Journal of Gerontology: PSYCHOLOGICAL SCIENCES Copyright 2001 by The Gerontological Society of America

2001, Vol. 56B, No. 5, P301–P313

A Longitudinal Study of the Relationship Between

Levels of Depression Among Persons With
Alzheimer’s Disease and Levels of Depression Among
Their Family Caregivers
Marcia M. Neundorfer, McKee J. McClendon, Kathleen A. Smyth, Jon C. Stuckey,
Milton E. Strauss, and Marian B. Patterson

University Alzheimer Center, University Hospitals of Cleveland/Case Western Reserve University, Cleveland, Ohio.

The purpose of this study was to examine the effects, over time, of depressive symptoms in persons with Alz-
heimer’s disease on depression in their family caregivers. In a sample of 353 patients and caregivers, multilevel
longitudinal analysis was used to accommodate an observational design in which the number of observation
points and the intervals between points varied across caregivers. The rate of change (increase) in caregiver de-
pression was predicted by the rate of change (increase) in patient depressive symptoms and by increase in patient
dependency in instrumental activities of daily living (ADLs). Acceleration of the increase in caregiver depression
was predicted by acceleration in patient dependency in instrumental and basic ADLs but not by acceleration in
patient depressive symptoms. These findings indicate the importance of measuring the rate and acceleration of
change in patient characteristics in order to understand caregiver depression. They also support early interven-
tions for caregivers.

I T is well documented that family caregivers of persons

with dementia have significantly more depressive symp-
toms than age- and gender-matched noncaregivers (Gal-
Between 40% and 50% of persons with Alzheimer’s dis-
ease have depressive symptoms, whereas the proportion
with actual depressive disorders ranges between 10% and
lagher, Rose, Rivera, Lovett, & Thompson, 1989; Schulz, 20% (Alexopoulos & Abrams, 1991; Katz, 1998; Wragg &
Visintainer, & Williamson, 1990). Furthermore, caregivers Jeste, 1989). However, few investigators have examined the
of persons with dementia report more emotional strain and effect of depressive symptoms in persons with dementia on
depressive symptoms than caregivers of persons who are depression in their caregivers. Drinka, Smith, and Drinka
not demented (Hooker, Monahan, Bowman, Frazier, & Shi- (1987) reported that severity of patient depression was posi-
fren, 1998; Ory, Hoffman, Yee, Tennstedt, & Schulz, 1999). tively associated with severity of caregiver depression, but
Prevalence rates for depressive symptoms among caregivers severity of patient dementia and dependency in ADLs were
of persons with dementia are reported to range from 28% to not. Their sample (N  127), however, was from a geriatric
55% (Schulz, O’Brien, Bookwala, & Fleissner, 1995). The clinic where the patients were of advanced age (62% were
characteristic of the person with dementia most consistently older than 80), had multiple medical problems, were highly
found to be associated with caregiver depression is the ex- depressed (69% met the criteria for major depression of the
tent to which the person exhibits problem behaviors (Schulz third edition of the Diagnostic and Statistical Manual
et al., 1995). Problem behaviors include a wide range of of Mental Disorders; American Psychiatric Association,
symptoms, including mood disorders (such as depression 1980), and were not all demented (73% met criteria for de-
and anxiety), activity disturbances (such as wandering and mentia). Also, their caregivers had an atypically high preva-
restlessness), disruptive and demanding behaviors (such as lence (83%) of major depression, all of which limits the
physical and verbal aggression), and psychotic symptoms generalizability of their findings.
(such as paranoid delusions). The purpose of this study was Hadjistavropoulos, Taylor, Tuokko, and Beattie (1994),
to examine one type of problem behavior in persons with in a sample (N  136) from an outpatient diagnostic clinic
dementia, depressive symptoms, and to determine the effect where 77% of patients met criteria for dementia, found that
over time, using multilevel analysis, of change in patient de- patient self-report of depressive symptoms was weakly as-
pressive symptoms on change in caregiver depression. To sociated with caregiver burden, but caregiver report of pa-
account for general decline in patient functioning and its tient dysphoria was strongly associated with caregiver bur-
contribution to caregiver depression, we added as control den. They did not measure caregiver depression. Caregiver
variables patient dependency in basic and instrumental ac- burden, however, has been shown to be positively associ-
tivities of daily living (BADLs and IADLs, respectively). ated with caregiver depression (Pruchno, Kleban, Michaels,
Additional control variables included duration of illness, & Dempsey, 1990; Stuckey, Neundorfer, & Smyth, 1996).
caregiver relationship to the patient (spouse vs nonspouse), The weaker correlation may be due to patient underreport-
and caregiver gender and self-reported health. ing of depressive symptoms; patient underreporting when

P301
P302 NEUNDORFER ET AL.

compared with reports of caregivers and clinicians is well longitudinal studies that tracked patient depression and de-
documented (Burke et al., 1998; Mackenzie, Robiner, & mentia beyond 1 year showed little progressive worsening
Knopman, 1989; Teri & Wagner, 1991). The only other patient of patient depression (Devanand et al., 1997; Marin et al.,
deficit that was significantly correlated with caregiver bur- 1997).
den was caregiver perception of greater dependency in IADLs. From longitudinal studies of depression in caregivers
Brodaty and Luscombe (1998) examined the association of community-residing persons with dementia (Alspaugh,
between caregiver psychological morbidity and patient de- Stephen, Townsend, Zarit, & Greene, 1999; Aneshensel,
pression in 193 patients diagnosed with dementia at a mem- Pearlin, Mullan, Zarit, & Whitlatch, 1995; Goode, Haley,
ory disorder clinic and their caregivers. Caregiver psycho- Roth, & Ford, 1998; Pruchno et al., 1990; Schulz & Wil-
logical morbidity (including depressive symptoms) was liamson, 1991; Vitaliano, Russo, Young, Teri, & Maiuro,
significantly associated with patient depression, both clini- 1991), the general picture is one of caregivers as a group
cian rated and patient self-rated. Caregiver psychological showing stability in depression. However, subsets of care-
morbidity was also significantly higher for women, spouses, givers either decrease or increase in depression. For exam-
and caregivers living with the patient, as well as for care- ple, Schulz and colleagues reported that, over 2 years, the
givers of patients with demanding problem behaviors, majority of caregivers (59%) had little change in depres-
higher severity of impairment in IADLs, and greater cogni- sion, with three quarters of these stable individuals consis-
tive impairment. A logistic regression analysis identified tently asymptomatic (below the threshold for risk for clini-
patient depression (clinician rated, but not patient self-rated) cal depression), whereas one fourth (13.7% of the whole
and demanding problem behaviors as being independently sample) were consistently symptomatic. The remaining care-
and significantly associated with caregiver psychological givers (41%) exhibited fluctuating patterns of depression,
morbidity. In contrast, results from three studies (Cum- with an increase in depression over time. Cross-sectional
mings, Ross, Absher, Gornbein, & Hadjiaghai, 1995; Moye, multivariate analyses at three time points revealed signifi-
Robiner, & Mackenzie, 1993; Ott & Fogel, 1992) showed cant positive relationships between caregiver depression
no correlations between patient depression and caregiver and patient behavior problems, but not dependency in either
depression. All samples, however, were small (Ns  33, 31, BADLs or IADLs.
and 50, respectively). Aneshensel and associates (1995) reported the most de-
Conclusions from these studies of the relationship be- tailed analysis of the impact of both patient and caregiver
tween patient depressive symptoms and caregiver depres- characteristics, and the change in these characteristics, on
sion are limited because of several factors. Samples from caregiver depression. Examining the caregiving career over
some studies included patients not diagnosed with dementia 3 years duration, they concluded, “It is evident that the pa-
(Drinka et al., 1987; Hadjistavropoulos et al., 1994), and tient’s impairment, its behavioral manifestations, and the re-
some sample sizes might have been too small to detect ef- sulting dependences are a core catalyst in the stress-prolifer-
fects (Cummings et al., 1995; Moye et al., 1993; Ott & Fo- ation process, especially the rate of decline exhibited as
gel, 1992). Also, the measures of patient depression and time progresses” (p. 148). Although they did not examine
caregiver depression, and their sensitivities, differed among patient depressive symptoms per se, they emphasized that it
studies. Only three studies (Brodaty & Luscombe, 1998; is not so much the level of patient impairment but the rate of
Drinka et al., 1987; Hadjistavropoulos et al., 1994) con- decline that is important to caregivers’ feelings of depres-
trolled for other patient characteristics, such as dementia se- sion. This supports the emphasis in our study on analysis of
verity and dependency in ADLs, and only Brodaty and Lus- rate of change in patient characteristics for their impact on
combe controlled for caregiver demographic characteristics, rate of change in caregiver depression.
such as gender, relationship to the patient, and coresidence To examine change in patient characteristics in order to
with the patient. Finally, all studies were cross sectional. understand change in caregiver depression, we used multi-
Thus, this study sought to examine the relationship between level analysis for statistical analysis in this study. This tech-
patient depression and caregiver depression among well- nique is well suited to examination of the extent to which
characterized dementia patients, while controlling for de- patients change in their illness characteristics, the extent to
pendency in ADLs and duration of illness, as well as rela- which caregivers change in depression over time, and the
tionship (spouse vs nonspouse), caregiver gender, and self- extent to which these two processes vary. It can account for
rated health. Caregiver self-rated health was added because different starting points, different duration, different rates of
of its strong reciprocal relationship with depressive symp- change, and multiple correlates of status and change, which
toms (Aneshensel, Frerichs, & Huba, 1984). Further, in this themselves might be changing. The hypothesis for the
study, the relationship between patient depression and care- present study was that as patient depressive symptoms in-
giver depression was examined over time to answer (a) to crease, caregiver depression increases, regardless of dura-
what extent does patient depression change? (b) to what extent tion of illness and patient dependency in ADLs, caregiver
does caregiver depression change? and (c) to what extent do gender, relationship, and self-reported health.
patient depression and caregiver depression vary together?
Most, but not all, cross-sectional studies of depressive METHODS
symptoms in dementia patients have indicated a decrease in
the frequency of depression in patients with more advanced Sample
dementia, but findings have depended on assessment crite- Participants in this study were 353 persons with dementia
ria (Katz, 1998; Payne et al., 1998). Findings from the few who were enrolled in the University Hospitals of Cleveland/
 DEPRESSION IN AD P303

