Background
Population aging and its associated challenges have been repeatedly reported by policy makers and health researchers across the globe [
1]. Much of the aging research has focused on the general population; the subset with intellectual and developmental disabilities (IDD) has received relatively little attention [
2].
Almost 1 % of the population has an IDD [
3,
4], which the World Health Organization defines as “a group of developmental conditions characterized by significant impairment of cognitive functions, which are associated with limitations of learning, adaptive behaviour and skills” (page 177) [
5]. These conditions originate before the age of 18 years [
6], with the expectation of lifelong disability.The risk of health problems is greater among persons with IDD [
7], and these contribute to the widely held notion that this group experiences premature aging, where 50 years may be considered “old” [
8‐
12].
While many illnesses and functional impairments become increasingly prevalent with age, age by itself is an insensitive and non-specific predictor of health vulnerabilities [
13,
14]. Frailty is frequently identified as an effective measure of health and vulnerability [
13,
15]. While a consistently applied definition does not exist in the literature, frailty is generally viewed as age-related decline associated with higher risk of adverse health outcomes [
16]. However, few studies have measured frailty in adults with IDD. A review of the literature revealed the existence of two research groups explicitly measuring frailty in this population, each using a different approach.
Brehmer and Weber (2010) created a frailty measure, the Vienna Frailty Questionnaire for Persons with Intellectual Disabilities – Revised (VFQ-ID-R) [
17,
18], which captures changes in 34 symptoms across four domains (physical, psychological, cognitive, and social). The presence of frailty is indicated if: (1) symptoms are present in at least three domains, and (2) a minimum of six symptoms are identified overall. Persons are described as “pre-frail” if only one of these criteria is met. In their sample of 147 Austrian adults with IDD aged 20–72 years, 17.7 % were frail, 17.7 % were pre-frail, and the remainder was non-frail [
17]. The small sample size, however, hampers the generalizability of results. In addition, the difficulty in applying the components in the VFQ-ID-R to other studies or databases further limits its use.
In the Netherlands, Fried et al.’s (2001) five frailty phenotype symptoms (weight loss, weakness, poor endurance and exhaustion, low physical activity and slowness [
19]) were measured in 848 individuals with IDD over 50 in the Healthy Aging and Intellectual Disabilities (HA-ID) study cohort [
20]. Here, persons were frail if at least three of the five symptoms were present, and pre-frail if one or two were present. Evenhuis et al. (2012) [
20] reported that 13 % of their cohort was frail and 60 % was pre-frail. Approximately 11 % of their sample under the age of 65 years was considered frail, which is similar to the published prevalence estimates of frailty in the general Dutch population aged 65 years or older. The findings support the hypothesis that persons with IDD experience frailty earlier than the general population [
20].
The frailty phenotype approach to measuring frailty is limited in that it focuses on physical limitations, which are known to be more common in individuals with IDD regardless of age [
10]. Others have noted that the strong influence of low mobility on frailty phenotype scores may lead to misclassification in the general population [
21]; this may also lead to misclassification among those with IDD [
10], who have higher rates of mobility limitations across the lifespan.
In a follow-up study, the HA-ID research group used an accumulation of deficits approach [
22] to measure frailty in 982 adults 50 years or more using formal care in the Netherlands [
10]. The accumulation of deficits approach emphasizes the proportion of health deficits a person has, rather than the presence of specific symptoms. The mean frailty score was 0.27 (SD = 0.13), with an upper limit of 0.69 (on a continuum ranging from 0 to 1, with a higher score representing frailty), and approximately 66 % of the sample was frail (score above 0.20). They also found that individuals with IDD had similar rates of frailty at age 50 as did the European population at age 70, again supporting previous work showing that persons with IDD are frail at earlier ages than in the general population.
While these are the only studies reporting on frailty measures developed specifically for populations of persons with IDD, generic frailty measurements have also been applied to this population. For example, in Ontario, Canada, the Frailty Marker, derived from the Johns Hopkins University Adjusted Clinical Group (ACG) System, is used to identify frailty in population-based studies relying on administrative data [
23]. This marker categorizes individuals as frail based on 81 diagnostic codes. As the actual diagnostic codes are unknown (i.e., not published), it is suspected that the Frailty Marker relies on the presence of specific medical conditions and fails to capture the multiple domains required for a well-balanced frailty measure. As such, it may not be the most appropriate measure of frailty, which limits its use for both individual-level and population-level health service planning – especially among persons with IDD.
The objective of this study is to describe the process of applying the accumulation of deficits approach to develop a frailty index (FI) based on administratively held clinical data for use among community-dwelling adults with IDD receiving home-based health care services (i.e., home care), and to identify individual characteristics associated with frailty. Using associations with 1-year admission to long-term care (LTC), appropriate cut-off scores for the newly developed FI are determined, and the prevalence of frailty in this population is described.
