Development and Validation of Caregiver Risk Evaluation (CaRE): A New Algorithm to Screen for Caregiver Burden

Abstract

Objective: The main objective was to develop a decision-support tool to assess the risk of caregiver burden, the Caregiver Risk Evaluation (CaRE) algorithm. Methods: Home care clients were assessed using the Resident Assessment Instrument for Home Care (RAI-HC). Their caregiver completed the 12-item Zarit Burden Interview (ZBI), the main dependent measure, which was linked to the RAI-HC. Results: In the sample (n = 344), 48% were aged 85+ years and 61.6% were female. The algorithm can be collapsed into four categories (low, moderate, high, and very high risk). Relative to the low-risk group, clients in the very high-risk group had an odds ratio of 5.16 (95% confidence interval: [2.05, 12.9]) for long-term care admission, after adjusting for client age, sex, and regional health authority. Discussion: The CaRE algorithm represents a new tool to be used by home care clinicians as they proactively plan for the needs of clients and their caregivers.

Keywords

analysis-regression models caregiving-informal home and community based care and services interRAI

Unpaid caregivers in Canada form the backbone of care provided to older adults in the community. About 18% of Canadians act as informal caregivers (e.g., family/friends—referred to simply as “caregivers” throughout the remainder of this article) to older adults (Canadian Institute for Health Information, 2010). Many caregivers provide this support willingly and find it to be a positive experience (Kruithof et al., 2012); however, as care needs increase, many also find it stressful, making it difficult to continue. In particular, caregivers report that their caregiving responsibilities can negatively affect their ability to manage their working lives and their social/leisure activities (Health Quality Ontario, 2016).

Caregiver burden has been described as a negative affective outcome which results as a perceived inability to meet the demands of the caregiving role (Garlo et al., 2010; Miller et al., 2012). It encompasses the physical, psychological, emotional, social, or financial problems experienced by caregivers (Zarit et al., 1980). Caregiver burden is correlated with a deterioration in the caregiver’s mental (Duggleby et al., 2016) and physical health (Vitaliano et al., 2003) and is a strong risk factor for the care recipient’s admission to a long-term care (LTC) facility (Cepoiu-Martin et al., 2016; Gaugler et al., 2009; Rockwood et al., 2014; Verbeek et al., 2015). Furthermore, the Alzheimer Society of Canada forecasts a 7% increase in dementia among older adults living in the community over the next 30 years (Alzheimer Society of Canada, 2010). This fact, together with population aging and an increase in the number of older adults living with chronic conditions, will result in more families and friends providing care to their loved ones making it critical to identify caregivers who may be experiencing the early signs and symptoms of burden.

Several assessments exist to measure caregiver burden (Vitaliano et al., 1991) with the Zarit Burden Interview (ZBI) being used most often (Mosquera et al., 2016). The original ZBI has 22 items and was developed for use with caregivers of individuals with Alzheimer’s dementia (Zarit et al., 1980). It has documented validity and reliability (Bachner & O’Rourke, 2007; Hebert et al., 2000), and higher scores are correlated with an increased risk of admission to an LTC facility (Miller et al., 2012). It is considered to be the most useful instrument based on its validity, ease of use, and ability to predict change over time (Van Durme et al., 2012).

A shorter 12-item version is also widely used and was chosen for the purposes of this study. This shorter version has strong internal consistency (Bedard et al., 2001; O’Rourke & Tuokko, 2003), is highly correlated with the longer version (Bedard et al., 2001) and with dementia severity (Branger et al., 2016), and has demonstrated predictive validity for caregiver depression (O’Rourke & Tuokko, 2003). The 12-item version has also been evaluated among older adults with advanced cancer and among older people with no cognitive challenges (Higginson et al., 2010; Iecovich, 2012), and in both studies, the results were positive in terms of internal consistency and validity.

Given the increasing demand for home care and individual preferences to remain at home and “age in place,” there is the potential for increased reliance on caregivers. This can lead to the caregiver experiencing negative health-related outcomes (Duggleby et al., 2016). As such, there is a growing need to find efficient ways to assess caregivers and flag those, within the home care sector, who may be at risk.

In Canada, the Resident Assessment Instrument for Home Care (RAI-HC) is being used in seven provinces/territories, in multiple states in the United States, and in multiple countries around the world (e.g., the United Kingdom, Belgium, Germany, Australia). The RAI-HC is a standardized assessment that is useful in assisting home care clinicians in care planning and in making decisions regarding placement in LTC. Completing these assessments represents a significant investment in time and human resources for the participating region or country. As such, it is important to find ways in which to maximize the efficient use of these data. One way to do this is to create electronic decision-support tools for use by clinicians. This would then eliminate the need to complete additional caregiver burden assessments thus reduce the assessment burden for both clinicians and informal caregivers. This study sought to develop and validate a new decision-support tool, termed the Caregiver Risk Evaluation, or CaRE, to estimate the risk of caregiver burden among home care clients, using items from the RAI-HC.

Design and Methods

Population, Setting, and Study Context

This study utilized secondary data collected for home care clients using the RAI-HC, a government-mandated assessment in both the Winnipeg Regional Health Authority (WRHA) in Manitoba and Island Health (located on Vancouver Island, BC) (British Columbia Ministry of Health, 2016; Toews, 2016). It contains just over 300 items, including domains such as cognitive and sensory functioning, communication, diagnoses, and functional abilities. Individual items on the RAI-HC are valid and reliable (Landi et al., 2000; Poss et al., 2008). Assessments are completed in a client’s home by trained assessors (typically a registered nurse) following guidelines published by interRAI, the nonprofit group that holds the copyright to the instrument. The assessor interviews the client and their caregiver, can consult with other health care professionals (e.g., primary care physician), or observe client’s abilities, as needed. Assessments are completed every 6 to 12 months after admission to the home care service, or after a significant change in clinical status.