Case Western Reserve University Alzheimer’s Disease Re- depressive symptoms were measured. The time interval
search Center (ADRC) and their family caregivers. Care- from the first to the last observation on patients ranged from
giver and patient depression measures were collected from 4 months to 5 years, with a mean time interval of 2 years.
October 1992 through March 1999. All patients were living All data points on patients preceded or corresponded to the
in the community throughout the study. Using criteria of the last data point on their caregivers. We used multilevel anal-
National Institute of Neurological and Communicative Dis- ysis to accommodate this flexible observational design.
orders and the Stroke/Alzheimer’s Disease and Related Dis-
orders Association (McKhann et al., 1984), 63% of patients Measures
were diagnosed with probable Alzheimer’s disease (AD), Descriptive statistics and alphas at baseline, and the mean
31% were diagnosed with possible AD, and 6% were diag- number of data points for each measure, are displayed in
nosed with other dementias. Among those diagnosed with Table 1. Measures on the patients included duration of ill-
possible AD, 37% were diagnosed with possible AD with de- ness, BADLs and IADLs, and depressive symptoms. No
pression. Duration of illness ranged from 1 year to 11 years, separate measures of cognitive functioning were used be-
with a mean of 4 years. For dementia severity at the begin- cause scores on the Mini-Mental State Exam (MMSE; Fol-
ning of data collection, as measured with the Clinical De- stein, Folstein, & McHugh, 1975) were missing for more se-
mentia Rating scale (Hughes, Berg, Danziger, Coben, & verely impaired patients and correlations between MMSE
Martin, 1982), 4% were rated 0.5 (uncertain dementia, all of and total scores on ADLs were fairly high (r  .63).
whom were rated as 1 at the next annual evaluation), 61% Duration of illness was based on the date of the earliest
were rated 1 (mild dementia), 25% were rated 2 (moderate possible symptoms, as identified by the caregiver in the ini-
dementia), 10% were rated 3 (severe dementia), and fewer tial interview with the physician. This date was subtracted
than 1% were rated 4 (profound dementia). from the date of the baseline medical evaluation to compute
The patients were mostly female (53%), with a mean age duration of illness at baseline, which was used as the one
of 73 years (range  50 to 95 years) at the beginning of data measure of duration of illness in all analyses. Mean duration
collection. Eighty-seven percent of the patients lived with of illness at baseline was 4 years, with a range from 1 year
their caregivers at the beginning of data collection, and 88% to 11 years.
were living with their caregivers at the last data collection ADLs were measured with the Cleveland Scale for Activ-
point. African Americans made up 10% of the sample. Care- ities of Daily Living (Patterson et al., 1992), which is ad-
givers were mostly female (67%), with a mean age of 64 ministered by a trained examiner to the caregiver. This scale
years (range  31 to 86 years). The majority were spouses was designed to measure both BADLs and IADLs in per-
(43% wives and 28% husbands), and 18% were daughters, sons with dementia, across the range of dementia severity.
5% were sons, and the rest were other relatives. The particular strength of this scale is that it includes higher
The design was to collect data on patients at their entry level tasks, such as initiating, planning, and organizing
and subsequent annual ADRC visits. At these visits, care- (e.g., “initiates dressing at appropriate times,” “selects
givers were given self-administered questionnaires to com- clothing,” “prepares bath”), on which persons with early de-
plete and mail back. The response rate for caregivers was mentia may have deficits. It also includes lower level tasks
consistently 90%; all questionnaires were returned within that are more automatic (e.g, “cleans self,” “feeds self”), on
90 days of the patient’s visit; and, for each patient, the care- which persons with more severe dementia have deficits.
giver was the same person throughout this study. Although Items are rated on a scale ranging from 0 (completely inde-
the basic design called for the assessment of patient and
caregiver characteristics on an annual basis for as long as
the patient remained in the community, a subset of care- Table 1. Descriptive Statistics at Baseline and Number of Data
givers (46%) were enrolled in a separate study of caregiver Points (N  353)
depression and were quarterly asked to complete the ques-
tionnaires and answer questions about patients’ ADLs and Number of
Baseline Data Points
behavioral symptoms. This supplementary study resulted in
up to four additional measurements per caregiver and pa- M (SD) Range  M (SD)
tient on these variables. Patient
Our data collection design resulted in a variable number Duration of illness (years) 4.0 (2.6) 1–11 1.0
of data points and variable time intervals between data col- Basic ADL 8.0 (12.5) 0–57 .95 4.5 (2.8)
lections. For 40% of all caregivers, there was 1 data point at Instrumental ADL 17.2 (11.5) 0–45 .89 4.4 (2.7)
which the outcome variable, caregiver depression, was mea- Depressive symptoms 4.1 (4.8) 0–26 .75 4.1 (2.6)
sured; for the rest of the caregivers, there were between 2 Caregiver
Gendera 0.7 (0.5) 1.0
and 10 data points (mean of 3.2 data points) at which care- Spouse/nonspouseb 0.7 (0.5) 1.0
giver depression was measured. The time interval from the Self-rated healthc 2.6 (1.0) 1–5 1.0
first to the last observation on caregivers ranged from 4 Depression (CES–D) 12.6 (9.6) 0–52 .89 3.2 (2.5)
months to 5 years, with a mean time interval of 1 year. For
Note: ADL  activity of daily living; CES–D  Center for Epidemiologic
20% of all patients, there was 1 data point at which the main Studies–Depression scale.
independent variable, patient depressive symptoms, was aMale  0; female  1.

measured; for the rest of the patients, there were between 2 bNonspouse  0; spouse  1.

and 10 data points (mean of 4.1 data points) at which patient c1  Excellent; 5  poor.
P304 NEUNDORFER ET AL.

