Introduction
Self-reported unmet health care need is often used as a marker of access to health care and has been included as an indicator by various national and international organizations and institutions monitoring health.Reference Allin, Grignon and Le Grand1 It is defined as the difference between health services that are considered necessary for a particular health problem compared to services that are received.Reference Carr and Wolfe2 Stroke is one of the major causes of disability worldwide.Reference Feigin, Roth and Naghavi3 Previous research suggests that adult stroke survivors have higher rates of unmet social and health care needs compared to the general population.Reference Duxbury, Depaul, Alderson, Moreland and Wilkins4–Reference Ronksley, Sanmartin and Quan6 Furthermore, perceived poor access to care has been associated with higher rates of hospitalization for chronic diseases.Reference Bindman, Grumbach and Osmond7 Thus, evaluating the burden of unmet health care needs in stroke survivors is important for health care planning and delivery.
Integrated systems of stroke care that aim to coordinate optimal care across an entire region are recommended by professional stroke societies and may be associated with a reduction in the risk of death and disability after stroke.Reference Schwamm, Pancioli and Acker8–Reference Ganesh, Lindsay and Fang10 In 2000, the province of Ontario was among the first North American jurisdictions to launch an integrated system of stroke care delivery, with full implementation in 2005. Similar organized stroke systems were implemented in five other provinces by 2006. Although there were variations in the way that the stroke systems were implemented in different provinces, common elements included the development of regional strategies for providing stroke care, the designation of certain hospitals as “comprehensive stroke centers” capable of providing thrombolysis and other advanced interventions, and the promotion of evidence-based stroke care. We aimed to study temporal trends in the self-reported unmet health care needs of adult stroke survivors and to determine the impact of integrated systems of stroke care on these needs.
Methods
Data Sources
We used public-use microdata files of the Canadian Community Health Survey (CCHS) cycles 2000–2001, 2003, 2005, 2010, 2012, and 2014 available under a data liberation initiative at the University of Toronto and pooled them to form the study cohort. CCHS is a nationwide cross-sectional survey of households that uses three sampling frameworks: area frame, telephone frame, and random digit dialing.Reference Béland11 People living on Indian Reserves and Crown Lands, institutional residents, and full-time members of the Canadian forces are not included in the survey. Information on various factors is collected over telephone by a trained interviewer using well-developed survey questionnaire. We included information from those over the age of 40 years and excluded information from respondents in the Northwest Territories, Yukon and Nunavut due to lack of information on stroke services in these jurisdictions.
Exposure, Outcome, and Intervention
Information on stroke was obtained by asking, “Do you suffer from effects of a stroke?.” The respondent could answer “yes,” “no,” or not respond. This was dichotomized and observations with missing values were excluded in the final analysis.
The primary outcome was self-perceived unmet health care need, which was ascertained by asking, “During the past 12 months, was there ever a time when you felt that you needed health care but you didn’t receive it?” and the respondent could answer, “yes,” “no,” or not respond. This was dichotomized and observations with missing values were not included in the final analysis. In the 2012 survey, questions on unmet health care needs were only included in selected provinces: Ontario, British Columbia, and Nova Scotia. In a sensitivity analysis, the data from this cycle were excluded.
A binary variable, intervention, was created based on the province of residence of the respondent. Those in the six provinces that implemented integrated systems of stroke care (Ontario, British Columbia, Alberta, Nova Scotia, Prince Edward Island, and Quebec) were deemed to have received the intervention, while the four that did not have such systems (Newfoundland, New Brunswick, Manitoba, and Saskatchewan) formed the comparison group without the intervention. In addition, a binary variable, time, corresponding to before and after the implementation of stroke systems, was created by classifying data from CCHS cycles 2000–2001, 2003, and 2005 as before and from CCHS cycles 2007–2008, 2010, 2012, and 2014 as after.
Statistical Analyses
We compared the characteristics of stroke survivors before the intervention in provinces with the intervention compared to those without. The same was repeated for the time after intervention. We used t-tests or Wilcox rank sum tests to assess differences in continuous variables, and χ 2 tests to assess differences in categorical variables.
