Background
Stroke is the second most common cause of death Reference Lozano, Naghavi and Foreman1 and a leading cause of disability worldwide. It is estimated that the prevalence of disability due to stroke will increase by 24% between now and 2030. Reference Feigin, Krishnamurthi and Parmar2 Patient-reported physical and social well-being have been recognized as important outcomes after stroke. Reference Carod-Artal and Egido3,Reference De Wit, Theuns and Dejaeger4 There has been increasing interest in the use of health utility scores to capture health-related quality of life (HRQoL) in stroke survivors Reference Rebchuk, O’Neill, Szefer, Hill and Field5,Reference Reeves, Lisabeth and Williams6 as utility scales provide an assessment of HRQoL that reflects the preference valuation of health states by community members. Prior studies have shown reduced HRQoL in patients with stroke using measures such as the EuroQol-5 Dimension (EQ-5D) Reference De Wit, Theuns and Dejaeger4,Reference Delcourt, Zheng and Chen7–Reference Cadilhac, Andrew and Lannin11 and Health Utilities Index Mark 3 (HUI3), Reference Haacke, Althaus, Spottke, Siebert, Back and Dodel12–Reference Pickard, Johnson and Feeny14 and more recently major clinical trials in stroke have incorporated HRQoL assessments. Reference Goyal, Demchuk and Menon15,Reference Berkhemer, Fransen and Beumer16
The EQ-5D and HUI3 both correlate strongly with the Barthel index and Rankin scale, Reference Pickard, Johnson and Feeny14 although contain additional important domains such as mood and cognition. Prior studies have assessed the HUI3 in stroke, Reference Edwards Jodi, Koehoorn, Boyd Lara and Levy Adrian13,Reference Pickard, Johnson and Feeny14,Reference Grootendorst, Feeny and Furlong17,Reference Kapral, Fang and Hill18 although many are outdated, relied on self-report of stroke, and did not assess differences by age. There is also a lack of population-based studies on HRQoL after stroke comparing to controls without stroke.
An updated understanding of post-stroke dysfunction and HRQoL is needed to design and implement programs to better address the needs of stroke survivors. To overcome the limitations of prior survey research that relied on patient-reported stroke history, we linked the Canadian Community Health Survey (CCHS) to population-based universal healthcare administrative databases to ascertain physician-diagnosed stroke cases, and then compared HRQoL to those without a stroke linkage. We assessed differences by age, sex, and health attribute. We hypothesized that younger individuals with stroke would suffer greater losses in HRQoL compared to older individuals, and that impairments would be most pronounced in domains of mobility, cognition, and emotional health.
Methods
Study Sample: Canadian Community Health Survey
The CCHS is an annual cross-sectional survey, representing 97% of the Canadian population aged 12 years and older. The survey randomly samples households nationwide and selects one respondent from each household. The CCHS collects information about health status, health determinants, and health care utilization of the household population. We used CCHS years 2009, 2010, 2013, 2014, and 2015 as the HUI3 was a mandatory component of the questionnaire in these years. Interviews are conducted using computer-assisted personal and telephone interview software. 19 Before releasing the CCHS for use, data are assessed for quality and compared to previous cycles to avoid errors. Approximately 3% of the Canadian population is excluded from the survey target population, which includes those living on Indigenous reserves, those living in foster care, full time members of the Canadian Armed Forced, the institutionalized population and the remote Région du Nunavik and Région des Terres-Cries-de-la-Baie-James. Data were collected by Statistics Canada using a multistage sample allocation strategy to support estimation at the health region and provincial level. 19
While the CCHS contains an item on the questionnaire regarding a history of stroke diagnosis, self-report of stroke is subject to recall bias and there is a large discordance between individual self-report and medical diagnosis of stroke derived from administrative data. Reference Wu, Lai, Gau, Wang and Tsai20,Reference Muggah, Graves, Bennett and Manuel21 The CCHS has recently been linked to administrative databases of hospital discharge and emergency department visits, 22 allowing more accurate determination of stroke cases.
