VALUE IN HEALTH 17 (2014) 266–274
Available online at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/jval
A Multistate Model and an Algorithm for Measuring Long-Term
Adherence to Medication: A Case of Diabetes Mellitus Type 2
Majken Linnemann Jensen, MSc1,2,*, Marit Eika Jørgensen, MD, PhD1, Ebba Holme Hansen, MSc2,
Lise Aagaard, MSc, PhD3, Bendix Carstensen, MSc1
1
Steno Diabetes Center A/S, Gentofte, Denmark; 2Faculty of Health and Medical Sciences, University of Copenhagen, Copenhagen,
Denmark; 3Faculty of Health, University of Southern Denmark, Odense, Denmark
AB STR A CT
Objectives: To develop a multistate model and an algorithm for
calculating long-term adherence to medication among patients with
a chronic disease. Methods: We propose definitions of the different
states of waiting, persistence, with sufficient supply to implement the
prescribed dosing regimen, gaps, nonpersistence, and nonacceptance
and an algorithm for transitions between states to describe long-term
adherence to medication treatment. The model and algorithm are
operationalized for use in a case with a retrospective cohort of
patients with type 2 diabetes mellitus, with access to records of
prescribed drugs from a Danish diabetes research hospital and records
of filled prescriptions at Danish pharmacies from the Danish Health
and Medicines Authority. Results: Calculations of long-term adher-
ence to medication are shown for patients with type 2 diabetes
mellitus on metformin and/or simvastatin. The study shows how
Introduction
Adherence to medication defined as “the extent to which a
person’s behavior with regard to taking medication corresponds
with agreed recommendations from a health care provider” [1]
has been explored extensively for more than four decades [2,3].
Adherence to long-term therapy for chronic illnesses in devel-
oped countries is estimated to average 50% by the World Health
Organization (WHO) [4]. Studies have shown that poor adherence
is associated with worsening of the patients’ clinical status/
health [5,6], higher risk of hospitalization [7], risk of preventable
drugs-related hospital admissions [8], and higher mortality risk
[9]. Losing adherence has been shown to be associated with
higher outcome of hospitalizations and emergency department
visits [10].
It has been argued, however, that “The term ‘adherence rate’,
as used by the WHO, is ill considered and meaningless” [11]
because it cannot distinguish whether poor adherence stems
from poor acceptance, poor execution, or early discontinuation of
treatment. The need for methods and models on how to improve
the consistency and quality of measures of adherence to
* Address correspondence to: Majken Linnemann Jensen, Steno Diabetes Center A/S, Niels Steensens Vej 1, (NLD2.06-MLiJ), DK-2820
Gentofte, Denmark.
E-mail: MLiJ@steno.dk.
1098-3015/$36.00 – see front matter Copyright & 2014, International Society for Pharmacoeconomics and Outcomes Research (ISPOR).
Published by Elsevier Inc.
http://dx.doi.org/10.1016/j.jval.2013.11.014
the prevalence of patients waiting to initiate treatment, patients with
supply to implement the prescribed dosing regimen, patients not
accepting treatment, and patients discontinuing treatment varies
over time. Conclusions: The proposed multistate model and algo-
rithm can easily be translated and used for the calculation of
adherence to medication in any chronic disease. The model and
algorithm take time into account, and thus, changes in incidence
rates and prevalence of the different states over time can be estimated
on several time scales (calendar time, age of the patient, and time
since indication for medication).
Keywords: acceptance, adherence to medication, diabetes mellitus,
implementation, multistate model, persistence.
Copyright & 2014, International Society for Pharmacoeconomics and
Outcomes Research (ISPOR). Published by Elsevier Inc.
medication has been addressed [3,12,13] with efforts to meet
these needs [14,15], but there is still no uniformity in the
terminology used to describe how patients deviate from pre-
scribed medication therapy [16]. It can be argued that poor
adherence among some patients probably is a relapsing condition
over time similar to relapses in attempting smoke cessation
among smokers and relapses in retaining weight loss among
overweight persons. Whether intentionally or unintentionally
[17] on the part of the patient, a mismatch between the health
care professionals’ assumptions about the drug intake based on
the electronic medical records of prescribed medication and what
in reality has been collected and ingested by the patient signifies
a problem in the pharmacotherapeutic chain. Extended pharma-
coepidemiological research to devise methods to identify ther-
apeutic lapses and to estimate the extent and magnitude of
discrepancies between written prescriptions issued by the
physician and prescriptions filled by the patient has been called
for [11].
Long-term adherence to medication should comprise meas-
ures of initiation, acceptance, implementation of dosing regimen,
and persistence [11,16,18], which entails that more than one type
 VALUE IN HEALTH 17 (2014) 266–274
of event is of interest: when the patient is prescribed with the
drugs for the first time, when he or she initiates treatment, when
he or she runs out of supply, or when he or she or the prescribing
doctor discontinues the treatment. We set up a multistate model
to describe the events of interest allowing the same type of event
to occur more than once [19].
Along with development of the multistate model, rules for an
algorithm describing how and when a patient moves between the
different states must be defined.
Type 2 diabetes mellitus (T2DM) has been chosen as the
case of a chronic illness because it is the main focus of
the research team, and we have access to extensive data. T2DM
is a complex metabolic disorder and prevalent in both industrial-
ized and developing countries [20]. Guidelines [21] for the treat-
ment of T2DM recommend long-term multifactorial medication
therapy for hyperglycemia, hypertension, hypercholesterolemia,
