 1.01. However, note that there is more medication on hand than days in the study period (181 versus 180) because both the second and fifth refills were obtained earlier than when the supply of the previous refill was exhausted. This issue highlights a limitation of the MPR: it can overestimate adherence if patients regularly refill their prescriptions early or if they switch medications before consuming all the prior supply. Some investigators handle this problem by capping the adherence rate at 1.00 (Hess et al. 2006).

2.2 Proportion of days covered

The PDC is generally defined as the number of days in which the patient has the medication available divided by the number of days in a specified study period (Benner et al. 2002). In contrast with MPR, which sums the total supply of medication available, PDC sums the number of days in which the medication is available. More specifically, each day is categorized as either 1 or 0 representing whether the patient had medication available on that day or not, respectively.

Table 2 illustrates how the daily supply is quantified for the PDC calculation when there are either overlapping days of supply or no supply available. The first seven observations highlight the overlapping days of medication supply from January 23rd to January 29th for the patient data presented in table 1. Here the supply equals 1, even though the patient actually had more medication on hand for those days because they refilled the prescription early. The next six observations (February 22nd to February 26th) highlight days in which no supply was available and therefore are coded as 0. On February 27th, the patient refilled her prescription, and therefore a value of 1 is assigned to that day’s supply.

Table 2.

Quantifying daily supply for the PDC calculation using the patient data in table 1

Date	Study day	Supply
23Jan2013	23	1
24Jan2013	24	1
25Jan2013	25	1
26Jan2013	26	1
27Jan2013	27	1
28Jan2013	28	1
29Jan2013	29	1
…	…	…
22Feb2013	52	0
22Feb2013	53	0
23Feb2013	54	0
24Feb2013	55	0
25Feb2013	56	0
26Feb2013	57	0
27Feb2013	58	1

Using the same study period as in the MPR example above (January 1, 2013 through June 29, 2013), the sum of the number of days in which the medication is available is 170. As such, the PDC is 0.94 (170/180), which is more conservative than the 1.01 derived using the MPR.

A modification to the PDC can be implemented that involves shifting refill dates forward to the day after the supply of medication is exhausted. This procedure essentially credits the patient for having additional supply on hand (due to early refilling). Table 3 illustrates how the original fill dates and fill end dates (from table 1) are modified to separate overlapping days of supply. For example, whereas the second medication fill originally occurred on January 23rd (five days before the existing supply was exhausted), the modified refill date has been shifted forward to the day after that refill was exhausted (January 29th). Because of this shift, the supply for this second refill is now exhausted on February 27th rather than on the original date (February 21st). All other subsequent overlapping refill periods are shifted forward accordingly. Note that shifting is only correct when all data are known and does not credit dispensed medication before the study period or when it is unavailable.

Table 3.

Modifying the original fill dates (from table 1) to credit the patient for additional supply

	Original dates		Modified dates
Fill no.	Fill date	Fill end date	Fill date	Fill end date
1	30Dec2012	28Jan2013	30Dec2012	28Jan2013
2	23Jan2013	21Feb2013	29Jan2013	27Feb2013
3	27Feb2013	28Mar2013	28Feb2013	29Mar2013
4	03Apr2013	02May2013	03Apr2013	02May2013
5	30Apr2013	29May2013	03May2013	01Jun2013
6	29May2013	27Jun2013	02Jun2013	01Jul2013
7	27Jun2013	26Jul2013	02Jul2013	31Jul2013

Using this modification while retaining the same study period as the previous examples (January 1, 2013 through June 29, 2013), the sum of the number of days in which the medication is available is 176. Thus, the PDC is now 0.98 (176/180), which lies somewhere between the values derived for the MPR and the basic PDC (1.01 and 0.94, respectively).

3 The medadhere command

This section describes the syntax of the medadhere command, which computes the MPR and the PDC.