CCHS Sharelink
CCHS Sharelink is a CCHS sub-sample with about 85% of total respondents, who agreed to have their responses linked to administrative records. Statistics Canada created sample weights for CCHS Sharelink to retain population representativeness. CCHS Sharelink resembles CCHS in sociodemographic characteristics. 22 Linkages were performed by Statistics Canada and included the Canadian Institutes of Health Discharge Abstract Database (CIHI-DAD) for hospitalizations and the National Ambulatory Care Reporting System (NACRS) for emergency department visits. The DAD is an administrative database which collects demographic, diagnostic, and treatment information from all admitted patients in Canada, excluding the province of Quebec. NACRS collects similar information for visits to the emergency department. CIHI-DAD was linked back to 1999–2000 and NACRS to 2002–2003.
Case Definition
We first excluded respondents from Quebec, as Quebec does not contribute data to CIHI-DAD. We also excluded respondents under 40, due to very few stroke cases in that age group. All cases of first stroke were identified in CIHI-DAD and NACRS from 1999 to 2015 using International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, Canada (ICD-10-CA) codes (ischemic stroke: I63.x, I64.x, H34.1; intracerebral hemorrhage [ICH]: I61.x). ICD-9 codes [ischemic stroke: 434.01, 434.11, 434.91, 436; ICH: 431] were used to obtain cases prior to 2003. These case definitions used have high validity in Canada. Reference Porter, Mondor, Kapral, Fang and Hall23,Reference Kokotailo and Hill24 Individuals who had a prior episode of stroke captured in the administrative databases and who participated in the CCHS in the selected years were included in the stroke cohort. Control subjects without prior identification of stroke comprised the remainder of individuals in the CCHS in the selected years.
Covariates
Covariates were obtained from the CCHS, including age, sex, year of survey, province of respondent, rural residence, ethnicity, education level, total household income quartile, marital status, and self-reported body mass index (BMI), smoking status, hypertension, diabetes, heart disease, chronic obstructive pulmonary disease (COPD), arthritis, and asthma. Age was categorized as <60, 60–74, and 75+. Categorization of the remainder of variables can be seen in Table 1. Due to known biases in self-report of BMI, we employed a correction suggested by Statistics Canada. Reference Shields, Connor Gorber, Janssen and Tremblay25 Chi-square test was used to assess differences in proportions between those with and without stroke. All numerical estimates were rounded as per Statistics Canada confidentiality policies.
Weighted percentages and means of sample characteristics by stroke status
CI = confidence interval; COPD = chronic obstructive pulmonary disease; HUI3 = Health Utilities Index Mark 3.
Proportions are obtained by following Statistics Canada procedures on weighting and rounding and categories may not always sum to 100.
Health Utilities Index Mark 3
The HUI3 is a generic, preference-based measure of health status, which includes the attributes of vision, hearing, speech, ambulation, dexterity, emotion, cognition and pain, with five or six levels for each attribute (Supplemental Table 1) and incorporates utility weights based on standard gamble studies in a general population sample. The HUI3 has been validated for use in the Canadian stroke population. Reference Grootendorst, Feeny and Furlong17 The summary index ranges from −0.36 (state worse than dead) through 0 (similar to dead) to 1 (perfect health) Reference Horsman, Furlong, Feeny and Torrance26 and is obtained by multiplying each individual attribute rank by its multi-attribute weight, and then using the following equation: HUI3 score = 1.371*(Vision*Hearing*Speech*Dexterity*Mobility*Emotion*Cognition*Pain) − 0.371. The single attribute weights were used when analyzing individual attributes. Reference Feeny, Furlong and Torrance27
Analysis
HUI3 index score means with 95% confidence intervals (CIs) were stratified by age and sex and t-tests were used to evaluate the difference between those with and without stroke. Chi-squared tests were used to compare proportions across categories of global impairment Reference Feng, Bernier, McIntosh and Orpana28 [none (1), mild (0.89–0.99), moderate (0.7–0.88), severe (>0–0.69), or equal to or worse than dead (0 or below)]. We used multivariable linear regression to determine the association between history of stroke and mean difference in HUI3 score using the multi-attribute weights, separately for each CCHS survey year. Because the CCHS uses a multi-stage sampling procedure and several sampling frames, a replicate bootstrap weighting procedure is recommended to deal with design effects. A set of 500–1000 bootstrap weights provided by Statistics Canada were used to account for clustering and variation in selection probabilities. These bootstrap weights also include adjustments for non-response to help minimize selection bias. We used bootstrap weights for the linear regression in each survey year separately, and then combined the coefficients at the survey level using random effects meta-analysis.