Home care in Canada typically offers services such as health promotion, rehabilitation, support and maintenance, and end-of-life care (Accreditation Canada and the Canadian Home Care Association, 2015). All provinces and territories provide home care to their residents; however, policies vary regarding eligibility and service intensity, as home care is not covered under the national Medicare program. In general, a referral from a physician is not required to receive home care. In most regions, a trained case manager (usually a registered nurse) will complete a needs assessment and determine the level and intensity of care based on their clinical judgment. It is possible that the family could assume some financial responsibility for care that is not covered by the publicly funded system. This would likely depend, in large part, on the level of home care provided and whether the family felt that this was adequate.

Within Island Health, the ZBI was routinely used in the two sites involved in the study (e.g., Victoria, Campbell River), regardless of the cognitive status of the home care client. As it was used as part of routine clinical practice, consent from informal caregivers was not required. However, this was not the case within WRHA. During the study, a designated staff member approached the home care client’s primary caregiver about the project to see if they would be interested in completing the ZBI. They were selected based on the fact that their relative was receiving home care and the RAI-HC assessment was scheduled to be completed in the near future. This was done to ensure that the RAI-HC was completed within approximately 1 month of the ZBI. No home care clients were excluded based on their cognitive status. As completion of the ZBI was not part of the WRHA’s normal clinical practice, individual consent was obtained from each primary caregiver.

Description of Key Outcome Measures

The main outcome measure was the score on the ZBI, which was used as the dependent variable in all decision tree analyses. The client’s primary caregiver completed the 12-item ZBI. Each item is scored from zero to four for a maximum score of 48. Higher scores indicate greater levels of perceived burden (Bedard et al., 2001). A secondary outcome was the client’s status at 12 months following the RAI-HC assessment. Based on the literature, we anticipated that there would be a relationship between the risk of admission to an LTC facility (dependent variable) and the level of risk for caregiver burden, based on the CaRE algorithm. Both health authorities tracked client status to determine if the person was still receiving home care at that time, had died, or was placed in an LTC facility. The health authorities sent de-identified RAI-HC data, linked with both the items on the ZBI, and also linked to the client’s status at 12-months, to the research team.

Sample

The sample included 344 unique home care clients (226 from WRHA and 118 from Island Health), aged 65 years and older. Within WRHA, a total of 297 caregivers were approached and 230 consented and completed the ZBI, either in person or over the phone. Within Island Health, the ZBI was completed as part of normal practice, either in person or on the telephone. Evidence suggests that reliability is good when using either type of administration (Lin et al., 2017; Tremont et al., 2013). Four clients in the WRHA did not have a timely RAI-HC assessment to link to the ZBI and were removed from the study. If a client had multiple RAI-HC assessments within Island Health, the assessment that was closest in date to the completion of the ZBI was selected. Data collection took place between April 2010 and March 2016.

Health Index Scales and Other Items From the RAI-HC

There are six validated health index scales generated from items within the RAI-HC that were used in the analysis. Higher scores on these scales indicate a greater degree of impairment.

Activities of Daily Living Self-Performance Hierarchy (ADL-SHS) is a 6-point scale that measures performance on items such as toilet use and personal hygiene. A cut point of two or greater indicates at least mild functional impairment. The ADL-SHS has been validated against the Barthel Index (Morris et al., 1999).

Instrumental Activities of Daily Living (IADL) Involvement Scale is a 21-point scale that assesses items such as independence in housework, managing finances, and medication management. A cut point of 14 or higher was used to indicate at least moderate impairment in performing these tasks, in line with previous studies (Guthrie et al., 2018; Williams et al., 2018).

Cognitive Performance Scale (CPS) is comprised of four items (cognitive skills for daily decision-making, short-term memory, making self understood, and eating) and ranges from zero to six, where a cut point of two or more represents at least mild cognitive impairment. The CPS has been validated against the Mini-Mental State Examination (MMSE) (Morris et al., 1994) and the Montreal Cognitive Assessment (Jones et al., 2010).

Depression Rating Scale (DRS) is a 14-point summative scale that has been validated against the Cornell Scale for Depression and the Hamilton Depression Scale (Burrows et al., 2000). A score of 3+ is typically used as a cut point as it can accurately predict a clinical diagnosis of depression (Martin et al., 2008). In the current analysis, we considered this cut point, but also others (e.g., DRS of 0 vs. 1+), as any score above zero indicates the presence of signs/symptoms of depression.

Pain Scale measures the frequency and intensity of pain on a 4-point scale ranging from 0 (no pain) to 4 (daily/severe) pain, where a cut point of two or greater indicates daily pain (Fries et al., 2001).

Changes in Health, End-Stage Disease and Signs and Symptoms (CHESS) scale assesses the level of health instability. The scale ranges from 0 (no instability) to 5 (severe health instability), and for each one-point increase on the CHESS scale, there is nearly a twofold increased risk of death (Hirdes et al., 2003). A score of two or more was used to indicate the presence of at least moderate health instability (Hirdes et al., 2008).

The Method for Assigning Priority Levels (MAPLe) score represents a decision-support tool to prioritize the need for community-based care or LTC. It can also be generated directly using items from the RAI-HC (Hirdes et al., 2008). The MAPLe can range from one to five, with higher scores indicative of higher priority/greater urgency. We collapsed the MAPLe score into two groups as a number of the groups had small sample sizes (<5). A score of four or higher was used to identify clients with a higher priority/greater urgency.

