Healthcare use and cost in dementia caregivers: Longitudinal results from the Predictors Caregiver Study

Abstract

OBJECTIVE:

To examine the effects of caregiver and patient characteristics on caregivers’ medical care use and cost.

METHODS:

147 caregiver/patient dyads were followed annually for 6 years in 3 academic AD centers in the US. Logistic, negative binomial, and generalized linear mixed models were used to examine overall effects of caregiver/patient characteristics on caregivers’ hospitalizations, doctor visits, outpatient tests and procedures, and prescription and over-the-counter medications.

RESULTS:

Patients’ comorbid conditions and dependence were associated with increased healthcare use and costs of caregivers. Increases in caregiver depressive symptoms are associated with increases in multiple domains of caregivers’ healthcare use and costs.

DISCUSSION:

Findings suggest that we should expand our focus on dementia patients to include family caregivers to obtain a fuller picture of effects of caregiving. Primary care providers should integrate caregivers’ needs in healthcare planning and delivery. Clinical interventions that treat patients and caregivers as a whole will likely achieve the greatest beneficial effects.

INTRODUCTION

In 2012, more than 5.4 million older adults in the US have dementia, two-thirds of whom live in the community and are cared for by family and friends.^[1] A recent report estimated the annual cost per person attributable to dementia, including medical care, nursing home care, and in home formal and informal care, between $41,689-56,290.^[2] If the proportion of individuals with dementia living in the community remains stable, the burden of caregiving on family and friends will increase exponentially.

The mental and physical health effects of caregiving on family caregivers have been well documented. Compared to non-caregivers, caregivers have higher rates of depression, stress, and other psychological problems.^[3-7] They also experience more physical problems, including lower self-rated health, weakened immune system, and increased cardiovascular problems.^[8-11] Within the caregiver community, effects of caregiving are particularly strong for dementia caregivers as they face unique challenges in providing care due to the cognitive and physical disability experienced by patients and long duration of care.^[3-7]

Because of these negative health effects of caregiving, economic and health consequences to dementia caregivers can be substantial. A large and growing body of literature has examined health and community services used by caregivers of individuals with cognitively impairment.^[12-15] However, the focus of the research has been on services that are potentially helpful for the caregivers in their role as caregivers. Results show that there is high need but often low use of these services. Main barriers to use include not knowing where to obtain services, and that many caregivers wait too long to use services, by which time care recipients often are too frail to benefit. With few exceptions, caregivers’ own healthcare utilization has received little attention.^[16] An early study examined a comprehensive set of health services use, including outpatient, inpatient, and home health care, but did not separate utilization by care recipient or caregiver.^{[12, 13]} Another study reported slightly higher cost from hospitalizations but lower cost from doctor visits in a sample of dementia caregivers compared to non-caregivers with similar socioeconomic characteristics, but the differences in magnitude were small.^[17] Similarly small but statistically significant increases in caregivers’ medication cost as caregiving hours increased also have been reported.^[18] Caregivers' acute care utilization have been shown to increase as patients’ behavioral and psychological symptoms and functional status worsened.^[19] More recent analyses from the Resources for Enhancing Alzheimer's Caregiver Health (REACH) study showed that caregivers’ hospitalization rates and probability of multiple doctors’ visit increased significantly during an 18-months study period, but use of primary care and mental health service and select medication use (antidepressant, anti-anxiety medications) did not significantly increase.^[20] It is unclear however, whether patients’ clinical characteristics were included or how they might have affected caregivers’ medical care use in this study.

The relationship between caregiving and caregivers’ health is typically described by Pearlin’s model of caregiver stress in a process in which both care recipient and caregiver characteristics change over time but are both important determinants of caregiver outcomes.^[21] In this model, objective stressors are measured by the progression of care recipients’ chronic illness and decline. Typical outcomes include type and quantity of caregiving, caregivers’ psychological stress, burden, and psychiatric and physical health. Early models of the caregiving process emphasized increased burden and stress over time, but longitudinal studies suggested adaptations over time and stable caregiver burden and depression^{[22, 23]}. In this current study, we take advantage of a rare opportunity of having both caregiver and care recipient data in a longitudinal setting, and extend this model to examine caregivers’ medical care use and related costs as outcomes. Because of the modest sample size of our study, we estimated a set of reduced form equations to examine the overall effects of patient and caregiver characteristics on caregivers’ own medical care use and related costs.