and anticoagulation to prevent/postpone diabetes complications.
Metformin (ATC code: A10BA02, recommended as first-choice
blood glucose–lowering medication in T2DM [21]) and simvasta-
tin (ATC code: C10AA01, recommended as first choice for
dyslipidemia in T2DM in Denmark [22]) were chosen as
case drugs. ATC codes refer to WHO’s Anatomical Therapeutic
Chemical classification system codes [23]. The multistate
model and the corresponding estimation algorithm can be used
for any type of long-term medication used by patients with
a chronic disease, for example, human immunodeficiency
virus-positive patients, cancer patients, or organ transplanted
patients.
Thus, the aim of the current study was to develop a multistate
model to be used as a tool to describe and measure long-term
adherence to treatment for patients with a chronic disease on
lifelong medication. The model can cover either one drug at the
time (as in this study) or a group of drugs.
Case Material
Study Population
A cohort of patients with T2DM from the Danish diabetes
research hospital Steno Diabetes Center (Steno), a specialized
diabetes tertiary referral hospital for patients with both type 1
diabetes mellitus and T2DM, was used. A total of 4322 patients
with T2DM with at least one written prescription of either
metformin or simvastatin and enrolment at Steno during the
period 1998 to 2009 were included.
Daily Doses Prescribed by the Physician
Since 1998, the physicians at Steno have made electronic regis-
trations of their recommendations of medication to the patients.
The recommendations are recorded for each drug identified by
the ATC code with a starting date and a daily dose. Cessation of a
recommendation is recorded with an ending date. These recom-
mendations are the basis for the actual (electronic) written
prescription of the drugs issued by the physician, which enables
the patients to pick up the drugs at any Danish pharmacy of their
choice.
Prescriptions Filled by the Patient
Every filled prescription (Rx) is registered with information about
the date of purchase, purchased amount of the drug in defined
daily doses (DDDs) [24], ATC code, and the patient’s identity
number. Since 1995 this information is stored in the Register of
Medicinal Product Statistics (RMPS) owned by the Danish Health
and Medicines Authority and was made available to us via
Statistics Denmark.
267
Every person in Denmark is identified by a unique identity
number (the Danish Civil Registration number), which is used for
linkage of registers in this analysis.
Methods
Patients enter the study on the day of the first written prescrip-
tion at Steno after January 1, 1998, for the drugs in question. End
of follow-up per patient per type of drug is the earliest of date of
death, the last date of referral of the patient from Steno to a
physician or another specialist unit before December 31, 2009, or
December 31, 2009.
The drugs are classified according to WHO’s ATC codes at
level 5 as of January 1, 2010. Amounts of prescribed and
purchased drugs are transformed into units of DDD by using
WHO’s DDD values as of January 1, 2010.
For the definition of this multistate model and algorithm,
adherence to medication is restricted to prescriptions of drugs in
which the prescribed daily dose is known as well as the amount
of the drug being filled. Drugs prescribed to be used “as needed”
(e.g., insulin) and over-the-counter drugs can therefore not be
covered by this model and algorithm.
For each type of drug for each patient, records with prescribed
daily dose including starting and ending dates are extracted from
Steno’s Electronic Medical Record (EMR). All records on Rx from
the RMPS are sorted by patient, by type of medicine, and by
day of purchase in ascending order and handled one by one
consecutively.
Duration of the first Rx starts on the day of purchase and lasts
until the amount has been spent according to the daily prescribed
dose. The daily prescribed dose can in theory vary every day. This
information is used to compute the duration of a given Rx. If
subsequent prescriptions are filled within the previous Rx’s
duration time, the duration of these will be pasted to the end of
the previous Rx. Adjustment for waste of the drug is not included.
Any oversupply is capped only at the very end of the follow-up
period.
The medication process, shown graphically in Figure 1 and
concepts detailed in Figure 2, starts with the first written
prescription of a daily dose of a given drug to the patient by the
physician. The patient can adhere to the timing and dosage of
taking the drug only if he or she initiates treatment by filling the
prescription and hereafter refills the prescription with a fre-
quency that allows for a sufficient supply of the drug. If the
patient runs out of supply before refilling the next prescription,
we say that a gap occurs. In the absence of a Medication Event
Monitoring System in register-based studies, we are forced to
assume that medication corresponding to the prescribed daily
dose is taken every day until the available supply has been used.
A scenario in which a patient from time to time skips days to
extend supply—perhaps to save money—will postpone the day
for when the next refill is needed. But a date for when the next
refill is needed cannot be modeled or estimated without unrea-
sonable extra assumptions. Instead, such scenarios will present
themselves as one period with sufficient supply followed by a
gap. A patient is defined as persistent when he or she has a
sufficient supply to implement the prescribed dosing regimen
with gaps between refills that are smaller than a predefined
number of days (here 180 days). If the patient has been without
supply for more than the predefined number of days, then we
assume that treatment has been discontinued by the patient and
he or she is no longer persistent. This is shown with the red
dotted line in Figure 1. Discontinuation may be followed by a later
resumption of treatment, starting the process over again. Treat-
ment can also be discontinued by the physician, and then in
268 VALUE IN HEALTH 17 (2014) 266–274