Several other individual items on the RAI-HC provide detailed information regarding the client’s primary caregiver, including their relationship to the client (spouse/child/neighbor), whether they reside with the client (yes/no), whether they experience feelings of distress, anger or depression, whether the caregiver or the client feels that the client would be better off in another living environment, and the number of informal care hours provided in the past week. We examined several dichotomous variables, including the presence of responsive behaviors (e.g., wandering), urinary incontinence (yes/no), and a diagnosis of dementia (i.e., a diagnosis of Alzheimer’s disease or another type of dementia). We also examined whether the client experienced an overall change in their care needs compared with 90 days ago, or since their last assessment (no change/improved and receiving less support/deteriorated and receiving more support).

Development of the Algorithm and Analyses

The algorithm was created in four main steps as outlined below.

Step 1

Demographic characteristics, the health index scales, the MAPLe score, characteristics of the caregiver, diagnoses, and whether there was a change in the client’s health status were examined in relation to the ZBI score. The t tests and a one-way analysis of variance (ANOVA) were used to calculate p values and determine statistical significance as the dependent variable (ZBI score) is a continuous measure. All significant variables, based on a two-tailed alpha level of .05, and those identified in the literature, were considered for inclusion in the algorithm.

Step 2

The algorithm was created using interactive decision tree software, such that each branch is informed from an explanatory variable, drawn from candidate items on the RAI-HC assessment, as described above. The dependent variable was the score on the ZBI. Decision trees are well-suited to analytic situations where interactions among explanatory variables are expected, such that subgroups of cases, based on a particular characteristic, can be modeled separately from others (Batra & Agrawal, 2018; Gulati et al., 2016).

The tool in SAS Enterprise Miner (SAS Institute Inc., 2016) presents the variables for a given potential split by rank order of the Gini coefficient, which provides a measure of the degree of inequality of the resulting groups. At each step in the tree development, the software automatically recommends splits for independent variables being considered, but one can override these suggestions. As such, the development of the algorithm was guided both by the statistical properties of the model and by clinical relevance through input from home care clinicians and informal caregivers (described in Step 3).

To protect from over-fitting that can occur with small datasets that may contain nonrepresentative sampling, a random 25% sample of the cases was held back to provide some independent confirmation of each split. The results of this 25% validation sample varied depending on the splitting variable and location in the various decision trees that were constructed. As the splitting process is manually controlled, the operator sees the evidence for a given split before accepting it, or considering a different splitting variable. For splits where both 75% and 25% samples showed similar results, this was taken as evidence that the choice was appropriate and unlikely to be influenced by a few chance observations. In the case where the 25% sample differed, this was cause to consider another item for this split. Just over a dozen tree variants were produced which were examined by the investigators for their face validity and explanatory power.

Step 3

We held four separate discussion groups with home care clinicians and caregivers to elicit feedback on the potential variables that should be included in the final algorithm. A total of 32 case managers and eight caregivers participated.

Based on the preliminary tree models, 23 potential variables, from the RAI-HC, were presented to the home care clinicians and caregivers who provided feedback on each, namely whether or not they felt they were important and if anything critical was missing from the list. These discussions highlighted the importance of cognitive impairment, incontinence, responsive behaviors, hours of informal care and impairments in ADLs (e.g., bathing, dressing), and IADLs (e.g., telephone use, managing medications). Three of these areas (i.e., ADL impairment, IADL impairment, and responsive behaviors) were related to the ZBI score but were not kept in the final algorithm. This is likely explained by the fact that they added little explanatory power above and beyond the CPS, a measure of cognitive impairment, which was included in the final algorithm. They also felt the age and sex of the client and certain diagnoses were not as important, and none of these variables were included in the final algorithm.

Step 4

The research team used all of this input to create the final version of the algorithm. The final CaRE algorithm contained nine terminal nodes, which clearly fell into four distinct categories based on the mean Zarit score (mean Zarit score near 10, near 14, near 20, and near 25), which are provided for each node in Figure 1. This suggested a four-level scale (low risk, moderate, high, and very high). A larger sample might have produced a tree with more nodes, allowing for a scale with additional distinct points, but given the limitations of the sample, a 4-point scale was considered to be reasonable. A summary of how the items were used to create each of the four risk group is shown in Table 1.

Figure 1.

Schematic representation of the CaRE algorithm.

Table 1.

Description of the Items in the Algorithm and How They Are Used to Populate the CaRE Risk Groups.

	CaRE risk group^a
Items in the algorithm	Low (n = 81)	Moderate (n = 115)	High (n = 103)	Very high (n = 45)
Caregiver does not feel distressed, angry, or depressed	√	√	√
Caregiver feels distressed, angry, or depressed			√	√
Caregiver does not live with the client	√	√	√
Caregiver lives with the client		√	√	√
Caregiver is a spouse or child	√	√	√
Caregiver is another relative or neighbor	√
Caregiver provides <7 hours of care each week	√
Caregiver provides 7+ hours of care each week		√	√
Home care client has no symptoms of depression		√
Home care client has symptoms of depression			√
Home care client has none or only mild cognitive impairment			√
Home care client has moderate to severe cognitive impairment				√

Note. CaRE = Caregiver Risk Evaluation.

A check mark indicates that the item was used in the creation of that particular CaRE risk group.

We anticipated that there would be a relationship between the risk of admission to an LTC facility (dependent variable) and four-level CaRE algorithm. To examine this, we calculate the odds ratio (OR) and 95% confidence interval (CI) for LTC admission as a function of the CaRE risk groups, both unadjusted, and after adjusting for age, sex, and health authority. All analyses were completed using SAS Enterprise Guide, version 7.1 (SAS Institute Inc., 2016) and SAS Enterprise Miner, version 13. A two-tailed alpha level of .05 was used to determine statistical significance. The study followed the Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD) guidelines (Collins et al., 2015). The design of the study was reviewed and approved by the Research Ethics Board at Wilfrid Laurier University (REB #4085), the Health Research Ethics Board at the University of Manitoba (REB#: HS21587), the WRHA Research Access and Approval Committee (RAAC #2015-025), and the Health Research Ethics Board in Island Health (REB#: H2014-115).