METHODS

Sample

Subjects for this study were recruited from the Predictors 2 study, a longitudinal study of individuals with probable Alzheimer’s disease (AD) or dementia with Lewy bodies (DLB) prospectively followed from early stages of illness in three sites: Columbia University College of Physicians and Surgeons; Johns Hopkins University School of Medicine; and Massachusetts General Hospital. Recruitment of subjects in the Predictors 2 Study began in 1998 and ended in 2012. Inclusion/exclusion criteria and evaluation procedures of the Predictors 2 study have been fully described elsewhere.^{[24, 25]} Briefly, following an initial evaluation, all subjects were diagnosed in a consensus conference with at least two faculty neurologists or psychiatrists specializing in dementia and one faculty neuropsychologist. Subjects with AD met NINCDS-ADRDA criteria for probable AD^[26], and those with DLB were diagnosed according to the 1996 consensus guidelines for the disease.^[27] At study entry, all subjects were required to have a modified Mini Mental State Examination (MMSE) Score(21) ≥ 30, equivalent to a score of ≥ 16 on the standard MMSE^[28]. Subjects also were required to have at least one family member/informant available. Subjects with Parkinsonism, stroke, alcoholism, schizophrenia, schizoaffective disorder, and electroconvulsive treatments were excluded. After the baseline evaluation, all subjects were followed in person semi-annually until dropout or death, with annual assessments of resource utilization. If subjects were unable to travel to the clinic for evaluation, they were visited at their residence (e.g., home, nursing home, or health care facility). There is 94% follow up of subjects. The study was approved by each local Institutional Review Board.

The Predictors Caregiver Study was initiated six years after launching the Predictors 2 Study. A total of 180 caregivers were active in the Predictors Study at the time of, or subsequent to, the launching of the Predictors Caregiver Study. Throughout the Caregiver study, these 180 caregivers provided care for 160 subjects (care recipients hereafter). Of these 180 caregivers, healthcare use data were missing for 21 caregivers. Deleting these observations leaves use with 159 caregivers with valid cost data for 147 care recipients. Of these 147 care recipients, 12 (8.2%) had multiple caregivers who participated with the subject at different times. We identified the caregiver who participated most often during the study as the subject’s main caregiver. Data on other caregivers were dropped from the analysis. Thus, the longitudinal cohort used in this analysis included 147 caregiver/care recipient dyads (for a total of 394 observations). On average, each dyad completed 3 annual assessments.

Measures

Outcomes

At baseline and annually thereafter, caregivers reported their own use of the following six domains of medical care: overnight hospitalizations, doctor visits, outpatient tests and procedures (e.g., blood test, mammogram, urinalysis), and assistive devices (e.g., walkers, grab bars). For those who reported using a particular item, data on intensity of use (e.g., number of hospitalizations, number of doctor visits) also were collected. Detailed information on prescription and over the counter (OTC) medications the caregiver took during the past six months including the name of the drug, number of days taking the drug, and number of pills per day were recorded on a concurrent medications form. We annualized utilization rates when domains were reported for less than a year (e.g., three months for outpatient medical test and procedures and six months for medications). We converted physical quantities for each domain of care into monetary values using costing methods reported in earlier studies.^[29] We then summed across all domains to obtain a total annual medical care cost for each caregiver. All cost values were adjusted to constant 2010 dollars using the medical care component of the Consumer Price Index.^[30] Detailed descriptions on data collection and price conversion were reported earlier.^[31]

Caregiver Characteristics

Caregiver age, gender, ethnicity, highest level of education, and relationship to care recipient were recorded at the start of the Caregiver Study. Whether the caregiver lived with the care recipient, frequency of contact with care recipient, and length of time caregiver had known the care recipient were recorded at each annual visit. Whether caregiver assisted with basic activities of daily living (BADL), including eating, dressing, bathing, using toilet, instrumental activities of daily living (IADL), including shopping, household chores, personal business, transportation, social activities), average amount of hours the caregiver spent per day on providing care, whether a paid attendant assisted with care, and caregiver’s employment status were reported annually. Caregiver comorbidities at baseline were computed using a modified Charlson scale and included items for myocardial infarction, congestive heart failure, peripheral vascular disease, hypertension, chronic obstructive pulmonary disease, arthritis, gastrointestinal diseases, liver disease, diabetes, chronic renal disease, and systemic malignancy.^[32] Caregiver depressive symptoms were measured by the 6-item depression subsection of the Brief Symptom Inventory (BSI).^[33] The instrument asked the caregivers to rate on a five-point Likert scale how much during the past week they were bothered by the following: feeling lonely, feeling blue, feeling no interest in things, feeling hopeless about the future, feelings of worthlessness, and thoughts of ending his/her life. Response for each item ranged from “not at all” to “extremely”, with higher scores indicating higher depressive symptoms.

Care Recipient Characteristics

At the baseline visit, demographic characteristics (e.g., age, gender, ethnicity, highest level of education) were recorded. Because care recipients’ living arrangement may change over time and patterns of health services utilization and costs may differ substantially between individuals in different living arrangements,^[34] information on care recipients’ living arrangements (living at home, in retirement homes, in assisted living facilities, or in nursing homes) was collected at each visit.

