Costs of Medication Nonadherence in Patients withDiabetes Mellitus:A Systematic Review and CriticalAnalysis of the Literaturevhe_539 915..922

Maribel Salas, MD, DSc, MSc,1 Dyfrig Hughes, MSc, PhD, MRPharmS,2 Alvaro Zuluaga, MD,3

Kawitha Vardeva, MSc,4 Maximilian Lebmeier, MSc5

1Division of Preventive Medicine, University of Alabama at Birmingham, Birmingham,AL, USA; 2Centre for Economics and Policy in Health,Bangor University, Bangor, UK; 3University of Alabama at Birmingham, Birmingham,AL, USA; 4Amgen, Cambridge, UK; 5Wyeth in Taplow,Taplow, UK

ABSTRACT

Objectives: Information on the health care costs associated with nonad-herence to treatments for diabetes is both limited and inconsistent. Wereviewed and critically appraised the literature to identify the main meth-odological issues that might explain differences among reports in therelationship of nonadherence and costs in patients with diabetes.Methods: Two investigators reviewed Medline, EMBASE, Cochranelibrary and CINAHL and studies with information on costs by level ofadherence in patients with diabetes published between January 1, 1997and September 30th 2007 were included.Results: A total of 209 studies were identified and ten fulfilled the inclu-sion criteria. All included studies analyzed claims data and 70% werebased on non-Medicaid and non-Medicare databases. Low medicationpossession ratios were associated with higher costs. Important differences

were found in the ICD-9/ICD-9 CM codes used to identify patients andtheir diagnoses, data sources, analytic window period, definitions ofadherence measures, skewness in cost data and associated statistical issues,adjustment of costs for inflation, adjustment for confounders, clinicaloutcomes and costs.Conclusions: Important variation among cost estimates was evident, evenwithin studies of the same population. Readers should be cautious whencomparing estimated coefficients from various studies because method-ological issues might explain differences in the results of costs of nonad-herence in diabetes. This is particularly important when estimates are usedas inputs to pharmacoeconomic models.Keywords: costs, diabetes, economics, medication adherence, medicationcompliance.

Introduction

Nonadherence has a significant impact on the cost-effectivenessof pharmaceuticals [1], and has been estimated to cost the USeconomy up to $100 billion per year [2]. In diabetes, nonadher-ence to oral hypoglycemic medications [3,4] may partly explainwhy only 43% of patients with diabetes mellitus have glycosy-lated hemoglobin (HbA1c) below the 7% level [5,6] recom-mended by the American Diabetes Association [7].

Studies of adherence in diabetes have focused on its eco-nomic burden [8–10], its complications [11,12] and the cost-effectiveness of antidiabetic drugs [13–18]. Many have reportedwide variation in the percent of patients being “nonadherent,”ranging from 13% to 64% for oral agents and from 19% to 46%for users of insulin [19–21]. Additionally, important variations inthe coefficient estimations for costs have been reported [21,22],which might be related to differences in the design, population,variables included in the analysis and statistical analyses. There-fore, we reviewed and critically appraised the literature to iden-tify the main methodological issues that might explain differencesamong reports in the relationship of nonadherence and costs inpatients with diabetes.

Methods

Search StrategyWe conducted a systematic literature review using Medline,EMBASE, Cochrane Library, and the Cumulative Index to

Nursing and Allied Health Literature (CINAHL) from January 1,1997 to September, 30 2007.

The key terms used included: (compliance, adherence, persis-tence, nonadherence, concordance) AND (economics, costs,value, expenditures, resource utilization) AND (diabetes, hyper-glycemia, diabetes-related complications, antidiabetic medica-tions, insulin, oral hypoglycemic agents). We also hand-searchedmedical journals and reviewed the reference lists of otherreviews.

Selection CriteriaStudies that reported costs by different levels of medicationadherence or persistence were included. Adherence and persis-tence definitions were according to previous studies [23]. We alsoincluded studies that used HbA1c as a proxy of medicationadherence because HbA1c is a well-established measure of gly-cemic control [22,24,25] and a proxy for adherence [26]. Non-English studies, articles with insufficient data, and those withoutcosts or adherence information were excluded.

Extracted InformationAbstracts and full publications were reviewed by two researchersand disagreements were resolved by consensus. The extractedinformation included the study design, data source(s), methodsof adherence measurement, statistical analysis, and results. Studydesigns were classified as trials, cohort, case-control, or cross-sectional studies. Data sources for patient demographics, adher-ence, resource utilization, and costs, as well as observation andfollow-up periods, were recorded (Table 1). For statistical analy-sis, we included information on any statistical method usedto assess the relationship or association between medication

Address correspondence to: Maribel Salas, University of Alabama at Bir-mingham, 1530 3rd Avenue South MT 644, Birmingham, AL 35294-4410, USA. E-mail: msalas@dopm.uab.edu

10.1111/j.1524-4733.2009.00539.x

Volume 12 • Number 6 • 2009V A L U E I N H E A L T H

© 2009, International Society for Pharmacoeconomics and Outcomes Research (ISPOR) 1098-3015/09/915 915–922 915

Tabl

e1

Stud

ies

iden

tified

with

cost

sre

port

edby

adhe

renc

ele

veli

ndi

abet

icpa

tient

s

Ref

eren

ceD

esig

n

Sour

ceof

data

Incl

usio

ncr

iteri

aO

bser

vatio

npe

riod

Follo

w-u

ppe

riod

Popu

latio

nD

iagn

osis

Adh

eren

ceC

osts

Balk

rish

nan

R,

2003

[28]

Ret

rosp

ectiv

eco

hort

Med

icar

eH

MO

inN

orth

Car

olin

aIC

D-9

code

s25

0.xx

Pres

crip

tion

refil

lsR

eim

burs

emen

tby

the

HM

OPa

tient

sag

ed�

65ye

ars,

enro

lled

ina

Med

icar

eH

MO

inN

orth

Car

olin

aw

hore

ceiv

ed�

1an

tidia

betic

pres

crip

tion

disp

ense

dev

ery

6m

onth

s

1996

–200

2U

pto

5ye

ars

Cob

den

D,2

007

[29]