Results

The mean time between a client’s RAI-HC and ZBI assessment was 36.7 days (SD = 37.4 days). Just under half of the sample were aged 85+ (48.0%), 61.6% were female, and 39.4% were married. Across four of the health index scales (i.e., CPS, DRS, IADL, and ADL-SHS), clients with impairments in these areas had higher scores on the ZBI. For the other two scales (i.e., Pain Scale and CHESS), there was no significant association. The MAPLe score was also significantly related to the ZBI score (p < .0001). Higher scores on the ZBI were significantly associated with the caregiver co-residing with the client; the caregiver was a spouse; the caregiver was experiencing distress, anger, or depression; and the primary caregiver felt the client would be better off living in another environment. Caregivers providing at least 7 hours per week of care had higher ZBI scores versus those providing less care (Table 2).

Table 2.

Relationship Between Demographic Characteristics and Other Health Measures for the Home Care Clients and the Score on the ZBI.

Variable	% (n)	Mean ZBI score	p
Demographic characteristics of home care clients
Age group (years)
65–74	12.8 (44)	19.4	.03
75–84	39.2 (135)	17.5
85+	48.0 (165)	15.1
Sex
Male	38.4 (132)	18.2	.05
Female	61.6 (212)	15.6	.05
Marital status
Married	39.4 (135)	19.5	.0008
Never married	2.9 (10)	12.0
Widowed	46.4 (159)	15.0
Separated/divorced/other	11.4 (39)	14.6
Health index scales for clients
CPS
Intact/mild impairment (0–1)	53.2 (183)	14.5	.0002
Moderate/severe impairment (2–6)	46.8 (161)	19.0	.0002
DRS
No signs/symptoms of depression (0–2)	86.9 (299)	15.6	<.0001
Signs and symptoms of depression (3–14)	13.1 (45)	23.1	<.0001
ADL-SHS
No/mild impairment (0–1)	62.2 (214)	16.1	.03
Moderate/severe impairment (2–6)	37.8 (130)	19.4	.03
Instrumental Activities of Daily Living Involvement Scale
No/mild impairment (0–13)	41.3 (142)	14.2	.001
Moderate/severe impairment (14–21)	58.7 (202)	18.4	.001
Pain Scale
No pain/less than daily pain (0–1)	50.0 (172)	16.8	.76
Daily/severe pain (2–4)	50.0 (172)	16.5	.76
CHESS scale
Mild/moderate health instability (0–2)	85.5 (294)	16.4	.56
Severe health instability (3–5)	14.5 (50)	17.7	.56
MAPLe score
Low/mild/moderate (1–3)	32.3 (111)	13.2	<.0001
High/very high (4–5)	67.7 (233)	18.3	<.0001
Characteristics of the caregiving relationship
Living arrangement
Caregiver does not live with client	57.3 (197)	14.6	<.0001
Caregiver lives with client	42.7 (147)	19.4	<.0001
Caregiver’s relationship to client
Child or child-in-law	57.3 (197)	15.7	<.0001
Spouse	31.1 (107)	20.2
Other relative	6.4 (22)	13.2
Friend/neighbor	5.2 (18)	9.7
Primary caregiver experiences feelings of distress, anger, or depression
No	75.9 (261)	14.4	<.0001
Yes	24.1 (83)	23.5	<.0001
Primary caregiver feels that client would be better off in another living environment
No	81.1 (240)	15.0	.0003
Yes	18.9 (56)	21.5	.0003
Hours of informal care in the previous week^a
0–6 hr	24.4 (84)	10.8	<.0001
7–130 hr	75.6 (260)	18.5	<.0001
Responsive behaviors among clients
Wandering
No	96.2 (331)	16.5	.72
Yes	3.8 (13)	18.6	.72
Verbally abusive
No	96.5 (332)	16.4	.2
Yes	3.5 (12)	21.7	.2
Socially inappropriate behavior
No	97.1 (334)	16.4	.04
Yes	2.9 (10)	25.5	.04
Resists care
No	90.7 (312)	16.2	.05
Yes	9.3 (32)	20.3	.05
Diagnosis of a stroke
No	81.7 (281)	16.7	.74
Yes	18.3 (63)	16.1	.74
Other items reflecting client status
Diagnosis of dementia
No	58.1 (200)	15.4	.007
Yes	41.9 (144)	18.3	.007
Reduced social interactions
No decline	65.4 (225)	15.3	.01
Declined, client is not distressed	26.7 (92)	19.0
Declined, client is distressed	7.8 (27)	19.6
Overall change in client’s care needs
No change or improved and person receives fewer supports	48.8 (168)	14.6	<.0001
Declined and person receives more supports	51.2 (176)	18.6	<.0001
Bladder incontinence
No	44.2 (152)	15.4	.09
Yes	55.8 (192)	17.6	.09

Note. ZBI = Zarit Burden Interview; CPS = Cognitive Performance Scale; DRS = Depression Rating Scale; ADL-SHS = Activities of Daily Living Self-Performance Hierarchy Scale; CHESS = Changes in Health, End-Stage Disease and Signs and Symptoms; MAPLe = Method for Assigning Priority Levels; CaRE = Caregiver Risk Evaluation.

This cut point was chosen to be in line with the way in which the variable is presented in the CaRE algorithm.

The final CaRE algorithm included whether or not the caregiver experienced distress, anger, or depression as the initial split and captured information about the caregiver’s living situation, whether the caregiver was a spouse of a child, hours of informal care, symptoms of depression, and cognitive impairment. The mean ZBI score across the nodes ranged from 9.0 to 24.9. The final model explained 22.8% of the variation in the ZBI score, which was the highest level of explained variance of any of the decision tree models that were considered.