Time prescribed with medication

1st written 1st filled

prescription prescription (Rx) 2nd Rx 3rd Rx 4th Rx 5th Rx

Waiting time time

Time to Acceptance

Initiation Acceptance

180 days Gap Gap Gap

Gap Nonpersistent

Persistent Persistent
Discontinuation
Days with sufficient supply to implement dosing regimen
Days without supply

Fig. 1 – Example of a patient’s time line in different states of adherence to medication. The patient is in waiting from the
time the medication is prescribed and until the first Rx, when treatment is initiated. The patient is able to implement
the dosing regimen when he or she has a sufficient supply to cover the daily prescribed dose. Gaps occur when the
patient is without supply of medication. He or she is still persistent, if he or she does not exceed the cutoff value
(here 180 days) for a gap. When the number of consecutive days without the supply of medication exceeds the cutoff
value for the length of a gap, the patient has discontinued treatment and is no longer persistent. At point of second Rx, the
patient resumes treatment, because he or she has a sufficient supply to implement the dosing regimen again. Rx, filled
prescription.

some cases resumed with a new written prescription at a

later point.
With the above definitions, the medication process can
be transformed into a multistate model as described in
Figure 2. The figure explains the events leading to a change
of state and how cutoff values in an algorithm determine
when a gap turns into discontinuation of treatment and
termination of persistence. A patient’s acceptance of treatment
is defined as having filled at least two prescriptions, the first
fill of a prescription within 360 days after the physician pre-
scribed the medication and the second filled within 360 days
after fill of the first. If not, the patient is regarded as not
accepting treatment (sometimes denoted as early nonper-
sistence [25]).
As shown in Figure 3, during time at Steno, a patient, on any
given day for a particular medicine, will be in one of the following
six states: 1) waiting to initiate treatment; in treatment and
persistent; either 2) with a sufficient supply to implement the
dosing regimen; or 3) in a gap without supply of medication; 4)
nonpersistent, because treatment has been discontinued; 5) non-
accepting, because treatment has not been accepted with at least
two Rxs within the allowed span of time; or 6) without a
prescription for the drug, either because the physician has not
prescribed it or because the physician has discontinued the
prescription for the drug.
All the arrows in Figure 3 show the possible transitions
between states. With the described data, incidence rates can be
calculated for each transition, and is a way of describing the
variability in the refill behavior.
Calculations and analyses were performed by using SAS,
version 9.2 (SAS Institute, Inc., Cary, NC). Graphs were done in
R, version 2.15.2 (http://www.R-project.org/).
Results from the Case
The prevalence of the five different states while prescribed with
metformin and simvastatin, respectively, during up to 5 years of
follow-up, is shown in Figure 4. The date of entry (index date) per
patient per ATC code is the date of the first written prescription
in EMR after January 1, 1998, and time since this date is used as
our primary time scale. The graphs show that the proportion of
patients with sufficient supply to implement the prescribed
dosing regimen reaches a relatively stable level within the first
3 months. Some 10% of metformin patients fail to fill the first
prescription within 360 days and are categorized as nonaccept-
ing. Some of these patients gradually over time initiate treatment
shown by the decrease in the nonaccepting proportion of
patients.
The degree of persistence during the first year and the fifth year
among the patients prescribed with simvastatin is 77.0% (95%
confidence interval [CI] 72.7%–81.3%) and 84.4% (95% CI 80.2%–
88.2%), respectively. The difference in the degree of implementation
during the first year and the fifth year among patients persistent to
simvastatin is less pronounced with 85.1% (95% CI 81.5%–88.8%) and
88.3% (95% CI 85.0%–91.6%), respectively.
If the definition of a gap is changed from 180 days to 100 days
or 50 days, the prevalence for gaps and nonpersistence changes
substantially as is shown in the center and right panels in
Figure 4. Decreasing the cutoff value shifts person-time from
the gap state to the nonpersistent state, and vice versa when the
cutoff value is increased.
The median waiting time to fill a prescription for the first time
among those who filled were as follows: for metformin, 7 days
(interquartile range 1–28 days); for simvastatin, 9 days (inter-
quartile range 1–49 days).
 VALUE IN HEALTH 17 (2014) 266–274
Fig. 2 – Definitions of states and events related to adherence to medication.
Where the prevalence describes the adherence pattern for the
entire population, the incidence rates explain the volatility of the
patients’ transitions between states. Empirical incidence rates for
selected transitions (indicated with α, β, γ, and μ in Fig. 3) are
shown in Table 1.
Time to first gap and time to first discontinuation after the
date of initiation is shown in Figure 5. On comparing metformin
with simvastatin, we found that more patients on metformin
commence earlier on a gap than do patients on simvastatin. More
patients on simvastatin discontinue treatment earlier than do
patients on metformin, regardless of the choice of the cutoff
value (50, 100, or 180 days).
Discussion
Principal Findings
We developed a multistate model by defining the possible states
of a patient, and when transitions between states occur. The
model provides a tool to use prescription data to estimate
transition rates between states over time. Thus, changes in
transition rates and prevalence over time can be estimated on
several time scales (calendar time, age of the patients, and time
since indication for medication).
Adherence to medication has often been reported as a binary
outcome (poor adherence o80% or good adherence Z80%) [26,27].
269
It is hard to believe that a dichotomization of long-term patient
behavior related to adherence to medication (poor or good)
provides much useful information. The model we have proposed
overcomes the limitations that analyses based on summary
statistics face because it uses the time points for transitions
between states and differentiates between several types of poor
adherence: patients who do not accept treatment initially;
patients who discontinue early; or patients with a low degree of
implementation, which indicate many or long gaps. This is more
useful when identifying which patients or drugs to target for
counseling or other types of intervention to improve adherence.
The multistate model also has the potential for looking into