Methods
Study design
This study is part of a larger program of research focused on Health Care Access Research and Developmental Disabilities (H-CARDD) (see
www.hcardd.ca). In this study, the cohort includes 7,863 individuals living in Ontario in 2009/10, identified in one or more administrative health datasets (such as physician claims or hospital visits) as per a previous study [
4,
23] and who had at least one home care assessment between April 1
st, 2006 and March 31
st, 2014. Individuals were between the ages of 18 and 99 years in 2009/10 and at their first home care assessment. The study protocol was reviewed and approved by the institutional review board at Sunnybrook Health Sciences Centre, Toronto, Canada, and the Queen’s University Health Sciences Research Ethics Board.
Data linkage
The RAI-HC database contains the RAI-HC assessments provided by the Home Care Reporting System (HCRS). These data are stored with interRAI Canada and shared with the Institute for Clinical Evaluative Sciences (ICES). Data included the earliest home care assessment occurring in the study period for each cohort member. The Registered Persons Database (RPDB) contains information on all Ontario persons eligible for health coverage. For this study, it provided age (in years). The date of admission to LTC was retrieved from the Continuing Care Reporting System for Long-Term Care (CCRS-LTC) database, which provides demographic and clinical information about individuals receiving care in LTC homes. All accessed data had been de-identified using methods to ensure confidentiality and privacy. Datasets were linked using unique encoded identifiers and analyzed at the Institute for Clinical Evaluative Sciences (ICES).
Dataset creation
In Ontario, home care services are provided by the Ministry of Health and Long-Term Care’s fourteen regional Community Care Access Centres, who determine eligibility and coordinate providers. Services provided include visiting health professionals, help from personal support workers, homemaking services and community support services. Case managers use the Resident Assessment Instrument- Home Care (RAI-HC) [
24] as a needs assessment tool to assess those receiving, or about to receive, home care. InterRAI is an international collaboration that serves to collect and interpret data on health and social outcomes. The interRAI assessments, including the RAI-HC, have been implemented as routine and standardized measures in various care settings by several jurisdictions internationally, including several Canadian provinces [
25]. The RAI-HC assessment captures information related to demographic characteristics, home environment, functioning, health, medications, informal support, and formal health services. With respect to demographic characteristics, the current study used age, sex, living situation (i.e., alone, with a spouse and/or child(ren), with other family, or in a group setting with non-relatives), residential care history (i.e., lived in a residential care facility in the last 5 years), and rural status. Following the Statistics Canada definition, rural status was defined as living in a “location not included in a [Canadian census metropolitan area or census agglomeration], living in an urban centre of fewer than 10,000, or living in a rural area” (p. 157) [
26]. Rural status was determined from postal codes identified at time of assessment.
Following Brehmer and Weber (2010)’s [
18] categorization of frailty domains, the health-related items of interest were categorized into cognitive, physiological, psychological, and social domains. An additional domain of “service use” captures other service-related items [
10,
27]. The items for each domain were all selected from information available in the RAI-HC. Twenty-seven items related to cognitive patterns (e.g. memory loss, delirium), communication (e.g. ability to understand others), and practical skills (i.e., instrumental activities of daily living; e.g. help needed with ordinary housework, managing finances, shopping) were used to inform on the person’s cognitive functioning. One-hundred fourteen items were available to assess physiological health, such as: hearing, vision, bladder and bowel continence, health conditions (e.g. diarrhea, shortness of breath), nutritional status (e.g. morbid obesity, insufficient fluid intake), dental and oral status (e.g. chewing problems), skin conditions (e.g. pressure ulcers, wound care required), medical diagnoses (e.g. hypertension, Alzheimer’s disease, hip fracture, diabetes), and medications (e.g. use of anxiolytics, medication compliance). Psychological status was informed by seventeen items related to mood (e.g. feelings of sadness or depression, repetitive anxious complaints) and behaviours (e.g. wandering, verbal abuse). Items from the social domain included social isolation, withdrawal from activities of interest, and five other items.
In addition to these domains, items indicating home environment (e.g. difficulty accessing rooms in house, inhabitable heating/cooling) and service utilization (e.g. recent hospital admissions, unmet treatment goals) were used. Two global health status indicators were also accessed: self-reported health (asking if the individual feels he/she is in poor health) and the presence of diseases or conditions that make cognition, activities of daily living (ADL), mood, or behaviour patterns unstable.
Three variables related to informal support were included in this study: (1) caregiver is unable to continue caring (“caregiver inability”), (2) caregiver is unsatisfied with support from family and friends (“caregiver unsatisfied”), and (3) caregiver expresses feelings of distress, anger or depression (“caregiver distress”).