Compared with the low-risk CaRE group, clients in the very high-risk group were more likely to experience difficulties completing ADLs (33.3% vs. 8.6%), and the primary caregiver and the client were both more likely to feel that the client would be better living in a different environment (51.2% vs. 9.6%). A similar pattern was also found for the MAPLe score. In the low-risk CaRE group, only 51.8% of clients fell within the high MAPLe category (score of 4–5), compared with 95.6% in the very high-risk group (Table 3). Finally, we found a higher proportion of clients in the very high-risk group were residing in LTC after 12 months (37.8%) compared with 15% in the low-risk group (Table 4). After adjusting for age, sex, and health authority, the odds of LTC admission for clients in the very high-risk group was 5.16 times greater than for those in the low-risk group (OR = 5.16; CI = [2.05, 12.9]; Table 5).

Table 3.

Comparison of Selected Variables by CaRE Risk Group.

Items/scales from the RAI-HC	CaRE risk group
Items/scales from the RAI-HC	Low (n = 81)	Moderate (n = 115)	High (n = 103)	Very high (n = 45)	p
Mean ZBI score (SD)	9.8 (8.0)	14.4 (9.6)	20.9 (9.3)	24.9 (11.7)	<.0001
	% (n)
Health index scales for clients
ADL impairment (ADL-SHS score of 3–6)	8.6 (7)	11.3 (13)	21.4 (22)	33.3 (15)	.001
IADL impairment (IADL Involvement Scale score of 14+)	33.3 (27)	60.9 (70)	63.1 (65)	88.9 (40)	<.0001
Cognitive impairment (CPS score of 2+)	51.9 (42)	68.7 (79)	69.9 (72)	100.0 (45)	<.0001
Depressive symptoms (DRS score of 3+)	6.2 (5)	0.0 (0)	22.3 (23)	37.8 (17)	<.0001
MAPLe score for clients
Low/mild/moderate (1–3)	48.2 (39)	37.4 (43)	26.2 (27)	4.4 (2)	<.0001
High/very high (4–5)	51.8 (42)	62.6 (72)	74.8 (76)	95.6 (43)	<.0001
Characteristics of the caregiving relationship
Client would be better off in another living environment
Neither client nor caregiver agrees with this statement	90.4 (66)	93.2 (96)	73.1 (57)	48.8 (20)	<.0001
Client and caregiver both agree with this statement	9.6 (7)	6.8 (7)	26.9 (21)	51.2 (21)	<.0001
Mean (SD) hours of informal support in the last week	5.8 (9.6)	21.5 (21.5)	23.0 (23.5)	40.3 (29.8)	<.0001
Caregiver lives with client	0.0 (0)	53.0 (61)	39.8 (41)	100 (45)	<.0001

Note. CaRE = Caregiver Risk Evaluation; ZBI = Zarit Burden Interview; ADL-SHS = Activities of Daily Living Self-Performance Hierarchy; CPS = Cognitive Performance Scale; DRS = Depression Rating Scale.

Table 4.

Outcome Status 12-Months Post RAI-HC Assessment by CaRE Risk Group.

Client discharge status	CaRE group^a
Client discharge status	All clients (n = 344)	Low (n = 80)	Moderate (n = 115)	High (n = 102)	Very high (n = 45)	p
	% (n)
Still receiving home care	50.9 (174)	70.0 (56)	53.9 (62)	47.1 (48)	17.8 (8)	.001
Discharged elsewhere^b	13.5 (46)	7.5 (6)	16.5 (19)	14.7 (15)	13.3 (6)
Waiting for LTC	3.5 (12)	0.0 (0)	5.2 (6)	2.0 (2)	8.9 (4)
LTC admission	19.6 (67)	15.0 (12)	11.3 (13)	24.5 (25)	37.8 (17)
Died	12.6 (43)	7.5 (6)	13.0 (15)	11.8 (12)	22.2 (10)

Note. RAI-HC = Resident Assessment Instrument for Home Care; CaRE = Caregiver Risk Evaluation; LTC = long-term care.

12-month outcome status was missing for two clients.

Includes clients no longer receiving home care services, clients admitted to assisted living/hospital.

Table 5.

Odds of Admission to LTC as a Function of CaRE Risk Group.

CaRE risk groups	Unadjusted model	Adjusted for client age and sex	Adjusted for client age, sex, and health authority
CaRE risk groups	Odds ratio (95% confidence interval)
Low	Reference	Reference	Reference
Moderate	0.72 [0.31, 1.68]	0.80 [0.34, 1.87]	0.77 [0.33, 1.80]
High	1.84 [0.86, 3.94]	2.01 [0.93, 4.36]	1.84 [0.83, 4.03]
Very high	3.44 [1.46, 8.13]	4.11 [1.64, 10.3]	3.68 [1.45, 9.30]

Note. LTC = long-term care; CaRE = Caregiver Risk Evaluation.

Discussion and Implications

The CaRE algorithm represents a new decision-support tool for home care, generated directly from items within the RAI-HC. It was developed with significant input from home care clinicians and caregivers, and differentiates the level of risk of caregiver burden for caregivers in home care. It represents a new measure that can be used along with other outputs from the RAI-HC (e.g., health index scales) to guide care planning and service provision. Like other scales and algorithms generated from the RAI-HC, it should be considered as a measure to support, but not as a replacement for clinical judgment. Case managers using the CaRE score should do so in conjunction with the client, their family, and other formal service providers to ensure a client-specific response tailored to the unique needs and preferences of the caregiver and care recipient dyad.