3.1 Syntax

medadhere fill_date days_supply [if] [in] [, id(string) drug(string) start( string) [end(string)| length( #)] credit]

fill_date is the date when the prescription was filled, and days_supply is the number of days that the medication was intended to last (for example, if one has 60 tablets and should take a tablet twice per day for 30 days, days_supply is 30).

3.2 Options

id(string) specifies the patient identifier if the data contain multiple patients. By default, medadhere assumes that there is a single patient in the data.

drug(string) specifies the drug if the data contain multiple medications per patient. By default, medadhere assumes that there is only one medication per patient in the data.

start(string) specifies the desired start date of a study period as a literal date, such as start(01jan2013) or start(01/01/2013); see [D] Datetime translation. By default, medadhere uses the first fill_date.

end(string) specifies the desired end date of a study period as a literal date, such as end(31dec2018) or end(31/12/2018); see [D] Datetime translation. By default, medadhere uses the last refill’s end date (fill_date + days_supply) − 1. Only one of end() or length() may be specified.

length(#) specifies a study duration as an alternative to specifying end(). Only one of length() or end() may be specified.

credit specifies that overlapping refill periods be spaced out by shifting the next fill_date forward to the day after the previous refill has been exhausted. This essentially gives the individual “credit” for having more medication on hand. The credit option affects only the PDC.

3.3 Data generated by medadhere

medadhere replaces the data in memory with variables relevant to adherence rates computed for the MPR and the PDC. Table 4 describes these variables and indicates under which circumstances they are generated.

Table 4.

Variables that replace the data in memory when medadhere is executed

Variable	Description	When generated
id	Patient identifier	When id() specified
drug	Medication identifier	When drug() specified
study_start_dt	Start of study period	Always
study_end_dt	End of study period	Always
study_days	Number of days in study period	Always
supplyMPR	Total supply on hand (MPR)	Always
supplyPDC	Total supply on hand (PDC)	Always
mpr	Computed MPR	Always
pdc	Computed PDC	Always

The study_start_dt and study_end_dt variables are drawn directly from the options start() and end() when they are specified. If start() is not specified, then the first fill_date in the data is used as the study_start_dt. Likewise, if end() is not speci- fied, then the last (fill_date + days_supply) − 1 is calculated as the study_end_dt unless length() is specified, in which case the study_end_dt is calculated as (study_start_dt + length()) − 1. study_days is calculated as (study_end_dt− study_start_dt) + 1. Both the mpr and the pdc variables are computed as their respective supply/study_days.

4 Examples

medadheredata.dta contains pharmacy data for eight patients spanning the period of January 7, 2012 to December 31, 2013. While all patients refilled prescriptions for at least two medications, five of the patients refilled prescriptions for three medications. The scope of these data allows us to illustrate all the salient features of the medadhere command.

The medadhere command can be implemented in four scenarios: 1) a single patient on a single medication, 2) a single patient on multiple medications, 3) multiple patients on a single medication, and 4) multiple patients on multiple medications. The examples below are described accordingly.

4.1 A single patient on a single medication

In this example, we assume there is only one patient in the dataset. First, we load medadheredata.dta and keep only the observations for the first patient and drug 1:¹

. use medadheredata, clear

. keep if id==1 & drug==1 (290 observations deleted)

If the data are for a single patient on a single medication, we do not need to specify id() or drug(). Here we specify the start and end dates of the study period to span all days in 2013. We then list the results in the Results window (alternatively, you can review the data in the Data Editor):

For this patient, the results indicate that the PDC is more conservative than the MPR (0.90 versus 0.97). As a reminder, a value of 1.0 indicates perfect adherence.

4.2 A single patient on multiple medications

In this example, we again assume there is only a single patient in the dataset, but this patient is refilling prescriptions for multiple medications.

First, we load medadheredata.dta and keep only the observations for the first patient (and all of their drugs):

. use medadheredata, clear

. keep if id==1

(267 observations deleted)

If the data are for a single patient on multiple medications, we specify drug() but not id(). Here we specify the start date and length of the study period to span 365 days in 2013 (as an alternative to specifying end()).