Data on several clinical characteristics of the patient that can be assessed reliably by a clinician were recorded at each visit. Disease progression was characterized by transition from milder stages of dementia to more severe stages, measured by MMSE.^[28] Care recipients dependence on others was measured by the Dependence Scale (DS),^[35] a continuous index measuring the progressively greater dependence on others. Columbia University Scale for Psychopathology in Alzheimer’s Disease (CUSPAD), a semi-structured interview administered by a physician or a trained research technician, was used to measure the presence or absence of psychotic symptoms, behavior problems, and depressive symptoms.^[36] From the CUSPAD, we constructed a dichotomous variable to indicate the presence of psychotic symptoms if the patient had any delusions, hallucinations, or illusions.^{[37, 38]} We constructed a dichotomous variable to indicate the presence of behavioral problems if the patient had any of the following five symptoms: wandering away from home or caregiver, verbal outbursts, physical threats or violence, agitation or restlessness, or sundowning (more confusion at night or during evening, compared to the day). We also constructed a dichotomous variable to indicate the presence of depressive symptoms if the patient had any depressed mood (i.e., sad, depressed, blue, down in the dumps) and either had difficulty sleeping or had a change in appetite. Comorbidities included items for myocardial infarction, congestive heart failure, peripheral vascular disease, hypertension, chronic obstructive pulmonary disease, arthritis, gastrointestinal diseases, liver disease, diabetes, chronic renal disease, and systemic malignancy from the baseline visit.

Analysis

We estimated a set of reduced form equations in order to provide estimates of the overall effects of patient and caregiver characteristics on caregivers’ healthcare use and costs. We examined three sets of outcomes: any medical care use, intensity of use, and costs.

For any medical care use, estimations were performed using random effects logistic regressions. Dependent variables included any hospitalizations, any doctor visits, any prescription medication, any outpatient treatment/procedures, and any OTC medication. Because over 93% of the caregivers reported using some medical care, we did not estimate an overall model for any medical care use. Estimates for any use models are exponentiated and represent adjusted odds-ratios (OR).

For intensity of use, dependent variables included number of doctor visits, number of prescription medications, number of outpatient treatment/procedures, and number of OTC medications. We did not estimate a model for number of hospitalizations because only 46 hospitalizations were reported throughout the study. Intensity of use was examined using negative binomial regressions. We chose negative binomial over Poisson regressions because with count data, it is more robust when the variance is not equivalent to the mean of the distribution and it models between subject heterogeneity.^[39] It is unclear whether different processes govern zero and non-zero counts. We compared our results from our models, which assume the same processes for zero and non-zero counts to those obtained from zero-inflated negative binomial models, which assume different processes for zero and non-zero counts. Results were substantively similar and we therefore chose the simpler and more familiar negative binomial models. Estimates for intensity of use models are exponentiated and represent incidence rate ratios (IRR).

For caregivers who used a particular domain of medical care, we then estimated a generalized linear mixed regression model for its related cost. Similar to models on intensity of use, we did not estimate models for hospitalizations because of its low use rates. In addition to the four models of medical care use, a model estimating total cost of medical care also was estimated. Because the distributions of the cost variables were highly positively skewed, we constructed log-transformed costs as dependent variables. Estimates for the cost models are interpreted as semi-elasticities, i.e., percentage change in cost from a unit change in the independent variable.

For all three sets of models, we first examined bivariate association between caregivers’ medical care use and each caregiver/care recipient characteristic. Variables that were not significantly associated with caregivers’ medical care use in bivariate analyses were dropped from our final models. We tested several interaction terms between care recipient characteristics (dependence, chronic conditions) and caregiver characteristics (chronic conditions, depressive symptoms). They were statistically insignificant and were dropped from the analyses. Our final model therefore included the following independent variables for the caregiver: baseline age, gender, education, comorbid conditions, depressive symptoms (measured by the depression subsection of the BSI), and an indicator for a spouse caregiver. Care recipient characteristics included in the final models were baseline comorbid conditions, Dependence Scale (DS), indicator for whether the care recipient was depressed, and indicator for the presence of psychiatric problems. All caregiver and care recipient clinical characteristics, except baseline comorbid conditions, were time varying, current values.

Because of the longitudinal nature of the study, caregivers could contribute multiple observations in the study. We included a random intercept term to allow between person variations and a random slopes term to allow variations within person over time.^[40] Fixed effects of time were estimated by including indicators for assessment intervals.^[41] All models controlled for indicators for site as fixed effects. All analyses were performed using Stata 9.0.^[42]

RESULTS

Caregiver and Care Recipient Characteristics at Baseline

The typical caregiver in this sample was 65 years old (SD=13.4), female (76%), white (84%), and a college graduate with 16 years of schooling (SD=3.3) (Table 1). Slightly more than half of the caregivers were spouses (55%), a third were adult children (35%). The caregivers were relatively healthy, reporting fewer than one chronic condition at baseline. The most prevalent chronic conditions reported were hypertension (24%), depression (10%), diabetes (6%) and cancer (6%). Other chronic conditions were reported by fewer than 5% of the caregivers. The most commonly reported medications included anti-hypertensive agents (26%), medications for dyslipidemia (18%), and psychotropic medications (12%). Other medications were reported by fewer than 10% of the caregivers. Slightly over a third of the caregivers reported being employed at the time (37%). Those who were working reported an average of 32 work hours per week (SD=18). Three-fourths of the caregivers helped with instrumental activities of daily living (78%) and about half with basic activities of daily living (47%). Among those who provided help, caregivers reported spending an average of 3 hours a day helping with IADLs (SD=2.6) and 2 hours on ADLs (SD=1.9). A small fraction of caregivers also reported having other helpers who provided IADL (18%) or ADL help (14%). Among those with other helpers, caregivers reported that other helpers spent an average of 3 hours a day helping with either ADLs or IADLs.

Table 1.

Baseline Demographic Characteristics of Caregiver/Care Recipient Dyads (n=147)

Variables	Mean	(SD)
Caregiver Characteristics
Age, years	65.2	(13.4)
Younger than 65 (%)	47.6
65-74 (%)	24.5
75 or older (%)	27.9
Female (%)	75.5
Race (%)
White	83.7
Black	7.5
Years of schooling	15.9	(3.3)
Relationship to care recipient (%)
Spouse	55.1
Child	35.4
Other	9.5
Number of comorbid conditions	0.5	(0.9)
Hypertension (%)	23.8
Depression (%)	9.5
Diabetes (%)	5.7
Cancer (%)	5.7
Medication Use (%)
Anti-hypertensive	26.0
For dyslipidemia	17.9
Antidepressant	8.9
Other psychotropic medications	12.2
Vitamins	8.1
Depressive symptoms a	8.1	(3.1)
Employment status (%)
Currently working	37.4
Hours of work per week b	31.9	(18.2)
Provided help with IADL/BADLs (%)
IADL	78.2
ADL	46.9
Other helpers helped with IADL	17.7
Other helpers helped with ADL	13.6
Hours of help per day b
IADL	3.1	(2.6)
ADL	2.4	(1.9)
Other helpers helped with IADL	0.6	(1.6)
Other helpers helped with ADL	0.5	(1.7)
Total hours from all helpers b	5.4	(6.2)
Care Recipient Characteristics
Age, years	75.6	(7.3)
Female (%)	55.1
Race (%)
White	89.8
Black	8.2
Years of schooling	14.7	(3.2)
Living arrangement (%)
Home	68.7
Retirement home	15.6
Nursing home	15.0
Number of comorbid conditions c	0.8	(1.1)
Hypertension (%)	34.0
Depression (%)	13.2
Diabetes (%)	7.6
Angina (%)	7.0
COPD (%)	7.0
MMSE, range=(0-30)	17.4	(7.7)
Mild AD: MMSE≥21 (%)	42.9
Moderate AD, MMSE 10-20 (%)	38.4
Severe AD, MMSE<10 (%)	18.8
Dependence Scale, range=(0-15)	6.7	(3.1)
Presence of psychotic symptoms (%)	44.2

^aDepressive symptoms are measured by the 6-item depression subsection of the Brief Symptom Inventory (BSI), range = (0-30).

^bHours of work and hours of help are computed among those who were employed and provided help only.

^cPrevalence of other comorbid conditions were all <5%.

The typical care recipient in this sample was 76 years old (SD=7.3), female (55%), white (90%), had an average of 15 years of schooling (SD=3.2), and had fewer than one chronic condition at baseline (mean=0.8, SD=1.1) other than dementia. The most prevalent chronic conditions for the care recipients included hypertension (34%), depression (13%), diabetes (8%), angina (7%) and COPD (7%). Rates of other chronic conditions were low: myocardial infarction, congestive heart failure, and cancer were present in 4% of the sample, and chronic liver disease and renal disease were present in 2% of the sample. At baseline, most of the care recipients were living at home (69%), the rest was evenly split between living in retirement homes (16%) or in nursing homes (15%). Average MMSE score at baseline was 17.4 (SD=7.7) and DS score was 6.7 (SD=3.1).

Caregiver Baseline and Overall Medical Care Use and Cost

At baseline, 93% of the caregivers reported using some type of medical care; 10 (6.8%) reported not using any medical care (Table 2). During the entire study period, 4 of the 147 caregivers reported not using any medical care at all. Across all assessments (n=401), no medical care use was only reported 6.6% of the time (n=26).

Table 2.

Caregivers’ Baseline and Overall Medical Care Utilization

Variables	Baseline (n=147)		Across all assessments (n=394)
Utilization Rate (%)
Any use	93.2		93.4
Hospitalizations	10.2		11.7
Prescription medications	75.5		70.1
Doctor Visits	74.1		75.4
Over the Counter (OTC) medications	57.8		48.2
Outpatient tests/procedures	64.6		67.0
Assistive devices	25.9		24.1
Intensity of use among users, mean (SD)
Number of hospital admissions	1.1	(0.5)	1.2	(0.6)
Number of nights at hospital	3.9	(5.1)	4.8	(4.7)
Number of doctor visits	2.8	(4.6)	2.7	(4.4)
Number of assistive devices	1.4	(1.0)	1.5	(1.0)
Number of outpatient tests/procedures	2.8	(2.3)	2.7	(2.2)
Number of prescription medications	3.2	(2.4)	3.3	(2.6)
Number of OTC medications	3.3	(2.5)	3.2	(2.4)
Cost per person, mean $ (SD)
Total	2,585	(6,386)	3,068	(6,993)
Hospitalizations	1,122	(5,754)	1,503	(6,088)
Prescription medications	798	(994)	868	(1,253)
Outpatient tests/procedures	412	(851)	448	(903)
Doctor Visits	154	(301)	160	(298)
Assistive devices	53	(178)	54	(204)
OTC medications	46	(110)	34	(85)

At baseline, 10% of caregivers reported being hospitalized at least once during the previous assessment interval, three-fourths (74%) had a doctor visit, 65% received outpatient tests and procedures. A quarter of caregivers (26%) used assistive devices. Prescription (76%) and OTC (58%) medications were common. Average annual medical care cost was estimated at $2,585 (SD=6,386) in the year prior to baseline assessment, with the most expensive component being hospitalizations (mean=$1,122, 43.4% of total cost) and prescription medications ($798, 30.9%).

Table 2 also presents data on overall use and cost throughout the study period. Time trend in caregiver medical care use and costs was statistically insignificant for all domains except for prescription and OTC medications use, where they both decreased slightly (both p<0.05).

Multivariate Results on Utilization

Not surprisingly, caregiver’s own chronic conditions were associated with increased likelihood of having any doctor visits (OR=2.493, p<0.01) and hospitalization (OR=1.67, p<0.10), although the association with hospitalization was marginally significant (Table 3). Each point increase in caregiver’s BSI-depression subsection score was associated with increased likelihood of prescription (OR=1.112, p<0.05) and OTC (OR=1.117, p<0.05) medication use. After controlling for caregiver health, demographic characteristics were largely not associated with medical care use, except that older caregivers were more likely to use outpatient tests (OR=1.192, p<0.01).

Table 3.

Random Effects Logistic Regression Results on Caregivers’ Medical Care Use Over Time (n=394).

	Any Hospitalizations	Any Doctor Visits	Any Outpatient Treatment	Any Prescription Medications	Any Over the Counter (OTC) Medications
Caregiver	OR (SE)	OR (SE)	OR (SE)	OR (SE)	OR (SE)
Age in 5 year increments	1.099	1.145	1.192 ***	1.145	1.127
	(0.140)	(0.101)	(0.074)	(0.079)	(0.087)
Female	1.131	1.831	1.486	1.850 *	1.640
	(0.651)	(0.815)	(0.440)	(0.609)	(0.622)
Year of schooling	0.967	1.069	1.000	0.929	1.061
	(0.068)	(0.068)	(0.041)	(0.046)	(0.054)
Baseline comorbid conditions	1.670 *	2.493 ***	1.102	1.171	1.380
	(0.454)	(0.870)	(0.203)	(0.259)	(0.296)
Brief symptom inventory (BSI)	1.025	1.079	1.052	1.112 **	1.117 **
	(0.056)	(0.064)	(0.042)	(0.053)	(0.050)
Care Recipient
Baseline comorbid conditions	1.185	1.557 **	1.319 **	1.014	0.961
	(0.244)	(0.328)	(0.171)	(0.125)	(0.135)
Dependence Scale (DS)	0.962	1.160 **	1.106 **	1.040	0.958
	(0.074)	(0.071)	(0.046)	(0.048)	(0.048)
Depressed	0.533	1.495	0.860	0.636	0.525
	(0.407)	(0.839)	(0.304)	(0.248)	(0.245)
Presence of psychiatric problems	2.083 *	0.892	1.244	1.801 **	0.788
	(0.914)	(0.299)	(0.308)	(0.493)	(0.227)
Spouse Caregiver	2.247	1.758	1.428	1.841 *	2.043 *
	(1.507)	(0.844)	(0.464)	(0.681)	(0.842)

^*p<0.10,

^**p<0.05,

^***p<0.01

Several care recipient characteristics were associated with caregiver medical care use. In particular, care recipients’ chronic conditions was associated with caregivers’ increased likelihood of doctor visits (OR=1.557, p<0.05) and use of outpatient tests and procedures (OR=1.319, p<0.05). Care recipients’ DS Score also was associated with caregivers’ increased likelihood of doctor visits (OR=1.116, p<0.05) and use of outpatient tests and procedures (OR=1.106, p<0.05). Care recipients’ psychiatric problems was associated with caregivers’ increased likelihood of prescription medication use (OR=1.801, p<0.05) and hospitalization (OR=2.083, p<0.10), although the association with hospitalization was marginally significant. Compared to adult children and other caregivers, spouse caregivers were more likely to use prescription (OR=1.841, p<0.10) and OTC (OR=2.043, p<0.10) medications, although the effects were only marginally significant.

Multivariate Results on Intensity of Use

Among caregivers who reported using medical care, caregiver’s own chronic conditions were associated with increases in the number of doctor visits (IRR=1.271, p<0.01), number of prescription medications (IRR=1.432, p<0.01), and to a lesser extent, number of outpatient tests and procedures (IRR=1.073, p<0.10). Caregiver’s BSI-depression subsection score was significantly associated with all four domains of medical care: number of doctor visits (IRR=1.036, p<0.01), number of outpatient tests and procedures (IRR=1.044, p<0.01), number of OTC medications (IRR=1.038, p<0.05), and to a lesser extent, prescription medications (IRR=1.023, p<0.10). As expected, caregivers who were older had more doctor visits (IRR=1.068, p<0.10), outpatient tests and procedures (IRR=1.100, p<0.05), and prescription (IRR=1.061, p<0.10) and OTC (IRR=1.102, p<0.05) medications. In addition, female caregivers and caregivers with more schooling also had more doctor visits and OTC medications, although their effects were only marginally significant.

Care recipients’ characteristics were largely not associated with the intensity of caregivers’ medical care use, except for care recipient’s chronic conditions (IRR=1.094, p<0.10) and DS score (IRR=1.046, p<0.05), which increased the number of outpatient tests and procedures the caregivers had.

Multivariate Results on Costs

By far the most influential variable that affects caregivers’ total medical care is caregiver’s own chronic conditions (Table 5). Each additional chronic condition was associated with a 33.2% increase in total medical care cost. Caregivers’ chronic conditions also was associated with a 13.4% increase in the cost of doctor visits and 33.3% increase in the cost of prescription medications (both p<0.05). Increases in caregivers’ depressive symptoms, as measured by the BSI-depression subsection score, also was associated with increases in caregiver’s medical care cost. Each point increase in caregivers’ BSI score was associated with a 3.6% increase in the cost of doctor visits, a 10% increase in the cost of outpatient treatment/procedures (both p<0.01), and a 4% increase in total medical care cost (p<0.10). Care recipients’ characteristics were largely not associated with caregivers’ medical care cost, except for care recipient’s chronic conditions and DS score, which was marginally associated with higher cost of outpatient tests and procedures (p<0.10).

Table 5.

Generalized Linear Mixed Regression Results on Caregivers’ Medical Care Costs (Dependent Variable = log cost).

	Total Costs	Doctor Visits	Outpatient Treatments	Prescription Medications	Over the Counter (OTC) Medications
Caregiver	Coeff. (SE)	Coeff. (SE)	Coeff. (SE)	Coeff. (SE)	Coeff. (SE)
Age in 5 year increments	0.078	0.041 *	0.074	0.044	0.082
	(0.057)	(0.025)	(0.056)	(0.057)	(0.068)
Female	−0.141	0.252 **	0.407	−0.251	0.521
	(0.284)	(0.122)	(0.270)	(0.283)	(0.339)
Year of schooling	−0.032	0.022	0.003	−0.008	0.061
	(0.038)	(0.016)	(0.034)	(0.038)	(0.043)
Baseline comorbid conditions	0.332 **	0.134 **	−0.038	0.333 **	−0.067
	(0.158)	(0.062)	(0.141)	(0.159)	(0.173)
Brief symptom inventory (BSI)	0.042 *	0.036 ***	0.101 ***	0.021	0.017
	(0.024)	(0.013)	(0.028)	(0.019)	(0.029)
Spouse caregiver	0.444	0.007	−0.096	−0.106	0.444
	(0.299)	(0.135)	(0.283)	(0.296)	(0.380)
Care Recipient
Baseline comorbid conditions	0.057	0.000	0.150 *	−0.033	−0.104
	(0.106)	(0.041)	(0.093)	(0.110)	(0.127)
Dependence Scale (DS)	−0.003	−0.007	0.063 **	−0.013	0.016
	(0.033)	(0.017)	(0.036)	(0.029)	(0.042)
Depressed	0.121	−0.205	0.342	0.403	−0.422
	(0.345)	(0.139)	(0.321)	(0.366)	(0.449)
Presence of psychiatric problems	0.267	0.064	0.277	−0.065	−0.030
	(0.171)	(0.092)	(0.208)	(0.141)	(0.215)
Number of observations	368	296	264	276	190

^*p<0.10,

^**p<0.05,

^***p<0.01

DISCUSSION

Providing care to individuals with dementia often has negative health consequences for caregivers. In this study, we extend Pearlin’s model to examine caregivers’ own healthcare use in a longitudinal setting. We followed a sample of dementia caregivers who participated along with their care recipients in a natural history study of dementia and estimated in a set of reduced form models overall effects of caregivers’ own socio-demographic and health characteristics and care recipients’ clinical characteristics on caregivers’ medical care use and costs. Not surprisingly, the most consistent predictors of caregivers’ medical care use and costs were caregivers’ own physical and mental health. Higher comorbid conditions were associated with caregivers’ increased likelihood of hospitalizations and doctor visits, and also with increases in the number of doctor visits and prescription medications and their related costs. Each additional comorbid condition was associated with a 33% increase in the caregivers’ total medical care cost. Consistent with existing literature ^[43], we found caregivers’ depressive symptoms were independently associated with caregivers’ likelihood of using prescription and OTC medications, number of doctor visits and outpatient tests and procedures as well as related costs.

The relationship between caregivers’ depressive symptoms and healthcare cost is of note. Improvements on caregiver depression have been consistently reported in dementia caregiver intervention programs.^{[44, 45]} However, little is known of the effects of reduced depressive symptoms on caregiver healthcare costs. This lack of information has impeded widespread adoption of successful interventions. While the Predictors Study is a natural history study of dementia, our results suggest that improvements in caregiver depressive symptoms are associated with modest reductions in multiple domains of caregivers’ use medical care and costs and fill an important gap toward understanding intervention's full benefit.

In addition to caregivers’ own characteristics, our results showed an independent effect of caregiving on caregivers’ healthcare use. After controlling for caregivers’ own socio-demographic and health characteristics, objective stressors of care recipients’ disease severity and clinical characteristics remain important predictors of caregivers’ medical care utilization. In particular, care recipients’ comorbidities and dependence on others were associated with increases in caregivers’ likelihood of doctor visits and outpatient tests and procedures. Presence of psychiatric symptoms in the care recipient was associated with an almost two-fold increase in caregivers’ likelihood of using prescription medications. These results are consistent with an earlier study that showed dementia patients’ behavioral and psychological problems and functional status were associated with caregivers’ emergency care and hospitalizations.^[19] Interestingly, both studies also reported that cognitive impairment itself was not associated with caregiver’s medical care use. These results are consistent with the literature that behavioral symptoms in dementia patients are important predictors of caregiving time, caregiver burden, quality of life, and other negative outcomes.^{[22, 46-48]} In other studies however, behavioral problems have been shown to be unrelated to caregiver services use, and the effects of care recipient behavioral problems were small compared to the effects of ADL problems.^[14] Together these results highlight the difficulties of managing behavioral symptoms of dementia, and suggest that behavioral symptoms have differential effects on different dimensions of caregiver health and healthcare use.

Our study did not however, find strong associations between care recipient characteristics and caregivers’ intensity of medical care use and costs. Care recipients’ dependence on others had a small significant effect on the number and costs of outpatient tests and procedures caregivers had. Care recipients’ comorbidities also were only marginally associated with the number and costs of outpatient tests and procedures caregivers had. It is unclear why care recipient characteristics were not more strongly associated with caregivers’ intensity of medical use.^[49] As a secondary analysis, we examined high intensity use using several different cut off points (e.g., ≥3, ≥5 prescription medications) and found substantively similar results. These results are consistent with an earlier report comparing costs of hospitalization and doctor visits using data from a sample of caregivers of veterans with dementia, to those from non-caregivers with similar socioeconomic characteristics.^[17] In the current sample of caregivers whose provision of caregiving hours were relatively low, perhaps the burden of caregiving has not led the caregivers to neglect their own medical care needs. Perhaps this sample of well-educated caregivers have more resources available to them, limiting our ability to detect differences in their health care use and expenditures. It is worth noting that these data are based on caregiver self-report of utilization and are subject to recall bias. Such non-differential misclassification is likely to result in reducing estimated effect. It may also be possible that the overall burden of caregiving is more likely to impact overall utilization patterns than intensity of use. It also is possible that, instead of caregiver and patient characteristics independently affecting caregiver healthcare utilization as modeled in this study, caregiver health (e.g., chronic conditions) is a consequence of patients’ dementia. More research in this area is clearly needed.

There are several limitations to our study. First, patients were selected from tertiary care university hospitals and specialized AD centers and are a nonrandom sample of individuals affected by the disease in the general population. Patients in our sample were predominantly white, highly educated, and healthier than community based samples. Caution is needed in generalizing the results of this study to patients of other ethnicities, patients at lower education and income levels, and to community AD patients. Second, it should be noted that for analysis of healthcare use and costs, our sample may be somewhat under powered. With this in mind, we attempted to include a parsimonious set of independent variables in our estimation models. Third, because the Predictors Study is a natural history study of dementia, no interventions were conducted, and comparison data on individuals without dementia or their caregivers are not available. Therefore we chose to estimate a set of reduced form equations, providing estimates of total effects of exogenous patient and caregiver variables on caregiver healthcare use. Several hypotheses have been generated from Pearlin’s model. On the one hand, increasing caregiving demands, measured by care recipients’ clinical characteristics, could negatively affect caregiver health through wear and tear, thereby increase caregivers’ healthcare use. On the other hand, if caregivers adapt to their caregiving role, care recipients’ clinical characteristics may not strongly affect caregivers’ health or healthcare use. It is also possible that caregivers may ignore their own health and health care use because of high cost of care and the tremendous amount of time devoted to care provision. Testing these structural mechanisms of caregiving is beyond the scope of the current study. It is worth noting that in this cohort of caregivers, healthcare utilization remained stable over time, suggesting adaptation in caregiving, or more complex mechanisms that affect caregiving healthcare use are present. Important next steps include examining the structural mechanisms of caregiving, examination of caregiver healthcare use outcomes through intervention studies, and including comparison groups to estimate the effects of caregiving on caregiver healthcare use due to dementia.

The ultimate goal of health services research is to improve health outcomes of care recipients and caregivers. With the aging of the population and increases in the size of population with dementia, the effects of caregiving on caregiver health and healthcare use will continue to be important at both the individual and societal level. Beyond the direct measures and costs of dementia patients’ healthcare utilization, results from this study suggest that there is a direct effect of dementia on caregivers’ healthcare use. It is imperative that we expand our focus on dementia patients to include family caregivers. In terms of community services provided to the caregivers, access to and involvement of a social worker has been shown to be important in linking caregivers to available support. Primary care providers, although central to coordinating healthcare delivery to the care recipients, could be an important link in integrating caregivers’ needs while planning and delivering healthcare services to patients. Clinical interventions that treat the patients and their caregivers as a whole will most likely achieve the greatest beneficial effects.

Systematic review

Providing care to individuals with dementia can often have negative health consequences. As a result, economic and health consequences to dementia caregivers can be substantial. However, most studies have focused on the patient. With few exceptions, caregiver healthcare utilization has received little attention. The goal of this study is to examine the effects of caregiver and patient characteristics on caregivers’ medical care use and cost.

Interpretation

In this current study, we take advantage of a rare opportunity of having both caregiver and care recipient data in a longitudinal setting, and examine whether, and how caregivers’ medical care use and related costs are associated with their own health and socio-demographic characteristics as well as care recipients’ increasing dementia severity and clinical characteristics. We hypothesize that caregivers’ medical care use and costs will increase as care recipients’ health declines, and the effects on caregivers’ healthcare use will vary by objective stressors and by domain of care. Our results show that patients’ comorbid conditions and dependence were associated with increased healthcare use and costs of caregivers. Improvements in caregiver depressive symptoms are associated with modest reductions in multiple domains of caregivers’ use medical care and costs.

Future directions

Findings suggest that we should expand our focus on dementia patients to include family caregivers to obtain a fuller picture of effects of caregiving. For example, while improvements on caregiver depression have been consistently reported in dementia caregiver intervention programs, little is known of the effects of reduced depressive symptoms on caregiver healthcare costs. This lack of information has impeded widespread adoption of successful interventions. However, because the Predictors Study is a natural history study of dementia, no interventions were conducted, and comparison data on individuals without dementia or their caregivers are not available. Therefore we chose to estimate a set of reduced form equations, providing estimates of total effects of exogenous patient and caregiver variables on caregiver healthcare use. Important next steps include examination of caregiver healthcare use outcomes through intervention studies, and including comparison groups to estimate the effects of caregiving on caregiver healthcare use due to dementia.

Table 4.

Random Effects Negative Binomial Regression Results on Caregivers’ Intensity of Medical Care Use Over Time (n=394).

	# Doctor Visits	# Outpatient Treatment	# Prescription Medications	# Over the Counter (OTC) Medications
Caregiver	IRR (SE)	IRR (SE)	IRR (SE)	IRR (SE)
Age in 5 year increments	1.068 *	1.100 **	1.061 *	1.102 **
	(0.038)	(0.033)	(0.036)	(0.050)
Female	1.378 *	1.154	0.948	1.538 *
	(0.239)	(0.174)	(0.159)	(0.345)
Year of schooling	1.048 **	0.992	0.971	1.054 *
	(0.024)	(0.018)	(0.020)	(0.029)
Baseline comorbid conditions	1.271 ***	1.073 *	1.432 ***	1.121
	(0.110)	(0.079)	(0.121)	(0.128)
Brief symptom inventory (BSI)	1.036 **	1.044 ***	1.023 *	1.038 **
	(0.016)	(0.015)	(0.014)	(0.020)
Care Recipient
Baseline comorbid conditions	1.076	1.094 *	1.034	0.970
	(0.065)	(0.055)	(0.063)	(0.080)
Dependence Scale (DS)	1.026	1.046 **	1.017	0.995
	(0.023)	(0.021)	(0.020)	(0.028)
Depressed	0.999	0.938	0.864	0.699
	(0.205)	(0.174)	(0.178)	(0.199)
Presence of psychiatric problems	1.010	1.154	1.169	0.899
	(0.123)	(0.130)	(0.118)	(0.142)
Spouse Caregiver	1.072	1.150	1.398 **	1.862 ***
	(0.203)	(0.184)	(0.244)	(0.436)

^*p<0.10,

^**p<0.05,

^***p<0.01