As the RAI-HC is currently used widely in Canada, and internationally, the CaRE algorithm represents a very efficient way for these providers to assess the risk of caregiver burden without administering additional caregiver assessments. This is not only an efficient use of the existing information, but would also minimize assessment burden for home care clients and their families. Finally, the CaRE algorithm opens the door to cross-national comparisons of caregiver burden, and its correlates, among those countries routinely using the RAI-HC.

The items within CaRE were chosen based on both clinical input and feedback from caregivers and represent issues that are supported in the literature. For example, client depression (deAlmeida Mello et al., 2017; Hirdes et al., 2012; Mohamed et al., 2010), the caregiver co-resides with the client (deAlmeida Mello et al., 2017; Kim et al., 2012; Melis et al., 2009; Pauley et al., 2018), cognitive status (deAlmeida Mello et al., 2017; Oldenkamp et al., 2016), and the number of hours of informal support (Kim et al., 2012; Park et al., 2015; Pauley et al., 2018; Sutcliffe et al., 2016) are embedded in the CaRE algorithm and are also well-established risk factors for caregiver burden or distress. The algorithm is strongly associated with the risk of future LTC placement, even after adjusting for client age, sex, and the health authority where care was received. In algorithm development, the first split is key. Although we evaluated many possible variables for this initial split, we decided to use the item identifying whether or not the caregiver feels distressed, angry, or depressed. This is evidence of the validity of this single item with respect to caregiver burden.

A key strength of our approach is the use of input from multiple sources. The creation of the CaRE algorithm was undertaken using rigorous statistical methods, in conjunction with input from the existing literature and the opinions of caregivers and home care professionals. However, a possible limitation to this work is our limited sample size from two regions of Canada. In our sample, roughly 42% had a diagnosis of dementia, about twice the rate among long-stay home care clients in Canada. When compared with a very large sample (n = 163,527) of home care clients in Ontario (Williams et al., 2018), ours was very similar on demographic characteristics, such as age, sex, and marital status, but our sample was more likely to have a CPS score of two or higher (69% vs. 53%) and was somewhat less likely to co-reside with their primary caregiver (43% vs. 51%).

Using RAI-HC data from across Canada (n = 131,000), the Canadian Institute for Health Information found strong relationships between caregiver distress and the CPS score, hours of informal support, and with the DRS score. All relationships had unadjusted ORs of two or higher (Canadian Institute for Health Information, 2010). So, although our sample may not truly represent home care in Canada, other research strongly supports the variables included in the algorithm. In future studies, we plan to assess the validity of the CaRE algorithm in the large sample of long-stay home care clients in Canada, and where feasible, using RAI-HC data from other countries.

Conclusion

The CaRE algorithm represents a new decision-support tool generated from items within the RAI-HC assessment. It differentiates home care clients based on the level of risk for caregiver burden and can potentially reduce or eliminate the need for additional caregiver assessments. It represents an efficient method to identify enhanced risk for caregiver burden and can act as a “flag” to home care and other professionals to enable them to understand which families may be in need of additional care, education/training, respite support, or other services. By quickly identifying the caregivers who are most at risk and linking them with the necessary supports and services, this could maximize the likelihood of the person remaining in their own home and reducing the need for admission to LTC. Future research could use this algorithm to flag caregivers who may be at risk as part of a larger intervention study or randomized trial.

Footnotes

Acknowledgements

The authors gratefully acknowledge the financial support provided by the Alzheimer Society of Canada. We also offer our sincere thanks to the staff and clinicians within Island Health and the Winnipeg Regional Health Authority who gave generously of their time and helped to make this project possible.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Ethical Approval

The design of the study was reviewed and approved by the Research Ethics Board at Wilfrid Laurier University (REB #4085), the Health Research Ethics Board at the University of Manitoba (REB#: HS21587), the WRHA Research Access and Approval Committee (RAAC #2015-025), and the Health Research Ethics Board in Island Health (REB#: H2014-115).

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Alzheimer Society of Canada (Grant Number 15-10).

ORCID iD

Dawn M. Guthrie

References

Accreditation Canada and the Canadian Home Care Association. (2015). Home care in Canada: Advancing quality improvement and integrated care. http://www.cdnhomecare.ca/content.php?doc=44

Alzheimer Society of Canada. (2010). Rising tide: The impact of dementia on Canadian society. https://alzheimer.ca/en/suroit/Get-involved/Advocacy/Latest-info-stats/Rising-Tide

Bachner

Y. G.

O’Rourke

(2007). Reliability generalization of responses by care providers to the Zarit Burden Interview. Aging & Mental Health, 11(6), 678–685.

Batra

Agrawal

(2018). Comparative analysis of decision tree algorithms. In Panigrahi

Hoda

Sharma

Goel

(Eds.), Nature inspired computing (Vol. 652, pp. 31–36). Springer.

Bedard

Molloy

D. W.

Squire

Dubois

Lever

J. A.

O’Donnell

(2001). The Zarit Burden Interview: A new short version and screening version. The Gerontologist, 41(5), 652–657.

Branger

O’Connell

M. E.

Morgan

D. G.

(2016). Factor analysis of the 12-Item Zarit Burden Interview in caregivers of persons diagnosed with dementia. Journal of Applied Gerontology, 35(5), 489–507. https://doi.org/10.1177/0733464813520222

British Columbia Ministry of Health. (2016). RAI-HC clinical standards and “best practice” guidelines. https://www2.gov.bc.ca/assets/gov/health-safety/home-community-care/accountability/pdf/final_rai-hc_guidelines_-_2016.pdf

Burrows

A. B.

Morris

J. N.

Simon

S. E.

Hirdes

J. P.

Phillips

C. D.

(2000). Development of an MDS-based Depression Rating Scale for use in nursing homes. Age and Ageing, 29, 165–172.

Canadian Institute for Health Information. (2010). Supporting informal caregivers: The heart of home care. https://secure.cihi.ca/free_products/Caregiver_Distress_AIB_2010_EN.pdf

10.