Ret

rosp

ectiv

eco

hort

Phar

met

rics

ICD

-9C

Mco

de25

0.xx

excl

udin

gty

pe1

subc

odes

Pres

crip

tion

refil

lsPa

ymen

tsm

ade

byth

ird-

part

ypa

yers

tohe

alth

care

prov

ider

s(r

eim

burs

emen

t)

�18

year

s,ty

pe2

diab

etes

who

conv

erte

dto

BIA

sp70

/30

pen

devi

cean

dpr

evio

usly

trea

ted

with

hum

anor

anal

ogin

sulin

Janu

ary

1,20

01to

Apr

il30

,200

5A

tle

ast

2ye

ars

Balk

rish

nan

R,

2004

[30]

Ret

rosp

ectiv

eco

hort

Nor

thC

arol

ina

Med

icai

dpr

ogra

mIC

D-9

Pres

crip

tion

refil

lsR

eim

burs

emen

tTy

pe2

diab

etes

who

wer

ene

wly

star

ted

onth

iazo

lidin

edio

neth

erap

yor

othe

ror

alan

tidia

betic

drug

July

2001

toJu

ne20

022

year

s

Hep

ke,2

004

[31]

Ret

rosp

ectiv

eco

hort

Blue

Cro

ssBl

ueSh

ield

ofM

ichi

gan

ICD

-925

0,35

2.2,

362,

366.

41,6

48Pr

escr

iptio

nre

fills

Rei

mbu

rsem

ent

Non

-Med

icar

eel

igib

leM

ichi

gan

resi

dent

sen

rolle

dco

ntin

uous

lyin

1999

,at

leas

t1

inpa

tient

orEm

erge

ncy

room

clai

m,�

2pr

ofes

sion

alor

outp

atie

ntfa

cilit

ycl

aim

sw

ithdi

abet

esdi

agno

sis

and

afil

led

pres

crip

tion

for

antid

iabe

ticdr

ug.

1999

1ye

ar

Lee

WC

,200

6[1

7]R

etro

spec

tive

coho

rtw

ithpr

ean

dpo

stan

alys

is

Inte

grat

edm

edic

alan

dph

arm

acy

clai

ms

data

base

:Pha

rmet

ric

ICD

-9co

de25

0.xx

excl

udin

gty

pe1

subc

odes

Pres

crip

tion

refil

lsPa

ymen

tsto

the

heal

thin

sura

nce:

reim

burs

emen

t

�18

year

sof

age,

type

2di

abet

esw

hoin

itiat

edtr

eatm

ent

with

insu

linan

alog

uepe

nde

vice

betw

een

July

1,20

01an

dD

ec31

,200

2,an

dw

hose

trea

tmen

tw

asco

nver

ted

from

conv

entio

nalh

uman

oran

alog

uein

sulin

inje

ctio

n(v

ial/s

yrin

ge)

toa

prefi

lled

insu

linan

alog

uepe

n.

Janu

ary

2001

–Apr

il20

05U

pto

4ye

ars

Shen

olik

arR

A,

2006

[32]

Ret

rosp

ectiv

eco

hort

Nor

thC

arol

ina

Med

icai

dda

taba

seIC

D-9

CM

code

250.

xxPr

escr

iptio

nre

fills

Tota

lhea

lthca

reco

sts:

med

ical

and

dent

alca

re,

regu

lar

chec

kups

,offi

cevi

sits

,hom

ehe

alth

care

,in

patie

ntan

dou

tpat

ient

care

,lon

gte

rmca

refa

cilit

yca

rean

dpr

escr

iptio

ndr

ugs.

At

leas

ton

eIC

D9

code

for

diab

etes

,and

one

for

antid

iabe

ticm

edic

atio

nan

dM

edic

aid

elig

ibili

tyfo

r36

-mon

thfo

llow

-up

peri

od.A

fric

anA

mer

ican

sw

ere

anal

yzed

vs.o

ther

July

1,20

00to

June

30,2

003

1ye

ar

Soko

lMC

,200

5[3

3]R

etro

spec

tive

coho

rtA

dmin

istr

ativ

ecl

aim

sda

taba

sem

aint

aine

dby

ahe

alth

plan

orga

niza

tion

ICD

-9co

des

250.

xx,

357.

2,36

2.0x

,36

6.41

,648

.0

Pres

crip

tion

refil

lsA

ll-ca

use

cost

san

ddi

seas

e-re

late

dco

sts.

Patie

nts

aged

65an

dol

der

with

diag

nosi

sof

diab

etes

June

1997

toM

ay19

991

year

Wag

ner

EH,

2001

[34]

Ret

rosp

ectiv

eco

hort

Aut

omat

eddi

abet

esre

gist

ryfr

omth

eG

roup

Hea

lthC

oope

rativ

eof

Puge

tSo

und,

Seat

tleW

ashi

ngto

n

Dia

gnos

isof

diab

etes

and

HbA

1cfr

omdi

abet

esre

gist

ry

Pres

crip

tions

refil

lsan

dH

bA1c

Dec

isio

nsu

ppor

tsy

stem

that

isau

tom

ated

,st

ep-d

own

cost

acco

untin

gfo

rhe

alth

care

prov

ided

tom

embe

rs.

Dia

betic

sol

der

than

18ye

ars,

with

atle

ast

one

HB

A1c

,and

cont

inuo

usly

enro

lled

from

1992

–199

6Ja

nuar

y1,

1992

toM

arch

31,1

996

4ye

ars

Whi

te,T

J20

04[3

5]R

etro

spec

tive

coho

rtM

anag

edca

reor

gani

zatio

nda

taba

se

ICD

-9fo

rty

pe2

diab

etes

Perc

enta

geof

adhe

renc

eC

laim

sda

taPa

tient

sre

ceiv

ing

anor

alan

tidia

betic

med

icat

ion

and

have

adi

agno

sis

ofC

VD

,con

tinuo

usly

enro

lled

inth

ehe

alth

plan

,and

�30

year

sof

age

Apr

il1,

1998

toM

arch

31,2

000

1ye

ar

Shet

tyS,

2005

[36]

Ret

rosp

ectiv

eco

hort

US

Man

aged

care

orga

niza

tion

ICD

-9C

Mco

des

250.

x0or

250.

x2N

otre

port

edR

eim

burs

emen

tH

ad�

2cl

aim

sfo

rty

pe2

diab

etes

inei

ther

the

prim

ary

orse

cond

ary

posi

tion,

had

atle

ason

epr

escr

iptio

nfo

ran

oral

hypo

glyc

emic

agen

tan

d/or

insu

lin,h

adat

leas

ton

eav

aila

ble

HbA

1c,w

ere

com

mer

cial

lyin

sure

dw

itha

drug

bene

fit,a

ndha

dat

leas

t6

mon

ths

ofco

ntin

uous

enro

llmen

t.