health outcomes (e.g., long-term diabetes complications) related
to specific patterns of adherence to medication and to analyze
determinants of changes in transition rates and prevalence of the
different states over time (socioeconomic factors, medication
complexity, comorbidity, etc.).
If treatment of the chronic disease under study is
characterized either by (frequent) switches between drugs
or by add-ons of more drugs from the same therapeutic class,
which is common in T2DM, it would make sense to model the
entire group of drugs. This can be done by transforming
prescribed doses into DDD units and adding up to a total for
the group of drugs. A similar procedure must be used for the
filled amounts and then the model must be applied as
shown. Analyses of patients’ health outcomes related to
their pattern of adherence to medication would benefit from
270 VALUE IN HEALTH 17 (2014) 266–274
Transferred to Steno
Time at Steno
Patient is prescribed with medication
Persistent
Waiting
With supply to
implement dosing
regimen
No medication
Gap without
prescribed
supply, but
still persistent
Nonpersistent
Nonaccepting
Exit Steno Dead
Referred to GP or
other specialist unit
Fig. 3 – Multistate model and moves between states.
The five colored boxes inside the dotted lines represent the
five different states during time of follow-up for adherence
to medication. Follow-up for adherence to medication is
irrelevant for the brown box to the left
where the medication is not prescribed. The patient is
regarded as persistent when he or she has a sufficient
supply to implement the dosing regimen or experience
gaps without supply between refills that are smaller
than 180 days each. The arrows, regardless of color,
show all the possible moves between states. Between
some of the states, the patient can move both ways (a
bidirectional multistate model). Some of the moves (arrows)
have been highlighted with a color for further comments. α
next to the orange arrow is the incidence rate for patients
moving from being with sufficient supply to implement the
dosing regimen to a gap without supply. γ next to the red
arrow is the incidence rate for patients discontinuing
treatment by moving from a gap to being nonpersistent. The
βs next to the green arrows are the incidence rates for
changing status into being with sufficient supply to
implementing the dosing regimen again.The μs next to the
purple arrows are the incidence rates for changing status to
not having accepted treatment. Incidence rates for the
colored moves (arrows) are shown in Table 1. GP, general
practitioner.
applying the model to a group of drugs. If focus is differences in
adherence between two drugs, the model must be applied on
each drug and then compared, as shown with metformin and
simvastatin.
Strengths and Limitations
In contrast to most studies using registers with reimbursement
claims, Danish data offer a link between data on prescribed
medication and data on (re)filled medication. In the
Danish universal tax-funded nationwide health care system
[28], patients must pick up prescription drugs at a Danish
pharmacy, where all purchases are captured by the RMPS.
Reimbursements are subtracted at payment of the drug at
delivery, and so it is considered very unlikely that patients get
hold of their prescription drugs from sources other than the
Danish pharmacies.
This is a study based on electronic registers, and thus does
not depend on patients’ self-reported adherence to medication.
We argue that it must be difficult to remember something
you previously have forgotten, why self-reported adherence
to medication presumably can be heavily biased. Significant
differences between self-reported adherence and adherence
measured by Medication Event Monitoring Systems have
been reported [29]. Our study does not include data from a
Medication Event Monitoring System as is often seen in clinical
trials [30].
Our data record to which extent the drugs are available to the
patient, but not the extent to which the drugs are actually
ingested. Drugs wasted, lost, or shared with other users contrib-
ute to the difference between drugs available and drugs ingested
and we have no data on this.
Underestimating adherence using this methodology can be
triggered in three instances: 1) if the prescription of the drug
has been discontinued by a physician without a registration in
the EMR. This may also happen if physicians outside Steno
intervene. The RMPS captures transactions of all Rxs that have
been prescribed by physicians at outpatient clinics at other
Danish hospitals, but information on the prescribed doses
from these doctors outside Steno is not available in our data.
In case of overlapping prescriptions from medical doctors
outside Steno, which can occur, a discontinuation from another
doctor without Steno’s knowledge will erroneously induce
poor adherence or nonpersistence in this analysis; 2) if the
true prescribed dose is smaller than the actual registered
dose in EMR; 3) if the patient is supplied with drugs from a
hospital during hospitalization because the RMPS does not
capture information on drugs administered to patients from
the hospitals’ stocks during hospitalization at an individual
level.
False-positive adherence using this methodology may be
triggered if the true prescribed dose is larger than the actual
registered dose in the EMR.
Strengths and Weaknesses in Relation to Other Studies
By using these Danish data, it is possible to investigate long-term
adherence to medication among patients for up to 11 years.
Investigation of adherence to medication from initiation of
therapy to up to typically 3 years after has often been seen in
other studies [2,31–36].
Use of patients’ self-reporting of adherence to medication is
avoided in this study. Self-reported adherence to medication may
overestimate adherence to medication because a disproportion-
ate number of healthy users may sign up to the study [37,38],
because of recall and reporting bias [39], or because the attention
from the clinical trial staff may produce an improvement
in human behavior among the participants (the “Hawthorne”
effect) [40].
Studies using reimbursement claims data may underestimate
the adherence to medication for some of the individuals in the
population under investigation, if not all dispensed medication
has been reimbursed [40] or if the health care system does not
offer universal coverage [13,41]. These types of biases are absent
when using the Danish data.
This model captures all patients prescribed with the medi-
cation in question, including the patients who never fill a
prescription or fill only once, which is not always the case in
studies using reimbursement claims data, which may overesti-
mate the proportion of adherent individuals among the
 VALUE IN HEALTH 17 (2014) 266–274 271