pendent) to 3 (completely dependent). The possible range sample scored at baseline in the range at risk for clinical de-
for the 19 BADL items was 0 to 57; the mean was 8.0, with pression.
scores across the full range. The possible range for the 16 Mean caregiver depression at each annual visit is given in
IADL items was 0 to 48; the mean was 17.2, with a range Table 2. First, means for each pair of consecutive annual
from 0 to 45. visits (0–1, 1–2, 2–3, 3–4, and 4–5) are shown for care-
Depressive symptoms in the patient were measured by the givers with valid CES–D scores on each pair of years. These
Consortium to Establish a Registry for Alzheimer’s Disease means show the change in depression over a 1-year interval
Behavior Rating Scale for Dementia (BRSD; Mack & for the caregivers who were observed over that interval.
Patterson, 1996; Tariot et al., 1995). The BRSD is a struc- Second, means at each annual visit are given for all care-
tured interview completed by a trained interviewer with the givers who were observed in a given year. Neither set of
caregiver. The subscale for depressive symptoms includes means shows any clear-cut pattern of aggregate change in
seven items (feelings of anxiety, sad appearance, hopeless- depression. There is considerable variation between care-
ness, crying, guilt feelings, poor self-esteem, and feelings givers in rates of change in depression that is hidden by
that life is not worth living), plus a follow-up item asking if these aggregate means, as is seen.
the person ever made a suicide attempt. It does not include
symptoms of altered biological rhythms (e.g., changes in Statistical Analysis
appetite, energy, and sleep) or apathy (e.g., loss of interest The research objective was to investigate the effects of pa-
and reactivity), which are included in other depression tient depressive symptoms on the repeated measures of
scales (e.g., the Cornell Scale for Depression in Dementia; caregiver depression, controlling for duration of illness, de-
Harwood, Ownby, Barker, & Duara, 1998). Thus, the pendency in IADLs and BADLs, relationship (spouse–non-
BRSD Depressive subscale (BRSDDEP) is mood related spouse), caregiver gender, and caregiver self-reported
and has less overlap with symptoms associated with demen- health. Hierarchical modeling or multilevel analysis (Bryk
tia. Validity of the BRSDDEP has been supported by com- & Raudenbush, 1992; Goldstein, 1995) was selected to ana-
paring responses on BRSDDEP items between Alzheimer’s lyze the changes in depression. Multilevel analysis was cho-
disease patients who were diagnosed as either depressed or sen instead of repeated measures analysis of variance be-
nondepressed by use of semistructured clinical interviews cause the former specifies more directly the parameters of
(Jacobs, Strauss, Patterson, & Mack, 1998). On the basis of growth and change and the determinants of these parame-
the BRSDDEP score, 70% of the patients were correctly ters (Bryk & Raudenbush, 1992, p. 133). Multilevel analy-
classified as depressed or nondepressed. Caregivers rated sis also allows the use of repeated measures where both the
BRSDDEP items for frequency of occurrence in the past number of observation points and the intervals between the
month from 0 (has not occurred) to 4 (occurred 16 or more points vary across the units of analysis (i.e., caregivers).
days), plus a 0 or 1 for suicide attempt, yielding a possible This feature makes it possible to use all available data in-
range from 0 to 29. The mean for BRSDDEP was 4.1, with stead of restricting the analysis to cases that have the same
a range from 0 to 26. number of observations. It is even possible to include cases
Measures for the caregiver included gender, relationship with only one observation, as we have done, although these
(spouse vs nonspouse), self-rated health, and the main out- cases contribute very little to the longitudinal part of the
come variable, caregiver depression. Caregivers’ age, race, analysis (Snijders & Bosker, 1999, p. 181). Equations 1–4
income, and employment status were not related to care- illustrate a multilevel model for repeated measures.
giver depression at baseline, so they were not included in
subsequent analyses. There was a small but significant cor-
relation between caregiver education and caregiver depres- Table 2. Mean CES–D Scores (and n) at Each of Two Consecutive
sion at baseline (r  .17, p  .01), but caregiver educa- Annual Visits
tion was not included in the analyses in order to reduce the Year
number of independent variables. Coresidence of the care-
giver and patient was highly correlated with being a spouse 0 1 2 3 4 5
(r  .60), so spouse–nonspouse was used in the analysis. 11.99 → 11.91
Self-rated health was rated by a single item asking care- (149)
givers to rate their own health, from 1 (excellent) to 5 12.42 → 12.55
(69)
(poor). The mean was 2.6, with 45% rating their health as 11.57 → 11.80
excellent or good and 20% rating it as fair or poor. (35)
Depression in the caregiver was measured by the Center 14.70 → 11.20
for Epidemiologic Studies—Depression scale (CES–D; Rad- (20)
loff, 1977), one of the most frequently used measures of de- 11.89 → 15.33
pression among caregivers (Schulz et al., 1990). Caregivers (9)
were asked to rate the frequency with which they experi- Means for Total Yearly ns
enced each of 20 depressive symptoms during the past week 12.60 11.90 13.58 12.96 10.78 15.33
from 0 (less than one day) to 3 (5 –7 days), with a possible (351) (154) (91) (46) (23) (9)
range from 0 to 60. Scores of 16 or above suggest that an in- Notes: Quarterly measures are excluded from this table, and thus the ns do not
dividual is at risk for clinical depression. The mean was reflect the total number of observations used in the multilevel models. CES–D 
12.6, with a range from 0 to 52. Thirty-one percent of the Center for Epidemiologic Studies–Depression scale.
 DEPRESSION IN AD P305

Y ij = β 0i + β 1i t ij + β 2i t ij2 + e ij (1)

β 0i = γ 00 + γ 01 X 1i + … + γ 07 X 7i + u 0i (2)

β 1i = γ 10 + γ 11 X 1i + … + γ 17 X 7i + u 1i (3)

β 2i = γ 20 + γ 21 X 1i + … + γ 27 X 7i + u 2i (4)

Equation 1 is a within-person equation for Yij, caregiver de-

pression for person i measured at occasion j. In Equation 1,
Figure 1. Illustration of effects of 1i and 2i. A, 1i  0 and 2i  0;
Yij is a nonlinear function, named a quadratic function, of B, 1i  0 and 2i  0.
time tij. The term tij indicates the time in years from the first
occasion of measurement to each subsequent occasion,
where j  1 to mi and mi equals the number of occasions,
which varies from caregiver to caregiver. Thus, ti1  0 for t  0. If we wanted to evaluate 0i and 1i at a time later than
the first occasion and, for example, ti2  1.1 if the second t  0, we could subtract some reference value t0 from t and
measure occurred 1.1 years later. The term eij is the residual include tij  t0 in Equation 1. With t “centered” on t0, 0i
of Y for each caregiver at each point in time. The coeffi- and 1i would indicate the level of depression and the rate of
cients 0i, 1i, and 2i are the Y intercept, the linear slope or change in depression at tij  t0. One value that is often used
rate of change, and the acceleration of the rate of change, re- to center variables is the grand mean, in this case, t··. Not all
spectively, for each caregiver i. Thus, there are 353 different caregivers, however, have been observed long enough for
estimates for each coefficient. In other words, Equation 1 is them to have a value of tij  t··; some, for example, are mea-
actually 353 different equations, one for each caregiver. sured only at t  0. Therefore, we have chosen not to center
Because the Y intercept is the prediction of Y when t  0 t at the grand mean. Another value that is sometimes used to
(first measurement), 0i equals a caregiver’s predicted de- center variables in multilevel analysis is the within-unit
pression at the beginning of his or her participation in the mean, in this case ti·, the mean time for caregiver i. The con-
project. The variance of 0i across the 353 caregivers is the sequence of centering on ti· is that we would be evaluating
between-person variance of Yi1. The variance of the mi mea- 0i and 1i at different times for different caregivers, and the
sures of Yij for each caregiver around each caregiver’s mean fit of the model when t is centered on ti· is different than the
Yi· equals the within-person variance. fit when it is centered on t·· or not centered at all (Snijders &
1i is the rate of change in caregiver depression at t  0. Bosker, 1999). Snijders and Bosker (p. 81) stated that
To interpret the rate of change at a specific point in time within-unit mean centering should be avoided unless there
such as t  0, imagine that t increases by an infinitesimal is a clear theory that the relative score (tij  ti·) and not the
amount from that point. The product of 1i and this infini- absolute score (tij) is related to Y. Because we do not believe
tesimal change in t equals the amount that Y will change. that relative time is theoretically justified, we used the un-
Thus, 1i is the instantaneous rate of change at t  0. If 1i centered tij.
is positive, the caregiver’s depression is instantaneously in- The three parameters of our quadratic model of change,
creasing at baseline; if 1i is negative, the caregiver’s de- 0i, 1i, and 2i, are the focal parameters used to describe
pression is decreasing. each caregiver’s depression. Because there are no time-
In a nonlinear function such as the quadratic function varying covariates included in Equation 1, these are unad-
(Equation 1), the rate of change itself changes as t changes. 2i justed estimates for each caregiver and serve as the depen-
indicates the acceleration or deceleration of the rate of change dent variables in Equations 2–4. The objective of this study
in caregiver depression. It determines the change in the rate of was to determine whether the patient’s depressive symp-
change. In Figure 1A, 1i  2 and 2i  1. Thus, the initial toms help explain the between-person variance of 0i, 1i,
rate of change is positive or increasing. The negative accelera- and 2i, that is, variance in initial caregiver depression (the
tion parameter, however, means that the rate of increase de- intercept) and variance in the rate of change and accelera-
creases or decelerates as t increases. For example, the curve is tion of change in caregiver depression. Equations 2–4
steeper between A and B than between B and C. At C, how- show these parameters expressed as a linear function of
ever, the rate of increase equals 0, and beyond C it becomes seven independent variables, X1i, . . . X7i. These variables
negative. The negative acceleration parameter results in an ac- are a dummy variable for caregiver gender (female  1), a
celeration of the rate of decrease as t increases beyond t  1. dummy variable indicating whether the caregiver is a
In Figure 1B, there is an initial negative rate of change. The spouse of the patient (spouse  1), a self-report of the care-
positive acceleration parameter decreases or decelerates the giver’s health at baseline, the duration of AD at the time of
rate of decrease up to t  1. Beyond t  1, Y has a positive the patient’s first assessment, one of the parameters that de-
rate of increase, which accelerates as t increases further. scribes change in the patient’s IADLs, one of the parameters
As indicated above, 0i and 1i represent the level of de- for change in the patient’s BADLs, and one of the parame-
pression and the rate of change in depression at baseline, ters for change in patient depressive symptoms (BRSD-
P306 NEUNDORFER ET AL.