Janu

ary–

Dec

embe

r20

021

year

CV

D,c

ardi

ovas

cula

rdi

seas

e;H

MO

,Hea

lthM

aint

enan

ceO

rgan

izat

ion;

ICD

-9,I

nter

natio

nalC

lass

ifica

tion

ofD

isea

ses,

9th

Rev

isio

n;IC

D-9

CM

,Int

erna

tiona

lCla

ssifi

catio

nof

Dis

ease

sC

linic

alM

odifi

catio

n.H

bA1c

,gly

cosy

late

dhe

mog

lobi

n.

916 Salas et al.

nonadherence and costs, sample size, adjustment for inflationand/or discounting, adjustment for confounders or for the dayswhen patients were in institutionalized care settings such ashospital, and nursing home (Table 2).

Quality CriteriaA checklist for economic evaluation [27] was modified to assessthe quality of studies. The original checklist contained 35 items,but 5 of them were related to health economic models (12, 14 15,20, and 21), and were not considered applicable to the studiesincluded in the review. We assigned a score of 1 if an articleincluded the required item, and zero if it was not included.Therefore, the maximum score for an article that included allinformation related to study design, data collection, analysis andinterpretation of results was 30.

Results

Search ResultsTwo hundred nine titles were identified and their abstracts werereviewed. Fifty abstracts included information on both adherenceand costs in patients with diabetes, and their full articles wereretrieved. Ten studies [17,28–36] fulfilled the inclusion criteria(Fig. 1). All studies analyzed US claims data using retrospectivecohort studies designs [17,28–36] (Table 1). Three studiesutilized Medicare or Medicaid databases [28,30,32], whileall others used commercial or managed care organizations datasets.

Association between Medication Nonadherenceand CostThere were important variations in the items included in order toestimate costs. For example, one study included only claims forphysician office visits, outpatient services, and hospital stays [29],while another was more comprehensive, and included: costsfor hospitalization, outpatient care, emergency care, clinic visits,laboratory tests, professional services, and pharmaceuticals [31].Two studies took into account the net cost to the plan but theydid not include patients’ copayments and deductibles [33,35],while a third study included copayments and deductibles [36].The study by Wagner used its own internal accounting systemthat included overhead costs [34]. It was unclear in some studiesas to which specific costs were included [17,28,32].

Low medication possession ratios (MPRs) were generallyassociated with higher costs. For example, one study reported anassociation of MPR of 60% with mean total costs of $8699 [29].Balkrishnan et al. found that a 10% increase in MPR for anantidiabetic medication was associated with an 8.6% reductionin total annual health care costs [28]. Studies generally reportedincrements of mean annual costs according to baseline HbA1cvalues. For example, the mean annual costs for patients withbaseline HbA1c < 8% were $4475, while for those withHbA1c > 10 were $8088 [34] (Table 2).

Methodological IssuesThe specific International Classification of Diseases (ICD-9) orICD-9 Clinical Modification (ICD-9-CM) codes used to identifythe study population were not mentioned in three studies[30,34,35], and among those that were reported, there wereimportant variations in the codes included (Tables 1 and 2). Forexample, some studies included type 1 and type 2 diabetes[28,31,33], while others excluded type 1 diabetes [17,29,36].The population varied by study, as well by period of observation.

The maximum follow-up was 5 years, and half of the studiesfollowed patients for only 1 year.

Table 2 presents measures of adherence, costs, statisticalanalysis, results of each study, and quality score. All studies usedclaims data to collect drug utilization information. Five studiesused MPR as a measure of medication adherence [17,28–30,32],two studies did not report a specific medication adherencemeasure [34,36], and three used various measures of medicationadherence such as medication adherence rate [31], percentagedays supplies [33], and percentage of adherence [35]. All studiesused the total follow-up period to calculate adherence and costs,and used charges as proxy for costs. In terms of type of costs,some studies reported total health-care costs [17,28–30,34–36],while others focused on overall costs of health care [31], or costsrelated with diabetes care [32,33]. Two studies used Poissonregression models for costs [17,29], and the remainder usedmultivariate regression analysis for costs. Few studies log-transformed costs [32,34,36], and only one study [22] tried todeal with the skewed distribution of both health-care costs andMPRs. Seven studies were able to adjust for some potentialconfounders [17,28,29,31–33,36], while only one adjusted costsfor inflation and duration of hospitalizations [28]. Most studieswere assigned a low quality score (<50% of required informa-tion), ranging from 8/30 to 14/30.

Discussion

We identified various methodological issues that hinder compari-sons from being made across studies, and which might result insignificant differences in the reported associations between non-adherence to medicines and costs in patients with diabetes.

Based on the International Society for Pharmacoeconomicsand Outcomes Research (ISPOR) recommendations on improv-ing the quality of adherence studies [37,38], we found that thetype of study design was not clearly established, and studies wereunable to distinguish prevalent from incident cases. Incidentcases are more expensive than prevalent cases in terms of hospi-talization rates, length of stay, case mix, and service intensity, andhave higher discontinuation rates [39–41]. Studies included dif-ferent population groups, which has an impact on costs: somefocused on codes for type 2 diabetes only, type 1 and type 2,gestational diabetes, and/or diabetes-related complications. Forexample, gestational diabetes is more expensive than type 2diabetes because of the frequency and duration of hospitaliza-tions [42]. None of the studies described if primary, secondary, orboth codes were used. Previous studies have shown an increase incosts by up to twofold when both primary and secondary ICD-9codes were used [31].