In addition to individual items, two measures embedded in the RAI-HC were included. Clinical Assessment Protocols (CAPs) identify common risks for individuals using home care, such as abuse, functional decline or LTC placement [
28]. Algorithms using some RAI-HC items trigger CAPs, which then offer interpretations and potential interventions for case managers to include in home care planning [
28,
29]. The Institutional Risk CAP is triggered for individuals with a high risk of institutionalization and suggests ways of remaining in the community [
30]. The second embedded measure is the Cognitive Performance Scale (CPS), which provides the cognition level and characterizes individuals on a scale from 0 (intact cognition) to 6 (very severe impairment) [
31]. Individuals with CPS scores ≥3 (i.e., moderate or worse cognitive impairment) were grouped into one category [
32,
33].
Analysis
Selection of health deficits
The RAI-HC assessment provided 180 deficit variables. Previous work has suggested that deficits can be signs, symptoms, disabilities, diseases or abnormal laboratory measurements [
22]. The criteria for selecting variables, published first by Searle et al. (2008) [
22] and then modified by Schoufour et al. (2013) [
10] for persons with IDD, are described.
First, each deficit must be positively correlated with age. This was done by calculating Spearman’s correlation coefficients (r
s) between each deficit (ordinal variables) and age (as a continuous variable). Deficits that were not significantly and positively correlated with age were excluded, using a cut-off of r
s = 0.05 (
p < 0.0001). Second, deficits that were too saturated were excluded to prevent ceiling effects, using a prevalence cut-off of 80 %. For variables that were not dichotomous, deficits were considered present if any limitation existed. Third, the deficit must be associated with health status, which was determined using Chi-square tests (included if
p < 0.05) for association. Fourth, a wide range of health aspects should be included in the FI. A review of the literature was conducted to determine if the deficits appeared to cover different aspects of health, including all five domains of health [
18]. Searle et al.’s (2008) final criterion related to use of identical items over time is not relevant to this study, which assesses frailty at a single point in time [
22].
Schoufour et al. (2013) developed further inclusion criteria for the FI that are appropriate for persons with IDD [
10]. If a deficit variable has missing data for greater than 30 % of individuals, it should be excluded. Second, deficits were considered uncommon and excluded if prevalent in fewer than 5 % of individuals, to prevent floor effects. However, wherever possible, related variables were grouped to form multi-item deficits with a sufficient prevalence. Schoufour et al.’s (2013) [
10] criteria also suggested reducing the number of variables if they appear repetitive. Variables were grouped into twenty-seven categories, and correlations between remaining variables within categories were determined. Variables that were very highly correlated (r > 0.9) were either grouped into a multi-item deficit, or only the item with the highest correlation with age was included.
The list of excluded variables was screened by experts (the authors) for deficits unexpectedly omitted. This current study added one further stipulation. In an attempt to identify a change in deficits, which is crucial to capturing frailty [
14,
15], deficits were grouped whenever possible to create a “decline” variable to ensure that the FI included recent deficits, rather than long-standing functioning, health, or behaviour patterns.
Calculation of the continuous FI
Most variables were ordinal or dichotomous. Variables were recoded, if necessary, to scores of 0 (deficit not present), 0.5 (intermediate deficit), or 1 (full deficit present). One continuous variable (“falls frequency”) was recoded into an ordinal variable as 0, 0.25, 0.5, 0.75, and 1.0. The rescaling of deficits was congruent with previous publications [
34,
35], although some expert judgment was required. Variables were not weighted, therefore all selected deficits contributed to the final FI score equally [
36].
A FI score was calculated for each individual by dividing the sum of the deficit scores by the number of deficits measured, to create continuous values between 0 (no deficits present) and 1 (all deficits present).
Categorizing the FI
The FI can be informative as a continuous variable to describe and contrast populations’ distributions of vulnerability. However, to ease comparisons, the FI is often categorized into meaningful groups though cut-offs have not been consistently applied across studies [
34]. Hoover et al. (2013) [
34] reported methods to validate cut-off points for the FI, using stratum-specific likelihood ratios (SSLRs), to distinguish between frail and non-frail seniors and to identify “natural” ranges of frailty associated with different risks of adverse outcomes. SSLRs represent the likelihood that individuals in a specific frailty group (stratum) will experience an event (admission to LTC in 1-year follow-up) relative to their likelihood of not experiencing an event [
34,
37]. Using a subset of the cohort (
n = 7,115) with a home care assessment between April 1
st, 2006 and March 31
st, 2013, SSLRs were calculated. SSLRs are independent of the population prevalence [
34], and are less susceptible to spectrum bias (i.e. change in measure characteristics due to a different mix of severity) than a single cut-off. This process helps to ensure that lower and higher scores are correctly assigned to their own corresponding group [
38].