DEP). The terms u0i, u1i, and u2i are the residuals of 0i, 1i, β 0i = γ 00 + γ 01 X 1i + … + γ 07 X 7i + (2a)
and 2i, respectively, for each caregiver. γ 08 t i· + γ 09 t i· X 7i + u 0i
The independent variables (X1i, . . . X7i) are not identical
in each equation. Caregiver gender, spouse–nonspouse,
caregiver health, and duration of illness were included in β 1i = γ 10 + γ 11 X 1i + … + γ 17 X 7i + (3a)
each of Equations 2–4. The measures of IADL, BADL, and γ 18 t i· + γ 19 t i· X 7i + u 1i
BRSDDEP were different in each equation, based on the
following steps. First, we estimated the three parameters of
β 2i = γ 20 + γ 21 X 1i + … + γ 27 X 7i + (4a)
Equation 1 for each of these three patient measures. Thus,
each patient had an estimate of the initial level, the rate of γ 28 t i· + γ 29 t i· X 7i + u 2i
change, and the acceleration of change of IADL, BADL,
and BRSDDEP. That is, we estimated the three parameters Equations 2a–4a include a term for the main effect of ti·
of the quadratic model of change for each patient on each of (e.g., ti·) and one for its interactive effect with one X (e.g.,
the three variables. An appropriate parameter was selected ti· X7i). The main effect indicates the effect of ti· when X is at
for each variable to be used in predicting the caregiver de- its mean, and the interactive effect represents the change in
pression parameters, as follows. We used the initial levels the effect of X as ti· varies. If the effect of either ti· or ti·X7i is
(0i) of IADL, BADL, and BRSDDEP to predict the initial significant, it means that the variance in missing values (ti·)
level of caregiver depression (Equation 2); we used the rate is not random. The main effect of X (e.g., 07X7i) now equals
of change (1i) in IADL, BADL, and BRSDDEP to predict its effect when ti· is at its centered value (1.63). The first
the rate of change in caregiver depression (Equation 3); and step in testing ti· is to enter it into each of Equations 2a–4a in
we used the acceleration (2i) of IADL, BADL, and BRSD- order to examine any main effects it may have. Because we
DEP to predict the acceleration of caregiver depression are focusing on the effects of patient depression, a term rep-
(Equation 4). resenting the interaction of BRSDDEP with ti· is then in-
Although we did not center the time variable for the rea- cluded in each equation. For example, we include the prod-
sons given above, X1i  X7i were each centered at their uct of ti· and the baseline value of BRSDDEP in Equation
mean in order to improve the interpretation of the coeffi- 2a, the product of ti· and the rate of change in BRSDDEP in
cients 00, 10, and 20. Because these gammas indicate the Equation 3a, and the product of ti· and the acceleration of
baseline value, rate of change, and acceleration of caregiver BRSDDEP in Equation 4a. We also do the same for IADL
depression when all of the Xs equal 0, if the Xs are centered, and BADL, the patient control variables that may change
the gammas equal these parameters of change when all Xs over time. We do this one at a time for IADL, BADL, and
are at their means. The means of the dummy variables gen- BRSDDEP because otherwise there would be 12 terms in-
der and spouse–nonspouse indicate the proportion of cases volving ti· in the model.
that are coded 1, thus, 00, 10, and 20 represent proportion- We used the MLwiN software (Prosser, Rasbash, &
ately weighted averages of these parameters for men and Goldstein, 1991) for multilevel analysis to estimate the pa-
women and for spouses and nonspouses, adjusted for the rameters of our repeated measures models. MLwiN uses an
other Xs. iterative generalized least squares algorithm that provides
Multilevel models for repeated measures assume that consistent estimates of model parameters and standard er-
missing values of Yij are “missing at random.” This means rors. An alternative algorithm is also available to obtain re-
that missing values may be related to previous measures of stricted or unbiased estimates of the model parameters in
caregiver depression (Yij) and to the covariates included in small samples. As both algorithms provided similar esti-
the model (X1i, . . . X7i), but otherwise must be randomly mates for our models, we report the more commonly used
missing (Hedeker & Gibbons, 1997). In this study, missing unrestricted estimates for presentation purposes. Bryk and
values are primarily due to differences between caregivers Raudenbush (1992, p. 50) recommended comparing the pa-
and patients in the duration of their participation in the rameter estimate divided by its standard error (ˆ / ) to the t
study. Variance in the number of observations and the total distribution rather than the normal distribution for inferen-
time of participation is certainly related to the covariates tial purposes.
(e.g., duration of illness) and to measures of the caregiver’s
depression. It is a rather strong assumption, however, that RESULTS
missingness is otherwise random. The bivariate correlations at baseline of the variables in-
Variables representing patterns of missing values may be cluded in the multilevel analysis are displayed in Table 3.
included in the models to control for nonrandom missing- Looking first at the patient variables, as would be expected,
ness (Hedeker & Gibbons, 1997; Little, 1995). Because of longer duration of illness is related to greater dependency in
our desire to avoid overburdening an already complex both BADLs and IADLs. Although there is relatively high
model, we elected to represent variance in missing values collinearity between dependency in instrumental and basic
with a single variable, the mean time of observation of each activities at baseline (r  .71), which increases their stan-
caregiver (ti). Mean t is highly correlated with the total time dard errors as predictors of caregiver depression, the corre-
of observation (r  .983) and the number of observations (r  lations between the rate of change of instrumental and basic
.832). We centered ti· at 1.63 years, a value halfway between dependency and the acceleration of change of instrumental
the minimum and maximum values of ti·. The method of in- and basic dependency are only .20 and .21, respectively (not
cluding mean t in the model is shown in Equations 2a–4a. shown). Patient depressive symptoms are negatively related
 DEPRESSION IN AD P307