Cepoiu-Martin

Tam-Tham

Patten

Maxwell

C. J.

Hogan

D. B.

(2016). Predictors of long-term care placement in persons with dementia: A systematic review and meta-analysis. International Journal of Geriatric Psychiatry, 31, 1151–1171. https://doi.org/10.1002/gps.4449

11.

Collins

G. S.

Reitsma

J. B.

Altman

D. G.

Moons

K. G.

(2015). Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): The TRIPOD statement. Annals of Internal Medicine, 162(1), 55–63. https://doi.org/10.7326/M14-0697

12.

deAlmeida Mello

Macq

Van Durme

Ces

Spruytte

Van Audenhove

Declercq

(2017). The determinants of informal caregivers’ burden in the care of frail older persons: A dynamic and role-related perspective. Aging and Mental Health, 21, 838–843. https://doi.org/10.1080/13607863.2016.1168360

13.

Duggleby

Williams

Ghosh

Moquin

Ploeg

Markle-Reid

Peacock

(2016). Factors influencing changes in health related quality of life of caregivers of persons with multiple chronic conditions. Health and Quality of Life Outcomes, 14, 81. https://doi.org/10.1186/s12955-016-0486-7

14.

Fries

B. E.

Simon

S. E.

Morris

J. N.

Flodstrom

Bookstein

F. L.

(2001). Pain in US nursing homes: Validating a pain scale for the Minimum Data Set. Gerontologist, 41(2), 173–179.

15.

Garlo

O’Leary

J. R.

Van Ness

P. H.

Fried

T. R.

(2010). Burden in caregivers of older adults with advanced illness. Journal of the American Geriatrics Society, 58(12), 2315–2322. https://doi.org/10.1111/j.1532-5415.2010.03177.x

16.

Gaugler

J. E.

Krichbaum

Wyman

J. F.

(2009). Predictors of nursing home admission for persons with dementia. Medical Care, 47(2), 191–198.

17.

Gulati

Sharma

Gupta

(2016). Theoretical study of decision tree algorithms to identify pivotal factors for performance improvement: A review. International Journal of Computer Applications, 141, 19–25.

18.

Guthrie

D. M.

Davidson

J. G. S.

Williams

Campos

Hunter

Mick

. . . Wittich

(2018). Combined impairments in vision, hearing and cognition are associated with greater levels of functional and communication difficulties than cognitive impairment alone: Analysis of interRAI data for home care and long-term care recipients in Ontario. PLOS ONE, 13(2), Article e0192971. https://doi.org/10.1371/journal.pone.0192971

19.

Health Quality Ontario. (2016). The reality of caring: Distress among caregivers of home care patients. https://www.hqontario.ca/system-performance/reality-caring-report-en

20.

Hebert

Bravo

Preville

(2000). Reliability, validity, and reference values of the Zarit Burden Interview for assessing informal caregivers of community-dwelling older persons with dementia. Canadian Journal on Aging, 19(4), 494–507.

21.

Higginson

I. J.

Gao

Jackson

Murray

Harding

(2010). Short-form Zarit Caregiver Burden Interviews were valid in advanced conditions. Journal of Clinical Epidemiology, 63(5), 535–542.

22.

Hirdes

J. P.

Freeman

Smith

T. F.

Stolee

(2012). Predictors of caregiver distress among palliative home care clients in Ontario: Evidence based on the interRAI Palliative Care. Palliative and Supportive Care, 10, 155–163.

23.

Hirdes

J. P.

Frijters

D. H.

Teare

G. F.

(2003). The MDS-CHESS scale: A new measure to predict mortality in institutionalized older people. Journal of the American Geriatrics Society, 51(1), 96–100.

24.

Hirdes

J. P.

Poss

J. W.

Curtin-Telegdi

(2008). The Method for Assigning Priority Levels (MAPLe): A new decision support system for allocating home care resources. BMC Medicine, 6(9), 1–11.

25.

Iecovich

(2012). Psychometric properties of the Hebrew version of the Zarit Caregiver Burden Scale short version. Aging & Mental Health, 16(2), 254–253.

26.

Jones

Perlman

C. M.

Hirdes

Scott

(2010). Screening cognitive performance with the Resident Assessment Instrument for Mental Health Cognitive Performance Scale. Canadian Journal of Psychiatry, 55(11), 736–740. https://doi.org/10.1177/070674371005501108

27.

Kim

Chang

Rose

Kim

(2012). Predictors of caregiver burden in caregivers of individuals with dementia. Journal of Advanced Nursing, 68(4), 846–855. https://doi.org/10.1111/j.1365-2648.2011.05787.x

28.

Kruithof

W. J.

Visser-Meily

J. M.

Post

M. W.

(2012). Positive caregiving experiences are associated with life satisfaction in spouses of stroke survivors. Journal of Stroke and Cerebrovascular Diseases, 21(8), 801–807. https://doi.org/10.1016/j.jstrokecerebrovasdis.2011.04.011

29.

Landi

Tua

Onder

Carrara

Sgadari

Rinaldi

. . . Bernabei

(2000). Minimum Data Set for Home Care: A valid instrument to assess frail older people living in the community. Medical Care, 38(12), 1184–1190.

30.

Lin

C. Y.

Wang

J. D.

Pai

M. C.

L. E.

(2017). Measuring burden in dementia caregivers: Confirmatory factor analysis for short forms of the Zarit Burden Interview. Archives of Gerontology and Geriatrics, 68, 8–13. https://doi.org/10.1016/j.archger.2016.08.005

31.

Martin

Poss

J. W.

Hirdes

J. P.

Jones

R. N.

Stones

M. J.

Fries

B. E.

(2008). Predictors of a new depression diagnosis among older adults admitted to complex continuing care: Implications for the Depression Rating Scale (DRS). Age and Ageing, 37(1), 51–56. https://doi.org/10.1093/ageing/afm162

32.