To assess for the presence of modification by age or sex, we first ran a simple model with mean HUI3 as the outcome, stroke as the exposure, and age and sex included as interaction terms, using the Wald test to assess for a significant interaction (p < 0.05). Age was categorized as <60, 60–74, and 75 and older, with <60 serving as the reference group. As this model demonstrated modification by age, we then fit individual age-specific models, with the simple model adjusted for only sex. The full model included sex and the remainder of the covariates. All covariates demonstrated evidence of confounding in at least one age- and year-specific model defined as a change in the stroke-HUI3 regression coefficient by >10%. As NACRS cases were predominantly from provinces Ontario, Alberta, and British Columbia, we conducted a sensitivity analysis excluding stroke cases identified by NACRS only. We also did a sensitivity analysis excluding those with self-reported stroke from the control group.
For each of the individual attributes, we determined the proportions of respondents with and without stroke who had level 1, level 2, or level 3+ (see Supplemental Table 1 for description of all levels). We then determined the association between stroke and HUI3 for the eight individual attributes in the same fashion as the global analysis, using the single attribute weights.
Multiple Imputation
The HUI3 score was missing in 4.3% of controls and 8.5% of stroke respondents. The vast majority of cases were due to non-answer of one attribute on the HUI3. We conducted a multiple imputation analysis with ordinal regression to replace the missing HUI3 attribute values, using all covariates and the remaining seven attributes as predictors in the model. Visual inspection of the replaced values was done to ensure face validity. The linear regressions were then repeated with the imputed values.
Analyses were done in the Prairie Regional Data Centre at the University of Calgary using Stata 16.0 (College Station, TX). Threshold of significance for p-values was <0.05. Under Tri-Council guidelines, this analysis did not require approval by a research ethics board.
Results
There were 1240 people with prior stroke (91% with ischemic stroke and 9% with ICH), and 123,765 controls. Eleven percent of stroke records were identified from emergency room records only and the remainder from hospital admissions. Proxy interviews were conducted in 2.5% of controls and 13.0% of stroke respondents. The median time between stroke diagnosis and CCHS interview was 4.3 years (IQR 1.8–7.2). The I2 values in the meta-analyses ranged from 0% to 59.7%.
Compared to those without a history of stroke, those with stroke had older age, were more likely to be male, had a lower education, lower income and had a higher proportion of vascular risk factors. All baseline characteristics are shown in Table 1. Mean unadjusted HUI3 was lower for those with stroke, for females, and individuals at older age (Table 1). Compared to controls, there was a much lower proportion of stroke respondents with no or minimal impairment in health utility and a greater proportion with severe impairment (shown in Figure 1).
Weighted percentages in each Health Utilities Index Mark 3 (HUI3) index score category by stroke status.
In the linear regression, there was evidence of significant modification by age (p = 0.033 for age 60–74). There was no significant modification by sex (p = 0.34). In the stratified age- specific simple models, mean HUI3 difference between those with stroke and without stroke was greatest in those aged 60–74 years (−0.22, 95% CI −0.27 to −0.17), compared to those < 60 years (−0.17, 95% CI −0.24 to −0.10) and 75+ years (−0.13, 95% CI −0.19 to −0.07; Figure 2A). The age-specific pattern was similar in the fully adjusted model (Figure 2B), when excluding cases only identified in the emergency department, when excluding self-reported stroke from the control group, and when imputing missing HUI3 values (Supplemental Figure 1).
Adjusted mean difference in HUI3 index score for stroke respondents compared to controls, stratified by age group for simple (A) and full models (B).