We developed a multivariable log-binomial generalized estimating equations model, accounting for clustering among provinces, with stroke as the independent variable of interest and unmet health care needs as the dependent variable. The log-binomial model has been shown to be conservative compared to logistic models, especially for cross-sectional data.Reference McNutt, Wu, Xue and Hafner12 We adjusted for demographic information including age (10-year age groups, with 60–69 years as the reference category), sex (male as reference category), immigration status (non-immigrant as reference category), ethnic minority (Caucasian as reference category), marital status (married or common law as reference category, with single and widow or separated as other categories), education (3-level categorical scale: less than grade 12 as the reference category, greater than grade 12, postgraduate certificate or postgraduate degree), risk factors including smoking status (3-level categorical scale: current, former, and never; current as reference category), alcohol use (current vs. not currently drinking, the latter is reference), hypertension, diabetes, and heart disease (self-reported health conditions, binary variables). We chose these variables based on prior knowledge of factors that are related to both stroke and unmet health care needs, and to reflect what has been previously accounted when studying unmet health care needs.Reference Ronksley, Sanmartin and Quan6 We did not include sampling weights of the CCHS cycle. Key assumptions of model fit, multicollinearity, and impact of influential observations were assessed. Using subgroup analyses, we plotted the trend of adjusted risk ratio (aRR) of having unmet health care needs among stroke survivors compared to the general population in provinces with and without intervention for each CCHS cycle. Then, we obtained aRRs of unmet care needs of stroke survivors compared to the general population before and after 2006, in provinces with and without the intervention using subgroup analyses. Lastly, we obtained the difference in differences (DiD) estimate by adding a three-way interaction term of time and intervention and stroke. The assumption of parallel trends associated with the DiD approach was assessed by the trend plots described above.
We created a subsample by only including stroke survivors. Using this sample, we identified the predictors of unmet care needs in stroke survivors. We also performed the DiD analyses in this cohort using a two-way interaction between time and intervention.
To address the possibility that a change in unmet care needs of stroke survivors could be attributable to underlying temporal trends rather than the intervention, we performed the same analyses in patients with migraine and cancer, separately. The results of all analyses were reported with 95% CI around the estimates and P-values. A two-tailed alpha level of 0.05 was set as a cutoff for statistical significance. All data analyses were performed using SAS 9.4 Copyright© 2002–2012 by SAS Institute Inc., Cary, NC, USA.
Results
We included data from a total of 350,084 participants, with the majority (n = 283,373, 80.9%) living in provinces where an integrated stroke system was implemented. Provinces that did not receive the intervention had lower immigrant density and a lower proportion of respondents with postgraduate education than those that received the intervention, both before and after 2006 (Table 1). Before 2006, the proportion of stroke survivors who reported unmet health care needs was lower in provinces with the intervention compared to those without (30.7% vs. 36.5%, P < 0.001). This difference was eliminated after 2006 (16.2% vs. 13.5%, P = 0.19).
Characteristics of adult Canadian stroke survivors using data from the Canadian Community Health Surveys (2001–2014)
£ Before 2006, personal income categories were no income, <15,000, 15–29,000, 30–49,000, 50–79,000, and over 80,000.
$ Before here refers to before integrated stroke systems were implemented (2006), after refers to after their introduction (2006 onwards).
€ Present and absent refer to presence or absence of integrated stroke systems.
The aRR of having unmet health care needs in stroke survivors compared to the general population was 1.27 (95% CI, 1.22–1.32) (Supplementary Table 1), with similar findings in provinces with and without integrated systems of stroke care (Figures 1 and 2). We ensured that model assumptions for multicollinearity, model fit, and overspecification were appropriately met. Among stroke survivors, younger age (40–49 years), low income, diabetes, and heart disease predicted a higher risk of having unmet health care needs (Table 2).