Melis

R. J.

van Eijken

M. I. J.

van Achterberg

Teerenstra

Vernooij-Dassen

M. J. F. J.

van de Lisdonk

E. H.

Olde Rikkert

M. G. M.

(2009). The effect on caregiver burden of a problem-based home visiting programme for frail older people. Age and Ageing, 38, 542–547.

33.

Miller

E. A.

Rosenheck

R. A.

Schneider

L. S.

(2012). Caregiver burden, health utilities, and institutional service use in Alzheimer’s disease. International Journal of Geriatric Psychiatry, 27(4), 382–393.

34.

Mohamed

Rosenheck

Lyketsos

C. G.

Schneider

L. S.

(2010). Caregiver burden in Alzheimer disease: Cross-sectional and longitudinal patient correlates. American Journal of Geriatric Psychiatry, 18(10), 917–927. https://doi.org/10.1097/JGP.0b013e3181d5745d

35.

Morris

J. N.

Fries

B. E.

Mehr

D. R.

Hawes

Mor

Lipsitz

(1994). MDS Cognitive Performance Scale. Journals of Gerontology. Series A, Biological Sciences and Medical Sciences, 49(4), M174–M182. https://doi.org/10.1093/geronj/49.4.m174

36.

Morris

J. N.

Fries

B. E.

Morris

S. A.

(1999). Scaling ADLs within the MDS. Journals of Gerontology. Series A, Biological Sciences and Medical Sciences, 54(11), M546–M553. https://doi.org/10.1093/gerona/54.11.m546

37.

Mosquera

Vergara

Larranaga

Machon

del Rio

Calderon

(2016). Measuring the impact of informal elderly caregiving: A systematic review of tools. Quality of Life Research, 25(5), 1059–1092. https://doi.org/10.1007/s11136-015-1159-4

38.

Oldenkamp

Hagedoorn

Slaets

Stolk

Wittek

Smidt

(2016). Subjective burden among spousal and adult-child informal caregivers of older adults: Results from a longitudinal cohort study. BMC Geriatrics, 16(1), Article 208. https://doi.org/10.1186/s12877-016-0387-y

39.

O’Rourke

Tuokko

(2003). Psychometric properties of an abridged version of the Zarit Burden Interview within a representative Canadian caregiver sample. The Gerontologist, 43(1), 121–127.

40.

Park

Sung

Kim

Lee

(2015). Multidimensional determinants of family caregiver burden in Alzheimer’s disease. International Psychogeriatrics, 27(8), 1355–1364.

41.

Pauley

Chang

B. W.

Wojtak

Seddon

Hirdes

(2018). Predictors of caregiver distress in the community setting using the home care version of the Resident Assessment Instrument. Professional Case Management, 23(2), 60–69.

42.

Poss

J. W.

Jutan

N. M.

Hirdes

J. P.

Fries

B. E.

Morris

J. N.

Teare

G. F.

Reidel

(2008). A review of evidence on the reliability and validity of Minimum Data Set data. Healthcare Management Forum, 21(1), 33–39.

43.

Rockwood

J. K. H.

Richard

Garden

Hommick

Mitnitski

Rockwood

(2014). Precipitating and predisposing events and symptoms for admission to assisted living or nursing home care. Canadian Geriatrics Journal, 17(1), 16–21. https://doi.org/10.5770/cgj.17.93

44.

SAS Institute Inc. (2016). SAS enterprise guide.

45.

Sutcliffe

C. L.

Giebel

C. M.

Jolley

Challis

D. J.

(2016). Experience of burden in carers of people with dementia on the margins of long-term care. International Journal of Geriatric Psychiatry, 31(2), 101–108. https://doi.org/10.1002/gps.4295

46.

Toews

(2016). Future of home care services in Manitoba. https://www.gov.mb.ca/health/homecare/future_homecare.pdf

47.

Tremont

Davis

Papandonatos

G. D.

Grover

Ott

B. R.

Fortinsky

R. H.

. . . Bishop

D. S.

(2013). A telephone intervention for dementia caregivers: Background, design, and baseline characteristics. Contemporary Clinical Trials, 36(2), 338–347. https://doi.org/10.1016/j.cct.2013.07.011

48.

Van Durme

Macq

Jeanmart

Gobert

(2012). Tools for measuring the impact of informal caregiving of the elderly: A literature review. International Journal of Nursing Studies, 49(4), 490–504. https://doi.org/10.1016/j.ijnurstu.2011.10.011

49.

Verbeek

Meyer

Challis

Zabalegui

Soto

M. E.

Leino-Kilpi

. . . Hamers

J. P. H.

(2015). Inter-country exploration of factors associated with admission to long-term institutional dementia care: Evidence from the Right Time Place Care study. Journal of Advanced Nursing, 71(6), 1338–1350. https://doi.org/10.1111/jan.12663

50.

Vitaliano

P. P.

Russo

Young

H. M.

Becker

Maiuro

R. D.

(1991). The screen for caregiver burden. The Gerontologist, 31(1), 76–83.

51.

Vitaliano

P. P.

Scanlan

J. M.

Zhang

(2003). Is caregiving hazardous to one’s physical health? A meta-analysis. Psychological Bulletin, 129(6), 946–972.

52.

Williams

Jamal

Guthrie

(2018). The relationship between caregiver burden and depressive symptoms in Ontario home care clients. Home Health Care Services Quarterly, 37(1), 60–76. https://doi.org/10.1080/01621424.2018.1425647

53.

Zarit

S. H.

Reever

K. E.

Bach-Peterson

(1980). Relatives of the impaired elderly: Correlates of feelings of burden. The Gerontologist, 20(6), 649–655.