Contrary to accepted recommendations, none of the studiesvalidated ICD-9/ICD-9-CM codes [43]. Wilchesky et al. showed64% sensitivity of claims data to detect patients with diabetes[44], which means that an appreciable number of cases may bemissed. Similarly, none of the studies validated prescriptionclaims data that are vulnerable to errors from sampling, misiden-tification of newly treated patients, and misclassification ofadded versus switched medications [45,46]. ICD-9/ICD-9-CMcodes to measure utilization and costs also requires validation,because some studies have found that 9% of discharges incor-rectly omit codes for diabetes, and 13% of discharges are regis-tered without any foot-related diagnosis code [47].

Most studies used medication possession ratios, but therewere important variation in the definition. For some, MPR wasthe sum of days of antidiabetic prescription supply dispenseddivided by the number of days between prescription refills, fromthe first date of dispensing within each year until the dispensing

Costs of Nonadherence in Diabetes Mellitus 917

Tabl

e2

Con

tinua

tion

ofst

udie

sid

entifi

edw

ithco

sts

repo

rted

byad

here

nce

leve

lin

diab

etic

patie

nts

Ref

eren

ceN

Mea

sure

sm

etho

dSt

atis

tical

anal

ysis

Res

ults

Qua

lity

scor

eA

dher

ence

Res

ourc

eut

iliza

tion

Cos

ts/a

djus

tmen

tfo

rin

flatio

nSt

atis

tical

met

hod

Con

tinuo

usen

rollm

ent

inth

ehe

alth

insu

ranc

eA

djus

tmen

tby

conf

ound

ers

Adj

ustm

ent

byho

spita

lizat

ion

orot

her

loca

tion

Balk

rish

nan

R,

2003

[28]

775

MPR

defin

edas

the

days

ofan

tidia

betic

pres

crip

tion

supp

lydi

spen

sed

divi

ded

byth

enu

mbe

rof

days

betw

een

pres

crip

tion

refil

ls.T

heob

serv

atio

npe

riod

bega

nw

ithth

efir

stda

teof

disp

ensi

ngw

ithin

each

year

and

ende

das

the

disp

ensi

ngda

teof

the

last

pres

crip

tion

Adm

inis

trat

ive

clai

ms

data

ofth

eH

MO

Tota

lcos

tsno

tsp

ecifi

edSe

quen

tialm

ixed

-mod

elan

dre

gres

sion

anal

ysis

Yes

Cha

rlso

nin

dex

was

used

toad

just

byse

veri

ty

Num

ber

ofda

ysdu

ring

hosp

italiz

atio

nw

assu

btra

cted

from

the

deno

min

ator

MPR

for

1to

5ye

ars

offo

llow

upw

ere

0.70

,0.7

1,0.

75,0

.77,

and

0.78

;and

mea

nhe

alth

care

cost

sw

ere

$8,3

06,$

5,94

7,$5

,821

,$5

,043

,$5,

118.

10%

incr

ease

inan

tidia

betic

MPR

was

asso

ciat

edw

ithan

8.6%

decr

ease

into

tala

nnua

lhea

lthca

reco

sts

(P<

0.00

1).

Aft

er5

year

s,hi

ghad

here

nce

=$4

,000

whi

lelo

wad

here

nce

=$1

0,50

0

14/3

0

Cob

den

D,

2007

[29]

486

MPR

:sum

ofth

eda

ys’s

uppl

yof

drug

divi

ded

byth

enu

mbe

rof

days

betw

een

the

first

filla

ndth

ela

stre

fill

plus

the

days

’sup

ply

ofth

ela

stre

fill

Phys

icia

nvi

sits

,ho

spita

lizat

ion,

emer

genc

yde

part

men

tvi

sits

,pha

rmac

yda

ta

Tota

lhea

lthca

reco

sts,

annu

alad

just

edm

ean

all-c

ause

heal

thca

reco

sts/

adju

stm

ent

for

infla

tion

to20

05do

llars

Pers

on-t

ime

and

even

t-tim

ean

alys

isad

just

edby

leng

thof

follo

w-u

p.Po

isso

nre

gres

sion

mod

elan

dga

mm

are

gres

sion

mod

el

Yes

Hyp

ogly

cem

ia/

adju

stm

ent

for

Com

orbi

ditie

s

Not

repo

rted

MPR

of80

%or

grea

ter

was

asso

ciat

edw

ithsi

gnifi

cant

redu

ctio

nin

all-c

ause

heal

thca

reco

sts

(OR

0.55

,95%

CI

0.31

–0.8

0,P

<0.

05).

MPR

of68

%w

asas

soci

ated

with

tota

lmea

nco

sts

of$8

,056

�8,

559,

whi

lean

MPR

of59

%ha

dto

talm

ean

cost

sof

$8,6

99�

9,26

8.

14/3

0

Balk

rish

nan

R,

2004

[30]

3,48

3M

PRC

laim

sda

taTo

tala

nnua

lhea

lthca

reco

sts

Mul

tivar

iate

tech

niqu

esN

RN

RN

R13

%in

crea

sein

MPR

was

asso

ciat

edw

ith16

.1%

low

erto

tal

annu

alhe

alth

care

cost

s(P

<0.

001)

.

12/3

0

Hep

keK

L,20

04[3

1]57

,687

Med

icat

ion

adhe

renc

era

teca

lcul

ated

aspe

rcen

tage

ofda

ysth

atth

epa

tient

poss

esse

dan

yav

aila

ble

diab

etic

drug

duri

ngth

eye

ar

Inpa

tient

hosp

italiz

atio

n,ou

tpat

ient

care

,em

erge

ncy

care

,clin

icvi

sits

,lab

orat

ory

test

s,pr

ofes

sion

alse

rvic

esan

dph

arm

aceu

tical

s

Ove

rall

cost

ofhe

alth

care

and

cost

rela

ted

with

diab

etes

care

Leas

tsq

uare

sre

gres

sion

mod

elan

dm

ultiv

aria

telo

gist

icre

gres

sion

Yes

Illne

ssse

veri

tyus

ing

diag

nosi

sco

stgr

oup.