Respondents with stroke had greater proportion of moderate to severe impairment in all HUI3 attributes compared to controls, especially mobility (level 3 or above 34%), cognition (46%), and pain (35%; Figure 3). There was no significant interaction by age or sex for any of the individual HUI3 attribute utilities. In the simple model, stroke was significantly associated with reductions in all attributes except hearing, with the greatest reductions occurring in attributes emotion, cognition, pain, and mobility (Figure 4A). In the fully adjusted model, dexterity, emotion, cognition, and mobility remained significantly impaired in those with stroke versus controls, with most impairment in cognition and mobility (shown in Figure 4B). Results were similar after imputation of missing HUI3 attribute scores (Supplemental Figure 2).
Weighted percentage of impairment for levels of each HUI3 attribute (higher level is worse impairment), among stroke respondents and controls. See Supplemental Table 1 for description of all levels.
Adjusted mean difference of individual HUI3 attribute weights for stroke respondents compared to controls, for simple (A) and full model (B), plotted in order of greater HUI3 difference. H indicates hearing; S, speech; V, vision; D, dexterity; E, emotion; C, cognition; P, pain; and M, mobility.
Discussion
In our study of nationally representative cross-sectional surveys linked with administrative data, history of stroke was associated with a lower HRQOL as measured by the HUI3. We found modification by age, whereby those between 60 and 74 years experienced a greater adjusted reduction in HUI3 with stroke compared to those < 60 or 75+. Individuals with stroke had the greatest impairment in attributes of emotion, cognition, pain, and mobility.
Our study utilized administrative data for our case definition of stroke due to greater reliability than self-reported stroke. Reference Muggah, Graves, Bennett and Manuel21 In a prior study, health utility scores were higher for cases identified by health administrative data, suggesting that milder cases are detected. Prevalence of stroke in health administrative data was also lower as compared to self-report, implying the potential for misclassification bias in survey data. Reference Muggah, Graves, Bennett and Manuel21 Furthermore, the CCHS questionnaire asks whether the individual experiences the “effects of stroke,” which may result in misclassification from individuals with mild or resolved deficits, or misattribution of certain symptoms to other conditions or older age.
In this study, health utility scores for those with stroke based on hospital records were similar to prior published utility values for those with self-reported “effects of stroke” in Canada, Reference Edwards Jodi, Koehoorn, Boyd Lara and Levy Adrian13,Reference Mittmann, Trakas, Risebrough and Liu29,Reference Guertin, Humphries, Feeny and Tarride30 although these studies did not directly model the difference with control subjects. We found a significant reduction in HUI3 in stroke respondents compared to controls after adjustment for demographic, social, and co-morbid factors. The reason for the greater reduction in HUI3 score among stroke respondents aged 60–74 is unclear, although may relate to greater capacity for plasticity and recovery in younger individuals, and greater resilience in the elderly. Reference Jeste, Savla and Thompson31 Although floor effects may have played a role in the elderly due to lower baseline HRQoL, the HUI3 does not demonstrate substantial floor effects. Reference Luo, Johnson, Shaw and Coons32,Reference Davison, Jhangri and Feeny33 Survival bias may also result in higher than expected health utility scores in the elderly with stroke. Our results are nevertheless a meaningful and practical representation of community-dwelling stroke survivors with implications for resource needs. Stroke among 60–74 may occur when productivity and opportunity cost remain high, yet the adaptive mechanisms are lower, suggesting the potential need for greater post-stroke supports among survivors in the community in this age group.