4.3 Multiple patients on a single medication

In this example, we have multiple patients in the dataset, but we assume they are all taking a single medication. We reload medadheredata.dta and keep only the observations where drug = 1:

. use medadheredata.dta, clear

. keep if drug==1

(205 observations deleted)

If the data are for multiple individuals on a single medication, we specify id() but not drug(). Here we implement medadhere and do not specify the start date, which means that the command will use the earliest fill_date as the start date. We set the end date of the study period at 30June2013 (thereby allowing each patient to have a different study duration).

As shown, each patient’s study_start_dt corresponds to their first fill_date, resulting in different study periods for each patient.

4.4 Multiple patients on multiple medications

In this example, we use all the data—for all patients and all medications.

. use medadheredata.dta, clear

If the data are for multiple patients and multiple medications, we specify id() and drug(). Here we implement medadhere and do not specify the start or end dates but instead specify the observation period as being 180 days. We also specify the credit option.

As shown, each patient’s study_start_dt corresponds to their first fill_date, and their study_end_dt is 180 days beyond that date, resulting in 180-day study periods that are unique for each patient and each drug.

4.5 Dichotomizing the MPR and the PDC as measures of adherence

While both the MPR and the PDC are fractional measures of adherence ranging from 0 to 1 (except in cases where MPR overcounts the supply on hand, resulting in a value greater than 1.0), it is very common to find them dichotomized at a cutpoint of 0.80 in which patients above the cutpoint are considered adherent and those below the cutpoint are considered nonadherent (Cramer et al. 2008; Sikka, Xia, and Aubert 2005; Centers for Medicare & Medicaid Services 2018).

In this example, we use the PDC output from the previous implementation of the medadhere command (example 4.4) and generate a new variable called pdc80. We then tabulate this variable by drug type.

From the table, we see that adherence in this sample of patients was 75%, 75%, and 60% for drugs 1, 2, and 3, respectively.

5 Discussion

Measuring medication adherence is an essential component of any strategy to improve patient outcomes and increase the quality of care. Because big data have become more readily available (via administrative claims, electronic health records, and patientgenerated data through various applications and portals), capable software programs are needed to process these measures efficiently.

In this article, I introduced the medadhere command, which measures medication adherence using two common measures in practice—the MPR and the PDC. As demonstrated in the examples, the medadhere command provides users with great flexibility to tailor the measures to meet their specific needs. More-complex specifications can also be implemented in conjunction with the if or in qualifiers. While guidance on the use and reporting of these measures is beyond the scope of this article (see Peterson et al. [2007] for a comprehensive discussion), the medadhere command (and Stata, more generally) can easily support these approaches as well.

7 Programs and supplemental materials

Supplemental Material, st0578 - Assessing medication adherence using Stata

Supplemental Material, st0578 for Assessing medication adherence using Stata by Ariel Linden in The Stata Journal

Footnotes

6 Acknowledgment

I thank the anonymous reviewer and chief editor for their thoughtful reviews and recommendations for improving both the article and the medadhere command.

7 Programs and supplemental materials

To install a snapshot of the corresponding software files as they existed at the time of publication of this article, type

. net sj 19-4

. net install st0578 (to install program files, if available)

. net get st0578 (to install ancillary files, if available)

Notes

References

Andrade

S. E.

Kahler

K. H.

Frech

Chan

K. A.

. 2006. Methods for evaluation of medication adherence and persistence using automated databases. Pharmacoepidemiology and Drug Safety 15: 565–574.

Benner

J. S.

Glynn

R. J.

Mogun

Neumann

P. J.

Weinstein

M. C.

Avorn

. 2002. Long-term persistence in use of statin therapy in elderly patients. Journal of the American Medical Association 288: 455–461.