NR

20%

to39

%ad

here

nce

leve

lwas

need

edbe

fore

med

ical

care

cost

sw

ere

redu

ced.

For

diab

etes

rela

ted

cost

s,th

eth

resh

old

was

seen

until

40%

to59

%ad

here

nce

leve

l.A

dher

ence

-tot

alav

erag

eex

pend

iture

s0%

=$6

,500

,1–

19%

=$7

,250

,20

–39%

=$7

,750

,40

%–5

9%=

$7,5

00,

60%

–79%

=$7

,700

,80

%–9

9%=

$7,3

00,

100%

=$7

,900

.

11/3

0

Lee

WC

,200

6[1

7]1,

156

MPR

:sum

ofth

eda

ys’s

uppl

yof

med

icat

ion

divi

ded

byth

enu

mbe

rof

days

betw

een

the

first

filla

ndth

ela

stre

fill

plus

the

days

’sup

ply

ofth

ela

stre

fill.

Phys

icia

nvi

sits

,ho

spita

lizat

ions

,ER

visi

ts

Tota

lhea

lthca

reco

sts/

Cos

tsad

just

edto

2005

dolla

rsus

ing

the

cons

umer

pric

ein

dex

Pers

on-t

ime

and

time-

even

tan

alys

is.

Pois

son

regr

essi

onm

odel

san

din

cide

ntra

tera

tios.

Enro

llmen

tfo

rat

leas

t6

mon

ths

befo

reth

ein

dex

date

and

atle

ast

2ye

ars

ofco

ntin

uous

enro

llmen

taf

ter

the

inde

xda

te

Hyp

ogly

cem

icev

ents

NR

62%

MPR

toin

sulin

pen

ther

apy

=m

ean

annu

alal

l-cau

sehe

alth

care

cost

s$1

4,76

9

11/3

0

918 Salas et al.

Shen

olik

arR

A,2

006

[32]

1,07

3M

PR:N

umbe

rof

days

ofan

tidia

betic

pres

crip

tion

supp

lydi

spen

sed

(e.g

.,a

30-d

aysu

pply

)di

vide

dby

the

num

ber

ofda

ysbe

twee

nth

efir

stan

dla

stdi

spen

satio

n.M

ed-T

otal

appr

oach

:rat

ioof

tota

lnum

ber

ofda

ysth

edr

ugw

assu

pplie

dto

the

diffe

renc

ein

the

num

ber

ofda

ysbe

twee

nth

efir

stan

dla

stpr

escr

iptio

nda

tes.

Med

ical

and

dent

alca

re,

inpa

tient

and

outp

atie

ntca

re,r

egul

arch

ecku

ps,o

ffice

visi

ts,

hom

ehe

alth

care

,lo

ng-t

erm

faci

lity

care

and

pres

crip

tion

drug

s.

Ann

ualt

otal

and

diab

etes

-rel

ated

heal

thca

reco

sts/

No

disc

ount

edra

tere

port

ed

Mul

tivar

iate

regr

essi

onan

alys

isad

just

edby

cova

riat

es.C

osts

wer

etr

ansf

orm

edus

ing

loga

rith

man

dth

eyw

ere

tran

sfor

med

back

usin

gan

tilog

arith

ms

ofth

epa

ram

eter

estim

ate

Yes

Adj

uste

dby

com

orbi

ditie

sN

oM

ean

rate

ofad

here

nce

tone

wm

edic

atio

nof

59%

=$9

,546

�$1

4,86

1m

ean

tota

lhea

lthca

reco

sts

for

year

2an

dm

ean

diab

etes

-rel

ated

cost

sfo

rye

ar2

of$4

,576

�$8

,208

;The

estim

ated

coef

ficie

nts

and

stan

dard

erro

rsfo

rto

tala

nnua

lhe

alth

care

cots

asa

func

tion

ofco

vari

ates

wer

e:m

ale

sex

1,11

7.35

�1,

001.

69,h

igh

tota

lnu

mbe

rof

pres

crip

tions

8,22

3.48

�1,

002.

38;A

fric

anA

mer

ican

race

1,12

5.49

�91

4.39

;ra

teof

adhe

renc

e—2,

721.

68(9

32.5

0),c

onst

ant

728.

82(1

,180

.29)

and

adju

sted

r2=

0.06

.

13/3

0

Soko

lMC

,20

05[3

3]13

7,27

7Pe

rcen

tage

ofda

ysdu

ring

the

anal

ysis

peri

odth

atpa

tient

sha

da

supp

lyof

1or

mor

em

aint

enan

cem

edic

atio

nsfo

rth

eco

nditi

on

Med

ical

and

drug

clai

ms:

hosp

italiz

atio

n,ER

serv

ice,

outp

atie

ntse

rvic

esin

clud

ing

phys

icia

nof

fice

visi

tsan

dou

tpat

ient

visi

ts.

Nur

sing

hom

ean

dho

me

care

serv

ices

wer

eno

tin

clud

ed

Tota

lhea

lthca

reco

sts

(Sum

ofm

edic

al—

outp

atie

ntse

rvic

es,E

Rse

rvic

es,h

ospi

taliz

atio

n-an

ddr

ugco

sts)

,and

dise

ase-

rela

ted

cost

s.N

etco

stto

the

plan

spon

sor,

patie

ntco

paym

ents

and

dedu

ctib

les

wer

eno

tin

clud

ed.

Cos

tsw

ere

adju

sted

for

age,

sex,

com

orbi

dity

,di

seas

esu

btyp

e,em

ploy

men

tgr

oup

and

med

ical

plan

type

.

Logi

stic

regr

essi

onm

odel

.No

deta

ilfo

rco

sts

tran

sfor

mat

ion

was

prov

ided

.

Yes

Com

orbi

ditie

sw

ere

incl

uded

inth

ean

alys

is.

NR

Adh

eren

cele

vela

ndto

talc

osts

:1–

19%

=$8

,867

;20–

39%

=$7

,124

;40–

59%

=$6

,522

;60

–79%

=$6

,291

;80–

100%

=$4

,570

.Diff

eren

ces

wer

est

atis

tical

lysi

gnifi

cant

for

mos

tad

here

nce

leve

lsw

hen

com

pare

dw

ithth

ehi

ghes

tle

velo

fad

here

nce

(P<

0.05

).