100

100

100
Waiting Nonaccepting
Nonpersistent
Gap
80

80

80
60

60

60
40

40

40
With sufficient supply to
implement dosing regimen
20

20

20
Max. length of a gap = 180 days Max. length of a gap = 100 days Max. length of a gap = 50 days
0

0
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
100

100

100
Waiting
Nonaccepting
80

80

80
Nonpersistent
Gap
60

60

60
40

40

40
With sufficient supply to
implement dosing regimen
20

20

20

Max. length of a gap = 180 days Max. length of a gap = 100 days Max. length of a gap = 50 days
0

0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5

Fig. 4 – Time course of adherence to metformin and simvastatin (prevalence over 5 years). The green area is the prevalence
of patients with a sufficient supply of medication to implement the dosing regimen. The blue area is the prevalence of
patients waiting to fill the first prescription after the medication has been prescribed. The prevalence of patients waiting
stabilizes around 3 to 4 months after the medication has been prescribed. After 1 year, a proportion of the patients (purple
color) are categorized as not accepting treatment. However, some of these do initiate treatment later, narrowing the purple
area. The orange area is the prevalence of the patients who no longer have a sufficient supply of medication available
and thus experience a gap. For simvastatin, an increase in the orange area and a decrease in the green area are seen
after 3 months, followed by a gradual increase in the prevalence of patients with sufficient supply (the second green peak).
This is an illustration of some patients filling the first prescription with medication for 3 months, but managing to
fill the second prescription only after having run out of supply after the first fill. The red area is the prevalence of patients
having discontinued treatment. The size of the area depends on the choice of the cutoff value for the maximum acceptable
length of a gap.