to BADLs (.11), which indicates patients with greater .001, but not for BRSDDEP,
2 (3, N  353)  3.0, p 
BADL impairment are somewhat less depressed. A possible .392. Therefore, Model 3 is used to estimate 0i, 1i, and 2i
reason for this negative correlation is that basic BADL im- for IADL and BADL dependency, and Model 2 is used for
pairment tends to occur at a later stage in the progression of patient depression.
dementia, when patient depressive symptoms are less, as The gammas for the constant terms in Model 2 of Table 4
some studies have shown (Katz, 1998; Payne et al., 1998). indicate the mean at Time 0 (0i), the mean rate of change
This negative correlation, however, could be due to other (1i), and the mean acceleration (2i). Because ti· is included
confounding factors, such as decline in ability to express de- in Model 3 (uncentered), the gammas for the constant terms
pressive thoughts as the disease progresses. Among the pa- in Model 3 are the values of the parameters when ti·  0
tient variables, the only ones related to caregiver depression (i.e., patients with only a baseline measure). Each gamma
were higher dependency in IADLs and more patient depres- that is significant in Model 2 is also significantly different
sive symptoms. Among the caregiver variables, those re- from zero in the same direction in Model 3 for each of the
lated to caregiver depression were being female and in poor three patient variables. For the IADL and BADL variables,
health. the rate of change (6.77 and 6.61, respectively, in Model 3)
To conduct the multilevel analysis of caregiver depres- is significantly greater than zero, indicating a significant
sion, we first estimated the three parameters describing the rate of increase in patient dependency at Time 0, on the av-
change in each patient’s IADL, BADL, and BRSDDEP erage. The acceleration parameter is significantly less than
measures. Table 4 shows the results of three models for zero for IADL (0.44 in Model 3), meaning the rate of in-
each of these three variables. The first model computes the crease in IADL dependency slows or decelerates over time.
mean for each patient (the 00 for the constant term equals The acceleration parameter for BADL, on the other hand, is
the grand mean or mean of means, e.g., 21.31 for IADL). only marginally significant (.76 in Model 3, p  .068), sug-
(When time is not included as a variable in the model, the  gesting little or no change in the rate of increase in BADL
parameter equals the mean of the across-time means for over time. For patient depressive symptoms (BRSDDEP),
each patient rather than the mean of the Time 0 measure- neither the rate nor the acceleration of change (.45 and .27,
ments.) With these means, the within-patient variance ( 2e, respectively, in Model 3) is significant. To summarize, the
e.g., 51.75 for IADL) and the between-patient variance average patient displays a decelerating rate of increase in
( 2u0, e.g., 97.46 for IADL) may be determined. For IADL, dependency of IADLs, a positive rate of change in depen-
BADL, and BRSDDEP, both within- and between-patient dency of BADLs that possibly is accelerating, and no signif-
variances are significant. The between-patient variance, icant change in depressive symptoms (Figure 2).
however, is larger than the within-patient variance for each The mean time of observation (ti·) has a negative effect on
variable. More important, these variances indicate that there the rate of increase of dependency in both types of ADLs
are substantial changes over time within patients, as well as (0.89 and 1.33 for IADL and BADL, respectively, in
substantial differences between patients. Model 3), indicating that the initial rate of increase of ADL
The second and third models in Table 4 provide estimates dependency is lower for those who were observed longer.
of the three parameters of change for each of the three pa- These negative effects of ti· suggest that lower rates of in-
tient variables. Model 3 contains the mean time of observa- crease in ADL dependency may delay nursing home place-
tion for each patient (ti·) as a predictor of the initial level, the ment and thus increase the time of participation in the study.
rate of change, and the acceleration (0i, 1i, and 2i, respec- Mean time of observation, however, positively affects the
tively) of each patient variable in order to control for non- acceleration of IADL dependency (0.11) but not BADL de-
random missingness. The addition of ti· significantly im- pendency. Thus, the longer the patient was studied, the less
proves the fit of the models for IADL,
2 (3, N  353)  the deceleration of the rate of increase in IADL dependency.
12.8, p  .005, and BADL,
2 (3, N  353)  17.1, p  The positive effect of ti· on IADL acceleration, but not
BADL acceleration, may indicate that caregivers expect or
are able to tolerate accelerating IADL dependency (e.g.,
Table 3. Correlation Matrix at Baseline
needing help in travel, shopping, using the telephone).
Variable 1 2 3 4 5 6 7 8 Without accelerating dependency in bathing, dressing, and
1. Duration of illness —
toileting, the patient remains in the community and care-
2. Basic ADL .32** — givers thus continue in the study.
3. Instrumental ADL .33** .71** — More important than whether the mean rate and the mean
4. Depressive .04 .11* .02 — acceleration of patient variables are significantly different
symptoms than zero is whether these parameters vary significantly be-
5. Caregiver gendera .07 .10 .09 .07 — tween patients. The between-patient variance in Time 0
6. Spouse/non-spouseb .01 .09 .07 .05 .20 —
7. Self-rated healthc .02 .06 .11* .02 .13* .07 —
measurement, rate of change, and acceleration are indicated
8. Cargiver CES–D .02 .10 .22** .16** .14** .09 .29** — in Table 4 by 2u0, 2u1, and 2u2, respectively. Each of
these variances is significant for each variable in both Mod-
Note: ADL  activity of daily living; CES–D  Center for Epidemiologic els 2 and 3. Thus, there is significant between-patient vari-
Studies–Depression Scale.
a Male  0; female  1. ance in the rate of change and acceleration of both basic and
b Nonspouse  0; spouse  1. instrumental dependency and of depressive symptoms.
c 1  Excellent; 5  poor. The covariances between the parameters of change for
*p  .05; **p  .01. each of the three patient variables are indicated by u0u1,
P308 NEUNDORFER ET AL.

Table 4. Gammas for Models of Change in Instrumental and Basic Activities of Daily Living (IADLs and BADLs) and Change in Patient
Depressive Symptoms (BRSDDEP)
IADLs BADLs BRSDDEP

Equation/Variables 1 2 3 1 2 3 1 2 3
Time 0 (0i)
Constant 21.31* 17.10* 17.92* 12.98* 7.97* 9.15* 3.89* 4.01* 4.17*
t 0.99 1.26 0.20
σ u2 97.46* 115.70* 114.47* 181.42* 140.02* 138.55* 14.86* 17.59* 17.58*
0
Rate of Change (1i)
Constant 5.45* 6.77* 5.58* 6.61* 0.13 0.45
t 0.89* 1.33* 0.07
σ u2 17.96* 16.51* 82.02* 79.16* 5.71* 5.85*
1
Acceleration (2i)
Constant 0.33* 0.44* 0.22 0.76 0.02 0.27
t 0.11* 0.06 0.11
σ u2 0.26* 0.24* 3.55* 3.16* 0.20* 0.22*
2

σu 9.33* 8.70 34.80* 932.28* 3.49* 3.49*

0 u1

σu 0.34 0.42 10.20* 8.94* 0.31 0.31

0 u2

σu 2.06* 1.86* 13.74* 12.96* 1.01* 1.07*

1 u2

σ e2 51.75* 20.04* 20.03* 100.58* 22.41* 22.47* 7.04* 5.37* 5.34*

2LL 9,239.96 8,573.27 8,560.47 10,365.22 9,357.50 9,340.41 6,099.37 6,036.64 6,033.66

Note: BRSDDEP  Behavior Rating Scale for Depression Depressive Subscale.

*p  .05.

u0u2, and u1u2. The covariance across patients between the givers (65.299) than over time within caregivers (28.062),
rate of change and the acceleration of change ( u1u2) is sig- as was the case with the three patient variables. However,
nificantly less than zero for all three variables (1.86 for both sources of variance are significant. Model 2 shows that
IADL, 12.96 for BADL, and 1.07 for BRSDDEP in the average caregiver, unlike the average patient, has a rate
Model 3). These negative covariances indicate that the
greater the rate of increase in IADL, BADL, and BRSD-
DEP, the less the acceleration in these variables. Thus,
when the increase in each variable is rapid initially, it tends
to slow or even level off over time.
The values of u0, u1, and u2 (patient residuals as shown in
Equations 2–4) were generated by Models 2–3 in Table 4
for the measurement at Time 0, the rate of change, and the
acceleration of change, respectively. These residuals indi-
cate how much each patient deviates from the mean value of
each parameter of change (adjusted by ti· in Model 3). A
positive u1 for IADL dependency, for example, would indi-
cate that a patient had a higher than average rate of increase
in this kind of dependency. A negative u1 would indicate a
slower than average rate of increase. The mean of each re-
sidual variable equals 0 (u0  u1  u2  0). The residuals
for IADL and BADL from Model 3 and for BRSDDEP
from Model 2 were used as independent variables in the
model for changes in caregiver depression.
Table 5 contains the estimated parameters of four models
for caregiver depression, namely, the within-caregiver means
model (Model 1), the model for the means of the parame-
ters of change (Model 2), a model with the independent
variables as predictors of the parameters of change as speci-
fied by Equations 2–4 (Model 3), and a model with the inde-
Figure 2. Changes in patient depression and dependency at ti·  1.
pendent variables plus the mean time of observation as BADL  basic activities of daily living; IADL  instrumental activi-
specified by Equations 2a–4a (Model 4). Model 1 shows ties of daily living; BRSDDEP  Behavior Rating Scale for Demen-
that there is greater variance in depression between care- tia Depressive Subscale.
 DEPRESSION IN AD P309

Table 5. Models for Change in Caregiver Depression (Center for Epidemiologic Studies–Depression Scale)
Model 1 Model 2 Model 3 Model 4

Equation/Variables  SE  SE  SE  SE
Time 0 (0i)
Constant 13.155* 0.476 12.529* 0.492 12.533* 0.440 12.483* 0.782
Sex 2.679* 0.970 2.664* 0.972
Spouse 3.110* 1.007 3.099* 1.018
Health 2.178* 0.445 2.160* 0.447
Duration 0.412* 0.184 0.415* 0.184
B0(IADL) 0.227* 0.066 0.235* 0.066
B0(BADL) 0.016 0.059 0.019 0.059
B0(BRSDDEP) 0.543* 0.116 0.634* 0.188
t 0.008 0.629
t  B0(BRSDDEP) 0.084 0.152
σ u2 65.299* 5.997 65.299* 5.997 48.607* 5.243 48.402* 5.239
0
R2 0.256 0.259
Rate of change (1i)
Constant 1.359* 0.515 1.166* 0.478 1.182† 0.617
Sex 1.163 0.993 1.248 0.980
Spouse 0.787 1.190 0.302 1.187
Health 0.204 0.479 0.155 0.475
Duration 0.048 0.198 0.002 0.197
B1(IADL) 0.271* 0.129 0.304* 0.127
B1(BADL) 0.076 0.048 0.079† 0.047
B1(BRSDDEP) 0.402† 0.211 0.589* 0.268
t 0.607 0.595
t  B1(BRSDDEP) 0.693* 0.266
σ u2 14.622* 5.167 9.653* 4.147 6.969 3.912
1
R2 0.340 0.523
Acceleration of change (2i)
Constant 0.166 0.151 0.064 0.128 0.106 0.205
Sex 0.388 0.262 0.376 0.273
Spouse 0.088 0.329 0.094 0.345
Health 0.133 0.136 0.091 0.141
Duration 0.047 0.049 0.037 0.053
B2(IADL) 0.505† 0.282 0.612* 0.289
B2(BADL) 0.219* 0.068 0.214* 0.076
B2(BRSDDEP) 0.290 0.316 0.536 0.519
t 0.182 0.173
t  B2(BRSDDEP) 0.424 0.489
σ u2 0.811* 0.345 0.403 0.213 0.339 0.212
2
R2 0.503 0.582