We found that stroke was associated with the greatest reductions in utility for emotion, cognition, pain, and mobility. Among stroke survivors, 41% have intermediate to high risk of depression, Reference Davison, Jhangri and Feeny33,Reference Swartz, Cayley and Lanctôt34,Reference Luo, Johnson, Shaw and Coons32 and symptoms of depression have been found to be predictive of impaired functioning and community re-integration and low quality of life post-stroke. Reference Kapoor, Lanctot, Bayley, Herrmann, Murray and Swartz35 Although best-practice guidelines endorse screening for post-stroke depression, Reference Eskes, Lanctôt and Herrmann36–38 under-recognition and under-treatment remains common. Reference Swartz, Bayley and Lanctôt39 The association with pain impairment was attenuated after full adjustment, although over a third of respondents with stroke reported moderate to severe pain. Post-stroke pain is also under-recognized and estimates of post-stroke pain prevalence range from 10% to 30%. Post-stroke pain may be due to central pain or peripheral neuropathic pain and is associated with greater functional dependence, depression, and cognitive decline. Reference O’Donnell Martin, Diener, Sacco Ralph, Panju Akbar, Vinisko and Yusuf40–Reference Harno, Haapaniemi and Putaala42 Finally, respondents with stroke had the greatest proportion of impairment in the cognitive attribute compared to all other attributes, with over half of respondents reporting some degree of impairment. In the Oxford Vascular Study, the incidence of dementia was 34% at 1 year and after severe stroke and 8.2% after minor stroke. Reference Pendlebury and Rothwell43 Individuals with stroke have faster declines in executive function and global cognition than controls. Reference Levine, Galecki and Langa44 Our study re-inforces the prominent and persistent impacts of emotional, cognitive, pain, and mobility disturbances on overall HRQoL among community dwellers in the chronic phase of stroke.
Our analysis has numerous strengths, particularly with the quality and reliability of the data and the availability of a large number of baseline variables. The linkages of CCHS with the DAD are of high quality, with few false links. Reference Sanmartin, Decady and Trudeau45 The CCHS provides a large sample size which can be generalized to the Canadian household population. There were some limitations to the study. First, self-reported variables may be associated with response biases, although variables like hypertension and diabetes have shown high reliability in self-report. Reference Muggah, Graves, Bennett and Manuel21 The wording of questions is modeled after items contained in the US Behavioral Risk Factor Survey and specifies that a self-reported chronic condition must represent a diagnosis from a health professional. Second, individuals with stroke underwent the survey at various times from the stroke, although this is a realistic portrayal of the heterogeneity of stroke survivors living in the community. Third, diagnosis of stroke was reliant on administrative data, although administrative data coding for acute stroke has a high positive predictive value in Canada. Reference Porter, Mondor, Kapral, Fang and Hall23 Finally, we could not identify individuals hospitalized with stroke prior to 1999, although this provided a 10-year window from the earliest survey year.
We used the HUI3 to demonstrate the overall burden and degree of reduction in global HRQoL and specific attributes among community-dwelling stroke survivors compared to those without stroke. Emotion, cognition, pain, and mobility are particularly impacted among those with stroke. This knowledge could be used for future population-based assessments, measuring effects of interventions in improving HRQoL among stroke survivors, and monitoring of trends over time. Further studies should address whether the HUI3 or other measures of health utility add value to more standard stroke disability measures in routine clinical care or clinical trials.
Acknowledgements
The analysis was conducted at the Prairie Regional Data Centre, which is part of the Canadian Research Data Centre Network (CRDCN). The services and activities provided by the CRDCN are made possible by the financial or in-kind support of the Social Sciences and Humanities Research Council (SSHRC), the Canadian Institutes of Health Research (CIHR), the Canadian Foundation for Innovation (CFI), Statistics Canada, and participating universities whose support is gratefully acknowledged. The views expressed in this paper do not necessarily represent the CRDCN’s or that of its partners. The authors would like to thank Dina Lavorato and Stephanie Cantlay for their assistance in the Prairie Regional Data Centre.
Funding
RAJ is supported by a Fellowship Grant from the CIHR (MFE 164702).
Disclosures
EES reports royalties from UpToDate, consulting fees from Alnylam, Biogen, and Javelin. The other authors report no disclosures.
Statement of Authorship
RAJ was involved with project conception, analysis, and writing of the manuscript. SP was involved with project conception and critical revision of the manuscript. AL was involved with project conception and critical revision of the manuscript. JW was involved with analysis and critical revision of the manuscript ES was involved with project conception and critical revision of the manuscript.
Supplementary Material
To view supplementary material for this article, please visit https://doi.org/10.1017/cjn.2021.119.