The ten stratum cut-off points identified by Hoover et al. (2013) [
34] were used in this study. Strata were collapsed if there was an insufficient number of events or non-events, or if 95 % confidence intervals clearly overlapped [
39]. Confidence intervals were calculated using equations presented by Peirce and Cornell (1992) [
37]. This process ensured that strata were significantly different from each other.
Statistical analysis
The mean, standard deviation, and the maximum and minimum scores for the continuous FI are reported. A histogram shows the distribution of FI scores. Goodness-of-fit tests (e.g. the Cramer von- Mises test) assessed whether the distribution fit a Weibull or gamma distribution [
40,
41]. The mean FI score per year was estimated by calculating the regression coefficient β [
10], and the relationship between the upper limit of the FI and age was determined by plotting the 99
th percentile of each 10-year age group. The slopes of these scores can indicate the presence of an age-invariant sub-maximal limit to the FI, demonstrating that even with advancing age, no further deficits are accumulated.
Bivariate multinomial logistic regression models were completed to calculate odds ratios and 95 % confidence intervals to report the odds of frailty (pre-frail or frail compared to non-frail), by individual characteristics. These groups included age (per 10 year increase), sex, rural status, caregiver status variables (i.e. caregiver inability, caregiver unsatisfied, caregiver distress), living situation, residential care history, cognition level and the Institutional Risk CAP.
An adjusted multinomial logistic regression model was developed to determine adjusted odds ratios of frailty (pre-frail or frail compared to non-frail) and individual characteristics (listed above). Using backwards elimination to select significant covariates, at a significance level of α = 0.05 using the Wald test, the model retained significant covariates.
An analysis of variance (ANOVA) was completed to compare frailty groups by age (continuous). The correlation between frailty and self-reported health was determined. Bivariate logistic regression models were also developed for each individual deficit in the FI and 1-year admission to LTC; odds ratios and 95 % confidence intervals were determined.
All tests were two-sided tests, with an alpha value of 0.05, unless otherwise stated, to indicate statistical significance. All analyses were done using SAS Enterprise Guide version 6.1.
Conclusion
Measuring frailty among persons with IDD using home care services is feasible. This study has identified health deficits applicable for those with IDD to include in a FI and has presented cut-off points for the FI to distinguish between risk groups.
Premature aging has frequently been reported in adults with IDD; however the increased vulnerabilities that come with aging are rarely quantified. Frailty may be a better way to understand the needs of the young old with IDD. In the general population, caring for elderly citizens is particularly challenging due to the blend of both medical and social problems [
14], however adults with IDD face these challenges throughout their lives and these may worsen as they age.
Next steps include applying the FI to predictive models. If the FI is associated with time to adverse events (e.g. admission to LTC), the potential exists to use this measure as a tool in the community.
Availability of supporting data
ICES is a prescribed entity under the Ontario Personal Health Information Protection Act. As such, ICES policies and procedures are approved by Ontario’s Information and Privacy Commissioner. These policies require that access to data be limited to persons who require such access to perform their role on an approved ICES Project or Third-Party Project. Thus, we are prohibited from making ICES data publicly available. Only the results of analysis of ICES data may be made available.
Acknowledgements
The research reported in this paper was funded by the Ontario Ministry of Health and Long-Term Care (MOHLTC) Health Systems Research Fund (Ministry Grant #06671) and is part of the Health Care Access Research and Developmental Disabilities (H-CARDD) Program. The first author (K. McKenzie) was supported by the Syd Vernon Foundation and Queen’s University. Analyses were conducted at the Institute for Clinical Evaluative Sciences (ICES), which is funded by an annual grant from the Ontario MOHLTC. The opinions, results and conclusions reported in this paper are those of the authors and independent of the funding sources. No endorsement by the Ontario MOHLTC, the Ontario Ministry of Community and Social Services or ICES is intended or should be inferred. Parts of this material are based on data and information compiled and provided by the Canadian Institute for Health Information (CIHI). However, the analyses, conclusions, opinions and statements expressed herein are those of the author, and not necessarily those of CIHI. The authors wish to acknowledge the contributions of the staff at the Institute for Clinical Evaluative Sciences at Queen’s University, especially Marlo Whitehead for assistance in programming and database management.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
The idea of developing a measure of frailty for adults with intellectual and developmental disabilities was HOK’s and LM’s. KM wrote the manuscript, performed the statistical analysis, and interpreted the results, with editorial feedback provided by HOK and LM. All authors read and approved the final manuscript.