population under investigation, because the person-years in the assume that every Rx lasts a fixed number of days or do require
waiting state are not recorded [25,42]. only a minimum number of refills during a prespecified period of
Studies investigating persistence/possession measures do not time. That may over- or underestimate adherence to medication
necessarily take the actual prescribed doses into account, but [32,42,43].
272 VALUE IN HEALTH 17 (2014) 266–274

Table 1 – Incidence rates for metformin and simvastatin.

Incidence rate (95% CI) Transition No. of Person-
transitions/person-year transitions years
To state From state

Metformin
Commencing a gap α 2.82 (2.77–2.87) Gap With supply 13,449 4,772.0
Discontinuation to γ 0.27 (0.24–0.31) Nonpersistent Gap 224 820.2
nonpersistent
Refilling of medication after a β1 15.46 (15.19–15.73) With supply Gap 12,682 820.2
gap
Refilling of medication after β2 1.29 (1.06–1.51) With supply Nonpersistent 125 97.0
being nonpersistent
Not accepting treatment μ 0.02 (0.02–0.03) Nonaccepting 4 states* 144 6,060.9
Simvastatin
Commencing a gap α 1.41 (1.38–1.44) Gap With supply 6,809 4,833.0
Discontinuation to γ 0.50 (0.45–0.55) Nonpersistent Gap 394 793.2
nonpersistent
Refilling of medication after a β1 7.43 (7.24–7.62) With supply Gap 5,891 793.2
gap
Refilling of medication after β2 1.04 (0.89–1.19) With supply Nonpersistent 183 176.2
being nonpersistent
Not accepting treatment μ 0.06 (0.05–0.06) Nonaccepting 4 states* 379 6,459.7
Note. Incidence rates for number of transitions per person-year from one state to another. Based on a maximum acceptable length of 180 d
for a gap.
* The four states are 1) with supply to implement dosing regimen, 2) a gap, 3) nonpersistent, and 4) waiting.

Generalizability the drug concentration in plasma [11]. For other types of medi-
cation, it will only make sense to operate with a very short
The algorithm requires definition of cutoff values for acceptance
maximum acceptable length of a gap before a patient must be
and gaps. It is outside the scope of this article to discuss
considered as nonpersistent. The choice of cutoff values must
the “correct” choice of the maximum acceptable length of a
always depend on the drugs under study and hence is a clinical
gap before a patient is considered to have discontinued treatment
decision.
and has become nonpersistent. Even rather long gaps may not
The multistate methodology, detailed here, is applicable
be a problem in cases of drugs with a long half-life or forgiveness,
in cases in which detailed information on prescribed and filled
or in cases in which the outcomes of the medication are linked
medication is available at an individual level. Thus, in Scandi-
to long-term medication effects [43]. It is known that the carry-
navian countries, Scotland, and Healthcare Management
over effect of a drug in many cases lags well behind changes in
2500