σu 0.328 4.880 2.119 3.926 0.505 3.826

0 u1

σu 0.617 1.400 0.095 0.956 0.567 0.968

0 u2

σu 3.221* 1.306 1.971* 0.924 1.497 0.895

1 u2

σ e2 28.062* 1.418 23.228* 1.373 23.871* 1.386 24.013* 1.392

2LL 7,658.060 7,612.758 7,488.759 7,480.031

Note: IADL  instrumental activity of daily living; BADL  basic activity of daily living; BRSDDEP  Behavior Rating Scale for Depression Depressive Sub-
scale.
*p  .05 (two-tailed); †p  .05 (one-tailed).

of increase in depression that is significantly different than ( u1u2) between the rate of change and the acceleration of
zero (1.359). The average value of the acceleration parame- change (as was the case with the three patient variables).
ter (.166), however, is not significant. Thus, on average, That is, caregivers with high rates of increase in depression
caregivers are characterized by a positive rate of increase in (positive 1s) tend to have low acceleration or deceleration
depression across the observed occasions, with little or no (negative 2s) in depression. The addition of ti· to Model 2
deceleration of that increase. Both the rate of change (1i) was not significant (not shown).
and the acceleration of change (2i), however, vary signifi- Turning to Model 3 in Table 5, we see that six out of
cantly across caregivers (14.622 and 0.811, respectively). seven independent variables have significant effects on
Furthermore, there is a significant negative covariance caregiver depression at Time 0. Females, spouses, and care-
P310 NEUNDORFER ET AL.

givers with poor health are significantly more depressed at the ti·  BADL interaction,
2 (3, N  353)  0.670, p 
the baseline measurement. The shorter the duration from .880. Thus, there is no evidence that the effect on caregiver
onset of AD to entry into the study, the more depressed is depression of patient ADL dependency is related to how
the caregiver initially. The initial levels of IADL depen- long the caregiver participated in the study.
dency and of patient depressive symptoms are also signifi- We conducted 12 tests for the effect of ti·, to wit, three
cantly related to caregiver depression; the greater the pa- main effects and nine interactions. Only one was significant
tient’s IADL dependency and depression, the greater the (ti· 
1 [BRSDDEP]), and this interaction is counter what
caregiver’s baseline depression. The baseline measure of we would have reasonably expected. Therefore, we believe
BADL, however, is not significantly related to depression that this result should be treated cautiously. With the single
independently of baseline IADL. exception of this interaction, the results for Model 4 are
None of the caregiver variables are significantly related nearly identical to those for Model 3.
to the rate of change in caregiver depression. The rate of We examined interaction effects for IADL  BRSDDEP
change in the patient’s IADL dependency at Time 0 (0.271) and BADL  BRSDDEP to see if the effects of patient de-
and the rate of change in the patient’s depressive symptoms pressive symptoms depended on the degree of patient de-
at Time 0 (0.402), however, have significant positive effects pendency in IADLs and BADLs, and vice versa. These in-
on caregiver depression (for patient depressive symptoms, teractions were tested for the baseline measures, the rates of
t  1.869, p  .031, one-tailed, not shown). The greater the change, and the accelerations of these variables for a total of
rate of increase in the patient’s IADL dependency and de- six tests. The only significant result was a negative interac-
pressive symptoms, the greater the rate of increase in the tion between the accelerations of patient depressive symp-
caregiver’s depression. toms and patient dependency in basic activities, 2(BRSD-
With respect to the acceleration of caregiver depression, DEP)  2(BADL) (not shown). This indicates that the
acceleration in both IADL (0.505) and BADL (0.219) de- greater the acceleration of patient depressive symptoms, the
pendency has significant positive effects on caregiver de- less the effect of acceleration of BADL dependency on
pression. The greater the acceleration in IADL and BADL caregiver depression, and vice versa (i.e., higher accelera-
dependency, the more the change in caregiver depression tion of BADL dependency reduces the effect of acceleration
accelerates. This is despite the earlier finding that the initial of BRSDDEP). We should be cautious about this interac-
level and the rate of change in BADL at baseline were not tion, however, because it is the only one of these six tests
significantly related to caregiver depression. Acceleration that reached significance and because nine additional inter-
of patient depressive symptoms, however, is not signifi- actions (those involving ti·, just discussed) have also been
cantly related to acceleration of caregiver depression. tested, of which only one was significant.
We now discuss the results of including ti·, centered at We have found that changes in patient IADL and BADL
1.63, in the model in order to control for nonrandom miss- dependency and in patient depressive symptoms affect
ing values. As previously outlined, ti· was first entered in changes in caregiver depression. To examine more precisely
Equations 2–4 to test its main effects. The addition of ti· did the pattern of these effects over time, Figure 3 shows trends
not significantly improve the fit of the model,
2 (3, N  in caregiver depression as a function of average, high, and
353)  1.065, p  .786 (estimated parameters not shown). low initial rates of change in patient depressive symptoms.
Model 4 in Table 4 gives the estimated parameters specified In Figure 3A, the values of the parameters for the constant
by Equations 2a–4a for the three interactions involving ti·  terms of Model 4 in Table 5 are used for 0, 1, and 2. Be-
BRSDDEP. The addition of the three interaction terms re- cause each of the Xs is centered at its mean and ti· is cen-
sulted in a slight but significantly improved fit,
2 (3, N  tered at 1.63, these parameters represent the baseline de-
353)  7.665, p  .054. Of the three possible interactions, pression, the initial rate of change in depression, and the
only the positive (.693) for ti·  1 (BRSDDEP) reached acceleration of depression for an average caregiver who has
significance. This means that the longer the time the care- a mean observation time of 1.63 years. With these values of
giver was observed, the greater the positive effect of the ini- the betas, we plotted the quadratic model of change (Equation
tial rate of change in patient depressive symptoms on care- 1) over a period of 6 years. The maximum number of years
giver depression. The direction of this interaction is counter that caregiver depression was measured is 5.2 years; only 5%
to what we would have expected, which is that caregivers of the 353 caregivers were followed for 4 years or longer.
who become the most depressed by a rapid increase in their Thus, the trend shown beyond the 4th or 5th year in the plots
family members’ depressive symptoms would not stay in may not be valid. Nevertheless, the first plot shows that care-
the study as long as caregivers who are less affected by their givers who are average on all of the independent variables are
family members’ depressive symptoms. In other words, we expected to have monotonically increasing levels of depres-
would have expected a negative interaction, if any. The sion throughout most of the period. The rate of increase, how-
main effect of the rate of increase in patient depressive ever, slows as time passes. The most rapid increase of care-
symptoms, which is evaluated at t  1.63, is positive and giver depression is in the first years, and the increase slows
significant (.589). sufficiently to show no further increase by 5.6 years.
The addition of three interaction terms for ti·  IADL did Figure 3B illustrates the effect of a high rate of increase in
not significantly improve the model’s fit,
2 (3, N  353)  patient depressive symptoms (a high rate is defined as one
2.312, p  .510; it was necessary to use the restricted gener- standard deviation above the mean of 1) on change in care-
alized least-squares algorithm to obtain an interated solution giver depression. Mean values of all other variables, except
for this model. Neither did the addition of three terms for ti·  1.63, were used to generate the expected parameters of
 DEPRESSION IN AD P311

Figure 4. Change in caregiver depression at high and low rates of

change (1) in instrumental activities of daily living (IADL), mean
t  1.63, and means of other predictors. A, M  1 SD 1 (IADL); B,
M  1 SD 1 (IADL).

