Effects of Changes in Number of Medications and Drug Burden

Index Exposure on Transitions Between Frailty States and

Death: The Concord Health and Ageing in Men Project Cohort
Study
Kris M. Jamsen, PhD,*† J. Simon Bell, PhD,*†‡ Sarah N. Hilmer, PhD,†§ Carl M. J. Kirkpatrick,
PhD,* Jenni Ilom€aki, PhD,* David Le Couteur, PhD,k#** Fiona M. Blyth, PhD,k#
David J. Handelsman, PhD,k** Louise Waite, PhD,** Vasi Naganathan, PhD,k# Robert G.
Cumming, PhD,#†† and Danijela Gnjidic, PhD#‡‡

ated with a 73% greater risk of transitioning from the

OBJECTIVES: To investigate the effects of number of robust state to the prefrail state (adjusted 95% CI = 1.30–
medications and Drug Burden Index (DBI) on transitions 2.31) and a 2.75 times greater risk of transitioning from
between frailty stages and death in community-dwelling the robust state to death (adjusted 95% CI = 1.60–4.75).
older men. There was no evidence of an adjusted association between
DESIGN: Cohort study. total number of medications or DBI and the other transi-
SETTING: Sydney, Australia. tions.
PARTICIPANTS: Community-dwelling men aged 70 and CONCLUSION: Although the possibility of confounding
older (N = 1,705). by indication cannot be excluded, additional medications
were associated with greater risk of mortality in robust
MEASUREMENTS: Self-reported questionnaires and
community-dwelling older men. Greater DBI was also
clinic visits were conducted at baseline and 2 and 5 years.
associated with greater risk of death and transitioning
Frailty was assessed at all three waves according to the
from the robust state to the prefrail state. J Am Geriatr
modified Fried frailty phenotype. The total number of reg-
Soc 64:89–95, 2016.
ular prescription medications and DBI (a measure of expo-
sure to sedative and anticholinergic medications) were
calculated over the three waves. Data on mortality over Key words: medication; frail elderly; mortality; epi-
9 years were obtained. Multistate modeling was used to demiological methods
characterize the transitions across three frailty states (ro-
bust, prefrail, frail) and death.
RESULTS: Each additional medication was associated
with a 22% greater risk of transitioning from the robust
state to death (adjusted 95% confidence interval
(CI) = 1.06–1.41). Every unit increase in DBI was associ-

From the *Centre for Medicine Use and Safety, Faculty of Pharmacy and
P olypharmacy, or multiple medication use, is highly
prevalent in older people, with nearly half taking one
or more unnecessary medications.1,2 Polypharmacy often
Pharmaceutical Sciences, Monash University, Parkville, Victoria;
†
Cognitive Decline Partnership Centre, Hornsby Ku-ring-gai Hospital, occurs because of continuation of long-term medications
Hornsby, New South Wales; ‡Sansom Institute, School of Pharmacy and for which the benefits no longer outweigh the risks.3
Medical Sciences, University of South Australia, Adelaide,South Australia;
§
Kolling Institute of Medical Research; kSydney Medical School,
Polypharmacy has been associated with excess morbidity
University of Sydney, Sydney; #Centre for Education and Research on and mortality in older people.1 Minimizing unnecessary
Ageing; **ANZAC Institute, Concord Hospital, Concord; ††Sydney School medication use may maintain quality of life and reduce
of Public Health; and ‡‡Faculty of Pharmacy, University of Sydney, hospitalization.4 In particular, the use of anticholinergic
Sydney, New South Wales, Australia.
and sedative medications has been associated with func-
Address correspondence to Kris M. Jamsen, Centre for Medicine Use and tional impairment in older people.5 High anticholinergic
Safety, Faculty of Pharmacy and Pharmaceutical Sciences, Monash
University (Parkville campus), 381 Royal Parade, Parkville, VIC 3052,
and sedative load has been associated with functional
Australia. E-mail: kris.jamsen@monash.edu impairment, hospitalization, and mortality in older
DOI: 10.1111/jgs.13877
people.6,7

JAGS 64:89–95, 2016

© 2016, Copyright the Authors
Journal compilation © 2016, The American Geriatrics Society 0002-8614/16/$15.00
90 JAMSEN ET AL.
Frailty is a syndrome characterized by low physiologi-
cal reserve and limited ability to respond to stressors
resulting from cumulative decline across multiple systems.8
Older age, poor cognition, lower education, specific medi-
cal conditions, and hospitalization have been associated
with worse frailty status.9,10 It was recently demonstrated
that polypharmacy and exposure to anticholinergic and
sedative medications at baseline is associated with onset of
frailty over 2 years in older men,11 but it is unknown
whether medication load or anticholinergic and sedative
exposure, and changes in levels of exposure, contributes to
the transitions between frailty stages and subsequent pro-
gression to death. This is of major clinical importance
because, although frailty is thought to be reversible, with
individuals able to move between robust, prefrail, and frail
states, nonpharmacological and pharmacological therapies
to treat or reverse frailty remain elusive.8,12–16
Frailty is a dynamic process,15,17 whereby develop-
ment of frailty can lead to a “spiral of decline” of increas-
ing frailty and greater risk of worsening disability, hospital
admission, falls, and death.15 This spiral of decline can
lead to taking more medications, which may further con-
tribute to frailty. Unlike other factors that are associated
with frailty, medication use can be reversed, which may
assist in halting or even reversing this downward spiral. It
is unknown whether medication use is associated with
frailty transitions (frailty progression or regression) and,
simultaneously, mortality in older people.18 In the absence
of specific pharmacological strategies to prevent or treat
frailty, research into the effects of multiple medication use
and anticholinergic and sedative load on frailty transitions
is needed. The objective of this study was to investigate
the effects of total number of medications and anticholin-
ergic and sedative load captured over three time-points on
transitions between frailty stages over time and death in
community-dwelling older men.
METHODS
Study Design, Setting, and Participants
Participants were community-dwelling men aged 70 and
older living in a suburban area of Sydney, Australia,
and enrolled in the ongoing cohort study Concord Health
and Ageing in Men Project (CHAMP).19 Baseline recruit-
ment occurred from January 2005 to June 2007. The New
South Wales electoral roll was used as the sampling frame.
Contact was made with 3,005 of the 3,627 men who were
sent invitation letters; 2,815 of these met the eligibility cri-
teria, and 1,511 (53.7%) consented to participate in the
study. An additional 194 eligible men volunteered before
receiving the invitation letter, resulting in 1,705 partici-
pants at baseline. Participants were invited to participate
in Year 2 assessments, conducted from February 2007 to
October 2009. The Year 2 follow-up rate was 80.2%
(n = 1,367), and 958 (56.2% of baseline) participants
completed Year 5 assessments from August 2010 to March
2013. Each assessment consisted of a self-reported study
questionnaire and a clinical examination that consisted of
physical performance measures, neuropsychological testing,
and medication inventory. Participants were contacted
every 4 months to collect information regarding falls, frac-
JANUARY 2016–VOL. 64, NO. 1 JAGS
tures, and hospitalizations and to identify those who had
been admitted to aged care facilities or died.19 The Sydney
South West Area Health Service human research ethics
committee and the Concord Hospital human research
ethics committee approved the study.
Medication Exposure
In the present study, the primary exposure variables were
total number of medications and Drug Burden Index (DBI;
a pharmacological risk assessment tool that assesses cumu-
lative exposure to sedative and anticholinergic medications
according to the principles of dose-response and maximal
effect20). These medication exposures were determined at
all three assessments. Participants were asked to bring all
of their medications to the clinic. Each participant was
asked to report prescription and nonprescription medica-
tions they had used during the past month. Trained per-
sonnel verified each participant’s self-reported list of
medications by taking a structured medication history. The
name, strength, frequency, duration of use, and prescrip-
tion pattern (regular or as required) was recorded for each
medication. Medication data at all three waves were coded
using Iowa Drug Information Service (IDIS) drug code
numbers. The total number of medications at each assess-
ment represented each participant’s actual number of regu-
lar prescribed medications taken.
For each participant, the DBI was calculated at each
assessment as follows:
X
nd
Di =ðdi þ Di Þ ð1Þ
i¼1
Di represents the daily dose of medication i (i = 1, . . ., nd)
with anticholinergic or sedative properties, and di indicates
the recommended minimum dose of medication i registered
by the Therapeutic Goods Association of Australia. More-
over, di is used as an estimate of the dose required to
achieve half of the maximum anticholinergic or sedative
effect. Medications with clinically significant anticholiner-
gic or sedative effects were identified using the Australian
registered product information.21 Complementary medica-
tions and medications used only as required were excluded
from the DBI calculation.
Outcome Measures
The outcomes in the present study were frailty and death.
Frailty was determined at the baseline and 2- and 5-year
assessments according to the Fried frailty phenotype used
in the Cardiovascular Health Study (CHS).22 The five cri-
teria were weight loss, weakness, slowness, exhaustion,
and low activity. Adapted criteria were used for weight
loss, exhaustion, and low activity because the exact mea-
surements used in the CHS were not available. For
detailed descriptions of these criteria, see previous publica-
tions.22,23 At each assessment, participants were classed as
robust if they met none of the criteria, prefrail if they met
one or two of the criteria, and frail if they met three or
more of the criteria. Participants who had two or more
JAGS JANUARY 2016–VOL. 64, NO. 1
missing criteria were assigned a missing value for frailty
status and were excluded from the analysis. The length of
time in years from the baseline assessment to each subse-
quent follow-up assessment was computed.
Data on mortality were obtained mainly from tele-
phone calls every 4 months with subjects or informants.
Persons not contactable over the telephone were sent let-
ters in the mail. For men that withdrew from the study
but agreed to passive follow-up, the New South Wales
Registry of Births, Deaths and Marriages was used to
ascertain deaths. Time to mortality was defined as years
between the baseline assessment and date of death.
Covariates
Age, dementia or mild cognitive impairment (MCI) diag-
nosis at baseline, comorbidity, education level, and living
status were used as covariates, selected a priori based on
their known association with frailty transitions.9,10,14
Information on age, education level (≥7 years), and living
status (living alone or not) was collected at the baseline
clinic visit. Details of the baseline dementia or MCI diag-
nosis have been published elsewhere.24 Cognitive status
was determined as dementia, MCI, unknown, or no cogni-
tive impairment. For the purposes of analysis, participants
with MCI or dementia at baseline were combined into one
group.
Comorbidity was measured using the Functional
Comorbidity Index (FCI) at each clinic visit (range
0–16).25 Data on arthritis, osteoporosis, asthma, chronic
obstructive pulmonary disease or emphysema, angina pec-
toris, congestive heart failure (or heart disease), myocardial
infarction, Parkinson’s disease, stroke, peripheral vascular
disease, back pain, diabetes mellitus type I and II, and
upper gastrointestinal disease were obtained from the self-
reported questionnaires. Objective assessments were used
to assess the presence of depression, anxiety or panic dis-
orders, visual impairment, hearing impairment, degenera-
tive disc disease, and obesity. Depressive symptoms were
assessed using the 15-item Geriatric Depression Scale with
a score of five or greater indicative of depressive symp-
toms. Anxiety symptoms were measured using the Gold-
berg Anxiety Scale, with a score greater than five
considered as presence of anxiety. Corrected visual acuity
was assessed using a Bailey-Lovie chart, and poor vision
was defined as 6/19 visual acuity or worse. Obesity was
defined as a body mass index (BMI) of 30.0 kg/m2 or
more. For Year 2 and 5 follow-ups, visual impairment was
defined using self-reported data on macular degeneration,
glaucoma, and cataracts. FCI score was computed as the
sum of present conditions. Participants with three or more
missing conditions were assigned a missing value for the
FCI score.
Statistical Analysis
Participant demographic data were summarized numeri-
cally, and the frequencies of transitions between the three
frailty states (robust, prefrail, frail) and death were calcu-
lated. Multistate modeling for panel data26–28 was used to
characterize the transitions between the three frailty states
and progression to death (Figure 1). This enabled quantifi-
MEDICATION USE, FRAILTY TRANSITIONS, AND DEATH 91
cation of frailty progression and regression and simultane-
ously allowed the risk of death to vary according to frailty
status.29 In Figure 1, each a represents the transition inten-
sity (synonymous with the hazard rate in a traditional
time-to-event framework) between adjacent states within
the four-state model. The model assumed that the transi-
tional process occurred in continuous time but was
observed at prespecified times (at each wave). If a partici-
pant was observed as robust at baseline and frail at the
Year 2 follow-up, it was assumed that the participant tran-
sitioned through the prefrail state at some unobserved time
before the Year 2 follow-up. Thus, only transitions
between adjacent frailty states were specified in the model.
This model specification is recommended for multistate
modeling of chronic diseases when using panel data.30 The
model also assumed that each participant’s frailty status
was unknown at the instant before death.
The frailty status at the time of admission of partici-
pants admitted to an aged care facility was not recorded.
Also, the frailty status of those who were alive at the end
of the study (April 12, 2014) was unknown. Therefore,
the frailty status for these participants at these times was
censored. In this framework, censoring means that frailty
status was known only to be robust, prefrail, or frail. It
was specified in the model that the times of admission to
aged care facilities, the end of follow-up, and death were
observed exactly (opposed to panel-observed, which was
specified for the clinic visits).
Number of medications, DBI, and covariates were
incorporated into the transition intensities as follows:
axy ¼ exp½b1 No of medications þ b2 Age
þ b3 Cognitive impairments þ b4 FCI
þ b5 Education level þ b6 Living status ð2Þ
axy ¼ exp½b1 DBI þ b2 Age þ b3 Cognitive impairments
þ b4 FCI þ b5 Education level þ b6 Living status ð3Þ
Thus, two models were constructed: one with number
of medications as the main exposure and another with
DBI as the main exposure. In Equations 2 and 3, axy rep-
resents the transition intensity related to moving from state
x to state y (e.g., from robust to prefrail). Because number
of medications, DBI, and FCI were not observed at time of
admission to aged care, death, or the end of follow-up, the
Figure 1. Structural form of the multistate model. Each a rep-
resents the transition intensity of moving from one state to
another, as indicated by the subscripts.
92 JAMSEN ET AL.
values for these variables at these times were carried for-
ward from the previous clinic visit. Number of medica-
tions, DBI, age, and FCI were treated as time-dependent
covariates. Dementia or MCI at baseline, education level,
and living status were treated as time-independent covari-
ates (baseline values only). The estimated effects of the
medication exposures and covariates on the transition
intensities (b1, b2, b3, b4, b5, b6) were expressed as hazard
ratios (HRs) and 95% confidence intervals (CIs). Model fit
was assessed by plotting the observed prevalence of each
state (robust, prefrail, frail, death) over time and superim-
posing the corresponding model-predicted prevalence.
All data management and analyses were performed in
R.31 Multistate modeling for panel data was performed
using the msm package.30 Results were interpreted as sug-
gested in32.
RESULTS
Participant Characteristics
Data were available for 1,705 participants at the baseline
assessment, 1,366 at 2 years, and 954 at 5 years
(Figure 2). Median age was 76.0 at baseline, 78.0 at
2 years, and 80.6 at 5 years (Table 1). Across the three
assessments, approximately three-quarters of the men were
married, 17% to 18% lived alone, and 85% to 86% had
at least 7 years of education. For all three waves, the med-
ian number of medications and DBI was 4 and 0, respec-
tively. Median FCI was 2 at baseline and 3 at the 2- and
5-year assessments. Approximately 13% of men were diag-
nosed with cognitive impairment at baseline (Table 1).
Participant Transitions
From the robust state, there were 856 (63.9%) stationary
transitions (no transition), 363 (27.1%) transitions to the
prefrail state, 20 (1.5%) transitions to the frail state, and
101 (7.5%) transitions to death. From the prefrail state,
there were 603 (55.4%) stationary transitions, 172
(15.8%) transitions to the robust state, 114 (10.5%) tran-
sitions to the frail state, and 200 (18.4%) transitions to
death. From the frail state, there were 73 (33.3%) station-
ary transitions, 35 (16%) transitions to the prefrail state,
108 (49.3%) transitions to death, and three (1.4%) transi-
tions to the robust state.
Effects of Medication Exposure on Transitions
Accounting for the covariates, there was evidence that
each additional medication was associated with a 22%
Figure 2. Numbers of participants over follow-up.
JANUARY 2016–VOL. 64, NO. 1 JAGS
greater risk of death from the robust state (95%
CI = 1.06–1.41; Table 2). There was no evidence of an
adjusted association between number of medications and
the other transitions.
After adjustment for covariates, each one-unit increase
in DBI was associated with a 73% greater risk of transi-
tioning from the robust to the prefrail state (95%
CI = 1.30–2.31; Table 2) and a 2.75 times greater risk of
transitioning from the robust state to death (95%
CI = 1.60–4.75; Table 2). There was no evidence of an
adjusted association between DBI and the other transi-
tions.
DISCUSSION
This is the first prospective study to examine the effects of
multiple drug use and sedative and anticholinergic load on
frailty transitions and death in a cohort of older men.
After adjusting for clinically important covariates, there
was evidence that each additional medication was associ-
ated with greater risk of transitioning from the robust state
to death. There was also evidence that higher DBI was
associated with greater risk of transitioning from the
robust to prefrail states and from the robust state to death
(after adjusting for the covariates). These findings are of
particular clinical relevance given the high polypharmacy
and anticholinergic and sedative exposures in older peo-
ple.1,2 However, there was no evidence of an effect of
medications on transitions between the prefrail and frail
states or from the prefrail or frail states to death.
The adjusted association between DBI and greater risk
of transitioning from the robust to prefrail states is consis-
tent with previous research in the same cohort concluding
that higher DBI at baseline was associated with incident
frailty over 2 years in high-functioning older men.11 The
association between the number of medications and DBI
and transitioning from a state of robustness to death is
also consistent with previous literature.1,6 The magnitude
of the HRs for this transition were large, but this is plausi-
ble because the number of medications and higher DBI for
participants who were robust may indicate advanced dis-
ease not attributable to frailty or entirely attributable to
other characteristics. These results suggest that clinicians
should exercise caution even when prescribing additional
and DBI medications to robust older men.
An interesting finding was that there was no statistical
evidence of the effects of number of medications and DBI
on transitions from the prefrail and frail states, which sug-
gests that, once a participant became prefrail or frail, the
covariates considered in this study, as well as frailty itself,
played more-substantial roles in these transitions than
JAGS JANUARY 2016–VOL. 64, NO. 1 MEDICATION USE, FRAILTY TRANSITIONS, AND DEATH 93

Table 1. Participant Characteristics at the Three Table 2. Results from the Multistate Models
Assessments
Hazard Ratio
Baseline 2-Year 5-Year Exposure (95% Confidence Interval)a
Assessment, Assessment, Assessment,
Characteristic n = 1,705 n = 1,366 n = 954 Number of medications
Robust—prefrail 1.04 (1.00–1.09)
Age, median (IQR) 76 (8) 78 (8) 81 (7) Robust—death 1.22 (1.06–1.41)
Married, n (%) 1,278 (75) 1,035 (76) 737 (77) Prefrail—robust 0.99 (0.93–1.05)
Living alone, n 318 (19) 239 (18) 163 (17) Prefrail—frail 1.06 (0.99–1.13)
(%) Prefrail—dead 1.02 (0.93–1.11)
Tertiary ≥7 years 1,429 (85) 1,159 (86) 819 (86) Frail—prefrail 0.95 (0.85–1.06)
of education, n (%) Frail—dead 1.01 (0.96–1.07)
Number of 4 (4) 4 (4) 4 (4) Drug Burden Index
medications, Robust—prefrail 1.73 (1.30–2.31)
median (IQR) Robust—dead 2.75 (1.60–4.75)
Drug Burden 0 (0.17) 0 (0.33) 0 (0.33) Prefrail—robust 0.90 (0.59–1.36)
Index, median Prefrail—frail 1.03 (0.76–1.40)
(IQR) Prefrail—dead 1.18 (0.89–1.56)
Functional 2 (3) 3 (2) 3 (3) Frail—prefrail 0.65 (0.33–1.27)
Comorbidity Frail—dead 0.92 (0.73–1.16)
Score, median
a
(IQR) Adjusted for age, diagnosis of dementia or mild cognitive impairment at
Diagnosis of 213 (13) 157 (12) 77 (8) baseline, comorbidity, education level, and living status.
dementia or mild
cognitive
impairment, n (%) multistate modeling framework allowed the research ques-
tion of interest to be addressed using a single, flexible,
IQR = interquartile range. pharmacologically plausible model.
There are some limitations to this study. As with other
medication exposure. Studies that show the independent observational studies, residual confounding by other medi-
effects of frailty on mortality and other clinically relevant cal conditions, disease severity, and indication needs to be
outcomes in older people support this observation,33 and a considered. Despite adjustment for clinically important
summary of four large prospective cohort studies showed covariates, it is possible that number of medications and
that the worst adverse outcomes (falls, disability, hospital- DBI are markers for underlying conditions and diseases
ization, care home admission, death) occurred in the frail- that drive the transitions. The clinical diagnosis of demen-
est people.15 Nevertheless, future longitudinal studies are tia or MCI was conducted at the baseline visit only, so it
warranted to investigate further the association between was not possible to include it as a time-dependent covari-
medication use and transitions between and from different ate. The modified measures for three components of the
stages of frailty. frailty scale were used in this study. The study was con-
ducted in older men living in the community in Australia,
Strengths and Limitations which may limit generalizability to older men living in
other areas or in aged care facilities.24 Moreover, the
This study had several strengths. CHAMP is a large, popu- results may not be applicable to older women. The base-
lation-based cohort study that uses objective and clinically line participation rate in CHAMP was approximately
validated cognitive and physical performance measures 50%. Nevertheless, the men in the study had self-reported
and frailty scales.24 A systematic medication inventory was health status similar to that of men of the same age in the
performed by checking all medications, including over-the- nationally representative Men in Australia Telephone
counter medications, that participants brought in during a Survey.35 With respect to the DBI, di for a particular medi-
clinic visit.34 The use of multistate modeling allowed char- cation may vary between participants because of partici-
acterization of the transitions (progression and regression) pant-specific pharmacokinetics and pharmacodynamics.20
between frailty states and, simultaneously, progression to Furthermore, the DBI measures exposure to regular doses
death. This approach used all available data and allowed of anticholinergic and sedative medications rather than as-
the risk of death to change according to frailty state, needed drugs. This may have led to underestimation of
which is relevant because it has been reported that frailty true exposure.
is associated with greater risk of death.29 Furthermore, the With regard to the statistical analysis, the models used
structure of the model is consistent with the “spiral of in this study relied on the Markov assumption, which
decline” mentioned in previous work.15 Time-varying means that future transitions depend on the current state.
covariates and participant-specific follow-up times were Because the clinic visit data were panel-observed (opposed
specified, providing means of accurate characterization of to observed in continuous time), it was not possible to
the transition process. A unique feature of the multistate assess formally whether future transitions depended on the
modeling approach is that the estimation of each transition time spent in each state (a semi-Markov process).
accounted for all other transitions, which may result in Although number of medications, DBI, age, and FCI were
less-biased estimates than those obtained using a series of declared to be time-dependent covariates, the model
independent time-to-event models. Moreover, using the assumed that their respective values were constant
94 JAMSEN ET AL.
between observations, which may not be entirely accurate.
Cognitive diagnosis, education level, and living status were
recorded at baseline only. To accommodate model conver-
gence, the model assumed that the frailty state before aged
care admission and the end of follow-up was the same as
the previous observation. Also, because the model was not
fitted in a mixed-effects framework (not available in msm),
the estimated HRs should be interpreted as typical values
for the population. As such, the model should not be used
as an individual prediction or simulation tool. Last, meth-
ods of addressing missing data due to missed clinic visits,
such as multiple imputation, were not performed because
the benefits of multiple imputation were likely to be mini-
mal because the exposures of interest (total number of
medications and DBI) would have required imputation.36,37
CONCLUSIONS
Although the possibility of confounding by indication can-
not be excluded, additional medications were associated
with greater risk of mortality in robust community-dwelling
older men. Higher DBI was also associated with greater risk
of mortality in these men, as well as transitioning from the
robust to the prefrail state. Future longitudinal studies, as
well as withdrawal or deprescribing trials, are warranted to
investigate further the relationships between high-risk medi-
cation exposures and frailty transitions.
ACKNOWLEDGMENT
The authors acknowledge Ms. Clara Leung, who con-
tributed to the medication data cleaning and coding. The
authors also thank A/Prof Heather Allore of Yale Univer-
sity for her comments on the statistical analysis.
The CHAMP study was funded by the Australian
National Health and Medical Research Council (Project
Grant 301916) and the Ageing and Alzheimer’s Research
Institute of the Concord Hospital, Sydney. Kris Jamsen is
funded by the Australian National Health and Medical
Research Council Cognitive Decline Partnership Centre at
the time this work was done. Jenni Ilom€ aki and Danijela
Gnjidic are supported by Australian National Health and
Medical Research Council Early Career Fellowships.
Conflict of Interest: The editor in chief has reviewed
the conflict of interest checklist provided by the authors
and has determined that the authors have no financial or
any other kind of personal conflicts with this paper.
Author Contributions: Jamsen, Bell, Hilmer, Kirk-
patrick, Ilom€aki, Gnjidic: study concept and design. Le
Couteur, Blyth, Handelsman, Naganathan, Cumming:
acquisition of subjects and data. Jamsen, Bell, Hilmer,
Kirkpatrick, Ilom€aki, Le Couteur, Blyth, Handelsman,
Naganathan, Cumming, Gnjidic: data analysis and inter-
pretation. Jamsen, Bell, Hilmer, Kirkpatrick, Ilom€ aki, Le
Couteur, Blyth, Handelsman, Naganathan, Cumming,
Gnjidic: preparation of manuscript.
Sponsor’s Role: None.
REFERENCES
1. Maher RL, Hanlon J, Hajjar ER. Clinical consequences of polypharmacy
in elderly. Expert Opin Drug Saf 2014;13:57–65.
JANUARY 2016–VOL. 64, NO. 1 JAGS
2. Jyrkka J, Enlund H, Korhonen MJ et al. Patterns of drug use and factors
associated with polypharmacy and excessive polypharmacy in elderly per-
sons: Results of the Kuopio 75 + study: A cross-sectional analysis. Drugs
Aging 2009;26:493–503.
3. Hajjar ER, Cafiero AC, Hanlon JT. Polypharmacy in elderly patients. Am J
Geriatr Pharmacother 2007;5:345–351.
4. Pitkala KH, Juola AL, Kautiainen H et al. Education to reduce potentially
harmful medication use among residents of assisted living facilities: A ran-
domized controlled trial. J Am Med Dir Assoc 2014;15:892–898.
5. Hilmer SN, Mager DE, Simonsick EM et al. Drug burden index score and
functional decline in older people. Am J Med 2009;122:1142–1149.
6. Gnjidic D, Hilmer SN, Hartikainen S et al. Impact of high risk drug use
on hospitalization and mortality in older people with and without Alzhei-
mer’s disease: A national population cohort study. PLoS ONE 2014;9:
e83224.
7. Nishtala PS, Narayan SW, Wang T et al. Associations of drug burden
index with falls, general practitioner visits, and mortality in older people.
Pharmacoepidemiol Drug Saf 2014;23:753–758.
8. Fried LP, Tangen CM, Walston J et al. Frailty in older adults: Evidence for
a phenotype. J Gerontol A Biol Sci Med Sci 2001;56A:M146–M156.
9. Hoogendijk EO, van Hout HP, Heymans MW et al. Explaining the associ-
ation between educational level and frailty in older adults: Results from a
13-year longitudinal study in the Netherlands. Ann Epidemiol
2014;24:538.e532–544.e532.
10. Lee JS, Auyeung TW, Leung J et al. Transitions in frailty states among
community-living older adults and their associated factors. J Am Med Dir
Assoc 2014;15:281–286.
11. Gnjidic D, Hilmer SN, Blyth FM et al. High-risk prescribing and incidence
of frailty among older community-dwelling men. Clin Pharmacol Ther
2012;91:521–528.
12. Boyd CM, Xue QL, Simpson CF et al. Frailty, hospitalization, and progres-
sion of disability in a cohort of disabled older women. Am J Med
2005;118:1225–1231.
13. Yun H, Delzell E, Ensrud KE et al. Predicting hip and major osteoporotic
fractures using administrative data. Arch Intern Med 2010;170:1940–1942.
14. Woods NF, LaCroix AZ, Gray SL et al. Frailty: Emergence and conse-
quences in women aged 65 and older in the Women’s Health Initiative
Observational Study. J Am Geriatr Soc 2005;53:1321–1330.
15. Clegg A, Young J, Iliffe S et al. Frailty in elderly people. Lancet
2013;381:752–762.
16. Crentsil V, Ricks MO, Xue QL et al. A pharmacoepidemiologic study of
community-dwelling, disabled older women: Factors associated with medi-
cation use. Am J Geriatr Pharmacother 2010;8:215–224.
17. Gill TM, Gahbauer EA, Allore HG et al. Transitions between frailty states
among community-living older persons. Arch Intern Med 2006;166:418–
423.
18. Gnjidic D, Johnell K. Clinical implications from drug-drug and drug-dis-
ease interactions in older people. Clin Exp Pharmacol Physiol
2013;40:320–325.
19. Cumming RG, Handelsman D, Seibel MJ et al. Cohort profile: The Con-
cord Health and Ageing in Men Project (CHAMP). Int J Epidemiol
2009;38:374–378.
20. Hilmer SN, Mager DE, Simonsick EM et al. A drug burden index to define
the functional burden of medications in older people. Arch Intern Med
2007;167:781–787.
21. Donohoo E. MIMS Annual 2007. Sydney, Australia: CMPMedica Aus-
tralia, 2007.
22. Fried LP, Ferrucci L, Darer J et al. Untangling the concepts of disability,
frailty, and comorbidity: Implications for improved targeting and care. J
Gerontol A Biol Sci Med Sci 2004;59A:255–263.
23. Blyth FM, Rochat S, Cumming RG et al. Pain, frailty and comorbidity on
older men: The CHAMP study. Pain 2008;140:224–230.
24. Gnjidic D, Le Couteur DG, Hilmer SN et al. Sedative load and functional
outcomes in community-dwelling older Australian men: The CHAMP
study. Fundam Clin Pharmacol 2012;28:10–19.
25. Groll DL, To T, Bombardier C et al. The development of a comorbidity
index with physical function as the outcome. J Clin Epidemiol
2005;58:595–602.
26. Kay R. A Markov Model for analysing cancer markers and disease states
in survival studies. Biometrics 1986;42:855–865.
27. Kalbeisch J, Lawless J. The analysis of panel data under a Markov assump-
tion. J Am Statist Assoc 1985;80:863–871.
28. Cox DR, Miller HD. The Theory of Stochastic Processes. London: Chap-
man and Hall, 1965.
29. Kulmala J, Nykanen I, Hartikainen S. Frailty as a predictor of all-cause
mortality in older men and women. Geriatr Gerontol Int 2014;14:899–
905.
JAGS JANUARY 2016–VOL. 64, NO. 1
30. Jackson CH. Multi-state models for panel data: The msm package for R. J
Statistical Software 2011;38:1–29.
31. R Development Core Team. R: A Language and Environment for Statistical
Computing. Vienna, Austria: R Foundation for Statistical Computing, 2013.
32. Sterne JA, Davey Smith G. Sifting the evidence-what’s wrong with signifi-
cance tests? BMJ 2001;322:226–231.
33. Morley JE, Vellas B, van Kan GA et al. Frailty consensus: A call to action.
J Am Med Dir Assoc 2013;14:392–397.
34. Lau HS, de Boer A, Beuning KS et al. Validation of pharmacy records in
drug exposure assessment. J Clin Epidemiol 1997;50:619–625.
MEDICATION USE, FRAILTY TRANSITIONS, AND DEATH 95
35. Holden CA, McLachlan RI, Pitts M et al. Men in Australia Telephone Sur-
vey (MATeS): A national survey of the reproductive health and concerns of
middle-aged and older Australian men. Lancet 2005;366:218–224.
36. Lee KJ, Carlin JB. Recovery of information from multiple imputation: A
simulation study. Emerg Themes Epidemiol 2012;9:3.
37. Karahalios A, Baglietto L, Lee KJ et al. The impact of missing data on
analyses of a time-dependent exposure in a longitudinal cohort: A simula-
tion study. Emerg Themes Epidemiol 2013;10:6.