2500
2000

2000

time to first gap

time to first gap

1500

1500
1000

1000

max gap of 50 days

max gap of 50 days

500

500

max gap of 100 days

max gap of 100 days

max gap of 180 days max gap of 180 days

0

0 1 2 0 1 2

Fig. 5 – Time to first gap and first discontinuation after date of initiation. Time to discontinuation is shown in three different
versions using a maximum acceptable length of a gap of 50, 100, and 180 days, respectively.
 VALUE IN HEALTH 17 (2014) 266–274
Organizations [44], it would be possible to conduct similar
analyses. Danish health care professionals may use the Medicine
Profile (at https://www.sundhed.dk/sundhedsfaglig/min-side/
patientdata/medicinprofilen/ [in Danish]) for online calculation
of a specific patient’s adherence to medication, provided the
patient has consented [45]. Flaws in the Medicine Profile, how-
ever, have been reported [46].
Future Research
This study and WHO’s estimates of a low degree of adherence to
long-term therapies in general indicate a large potential for
improved health and quality of life among patients with poor
adherence to medication, if identified and helped to improve-
ment of adherence. We will pursue to develop a tool for the
identification of patients with different types of poor adherence
to medication. Associations between measures of adherence to
medication based on this model and health outcomes are also to
be investigated.
Sources of financial support: This study has in part been
funded by The Danish Agency for Science, Technology and
Innovation (at The Ministry of Science, Innovation and Higher
Education) and “danmark” sygeforsikring (a mutual health insur-
ance company). The funders had no role in study design, data
collection and analysis, decisions to publish, or preparation of the
manuscript.
R EF E R EN CE S
[1] World Health Organization, Sabete E, ed. Adherence to Long-Term
Therapies: Evidence for Action. Geneva: World Health Organization,
2003.
[2] Cramer JA. A systematic review of adherence with medications for
diabetes. Diabetes Care 2004;27:1218–24.
[3] Vermeire E, Hearnshaw H, van Royen P, Denekens J. Patient adherence
to treatment: three decades of research. A comprehensive review. J Clin
Pharm Ther 2001;26:331–42.
[4] World Health Organization, Sabete E, ed. Adherence to Long-Term
Therapies: Policy for Action. Meeting Report. Geneva: World Health
Organization 2001.
[5] Rozenfeld Y, Hunt JS, Plauschinat C, Wong KS. Oral antidiabetic
medication adherence and glycemic control in managed care. Am J
Manag Care 2008;14:71–5.
[6] Schectman J, Nadkarni M, Voss J. The association between diabetes
metabolic control and drug adherence in an indigent population.
Diabetes Care 2002;25:1015–21.
[7] Lau D, Nau DP. Oral antihyperglycemic medication nonadherence and
subsequent hospitalization among individuals with type 2 diabetes.
Diabetes Care 2004;27:2149–53.
[8] Howard RL, Avery AJ, Slavenburg S, et al. Which drugs cause
preventable admissions to hospital? A systematic review. Br J Clin
Pharmacol 2007;63:136–47.
[9] Ho P, Rumsfeld J, Masoudi F, et al. Effect of medication nonadherence
on hospitalization and mortality among patients with diabetes
mellitus. Arch Intern Med 2006;166:1836.
[10] Jha AK. Greater adherence to diabetes drugs is linked to less hospital
use and could save nearly $5 billion annually. Health Aff
2012;31:1836–46.
[11] Urquhart J. Variable compliance and persistence with prescribed drug
dosing regimens: implications for benefits, risks, and economics of
pharmacotherapy. In: Strom BL, ed., Pharmacoepidemiology (4th ed.).
John Wiley & Sons, Ltd; 2005:767–89.
[12] DiMatteo M. Variations in patients’ adherence to medical
recommendations: a quantitative review of 50 years of research. Med
Care 2004;42:200–9.
[13] Schneeweiss S, Avorn J. A review of uses of health care utilization
databases for epidemiologic research on therapeutics. J Clin Epidemiol
2005;58:323.
[14] Cramer JA. Medication compliance and persistence: terminology and
definitions. Value Health 2008;11:44–7.
[15] Peterson A, Nau D, Cramer J, et al. A checklist for medication
compliance and persistence studies using retrospective databases.
Value Health 2007;10:3–12.
273
[16] Vrijens B, De Geest S, Hughes DA, et al. A new taxonomy for describing
and defining adherence to medications. Br J Clin Pharmacol
2012;73:691–705.
[17] Lehane E, McCarthy G. Intentional and unintentional medication
non-adherence: a comprehensive framework for clinical
research and practice? A discussion paper. Int J Nurs Stud
2007;44:1468–77.
[18] Edwards LD, Fox AW, Stonier PD, eds. Principles and Practice of
Pharmaceutical Medicine (3rd ed.). West Sussex: Wiley-Blackwell, 2011.
[19] Putter H, Fiocco M, Geskus RB. Tutorial in biostatistics: competing risks
and multi-state models. Stat Med 2007;26:2389–430.
[20] International Diabetes Federation. IDF diabetes atlas. Available from:
http://www.idf.org/diabetesatlas/5e/the-global-burden. [Accessed
September 4, 2013].