greater caregiver depression at baseline, and an increasing

rate of change in patient depressive symptoms predicted an
increasing rate of change in caregiver depression; however,
counter to the hypothesis, acceleration in patient depressive
symptoms did not predict acceleration in caregiver depression.
Additional predictors of caregiver depression at baseline
Figure 3. Change in caregiver depression at average, high, and low
rates of change (1) in patient depressive symptoms (Behavior Rating
were shorter duration of illness and greater patient depen-
Scale for Dementia Depressive Subscale, or BRSDDEP), mean t  dency in IADLs, but not BADLs. The effects of these two
1.63, and means of other predictors. A, M of 1 (BRSDDEP); B, M  patient variables were not expected, but they point to the
1 SD of 1 (BRSDDEP); C, M  1 SD of 1 (BRSDDEP). need to intervene early with caregivers who may be de-
pressed by the initial realization of their family members’
illness and early signs of decline. Caregiver variables that
change in caregiver depression. The plot shows monotoni- predicted caregiver depression at baseline were being a fe-
cally increasing caregiver depression throughout the entire male caregiver, being a spouse rather than a nonspouse, and
period. Unlike the trend in Figure 1A, average deceleration being in poor health themselves, as expected. In predicting
is insufficient to cause a leveling off of caregiver depression the rate of change in caregiver depression, only one variable
during the 6 years. In fact, the curve would not peak until in addition to patient depressive symptoms was a significant
9.2 years. predictor; increasing rate of change in patient dependency
The third graph (Figure 3C) is based on a low rate of in IADLs predicted increasing rate of change in caregiver
growth in patient depressive symptoms (one standard devia- depression. Acceleration of the increase in caregiver depres-
tion below the mean). Although it shows a slight overall de- sion was predicted by acceleration in patient dependency in
crease in caregiver depression, there is little or no change in IADLs and BADLs.
caregiver depression during the first 4 years. Different rates of change in initial patient depressive
Figure 4 shows analogous plots evaluated at high and low symptoms were examined for their effects on patterns of
rates of change in IADL dependency. The curves in Figures change in caregiver depression (as shown in Figure 3). If
4A and 4B are virtually identical to those in Figures 3B and caregivers experienced the average rate of change in all in-
3C. Thus, variance in the initial rate of change in IADLs dependent variables, their initial depression increased rap-
and in patient depression have the same effect on changes in idly in the first 2 years, continued to increase slowly, and
caregiver depression. then leveled off at about 5 years. However, if patient de-
pression was increasing at a high rate (and all other vari-
DISCUSSION ables were held at their means), caregiver depression ini-
To summarize, the focus of this study was the effects, tially increased rapidly, continued to increase, and leveled
over time, of depressive symptoms in persons with AD on off later. On the other hand, if patient depression increased
depression in their family caregivers. Additional variables at a low rate, there was little or no change in caregiver de-
accounted for were duration of illness, patient dependency pression during the first 4 years. Tracking rates of change in
in BADLs and IADLs, relationship (spouse or nonspouse), patient IADL dependency (as shown in Figure 3), a change
caregiver gender, and self-reported health. The effects of that occurs early in the illness, showed a similar pattern of
these variables on baseline caregiver depression, the linear effects on caregiver depression.
rate of change in caregiver depression, and the acceleration To conclude, results of this study add a seldom-examined
of the rate of change in caregiver depression were evaluated patient characteristic, depressive symptoms, to the list of
using multilevel analysis (see Model 3 in Table 5). As hy- what have been called primary stressors (Aneshensel et al.,
pothesized, more patient depressive symptoms predicted 1995) that contribute to caregiver depression. More patient
P312 NEUNDORFER ET AL.

depressive symptoms were associated with caregiver de- changes early in dementia, and this likely accounts for the
pression at baseline, and an increase in patient depressive strong effect of patient dependency in IADLs on caregiver
symptoms was associated with an increase in caregiver de- depression. The Consortium to Establish a Registry for
pression. These findings support the findings of the few Alzheimer Disease BRSD (Tariot et al., 1995), with its De-
other investigators (Brodaty & Luscombe, 1998; Drinka et pressive Symptoms subscale, picked up mood changes in
al., 1987; Hadjistavropoulos et al., 1994) who have reported patients separate from the vegetative symptoms and apa-
an association between patient depressive symptoms and thy that confound depressive symptoms with symptoms of
caregiver psychological morbidity or burden. They are also dementia. Its sensitivity to mood changes early in dementia,
consistent with the results of Tower and Kasl (1996), who again, probably contribute to the effect of rate of change in
found, among a probability sample of community-dwelling depressive symptoms on depression. Finally, rather than
older adults, that depressive symptoms in one older spouse summing across the wide range of behavioral symptoms ex-
influence depressive symptoms in the other, both cross sec- hibited in persons with dementia, and ignoring their ebb and
tionally and longitudinally. Their findings were stronger flow over the course of the illness, we chose to focus on one
when couples were close and held when known intraper- pattern of symptoms, patient depressive symptoms, and the
sonal risk factors and the health status of the spouse were changes in them.
controlled for. Findings from this study suggest that treatment of patient
Tower and Kasl (1996), in their discussion of why de- depressive symptoms may improve the quality of life for
pression in one spouse might contribute to depression in the patients and for their family caregivers. They also support
other, gave insights into what might be happening between interventions for caregivers early in their family members’
caregivers and the person with dementia for whom they illness when caregivers’ depression may be high due to their
care, especially if they have a close relationship. Extrapolat- family member’s depressive symptoms and initial signs that
ing from Tower and Kasl, it seems that persons with demen- their family member is becoming more dependent.
tia who exhibit depressive symptoms are likely to be nega-
tive about the present and future, make critical comments,
and express little appreciation of their caregivers. Like Acknowledgments
spouses of depressed persons in general, female caregivers This research was supported by the National Institute on Aging, Alzhei-
of depressed family members may feel increased loneliness mer Disease Research Center Grants AG08012 and AG RO1 12673. We are
grateful for the data collection and management support of Shannon O. Gal-
and a sense of futility about their efforts to help; male care- lagher, Linda R. Rechlin, and Nancy M. Catalani.
givers may feel inadequate and helpless. All these negative
Address correspondence to Marcia M. Neundorfer, Fairhill Center for
interactions and thoughts can contribute to a downward spi- Aging, University Alzheimer Center, 12200 Fairhill Road, Cleveland, OH
ral of depression for the caregiver. 44120-1013. E-mail: [email protected]
The findings from the present study support the impor-
tance not only of patient depressive symptoms in under-
References
standing caregiver depression, but of patient characteristics
Alexopoulos, G., & Abrams, R. (1991). Depression in Alzheimer’s dis-
in general, including the early phase of the illness and the ease. The Psychiatric Clinics of America, 14, 327–339.
rate of increase in initial dependencies. As patients became Alspaugh, M. E. L., Stephen, M. A., Townsend, A. L., Zarit, S. H., &
more dependent in IADLs, usually the skills most sensitive Greene, R. (1999). Longitudinal patterns of risk for dementia care-
to early cognitive deficits, they needed more reminders, su- givers: Objective and subjective primary stress as predictors.
pervision, and guidance with complex tasks such as travel- Psychology and Aging, 14, 34–43.
American Psychiatric Association. (1980). Diagnostic and statistical man-
ing and shopping. The more rapidly these changes occurred, ual of mental disorders (3rd. ed.). Washington, DC: Author.
the more depressed were the caregivers, even controlling for Aneshensel, C. S., Frerichs, R. R., & Huba, G. J. (1984). Depression and
their own health, gender, and relationship to the patient. Ex- physical illness: A multiwave, nonrecursive causal model. Journal of
cept for Aneshensel and colleagues (1995), no investigators Health and Social Behavior, 25, 350–371.
Aneshensel, C. S., Pearlin, L. I., Mullan, J. T., Zarit, S. H., & Whitlatch, C.
of caregiver well-being have addressed the rate of change in J. (1995). Profiles in caregiving: The unexpected career. San Diego:
patient characteristics, and this has led to the underestima- Academic Press.
tion of the importance of patient characteristics in the stress Brodaty, H., & Luscombe, G. (1998). Psychological morbidity in care-
process. As Aneshensel and associates concluded, the pa- givers is associated with depression in patients with dementia.
tient’s impairment is central in the stress process experi- Alzheimer Disease and Associated Disorders, 12, 62–70.
Bryk, A. S., & Raudenbush, S. W. (1992). Hierarchical linear models.
enced by caregivers, especially the rate at which the patient Newbury Park, CA: Sage.
declines over time. Burke, W. J., Roccaforte, W. H., Wengel, S. P., McArthur-Miller, D.,
Limitations of the present study include the focus on one Folks, D. G., & Potter, J. F. (1998). Disagreement in the reporting of
type of patient behavior problem, depressive symptoms. Al- depressive symptoms between patients with dementia of the Alzheimer
type and their collateral sources. The American Journal of Geriatric
though depressive symptoms are the most frequently occur- Psychiatry, 6, 308–318.
ring problem, other less frequent behaviors, such as disrup- Cummings, J. L., Ross, W., Absher, J., Gornbein, J., & Hadjiaghai, L.
tive behaviors, should be examined for their relative contri- (1995). Depressive symptoms in Alzheimer disease: Assessment and de-
bution to caregiver depression. terminants. Alzheimer Disease and Associated Disorders, 9(2), 87–93.
Significant strengths of this study include a sample of pa- Devanand, D. P., Jacobs, D. M., Tang, M., Del Castillo-Castaneda, C., Sano,
M., Marder, K., Bell, K., Bylsma, F. W., Brandt, J., Albert, M., & Stern,
tients who were carefully diagnosed and well characterized. Y. (1997). The course of psychopathologic features in mild to moderate
The Cleveland Scale for Activities of Daily Living (Patter- Alzheimer disease. Archives of General Psychiatry, 54, 257–263.
son et al., 1992) proved to be highly sensitive to functional Drinka, J., Smith, J., & Drinka, P. (1987). Correlates of depression and
 DEPRESSION IN AD P313