8/30

Wag

ner

EH,

2001

[34]

4,74

4N

otin

clud

eda

mea

sure

ofad

here

nce

and

HbA

1cw

asus

edas

apr

oxy

ofm

edic

atio

nad

here

nce

Ann

ualu

tiliz

atio

nra

tes

Tota

lhea

lthca

reco

sts

and

mea

nco

sts

per

pers

on

Cos

tda

talo

gari

thm

ical

lytr

ansf

orm

ed.R

egre

ssio

nan

alys

is.

Yes

No

NR

Base

line

HbA

1c,L

evel

,%an

dm

ean

annu

alco

sts$

,(p

valu

esw

ere

calc

ulat

edfo

rth

edi

ffere

nce

inlo

gco

sts)

<8=

$4,4

75(P

=0.

18);

8–10

=$5

,898

(P=

0.32

);>1

0=

$8,0

88(P

=0.

53)

11/3

0

Whi

teT

J,20

04[3

5]N

RPe

rcen

tage

ofad

here

nce

Hos

pita

lizat

ion

and

ERA

djus

ted

tota

lhe

alth

care

cost

sR

egre

ssio

nm

odel

Yes

NR

NR

Patie

nts

with

�75

,>75

to�

95an

d95

%ad

here

nce,

adju

sted

tota

lhe

alth

care

cost

sw

ere

$US

5,70

6,$5

,314

and

$4,8

35(P

<0.

001)

.

9/30

Shet

tyS,

2005

[36]

3,12

1N

otin

clud

edbu

tH

bA1c

was

used

asa

prox

yof

med

icat

ion

adhe

renc

e.

Cla

ims

data

Cos

tsof

6m

onth

spe

riod

Mul

tiple

linea

rre

gres

sion

anal

ysis

.Lo

gari

thm

ictr

ansf

orm

atio

nof

cost

data

was

done

prio

ran

alys

is.

Yes

Adj

ustm

ent

byag

e,ge

nder

,spe

cial

tyof

the

phys

icia

n,co

mor

bidi

tyan

dto

talb

asel

ine

cost

s.

NR

Pred

icte

dto

tald

iabe

tes-

rela

ted

cost

for

targ

etH

bA1c

leve

lgro

updu

ring

the

first

year

offo

llow

upw

as$1

,540

per

patie

nt,3

2%hi

gher

than

the

tota

ldia

bete

sre

late

dco

st($

1,17

1)fo

rth

esa

me

targ

etgr

oup

(P<

0.00

1).

11/3

0

ER,e

mer

genc

yro

om;H

MO

,hea

lthm

aint

enan

ceor

gani

zatio

n;M

PR,m

edic

atio

npo

sses

sion

ratio

;NR

,not

repo

rted

.HbA

1c,g

lyco

syla

ted

hem

oglo

bin.

Costs of Nonadherence in Diabetes Mellitus 919

date of the last prescription [28,32]. Others added the days’supply of the last refill to the denominator [17,29], or they usedthe percentage of days that the patient possessed any availablediabetic drug during the year [31]. None of the studies consid-ered the effects of censoring, which is important, because sixfilled prescriptions evaluated over 12 months equals an MPR of50%, but if they are evaluated over 6 months, the six filledprescriptions equals an MPR of 100%.

The non-MPR measures included were: Med-total approachdefined as the ratio of total number of days the drug was suppliedto the difference in the number of days between the first and lastprescription dates [32]; the percentage of days during the analysisperiod that patients had a supply of one or more maintenancemedications for the condition [33], and the percentage of adher-ence [35]. The problem is that these measures are not compa-rable. Hess [48] analyzed various adherence measures and foundthat only 4—Continuous Measure of Medication Acquisition;Continuous Multiple Interval Measure of Oversupply; MPR; andMedication Refill Adherence—out of 11 measures were identicalfor measuring adherence to prescription refills throughout thestudy period.

With regard to confounders, 6 out of 10 studies made someeffort to adjust their estimates by disease severity, but most didnot adjust by comorbidities, thereby potentially underestimatingthe real costs. None of the databases used by analysts containinformation of behavioral variables such as smoking and alcoholthat are closely related to adherence [49–53]. There was also lackof information on adverse drug events, such as hypoglycemia,which has been shown to be a costly component of diabetes-related treatment [54]. None of the studies were able to measurethe direct consequences of either nonadherence (e.g., hyperosmo-lar coma) or associated utilization-based outcomes. Costs were,therefore, not disaggregated according to the main driversthat are a consequence of loss of therapeutic effect throughnonadherence.

All studies used charges as proxy for costs. However, chargeshave been criticized because they do not reflect real costs [55],and they do not take into account the various levels of copay-ment, deductibles, and coinsurance for prescriptions and othermedical services, including physician office care, medical emer-gency care, and inpatient hospitalization.

Only one study tried to deal with skewed distribution ofhealth-care costs and MPR [22]. This is important, because inap-

propriate analysis of costs will produce biased estimates for themean. For costs, nonparametric bootstrap techniques or GLMregression analyses are recommended [56,57].

Conclusion

The research assessing the association between medicationadherence/nonadherence and health-care costs is limited and ofpoor quality. There are important methodological differencesamong studies of costs of adherence/nonadherence in patientswith diabetes, making robust comparisons difficult; and thosedifferences might explain the inconsistency in the reportedassociations between medication adherence and costs. Readersshould be cautious when interpreting or comparing the results ofsuch studies. More research is needed to validate measures ofmedication adherence using claims data and to determine theimpact of nonadherence on health-care costs.

Acknowledgments

This article is written by members of the International Society forPharmacoeconomics & Outcomes Research (ISPOR) Economicsof Medication Compliance Working Group; part of the Medica-tion Compliance and Persistence Special Interest Group.

Source of financial support: None.