[21] Nathan DM, Buse JB, Davidson MB, et al. Medical management of
hyperglycemia in type 2 diabetes: a consensus algorithm for the
initiation and adjustment of therapy—a consensus statement of the
American Diabetes Association and the European Association for the
Study of Diabetes. Diabetes Care 2009;32:193–203.
[22] Danish Endochrine Society, Institute for Rational Pharmacotherapy,
The Danish College of General Practitioners. Guidelines for type
2-diabetes [pamphlet]. 2011.
[23] WHO Collaborating Centre for Drug Statistics Methodology.
ATC: structure and principles. Available from: http://www.
whocc.no/atc/structure_and_principles/. [Accessed September 4,
2013].
[24] WHO Collaborating Centre for Drug Statistics Methodology. DDD:
definition and general considerations. Available from: http://www.
whocc.no/ddd/definition_and_general_considera/. [Accessed
September 4, 2013].
[25] Raebel MA, Carroll NM, Ellis JL, et al. Importance of including early
nonadherence in estimations of medication adherence. Ann
Pharmacother 2011;45:1053–60.
[26] Simpson SH, Eurich DT, Majumdar SR, et al. A meta-analysis of the
association between adherence to drug therapy and mortality. BMJ
2006;333(7575):15.
[27] DiMatteo M, Giordani P, Lepper H, Croghan T. Patient adherence and
medical treatment outcomes: a meta-analysis. Med Care
2002;40:794–811.
[28] Johannesdottir SA, Mægbæk ML, Hansen JG, et al. Correspondence
between general practitioner-reported medication use and timing of
prescription dispensation. Clin Epidemiol 2012;4:13–8.
[29] Christensen A, Christrup L, Fabricius P, et al. The impact of an
electronic monitoring and reminder device on patient compliance with
antihypertensive therapy: a randomized controlled trial. J Hypertens
2010;28:194–200.
[30] Blaschke T, Osterberg L, Vrijens B, Urquhart J. Adherence to
medications: insights arising from studies on the unreliable link
between prescribed and actual drug dosing histories. Ann Rev
Pharmacol Toxicol 2012;52:275–301.
[31] Steiner J, Prochazka A. The assessment of refill compliance using
pharmacy records: methods, validity and applications. J Clin Epidemiol
1997;50(1):105–16.
[32] Caetano PA, Lam JMC, Morgan S. Toward a standard definition and
measurement of persistence with drug therapy: examples from research
on statin and antihypertensive utilization. Clin Ther 2006;28:1411–24.
[33] Van Wijk BL, Klungel OH, Heerdink ER, de Boer A. Refill persistence
with chronic medication assessed from a pharmacy database was
influenced by method of calculation. J Clin Epidemiol 2006;59(1):11–7.
[34] Mantel-Teeuwisse A, Klungel O, Verschuren W, et al. Comparison of
different methods to estimate prevalence of drug use by using
pharmacy records. J Clin Epidemiol 2001;54:1181–6.
[35] Cramer JA, Benedict Á, Muszbek N, et al. The significance of compliance
and persistence in the treatment of diabetes, hypertension and
dyslipidaemia: a review. Int J Clin Pract 2007;62:76–87.
[36] Donnan P, MacDonald TM, Morrist AD. Adherence to prescribed oral
hypoglycaemic medication in a population of patients with type 2
diabetes: a retrospective cohort study. Diabet Med 2002;19:279–84.
[37] Osterberg L, Blaschke T. Drug therapy: adherence to medication. N Engl
J Med 2005;353:487–97.
[38] Gadkari AS, Pedan A, Gowda N, McHorney CA. Survey nonresponders to
a medication-beliefs survey have worse adherence and persistence to
chronic medications compared with survey responders. Med Care
2011;49:956–61.
[39] Hansen RA, Kim MM, Song L, et al. Comparison of methods to assess
medication adherence and classify nonadherence. Ann Pharmacother
2009;43(3):413–22.
[40] Hamilton RA, Briceland LL. Use of prescription-refill records to assess
patient compliance. Am J Hosp Pharm 1992;49:1691–6.
[41] Zhang Q, Zhao C, Davies M, et al. Compliance and persistence with
concomitant statin and oral antihyperglycemic therapy. Am J Manag
Care 2011;17:746–52.
274 VALUE IN HEALTH 17 (2014) 266–274
[42] Evans CD, Eurich DT, Remillard AJ, et al. First-fill medication
discontinuations and nonadherence to antihypertensive therapy: an
observational study. Am J Hypertens 2012;25:195–203.
[43] Greevy RA, Huizinga MM, Roumie CL, et al. Comparisons of persistence
and durability among three oral antidiabetic therapies using electronic
prescription-fill data: the impact of adherence requirements and
stockpiling. Clin Pharmacol Ther 2011;90:813–9.
[44] Singer SR. EMR-based medication adherence metric markedly
enhances identification of nonadherent patients. Am J Manag Care
2012;18:e372–7.
[45] Andersen LS, Kjeldsen LJ, Haugbølle LS. Anvendelighed af den
personlige elektroniske medicinprofil til vurdering af komplians [in
Danish] [Suitability of the Personal Electronic Medication profile for
estimation of medication compliance]. Ugeskr Læger 2009;171
(11):899–903.
[46] Harbig P, Barat I, Lund Nielsen P, Damsgaard EM. Instantaneous
detection of nonadherence: quality, strength, and weakness of an
electronic prescription database. Pharmacoepidemiol Drug Saf
2012;21:323–8.