burden for informal caregivers of patients in a geriatrics referral clinic. Prevalence and impact of caregiving: A detailed comparison between
Journal of the American Geriatrics Society, 35, 522–525. dementia and nondementia caregivers. The Gerontologist, 39, 177–
Folstein, M., Folstein, S., & McHugh, P. (1975). “Mini-Mental State”: A 185.
practical method for grading the cognitive state of patients for the clini- Ott, B. R., & Fogel, B. S. (1992). Measurement of depression in dementia:
cian. Journal of Psychiatric Research, 12(3), 189–198. Self vs. clinician rating. International Journal of Geriatric Psychiatry,
Gallagher, D., Rose, J., Rivera, P., Lovett, S., & Thompson, L. W. (1989). 7, 899–904.
Prevalence of depression in family caregivers. The Gerontologist, 29, Patterson, M. P., Mack, J. L., Neundorfer, M. M., Martin, R. J., Smyth,
449–456. K. A., & Whitehouse, P. J. (1992). Assessment of functional ability in
Goldstein, H. (1995). Multilevel statistical models (2nd ed.). New York: Alzheimer disease: A review and a preliminary report on the Cleveland
Halsted. Scale for Activities of Daily Living. Alzheimer Disease and Associated
Goode, K. T., Haley, W. E., Roth, D. L., & Ford, G. R. (1998). Predicting Disorders, 6(3), 145–163.
longitudinal changes in caregiver physical and mental health: A stress Payne, J. L., Lyketsos, C. G., Steele, C., Baker, L., Galik, E., Kopunek, S.,
process model. Health Psychology, 17, 190–198. Steinberg, M., & Warren, A. (1998). Relationship of cognitive and
Hadjistavropoulos, T., Taylor, S., Tuokko, H., & Beattie, B. L. (1994). functional impairment to depressive features in Alzheimer’s disease
Neuropsychological deficits, caregivers’ perception of deficits and and other dementias. Journal of Neuropsychiatry, 10(4), 440–447.
caregiver burden. Journal of the American Geriatrics Society, 42, 308– Prosser, R., Rasbash, J., & Goldstein, H. (1991). ML3 software for three-
314. level analysis: User’s guide for version 2. London: Institute of
Harwood, D. G., Ownby, R. L., Barker, W. W., & Duara, R. (1998). The Education.
factor structure of the Cornell Scale for Depression in Dementia among Pruchno, R., Kleban, M., Michaels, J., & Dempsey, N. (1990). Mental and
probable Alzheimer’s disease patients. American Journal of Geriatric physical health of caregiving spouses: Development of a causal model.
Psychiatry, 6(3), 212–220. Journal of Gerontology: Psychological Sciences, 45, P192–P199.
Hedeker, D., & Gibbons, R. D. (1997). Application of random-effects pattern- Radloff, L. (1977). The CES-D Scale: A self-report depression scale for re-
mixture models for missing data in longitudinal studies. Psychological search in the general population. Applied Psychological Measurement,
Methods, 2, 64–78. 1, 385–401.
Hooker, K., Monahan, D. J., Bowman, S. R., Frazier, L. D., & Shifren, K. Schulz, R., O’Brien, A. T., Bookwala, J., & Fleissner, K. (1995). Psychiat-
(1998). Personality counts for a lot: Predictors of mental and physical ric and physical morbidity effects of dementia caregiving: Prevalence,
health of spouse caregivers in two disease groups. Journal of Gerontol- correlates, and causes. The Gerontologist, 35, 771–791.
ogy: Psychological Sciences, 53B, P73–P85. Schulz, R., Visintainer, P., & Williamson, G. M. (1990). Psychiatric and
Hughes, C. P., Berg, L., Danziger, W. L., Coben, L. A., & Martin, R. L. physical morbidity effects of caregiving. Journal of Gerontology: Psy-
(1982). A new clinical scale for the staging of dementia. British Jour- chological Studies, 45, P181–P191.
nal of Psychiatry, 140, 566–572. Schulz, R., & Williamson, G. (1991). A 2-year longitudinal study of de-
Jacobs, M. R., Strauss, M. E., Patterson, M. B., & Mack, J. L. (1998). pression among Alzheimer’s caregivers. Psychology and Aging, 6,
Characterization of depression in Alzheimer’s disease by the CERAD 569–578.
Behavior Rating Scale for Dementia (BRSD). American Journal of Ge- Snijders, T., & Bosker, R. (1999). Multilevel analysis. Thousand Oaks,
riatric Psychiatry, 6, 53–58. CA: Sage.
Katz, I. R. (1998). Diagnosis and treatment of depression in patients with Stuckey, J. C., Neundorfer, M. M., & Smyth, K. A. (1996). Burden and
Alzheimer’s disease and other dementias. Journal of Clinical Psychia- well-being: The same coin or related currency? The Gerontologist, 36,
try, 59(Suppl. 9), 38–44. 686–693.
Little, R. J. A. (1995). Modeling the drop-out mechanism in repeated-mea- Tariot, P. N., Mack, J. L, Patterson, M. B., Edland, S. D., Weiner, M. F.,
sures studies. Journal of the American Statistical Association, 90, Fillenbaum, G., Blazina, L., Teri, L., Rubin, E., Mortimer, J. A., Stern,
1112–1121. Y., and the Behavioral Pathology Committee of the Consortium to Es-
Mack, J. L., & Patterson, M. B. (1996). Manual: CERAD Behavior Rating tablish a Registry for Alzheimer’s Disease. (1995). The Behavior
Scale for Dementia (2nd ed.). Durham, NC: Consortium to Establish a Rating Scale for Dementia of the Consortium to Establish a Registry
Registry for Alzheimer’s Disease. for Alzheimer’s Disease. American Journal of Psychiatry, 152(9),
Mackenzie, T. B., Robiner, W. N., & Knopman, D. S. (1989). Differences 1349–1357.
between patient and family assessments of depression in Alzheimer’s Teri, L., & Wagner, A. (1991). Assessment of depression in patients with
disease. American Journal of Psychiatry, 146(9), 1174–1178. Alzheimer’s disease: Concordance between informants. Psychology
Marin, D. B., Green, C. R., Schmeidler, J., Harvey, P. D., Lawlor, B. A., and Aging, 6, 280–285.
Ryan, T. M., Aryan, M., Davis, K. L., & Mohs, R. C. (1997). Noncog- Tower, R. B., & Kasl, S. V. (1996). Depressive symptoms across older
nitive disturbances in Alzheimer’s disease: Frequency, longitudinal spouses: Longitudinal influences. Psychology and Aging, 11, 683–697.
course, and relationship to cognitive symptoms. Journal of the Ameri- Vitaliano, P. P., Russo, J. M., Young, H. M., Teri, L., & Maiuro, R. D.
can Geriatrics Society, 45, 1331–1338. (1991). Predictors of burden in spouse caregivers of individuals with
McKhann, G., Drachman, D., Folstein, M., Katzman, R., Price, D., & Stad- Alzheimer’s disease. Psychology and Aging, 6, 392–402.
lan, E. (1984). Clinical diagnosis of Alzheimer’s disease: Report of the Wragg, R., & Jeste, D. (1989). Overview of depression and psychosis in
NINCDS-ADRDA Work Group under the auspices of Department of Alzheimer’s disease. American Journal of Psychiatry, 146, 557–587.
Human Health and Services Task Force on Alzheimer’s Disease. Neu-
rology, 34(1), 939–944.
Moye, J., Robiner, W. N., & Mackenzie, T. B. (1993). Depression in
Alzheimer patients: Discrepancies between patient and caregiver re- Received August 24, 1999
ports. Alzheimer Disease and Associated Disorders, 7, 187–201. Accepted August 30, 2000
Ory, M. G., Hoffman, R. R., Yee, J. L., Tennstedt, S., & Schulz, R. (1999). Decision Editor: Toni C. Antonucci, PhD