Supporting information for this article can be found at: http://www.ispor.org/publications/value/ViHsupplementary.asp

References

1 Hughes DA, Bagust A, Haycox A, Walley T. The impact ofnon-compliance on the cost-effectiveness of pharmaceuticals: areview of the literature. Health Econ 2001;10:601–15.

2 Lewis A. Noncompliance: a $100 billion problem. RemingtonReport 1997;5:14–5.

3 World Health Organization. Adherence to long-term therapies:evidence for action. 2003. Available from: http://www.who.int/chp/knowledge/publications/adherence_full_report.pdf [AccessedDecember 12, 2007].

4 Osterberg L, Blaschke T. Adherence to medication. N Engl J Med2005;353:487–97.

Titles related to adherence = 54,352

Titles related to diabetes = 129,187

Titles related to costs = 85,252

Identified abstracts= 209

Studies that fulfill inclusion criteria = 10

Reviewed abstracts and full articles that have information of adherence, costs, and diabetes =

50

Excluded: No costs = 87 Non-English language = 2 Not directly related to diabetes = 49 No adherence = 21

Excluded: No costs of adherence =9 Not directly related to diabetes = 4 No adherence = 27

Figure 1 Flow diagram showing the number ofreferences identified, retrieved and included in thereview.

920 Salas et al.

5 Saaddine JB, Engelgau MM, Beckles GL, et al. A diabetes reportcard for the United States: quality of care in the 1990s. Ann InternMed 2002;136:565–74.

6 Kerr EA, Gerzoff RB, Krein SL, et al. Diabetes care quality in theVeterans Affairs Health Care System and commercial managedcare: the TRIAD study. Ann Intern Med 2004;141:272–81.

7 American Diabetes Association. Standards of medical care indiabetes—2007. Diabetes Care 2007;30:S4–41.

8 Abegunde DO, Mathers CD, Adam T, et al. The burden and costsof chronic diseases in low-income and middle-income countries.Lancet 2007;370:1929–38.

9 Clarke P, Gray A, Legood R, et al. The impact of diabetes-relatedcomplications on healthcare costs: results from the UnitedKingdom Prospective Diabetes Study (UKPDS Study No. 65).Diabet Med 2003;20:442–50.

10 Pohar SL, Majumdar SR, Johnson JA. Health care costs andmortality for Canadian urban and rural patients with diabetes:population-based trends from 1993–2001. Clin Ther 2007;29:1316–24.

11 Stratton IM, Adler AI, Neil HA, et al. Association of glycaemiawith macrovascular and microvascular complications of type 2diabetes (UKPDS 35): prospective observational study. BMJ2000;321:405–12.

12 UK Prospective Diabetes Study (UKPDS) Group. Effect of intensiveblood-glucose control with metformin on complications in over-weight patients with type 2 diabetes (UKPDS 34). UK ProspectiveDiabetes Study (UKPDS) Group. Lancet 1998;352:854–65.

13 Salas M, Ward A, Caro J. Health and economic effects of addingnateglinide to metformin to achieve dual control of glycosylatedhemoglobin and postprandial glucose levels in a model of type 2diabetes mellitus. Clin Ther 2002;24:1690–705.

14 Balu S. Incremental treatment expenditure of diabetes in theUnited States. Manag Care Interface 2007;20:20–7.

15 Valentine WJ, Erny-Albrecht KM, Ray JA, et al. Therapy conver-sion to insulin among patients with type 2 diabetes treated withoral agents: a modeling study of cost-effectiveness in the UnitedStates. Adv Ther 2007;24:273–90.

16 Pawaskar MD, Camacho FT, Anderson RT, et al. Health carecosts and medication adherence associated with initiation ofinsulin pen therapy in Medicaid-enrolled patients with type 2diabetes: a retrospective database analysis. Clin Ther 2007;29:1294–305.

17 Lee WC, Balu S, Cobden D, et al. Medication adherence and theassociated health-economic impact among patients with type 2diabetes mellitus converting to insulin pen therapy: an analysis ofthird-party managed care claims data. Clin Ther 2006;28:1712–25.

18 Kalsekar I, Iyer S, Mody R, et al. Utilization and costs for com-pliant patients initiating therapy with pioglitazone or rosiglita-zone versus insulin in a Medicaid fee-for-service population.J Manag Care Pharm 2006;12:121–9.

19 Lee WC, Balu S, Cobden D, et al. Prevalence and economic con-sequences of medication adherence in diabetes: a systematic lit-erature review. Manag Care Interface 2006;19:31–41.

20 Cramer JA. A systematic review of adherence with medicationsfor diabetes. Diabetes Care 2004;27:1218–24.

21 Ho PM, Rumsfeld JS, Masoudi FA, et al. Effect of medicationnonadherence on hospitalization and mortality among patientswith diabetes mellitus. Arch Intern Med 2006;166:1836–41.

22 Gerstein HC. Glycosylated hemoglobin: finally ready for primetime as a cardiovascular risk factor. Ann Int Med 2004;141:475–6.

23 Cramer JA, Roy A, Burrell A, et al. Medication compliance andpersistence: terminology and definitions. Value Health 2008;11:44–7.

24 American Diabetes Association. Standards of medical care indiabetes—2009. Diabetes Care 2009;32(Suppl. 1):S13–61.Available from: http://care.diabetesjournals.org/cgi/reprint/32/Supplement_1/S13 [Accessed January 5, 2009].

25 Selvin E, Marinopoulos S, Berkenblit G, et al. Meta-analysis:glycosylated hemoglobin and cardiovascular disease in diabetesmellitus. Ann Intern Med 2004;141:421–31.

26 Rozenfeld Y, Hunt JS, Plauschinat C, Wong KS. Oral antidiabeticmedication adherence and glycemic control in managed care. AmJ Manag Care 2008;14:71–5.

27 Drummond MF, Jefferson TO. Guidelines for authors and peerreviewers of economic submissions to the BMJ. BMJ 1996;313:275–83.

28 Balkrishnan R, Rajagopalan R, Camacho FT, et al. Predictors ofmedication adherence and associated health care costs in an olderpopulation with type 2 diabetes mellitus: a longitudinal cohortstudy. Clin Ther 2003;25:2958–71.

29 Cobden D, Lee WC, Balu S, et al. Health outcomes and economicimpact of therapy conversion to a biphasic insulin analog penamong privately insured patients with type 2 diabetes mellitus.Pharmacotherapy 2007;27:948–62.

30 Balkrishnan R, Rajagopalan R, Shenolikar RA, et al. Healthcarecosts and prescription adherence with introduction of thiazo-lidinedione therapy in Medicaid type 2 diabetic patients: a retro-spective data analysis. Curr Med Res Opin 2004;20:1633–40.

31 Hepke KL, Martus MT, Share DA. Costs and utilization associ-ated with pharmaceutical adherence in a diabetic population. AmJ Manag Care 2004;10:144–51.

32 Shenolikar RA, Balkrishnan R, Camacho FT, et al. Comparisonof medication adherence and associated health care costs afterintroduction of pioglitazone treatment in African Americansversus all other races in patients with type 2 diabetes mellitus: aretrospective data analysis. Clin Ther 2006;28:1199–207.

33 Sokol MC, McGuigan KA, Verbrugge RR, Epstein RS. Impact ofmedication adherence on hospitalization risk and healthcare cost.Med Care 2005;43:521–30.

34 Wagner EH, Sandhu N, Newton KM, et al. Effect of improvedglycemic control on health care costs and utilization. JAMA2001;285:182–9.

35 White TJ, Vanderplas A, Chang E, et al. The costs of non-adherence to oral antihyperglycemic medication in individualswith diabetes mellitus and concomitant diabetes mellitus andcardiovascular disease in a managed care environment. DisManage Health Outcomes 2004;12:181–8.

36 Shetty S, Secnik K, Oglesby AK. Relationship of glycemic controlto total diabetes-related costs for managed care health planmembers with type 2 diabetes. J Manag Care Pharm 2005;11:559–64.

37 Peterson AM, Nau DP, Cramer JA, et al. A checklist for medica-tion compliance and persistence studies using retrospective data-bases. Value Health 2007;10:3–12.

38 Hughes D, Cowell W, Koncz T, Cramer J, International Societyfor Pharmacoeconomics & Outcomes Research Economics ofMedication Compliance Working Group. Methods for integrat-ing medication compliance and persistence in pharmacoeconomicevaluations. Value Health 2007;10:498–509.

39 Johnson JA, Pohar SL, Majumdar SR. Health care use and costsin the decade after identification of type 1 and type 2 diabetes: apopulation-based study. Diabetes Care 2006;29:2403–8.

40 Pohar SL, Majumdar SR, Jacobs P, Johnson JA. Healthcare Uti-lization and Direct Healthcare Costs of Diabetes in Urban andRural Saskatchewan, 1991–2001. Edmonton: Alliance for Cana-dian Health Outcomes Research in Diabetes, 2006 (Institute ofHealth Economics Working Paper WP 06-1).

41 Liu L, Hader J, Brossart B, et al. Impact of rural hospital closuresin Saskatchewan, Canada. Soc Sci Med 2001;52:1793–804.

42 Phillips PJ. Gestational diabetes. Aust Fam Physician 2006;35:701–3.

43 Newton KM, Wagner EH, Ramsey SD, et al. The use of auto-mated data to identify complications and comorbidities of diabe-tes: a validation study. J Clin Epidemiol 1999;52:199–207.

44 Wilchesky M, Tamblyn RM, Huang A. Validation of diagnosticcodes within medical services claims. J Clin Epidemiol 2004;57:131–41.

45 Grymonpre R, Cheang M, Fraser M, et al. Validity of a prescrip-tion claims database to estimate medication adherence in olderpersons. Med Care 2006;44:471–7.

46 Friedman DS, Quigley HA, Gelb L, et al. Using pharmacy claimsdata to study adherence to glaucoma medications: methodology

Costs of Nonadherence in Diabetes Mellitus 921

and findings of the Glaucoma Adherence and Persistency Study(GAPS). Invest Ophthalmol Vis Sci 2007;48:5052–7.

47 Ragnarson Tennvall G, Apelqvist J, Eneroth M. The inpatientcare of patients with diabetes mellitus and foot ulcers. A valida-tion study of the correspondence between medical records and theSwedish Inpatient Registry with the consequences for cost esti-mations. J Intern Med 2000;248:397–405.

48 Hess LM, Raebel MA, Conner DA, Malone DC. Measurement ofadherence in pharmacy administrative databases: a proposal forstandard definitions and preferred measures. Ann Pharmacother2006;40:1280–88.

49 Zeller A, Schroeder K, Peters T. Cigarette smoking and adherenceto antihypertensive medication in patients from primary care. EurJ Gen Pract 2007;13:161–2.

50 Vik SA, Hogan DB, Patten SB, et al. Medication nonadherenceand subsequent risk of hospitalisation and mortality among olderadults. Drugs Aging 2006;23:345–56.

51 Cook RL, Sereika SM, Hunt SC, et al. Problem drinking andmedication adherence among persons with HIV infection. J GenIntern Med 2001;16:83–6.

52 Palepu A, Horton NJ, Tibbetts N, et al. Uptake and adher-ence to highly active antiretroviral therapy among HIV-infected people with alcohol and other substance use problems:the impact of substance abuse treatment. Addiction 2004;99:361–8.

53 Tucker JS, Burnam MA, Sherbourne CD, et al. Substance use andmental health correlates of nonadherence to antiretroviral medi-cation in a sample of patients with Human ImmunodeficiencyVirus Infection. Am J Med 2003;114:573–80.

54 Jonsson L, Bolinder B, Lundkvist J. Cost of hypoglycemia inpatients with type 2 diabetes in Sweden. Value Health 2006;9:193–8.

55 Finkler SA. The distinction between cost and charges. Ann InternMed 1982;96:102–9.

56 Rutten-van Mölken MP, van Doorslaer EK, van Vliet RC. Statis-tical analysis of cost outcomes in a randomized controlled clinicaltrial. Health Econ 1994;3:333–45.

57 Dodd S, Bassi A, Bodger K, Williamson P. A comparison ofmultivariable regression models to analyze cost data. J Eval ClinPract 2006;12:76–86.

922 Salas et al.