Centers for Medicare & Medicaid Services. 2018. Part C and D performance data. https://www.cms.gov/Medicare/Prescription-Drug-Coverage/PrescriptionDrugCovGenIn/PerformanceData.html.

Cramer

J. A.

Benedict

Muszbek

Keskinaslan

Khan

Z. M.

. 2008. The significance of compliance and persistence in the treatment of diabetes, hypertension and dyslipidaemia: A review. International Journal of Clinical Practice 62: 76–87.

Dragomir

Côté

White

Lalonde

Blais

Bérard

Perreault

. 2010. Relationship between adherence level to statins, clinical issues and health-care costs in real-life clinical setting. Value in Health 13: 87–94.

Hess

L. M.

Raebel

M. A.

Conner

D. A.

Malone

D. C.

. 2006. Measurement of adherence in pharmacy administrative databases: A proposal for standard definitions and preferred measures. Annals of Pharmacotherapy 40: 1280–1288.

Kulkarni

A. S.

Balkrishnan

Anderson

R. T.

Edin

H. M.

Kirsch

Stacy

M. A.

. 2008. Medication adherence and associated outcomes in Medicare health maintenance organization-enrolled older adults with Parkinson’s disease. Movement Disorder 23: 359–365.

Lam

W. Y.

Fresco

. 2015. Medication adherence measures: An overview. BioMed Research International 2015: Article 217047.

Linden

Adler-Milstein

. 2008. Medicare disease management in policy context. Health Care Finance Review 29: 1–11.

10.

McMahon

J. H.

Jordan

M. R.

Kelley

Bertagnolio

Hong

S. Y.

Wanke

C. A.

Lewin

S. R.

Elliott

J. H.

. 2011. Pharmacy adherence measures to assess adherence to antiretroviral therapy: Review of the literature and implications for treatment monitoring. Clinical Infectious Diseases 52: 493–506.

11.

Osterberg

Blaschke

. 2005. Adherence to medication. New England Journal of Medicine 353: 487–497.

12.

Peterson

A. M.

Nau

D. P.

Cramer

J. A.

Benner

Gwadry-Sridhar

Nichol

. 2007. A checklist for medication compliance and persistence studies using retrospective databases. Value in Health 10: 3–12.

13.

Psaty

B. M.

Koepsell

T. D.

Wagner

E. H.

LoGerfo

J. P.

Inui

T. S.

. 1990. The relative risk of incident coronary heart disease associated with recently stopping the use of β-blockers. Journal of the American Medical Association 263: 1653–1657.

14.

Rasmussen

J. N.

Chong

Alter

D. A.

. 2007. Relationship between adherence to evidence-based pharmacotherapy and long-term mortality after acute myocardial infarction. Journal of the American Medical Association 297: 177–186.

15.

Roebuck

M. C.

Liberman

J. N.

Gemmill-Toyama

Brennan

T. A.

. 2011. Medication adherence leads to lower health care use and costs despite increased drug spending. Health Affairs 30: 91–99.

16.

Sabaté

, ed. 2003. Adherence to Long-Term Therapies: Evidence for Action. Geneva: World Health Organization.

17.

Sattler

E. L. P.

Lee

J. S.

Perri

III . 2013. Medication (re)fill adherence measures derived from pharmacy claims data in older Americans: A review of the literature. Drugs & Aging 30: 383–399.

18.

Sikka

Xia

Aubert

R. E.

. 2005. Estimating medication persistency using administrative claims data. American Journal of Managed Care 11: 449–457.

19.

Sokol

M. C.

McGuigan

K. A.

Verbrugge

R. R.

Epstein

R. S.

. 2005. Impact of medication adherence on hospitalization risk and healthcare cost. Medical Care 43: 521–530.

20.

Steiner

J. F.

Prochazka

A. V.

. 1997. The assessment of refill compliance using pharmacy records: Methods, validity, and applications. Journal of Clinical Epidemiology 50: 105–116.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB