Natural History of Dependency in the Elderly: A 24

American Journal of Epidemiology
© The Author 2016. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of
Public Health. All rights reserved. For permissions, please e-mail: [email protected].
Vol. 183, No. 4
DOI: 10.1093/aje/kwv223
Advance Access publication:
January 28, 2016
Original Contribution
Natural History of Dependency in the Elderly: A 24-Year Population-Based Study
Using a Longitudinal Item Response Theory Model
Arlette Edjolo*, Cécile Proust-Lima, Fleur Delva, Jean-François Dartigues, and Karine Pérès
* Correspondence to Dr. Arlette Edjolo, Université de Bordeaux–Institute de Santé Publique et de Développemont (ISPED), INSERM U1219,
Case 11, 146 rue Léo Saignat, 33076 Bordeaux Cedex, France (e-mail: [email protected]).
Initially submitted July 23, 2014; accepted for publication August 13, 2015.
We aimed to describe the hierarchical structure of Instrumental Activities of Daily Living (IADL) and basic Activities of Daily Living (ADL) and trajectories of dependency before death in an elderly population using item response
theory methodology. Data were obtained from a population-based French cohort study, the Personnes Agées QUID
(PAQUID) Study, of persons aged ≥65 years at baseline in 1988 who were recruited from 75 randomly selected
areas in Gironde and Dordogne. We evaluated IADL and ADL data collected at home every 2–3 years over a
24-year period (1988–2012) for 3,238 deceased participants (43.9% men). We used a longitudinal item response
theory model to investigate the item sequence of 11 IADL and ADL combined into a single scale and functional
trajectories adjusted for education, sex, and age at death. The findings confirmed the earliest losses in IADL (shopping, transporting, finances) at the partial limitation level, and then an overlapping of concomitant IADL and ADL,
with bathing and dressing being the earliest ADL losses, and finally total losses for toileting, continence, eating, and
transferring. Functional trajectories were sex-specific, with a benefit of high education that persisted until death in
men but was only transient in women. An in-depth understanding of this sequence provides an early warning of
functional decline for better adaptation of medical and social care in the elderly.
Activities of Daily Living; aging; dependency; functional trajectories; hierarchy of limitations; item response theory;
Instrumental Activities of Daily Living
Abbreviations: ADL, Activities of Daily Living; HE, high-educated; IADL, Instrumental Activities of Daily Living; LE, low-educated;
PAQUID, Personnes Agées QUID; SD, standard deviation; SE, standard error.
of Lawton and Brody (5). The Katz scale (4) refers to daily
self-care activities and was developed as an ordinal index
for assessing physical functioning in the elderly, in which
ADL losses were supposedly ordered by increasing severity,
although several studies have demonstrated alternate hierarchies (6). The Lawton scale (5) was designed to assess independent living skills necessary to live in the community, including
more complex and sensitive tasks than ADL, with consequent
restrictions arising earlier in the dependency process (7).
From the scales of activity limitations and despite a preexisting hierarchy, dependency is analyzed either as a degree,
while it is above all a continuous process, or by summary
scores which suppose an unlikely equal weighting for each
activity. Yet simple scores are not sufficiently accurate to discriminate subtle proximate levels of dependency between
Demographic aging combined with a constant increase in
life expectancy in developed societies makes dependency in
the elderly a major challenge. Dependency is a highly complex
and multidomain construct ( physical, psychological, cognitive, economic, etc.) (1) for which no consensus has been
reached. However, the common aspect of all of its components
is a need for human assistance to engage in daily functioning
(2). Because the entire picture of dependency is difficult to approach, due notably to its major psychosocial component, the
study of dependency in epidemiology is mainly restricted to its
functional aspects.
Dependency is often assessed by means of activity limitations (the formal and updated term for disability (3)) outlined
in the Activities of Daily Living (ADL) scale of Katz et al. (4)
and the Instrumental Activities of Daily Living (IADL) scale
277
Am J Epidemiol. 2016;183(4):277–285
278 Edjolo et al.
Table 1. Characteristics of Participants at Baseline, at the Last Study Visit With Information on IADL/ADL, and at
Death (n = 3,238), PAQUID Study, 1988–2012a
At Baseline (n = 3,238)
No.
%
Mean (SD)
At the Last Visit (n = 3,238)
No.
%
Mean (SD)
Sociodemographic Characteristics
Age, years
76.46 (6.83)
83.24 (7.56)
Male sex
1,420
43.87
1,420
43.87
High educational levelb
2,033
62.81
2,033
62.81
Marital status (cohabiting couples)
1,780
54.99
1,354
42.02
245
7.57
216
6.68
1,212
37.40
1,652
51.30
1,814
56.67
2,351
73.81
101
3.12
719
22.21
Single, divorced, separate, or other
Widowed
Polymedicated (use of >3 medications)
Dementia
MMSE score
25.21 (4.21)
22.56 (7.27)
No. of IADL limitations (partial or total)
0.89 (1.42)
2.26 (2.02)
No. of ADL limitations (partial or total)
0.38 (1.00)
1.45 (2.09)
Activity Limitation Items
Bathing
No limitation
2,938
90.82
2,141
Partial limitation
173
5.35
311
66.28
9.63
Total limitation
124
3.83
778
24.09
Dressing
No limitation
2,984
92.24
2,331
72.17
Partial limitation
128
3.96
204
6.32
Total limitation
123
3.80
695
21.52
3,158
97.62
2,703
83.68
Partial limitation
51
1.58
262
8.11
Total limitation
26
0.80
265
8.20
2,810
86.97
2,194
68.03
371
11.48
553
17.15
50
1.55
478
14.82
Toileting
No limitation
Continence
No limitation
Partial limitation
Total limitation
Table continues
individuals (8). An alternative is to consider dependency as a
latent continuum for which IADL/ADL are meaningful manifestations. This is the purpose of the item response theory,
which consists of linking a latent variable (here, the non–
directly observable dependency) to observable manifestations
(here, IADL/ADL limitations) (9). Dependency is not only a
continuum but also a dynamic process evolving over time,
and the description of its trajectories in the elderly population produces essential information about the natural history
of aging.
In the current study, we aimed to describe simultaneously
the hierarchy of the 5 IADL, common to both sexes, and 6
ADL occurrences and the trajectories of dependency before
death among elderly participants in a French cohort study,
the Personnes Agées QUID (PAQUID) Study, over 24 years of
follow-up using a longitudinal item response theory approach.
METHODS
Study cohort
The PAQUID Study, initiated in 1988, is an ongoing prospective epidemiologic study on cerebral and functional
aging conducted in the general population. The methodology
has been previously described in full (10). At baseline, 3,777
participants aged ≥65 years, initially all community-dwellers, were randomly recruited from the electoral rolls of 2 administrative areas in southwestern France. Participants were
interviewed face to face at home every 2–3 years by specially
trained neuropsychologists. Data were collected from the
participant or from a proxy, when self-assessment was
impossible or invalid. Sociodemographic, environmental,
and health-related information was prospectively collected
at each wave. Educational level (high-educated (HE) or
Am J Epidemiol. 2016;183(4):277–285
Dynamic Item Response Theory Model of IADL/ADL 279
Table 1. Continued
At Baseline (n = 3,238)
No.
%
3,142
Mean (SD)
At the Last Visit (n = 3,238)
No.
%
97.16
2,697
83.52
83
2.57
416
12.88
9
0.28
116
3.59
Mean (SD)
Transferring
No limitation
Partial limitation
Total limitation
Eating
No limitation
3,129
96.75
2,631
81.53
Partial limitation
92
2.84
435
13.48
Total limitation
13
0.40
161
4.99
Telephoning
No limitation
2,865
88.54
2,001
62.01
Partial limitation
233
7.20
642
19.89
Total limitation
138
4.26
584
18.10
Shopping
No limitation
2,214
68.40
1,220
37.76
Partial limitation
569
17.58
663
20.52
Total limitation
454
14.03
1,348
41.72
2,230
68.89
1,237
38.30
Partial limitation
866
26.75
1,462
45.26
Total limitation
141
4.36
531
16.44
2,989
92.45
2,095
64.96
Partial limitation
142
4.39
549
17.02
Total limitation
102
3.15
581
18.02
Transporting
No limitation
Medication
No limitation
Finances
No limitation
2,623
81.06
1,740
53.94
Partial limitation
362
11.19
514
15.93
Total limitation
251
7.76
972
30.13
Abbreviations: ADL, Activities of Daily Living; IADL, Instrumental Activities of Daily Living; MMSE, Mini-Mental State
Examination; PAQUID, Personnes Agées QUID; SD, standard deviation.
a
The mean age at death was 86.70 (SD, 6.88) years.
b
Defined as ≥7 years of schooling.
low-educated (LE)) was defined by duration of schooling,
with a threshold of 7 years (≥7 years vs. <7 years). The PAQUID program included a systematic and regular search for
all deaths that occurred in the cohort. Unlike the functional
assessments, which were conducted at planned follow-up
points over a period of 22 years, deaths took place in continuous time up to 24 years after inclusion.
Informed consent was obtained from all participants, and
the ethics committee of the university hospital in Bordeaux
approved the research according to the principles embodied
in the Declaration of Helsinki.
Functional assessment
The IADL were assessed using the French version of the
Lawton scale (5), including telephoning, shopping, using transAm J Epidemiol. 2016;183(4):277–285
portation, handling medication, and managing finances. To ensure comparability, the 3 items specific to females (laundry,
meals, and housekeeping) were removed. The basic ADL
were assessed using the French version of the Katz Index of
ADL (4), including bathing, dressing, toileting, continence, eating, and transferring. Whatever the scale, limitation was defined
here on a 3-point scale as having no limitation, partial limitation, or total limitation (coded 0, 1, and 2, respectively) in performing the task. ADL were originally scored in 3 grades, but
for IADL the new rating considers the minimum and maximum
grades of the Lawton scale, defined respectively as no limitation
(0) and total limitation (2). Partial limitation consequently includes the remaining intermediate grade (1). Based on previous
studies (11–13), we assumed that both the IADL and ADL
scales assessed a single underlying dimension of dependency.
280 Edjolo et al.
Selection procedure
In the current study, we investigated the item sequence of
IADL and ADL and the process of increasing dependency preceding death—that is, among participants who died during the
24 years of follow-up and had at least 1 available IADL/ADL
measurement. Death as an endpoint better reflects the complete
natural history of dependency from IADL and ADL, the first
scale being more sensitive to early deterioration and the latter
being more sensitive to terminal limitations (7). Participants
who were alive after the 24-year active research period for
deaths and those with completely missing data were excluded
from the present study.
variability in the latent dependency at death. These 2 constraints did not alter the hierarchy between items or the
effects of covariates on the latent trajectory. Multivariate
Wald tests were used to assess the statistical differences at a
5% significance level for location and discrimination parameters, as well as for associations with covariates.
Analyses were performed with SAS software, version
9.1.3 (SAS Institute, Inc., Cary, North Carolina), and the
Fortran90 HETMIXSURV program (free download at http://
www.isped.u-bordeaux.fr/biostat).
RESULTS
Participants’ characteristics
Statistical methods
Eligible and excluded participants were compared using
Student’s t test and the χ2 test.
Time-independent relationships between each repeated
item and the latent trait over time were specified in a 2parameter probit item response theory model for graded responses. Thus, each 3-level IADL and ADL limitation was
characterized by 3 parameters:
1. Two location parameters (αpartial and αtotal ) give the
threshold in the latent trait continuum defining the change
of severity level (between no limitation and partial limitation and between partial limitation and total limitation)
measured by the item. The location parameter—also called
difficulty—can be interpreted as a probability: A participant with a dependency at αpartial has a 0.5 (50%) probability of being at least partially limited and a 0.5 probability
of having no limitation. Similarly, with a dependency at
αtotal, the participant has a 0.5 probability of being totally
limited and a 0.5 probability of having at most a partial
limitation for a given item.
2. The discrimination parameter corresponds to the inverse
of the residual variability of the item. A higher discrimination value means that the item is more powerful for
positioning the individual between proximate levels of
dependency.
The sequence of items was established using the estimated
location values along the dependency continuum.
Dependency, defined as the unique latent process that generated the set of items, was simultaneously modeled according to number of years preceding death in a linear mixed
model with a quadratic trajectory shape to allow acceleration
of the functional decline towards death (14). Three correlated
individual random effects (on the intercept, the slope, and
the quadratic slope with years to death) captured the interparticipant variability. The trajectories were adjusted for age at
death, sex, education, sex × education, and the interactions of
education, sex, and education × sex with number of years to
death and squared years to death. The reference category for covariates was LE men who died at age 85 years (the median age at
death in this category). The dimensions (mean and dispersion)
of the latent dependency were defined as follows: Latent dependency was centered (level 0) around the mean latent dependency level at death in the reference category. Latent dependency
was scaled (1 standard deviation) to the interindividual
The present sample included 3,238 deceased participants
(86% of the cohort) after exclusion of the 539 participants
who were still alive after 24 years of follow-up. All of the deceased participants were included, because they all had at
least ADL and IADL information at baseline. At baseline,
the excluded participants were significantly younger (mean
age = 69.4 (standard deviation (SD), 3.4) years), more often
female (71.1%), more often highly educated (74.4%), more
often without limitations (48.1%), and less often demented
(0.2%) but were more often polymedicated (use of >3 medications) (66.2%) (P < 0.001). Participants’ characteristics are
presented in Table 1. The mean age was 76.5 (SD, 6.9) years
at baseline and 88.7 (SD, 6.9) years at death; 56.1% were female, and 62.8% were highly educated (58.6% for women,
68.3% for men). At baseline, only 43.3% were polymedicated and 3.1% were demented. The prevalence of limitations
was lower than 10.0% for all of the ADL except continence
(13.0%) and was between 7.6% and 31.6% for IADL limitations (with higher prevalences in shopping and transporting),
with 0.4 (SD, 1.0) ADL limitations and 0.9 (SD, 1.4) IADL
limitations on average.
The participants were not necessarily seen at each visit
from baseline to death. (See Web Table 1, available at http://
aje.oxfordjournals.org/, for visit-specific information about
response rates.) The median number of visits with IADL/
ADL information per participant was 5 (interquartile range,
3–7) as compared with the 11 planned visits. The median
length of follow-up until death was 9.82 years (interquartile
range, 4.97–14.93), but the median elapsed time between
baseline and the last visit with IADL/ADL information was
5.22 years (interquartile range, 1.02–10.29), and the median
elapsed time between the last visit and death was 1.86 years
(interquartile range, 0.93–3.64). (See Web Tables 2 and 3 for
further details according to education and sex.)
Item sequence
The sequence of IADL and ADL losses provided by the
estimated item locations is presented in Table 2 and Web Figure 1. At the location threshold, a participant has a 0.5 probability of having a strictly lower level of limitation. The
vertical axis on Web Figure 1 represents the continuum of dependency, centered on 0 as a standardized score. The 0 represents the mean level of dependency at the time of death for a
less educated man who died at age 85 years, and a 1-unit deviation represents the standard deviation of the sample at the
Am J Epidemiol. 2016;183(4):277–285
Dynamic Item Response Theory Model of IADL/ADL 281
Table 2. Item Location Values From a Normal Ogive Item Response
Theory Model of IADL/ADL (n = 3,238), PAQUID Study, 1988–2012
Activity Limitation Item
Table 3. Estimated Item Discrimination Values (δ) From a Normal
Ogive Item Response Theory Model of IADL/ADL (n = 3,238),
PAQUID Study, 1988–2012
Item Location (SE)
Total transferring
2.01 (0.06)
Activity Limitation Item
Total eating
1.77 (0.05)
Very high discrimination
Total continence
1.55 (0.06)
Bathing
2.00 (0.05)
Total toileting
1.38 (0.04)
Toileting
1.94 (0.06)
Partial transferring
0.82 (0.04)
Dressing
1.86 (0.05)
Total telephoning
0.80 (0.04)
Eating
1.79 (0.05)
Partial toileting
0.79 (0.05)
Shopping
1.77 (0.04)
0.79 (0.04)
Transferring
1.67 (0.05)
Total medication
0.73 (0.04)
Medication
1.61 (0.04)
Partial eating
0.66 (0.04)
Finances
1.49 (0.03)
Total dressing
0.52 (0.04)
Transporting
1.49 (0.03)
Total bathing
0.41 (0.04)
Telephoning
1.37 (0.03)
Total transporting
Partial continence
0.21 (0.04)
Total finances
0.16 (0.04)
Partial dressing
0.15 (0.04)
Partial medication
−0.02 (0.04)
Partial bathing
−0.06 (0.04)
Partial telephoning
−0.14 (0.04)
Total shopping
−0.28 (0.04)
Partial finances
−0.56 (0.04)
Partial transporting
−1.06 (0.04)
Partial shopping
−1.07 (0.04)
Abbreviations: ADL, Activities of Daily Living; IADL, Instrumental
Activities of Daily Living; PAQUID, Personnes Agées QUID; SE,
standard error.
Discrimination (δ) (SE)
Moderate discrimination
Continence
0.73 (0.02)
Abbreviations: ADL, Activities of Daily Living; IADL, Instrumental
Activities of Daily Living; PAQUID, Personnes Agées QUID; SE,
standard error.
Discrimination
According to the guidelines of Baker (14) for a normal
ogive model, item discrimination values were moderate for
continence and very high (discrimination value >1) for all
other items (Table 3). (Complete item behavior is illustrated
by item characteristic curves in Web Figure 3.)
Functional trajectories
time of death. The higher the vertical axis value, the more severe the dependency. Partial and total limitations are shown
on the left and right sides of this vertical axis, respectively.
Nonsignificantly different items are presented on the same
line (e.g., continence and dressing for partial limitation).
Firstly, with a range between −1.07 and 2.01 units, we confirmed that the dependency continuum begins with IADL
losses and ends with ADL losses, with an overlapping of IADL
and ADL losses in the middle. Regarding some key points, at
the beginning, simultaneous partial limitations in shopping
(location value, −1.07 (standard error (SE), 0.04) and transporting (−1.06 (SE, 0.04)) were observed; next, shopping was the
first task to be restricted at the total level of limitation (−0.28
(SE, 0.04)). At the “average” functional level of the reference
category (i.e., at 0), the first ADL loss occurred with partial
limitation in bathing (−0.06 (SE, 0.04)), followed by partial
limitation in managing medication (−0.02 (SE, 0.04)). Next
the first total limitations were in bathing (0.41 (SE, 0.04)) and
dressing (0.52 (SE, 0.04)). Finally, at the end of the continuum,
we found total limitations in toileting (1.38 (SE, 0.04)), continence (1.55 (SE, 0.04)), eating (1.77 (SE, 0.04)), and transferring (2.01 (SE, 0.04)). This predicted hierarchy of limitations
was confirmed by the comparison with the prevalent limitations
both at baseline and at the last visit (Web Figure 2).
Am J Epidemiol. 2016;183(4):277–285
Figure 1 displays the mean latent functional trajectories
according to sex and education and their interaction (P <
0.001) for an age at death of 85 years (the graphs for ages
at death of 80, 85, and 90 years are given in Web Figure 4).
Four functional profiles were identified 22 years before death
from the lowest level of dependency to the highest: HE men
(mean = −2.99 units (SE, 0.11)), LE men (−2.41 (SE, 0.18)),
HE women (−2.07 (SE, 0.12)), and LE women (−1.87 (SE,
0.10)). According to the 95% confidence intervals, the profile
for HE men differed significantly from that of the other groups
over the 22-year period (P < 0.001). Between 18 and 2.5
years before death, HE women were significantly (P < 0.001)
different from the LE women. From 6 years before death onward, only 2 profiles remained: HE men versus the other 3
groups. Finally, at death, the HE men kept the best mean
functional status (−0.43 (SE, 0.04)), corresponding roughly
to the thresholds of total limitation in bathing (Table 2). In
contrast, HE women (0.08 (SE, 0.05)), LE men (reference
category at 0), and LE women (who were at the most disadvantaged level of functioning (0.15 (SE, 0.04))) shared
the same mean level of latent dependency at death, which
corresponded approximately to a threshold between partial
limitations in handling medication and dressing. This is in accordance with the data: At baseline and at the last visit (Web
282 Edjolo et al.
1.00
0.50
0.00
Latent Dependency
–0.50
–1.00
–1.50
–2.00
–2.50
–3.00
LE Men
HE Men
LE Women
HE Women
–3.50
–4.00
22
20
18
16
14
12
10
8
6
4
2
0
Time Before Death, years
Figure 1. Twenty-two-year mean trajectories of latent dependency preceding death, by sex and educational level, PAQUID Study (n = 3,238),
1988–2012. A quadratic linear mixed model adjusted for age at death, education, sex, the interaction between education and sex, and the interactions of education, sex, and education × sex with number of years to death and squared years to death. A 1-point unit corresponds to the standard
deviation of the mean level of latent dependency at death. HE, high-educated; LE, low-educated; PAQUID, Personnes Agées QUID. Dashed
curves, 95% confidence intervals.
Table 4), HE men had globally the highest prevalence of performing daily activities for all activities (except for dressing,
transferring, and eating at baseline), the lowest being seen for
LE women.
Explained variation
Twenty-two years before death, the latent dependency explained more than 90.0% of the interindividual variability for
each item except continence (84.3%). Ten years before death,
it explained more than 80.0% of the variance for each item
and 54.2% for continence (Table 4).
DISCUSSION
We studied the natural history of dependency in the last 2
decades of life in the PAQUID cohort. We provided a synthetic
hierarchy of the combined Lawton (5) and Katz (4) scales
along the latent dependency continuum without prior assumptions. We simultaneously described the functional trajectories
preceding death and confirmed sex and education inequalities.
Highly educated men died at a lower level of dependency than
the other groups. This is one of the rare studies to show such a
hierarchy using longitudinal data from a large representative
sample of elders with regular follow-up, an essential prerequisite to study of the natural history of dependency.
The hierarchy of limitations
Along the continuum of dependency, functional losses
began with half of the IADL (transporting, shopping, and finances, then telephoning) and ended with ADL (eating and
transferring), with an overlapping of IADL and ADL limitations in the middle. As noted in previous publications (15–
17), bathing and dressing are the first ADL losses, also
defined by Katz et al. as the thresholds of disability (4). They
occurred here when the process was next precipitated into a
cascade of losses before the extreme stage. This predicted hierarchy of limitations was confirmed by a comparison with
the prevalent limitations both at baseline and at the last visit
(see Web Figure 2). The innovative feature of our analysis
(longitudinal, large sample, 3-point quotation) limits comparisons, but in a cross-sectional Rasch analysis conducted
on 372 volunteers, Doble and Fisher (18) described a scale of
14 dichotomous items with an overlap between ADL and
IADL and showed quite similar results (except for continence).
In the present study, the early partial limitations (transporting
and shopping) concerned 2 IADL for which physical skills
and cognitive competence are particularly required and may
actually be the early physical manifestations of aging and the
entry into the disablement process.
Finally, the combined scale and the 2 levels of limitation
( partial and total) allowed an extension of the spectrum of
Am J Epidemiol. 2016;183(4):277–285
Dynamic Item Response Theory Model of IADL/ADL 283
Table 4. Percentage of Explained Variance in a Longitudinal Item
Response Theory Model of IADL/ADL for Activity Limitations 20, 15,
10, and 5 Years Before Death and At Death (n = 3,238), PAQUID
Study, 1988–2012
Activity
Limitation
Item
% of Variance Explained,
by No. of Years Before Death
20
15
10
5
0a
Bathing
97.6
95.5
89.9
75.7
80.0
Dressing
97.2
94.9
88.5
73.0
77.7
Toileting
97.4
95.2
89.3
74.6
79.0
Continence
84.3
73.9
54.2
29.3
34.7
Eating
97.0
94.4
87.7
71.4
76.2
Transferring
96.6
93.7
86.1
68.4
73.5
Telephoning
95.0
90.9
80.7
59.4
65.3
Shopping
96.9
94.4
87.5
70.9
75.8
Transporting
95.7
92.2
83.1
63.2
68.8
Medication
96.3
93.3
85.2
66.9
72.2
Finances
95.8
92.2
83.2
63.5
69.1
Abbreviations: ADL, Activities of Daily Living; IADL, Instrumental
Activities of Daily Living; PAQUID, Personnes Agées QUID.
a
At time of death.
dysfunction along the continuum. However, some gaps persisted at the beginning of the continuum (between partial
shopping and partial finances) and at the early end (between
partial transferring and total toileting losses). So, either additional relevant items are necessary to sharpen the detection of
persons at these levels of dependency or these stages represent 2 threshold events on the continuum. In contrast, other
items were found to be redundant as reflecting similar levels
of dependency. Such item redundancy also underlines the
limits of simple scores that would overassess certain dependency levels. On the other hand, redundancy might allow
adaptive testing with items better suited to the individual
(19), provided good item performances.
Item characteristics
Whatever the distance from death, the latent dependency
captured the major part of the interindividual item variability,
with perhaps the exception of continence. This was induced
by the very good discrimination (small measurement error for
all items except continence), and it confirmed the hypothesis
that IADL and ADL limitations could be the manifestations
of a unique major process of dependency. The more discriminating items are definitely basic activities: bathing, toileting,
dressing, and eating. They enable discrimination of proximate levels of dependency. The lower IADL discrimination
values may be explained by the more important impact of
contextual factors (social/physical environment). Improvement of item discrimination in IADL would be useful to improve assessment and detect at-risk individuals in the earlier
stages of dependency for closer clinical follow-up and/or improved prevention.
Among the 11 items explored here, the behavior of the
transporting item was peculiar. Distance on the continuum
Am J Epidemiol. 2016;183(4):277–285
between its partial and total limitation thresholds was much
higher (1.85) than distance for the other activities (ranging
from 0.37 units (dressing) to 1.19 (transferring)), yet its discrimination was very high, showing that the item can correctly discriminate between individuals at the 2 levels. A
possible explanation is that this item combines very different
domains, both cognitive and physical (executive functions,
hearing, vision, etc.). Indeed, researchers in several studies
have reported on IADL categorized into physical and cognitive
domains (20, 21). The physical dimension might be involved
at the beginning of the process (e.g., fear of falling or not being
able to get up or get on/off the bus), whereas the cognitive dimension might appear at the end (e.g., knowledge and control
of one’s car and the understanding required to drive, which involves multiple abilities such as attention, memory, and visuospatial abilities). This item could constitute a candidate for
splitting into different clearcut dimensions.
Functional trajectories
The estimated functional trajectories highlighted the benefit of higher education, but the persistent postponement of
dependency afforded by education in men did not persist
in women. Even if effects of education on health fade with
age (22), education has been related to various medical conditions and behaviors (management of chronic conditions,
appraisal of health status, or access to health services and
prevention) (19) and covers several concepts, such as professional occupation (23), brain and cognitive reserve capacity,
and lifelong stimulating activities (24).
Our findings also confirmed that functional trajectories
were sex-specific; men retained a better functional status than
women whatever their age at death and educational level.
Indeed, with a higher prevalence of nonfatal disabling diseases (fractures, osteoporosis, back problems, arthritis, and
depression) (25), women reported more limitations, a greater
need for help, a higher dependency level, and a sharper
decline (26).
Some key points on the trajectory curves must be stressed.
Although data were available up to 24 years before death, we
did not emphasize the dependency trajectory beyond 18 years
before death, since only 5% of the sample had a follow-up
time longer than 17.59 years before death. Moreover, if we
consider the first partial losses in using transportation and
shopping as the entry point into the dependency process,
the same level of dependency is delayed by education for
nearly 4 years for women and 2 years for men.
Strengths and weaknesses
The innovative feature of our study was that it relied on a
longitudinal analysis of 24 years’ follow-up of a substantial
sample from a general elderly population. Moreover, an analysis of events preceding death allows investigation of a homogeneous population. Item response theory is then particularly
relevant for statistical inference about dependency, since it formalizes the multiple measurements of this latent construct and
makes use of all of the data without a priori assumptions, other
than the existence of a single underlying latent construct (27).
Indeed, item difficulties rely not on a clinical judgment but on
284 Edjolo et al.
the relationship with the dependency level of an individual,
which can be influenced by other factors (age, sex, education,
health, depression, etc.). Simultaneously, the linear mixed
model precisely models the change of this latent trait over
time by taking into account the correlation between the repeated and multiple measurements in each individual.
Several limits of this study could be exposed. Firstly, the
item response categories (no limitation, partial limitation,
and total limitation) could have been detrimental to finegrained data on IADL limitations (ADL being originally
scored in 3 grades). In particular, concerning items such as
telephoning, transporting, and shopping, the 4 or 5 original
response levels of the Lawton scale (5) were gathered into
3 levels to improve the comparability between items. Secondly, we have not yet investigated underlying intra-individual
(comorbidity, psychological aspects, etc.) and extra-individual
(social/material environment, such as marital status) factors
that would possibly explain some heterogeneity in dependency patterns (28). Thirdly, because of the dynamic nature
of dependency, more frequent assessment would have better
captured subtle changes in the process. Finally, it could be
possible that an item behavior varies between subgroups of
individuals, a source of bias liable to compromise accuracy.
Further analysis of such differential item functioning (29)
among subgroups (women/men, old/oldest old, demented/
nondemented, etc.) would be of great interest.
Understanding the natural history of dependency provides
an early warning of functional decline or a signal to continue
in-depth functional assessments or closer clinical follow-up.
It also allows the selection of candidates for rehabilitation or
the screening of individuals prone to a severe decline or incipient dementia. The study of functional trajectories is an essential prognostic tool for clinicians to anticipate the needs
related to functional decline in the elderly, insofar it could
allow better targeted interventions. Further research is needed
to explore the great complexity of dependency in the elderly,
in terms of the heterogeneity of functional trajectories and
their determinants.
ACKNOWLEDGMENTS
Author affiliations: Institut National de la Santé et de
la Recherche Médicale (INSERM), INSERM U1219 Bordeaux Population Health Research Center, Bordeaux, France
(Arlette Edjolo, Cécile Proust-Lima, Fleur Delva, JeanFrançois Dartigues, Karine Pérès); and Université de Bordeaux, Institut de Santé Publique et de Développement
Bordeaux (ISPED), Bordeaux, France (Arlette Edjolo, Cécile
Proust-Lima, Fleur Delva, Jean-François Dartigues, Karine
Pérès).
This work was supported by the Institut de Recherche en
Santé Publique.
The PAQUID Study was supported by the AGRICA
Group (Paris, France); the Association pour la Recherche
Médicale en Aquitaine (Bordeaux, France); Caisse Nationale
d’Assurance Maladie des Travailleurs Salariés (Paris); Caisse
Nationale de Solidarité pour l’Autonomie (Paris); the Conseil
Général de la Dordogne (Périgueux, France); the Conseil
Général de la Gironde (Bordeaux); the Conseil Régional
d’Aquitaine (Bordeaux); the Fondation de France (Paris);
France Alzheimer (Paris); GIS Longévité (Paris); INSERM
(Paris); IPSEN France (Boulogne-Billancourt, France); the
Mutuelle Générale de l’Education Nationale (Paris); the
Mutualité Sociale Agricole (Bagnolet, France); Novartis (RueilMalmaison, France); and SCOR Insurance (Paris).
All of the authors affirm that this article is an honest, accurate, and transparent account of the study being reported; that
no important aspects of the study have been omitted; and that
any discrepancies from the study as planned have been
explained.
F.D. reports receiving personal fees from Novartis outside
the parameters of the present work. J.-F.D. reports receiving
grants from the Caisse Nationale de Solidarité pour l’Autonomie during the conduct of the study and grants and personal
fees from IPSEN France and Novartis outside the parameters
of the present work.
REFERENCES
1. Baltes MM. The Many Faces of Dependency in Old Age.
New York, NY: Cambridge University Press; 1996.
2. Wilkin D. Conceptual problems in dependency research.
Soc Sci Med. 1987;24(10):867–873.
3. de Kleijn-de Vrankrijker MW. The long way from the
International Classification of Impairments, Disabilities and
Handicaps (ICIDH) to the International Classification of
Functioning, Disability and Health (ICF). Disabil Rehabil.
2003;25(11-12):561–564.
4. Katz S, Ford AB, Moskowitz RW, et al. Studies of illness in the
aged. The Index of ADL: a standardized measure of biological
and psychosocial function. JAMA. 1963;185(12):914–919.
5. Lawton MP, Brody EM. Assessment of older people:
self-maintaining and instrumental activities of daily living.
Gerontologist. 1969;9(3):179–186.
6. Lazaridis EN, Rudberg MA, Furner SE, et al. Do activities of
daily living have a hierarchical structure? An analysis using the
Longitudinal Study of Aging. J Gerontol. 1994;49(2):M47–M51.
7. Barberger-Gateau P, Rainville C, Letenneur L, et al. A
hierarchical model of domains of disablement in the elderly: a
longitudinal approach. Disabil Rehabil. 2000;22(7):308–317.
8. Sims T, Holmes TH, Bravata DM, et al. Simple counts of ADL
dependencies do not adequately reflect older adults’ preferences
toward states of functional impairment. J Clin Epidemiol. 2008;
61(12):1261–1270.
9. Thomas ML. The value of item response theory in clinical
assessment: a review. Assessment. 2011;18(3):291–307.
10. Dartigues JF, Gagnon M, Barberger-Gateau P, et al. The Paquid
epidemiological program on brain ageing. Neuroepidemiology.
1992;11(suppl 1):14–18.
11. Spector WD, Fleishman JA. Combining activities of daily
living with instrumental activities of daily living to measure
functional disability. J Gerontol B Psychol Sci Soc Sci. 1998;
53(1):S46–S57.
12. Kempen GI, Suurmeijer TP. The development of a hierarchical
polychotomous ADL-IADL scale for noninstitutionalized
elders. Gerontologist. 1990;30(4):497–502.
13. LaPlante MP. The classic measure of disability in activities of
daily living is biased by age but an expanded IADL/ADL
measure is not. J Gerontol B Psychol Sci Soc Sci. 2010;65(6):
720–732.
Am J Epidemiol. 2016;183(4):277–285
Dynamic Item Response Theory Model of IADL/ADL 285
14. Baker FB. The Basics of Item Response Theory. 1st ed. College
Park, MD: ERIC Clearinghouse on Assessment and Evaluation;
2001. (ERIC Document Reproduction Service no. ED 458 219).
15. Delva F, Edjolo A, Pérès K, et al. Hierarchical structure of the
Activities of Daily Living scale in dementia. J Nutr Health
Aging. 2014;18(7):698–704.
16. Morris JN, Berg K, Fries BE, et al. Scaling functional status
within the interRAI suite of assessment instruments. BMC
Geriatr. 2013;13:128.
17. Gerrard P. The hierarchy of the activities of daily living in the
Katz index in residents of skilled nursing facilities. J Geriatr
Phys Ther. 2013;36(2):87–91.
18. Doble SE, Fisher AG. The dimensionality and validity of the
Older Americans Resources and Services (OARS) Activities of
Daily Living (ADL) Scale. J Outcome Meas. 1998;2(1):4–24.
19. Meeks S, Murrell SA. Contribution of education to health and
life satisfaction in older adults mediated by negative affect.
J Aging Health. 2001;13(1):92–119.
20. Thomas VS, Rockwood K, McDowell I. Multidimensionality
in instrumental and basic activities of daily living. J Clin
Epidemiol. 1998;51(4):315–321.
21. Ng TP, Niti M, Chiam PC, et al. Physical and cognitive domains
of the Instrumental Activities of Daily Living: validation in a
multiethnic population of Asian older adults. J Gerontol A Biol
Sci Med Sci. 2006;61(7):726–735.
Am J Epidemiol. 2016;183(4):277–285
22. Robert S, House JS. SES differentials in health by age and
alternative indicators of SES. J Aging Health. 1996;8(3):
359–388.
23. Cambois E, Barnay T, Robine JM. Espérances de vie,
espérances de vie en santé et âge de départ à la retraite: des
inégalités selon la profession en France. Retraite et Société.
2010;59(3):194–205.
24. Foubert-Samier A, Catheline G, Amieva H, et al. Education,
occupation, leisure activities, and brain reserve: a populationbased study. Neurobiol Aging. 2012;33(2):423.e415–423.e425.
25. Murtagh KN, Hubert HB. Gender differences in physical
disability among an elderly cohort. Am J Public Health. 2004;
94(8):1406–1411.
26. van Houwelingen AH, Cameron ID, Gussekloo J, et al.
Disability transitions in the oldest old in the general population.
The Leiden 85-plus study. Age (Dordr). 2014;36(1):483–493.
27. Bertand D, El Ahmadi A, Heuchenne C. D’une échelle ordinale
de Guttman à une échelle de rapports de Rasch. Math Soc Sci.
2008;184(4):25–46.
28. Gill TM, Gahbauer EA, Han L, et al. Trajectories of disability in
the last year of life. N Engl J Med. 2010;362(13):1173–1180.
29. Fleishman JA, Spector WD, Altman BM. Impact of differential
item functioning on age and gender differences in functional
disability. J Gerontol B Psychol Sci Soc Sci. 2002;57(5):
S275–S284.