Unmet health care needs in Canadians with disease (stroke, cancer, or migraine) compared to the general population, measured as aRRs, before and after introduction of integrated systems of stroke care (year 2006) using data from Canadian Community Health Surveys (2000–2014 cycles). Note: Confidence intervals for cancer subgroup are not adjusted for clustering. P-values in the parenthesis represent the significance of interaction term (disease*time).
Unmet health care needs in Canadian stroke survivors compared to the general population over time represented as aRRs, using cross-sectional data from the Canadian Community Health Survey (2000–2014 cycles).
Results of log-binomial generalized estimating equation (GEE) model for unmet health care needs among Canadian stroke survivors (n = 5976)
Quasi-likelihood under the independence model criterion (QIC) = 6669.13.
* Adjusted for province of residence.
After accounting for temporal trends, there was no association between the implementation of integrated systems of stroke care and unmet health care needs of stroke survivors relative to the general population (P-value for three-way interaction term = 0.63) (Supplementary Table 2). These results did not differ significantly when data from the 2012 cycle were excluded in sensitivity analyses. When restricted to stroke survivors only, the improvement in unmet health care needs after 2006 was greater in provinces without integrated systems of care than in provinces with such systems(P-value for two-way interaction term = 0.024) (Supplementary Table 2).
Patients with migraine and cancer had a higher risk of unmet health care needs compared to the general population, with an aRR 1.21 (95% CI, 1.19–1.24) for migraine and an aRR 1.49 (95% CI, 1.45–1.53) for cancer. In provinces with integrated systems of stroke care, unmet health care needs of cancer survivors improved after 2006, but worsened for migraine sufferers (Figure 1).
Discussion
We found that Canadian stroke survivors had higher self-perceived unmet health care needs than the general population. Younger age, the presence of comorbid conditions, and low self-rated health predicted higher unmet health care needs among stroke survivors. While the proportion of stroke survivors with unmet health care needs decreased between 2000 and 2014, after accounting for temporal trends there was no association between implementation of integrated systems of stroke care in 2006 and unmet health care needs of stroke survivors.
Our findings of higher unmet health care needs in patients with a chronic medical condition (stroke, migraine, and cancer) compared to the general population are consistent with previous research with one study reporting a cumulative impact of chronic conditions on the odds of unmet health care needs.Reference Ronksley, Sanmartin and Quan6,Reference Reid, Metcalfe, Patten, Wiebe, Macrodimitris and Jetté13 Although a prior study did not find an association between stroke and unmet health care needs, it only included information from the 2001 to 2005 CCHS cycles and did not account for provincial clustering.Reference Ronksley, Sanmartin and Quan6 In contrast, a study from Australia reported that up to 84% of stroke survivors had unmet health care needs.Reference Andrew, Kilkenny and Naylor14 Lack of rehabilitation services and insufficient social supports were considered the most common causes of unmet health care needs by stroke survivors.Reference Andrew, Kilkenny and Naylor14,Reference de Haan, Scholte op Reimer, Rijnders, Limburg and van den Bos15
Consistent with previous research in Canada, we found a higher risk of unmet health care needs in stroke survivors who also had comorbid chronic conditions such as heart disease or diabetes.Reference Ronksley, Sanmartin and Quan6 A UK-based study found that younger stroke survivors were more likely than older ones to report unmet needs for intellectual fulfilment and family support.Reference Kersten, Low, Ashburn, George and McLellan16 Among Australian stroke survivors, age greater than 65 years was associated with a lower adjusted incidence rate ratio (aIRR) of unmet health care needs (aIRR 0.62; 95% CI 0.50–0.77).Reference Olaiya, Cadilhac and Kim17 Consistent with this, we found that younger age was associated with higher unmet health care needs among stroke survivors, confirming the social and mental health needs of young stroke survivors.Reference McKevitt, Fudge and Redfern5 While other studies administered survey questionnaires to stroke survivors in a particular jurisdiction, we used a population-based sample from national surveys which may explain the differences in the magnitude of association between age and unmet health care needs in stroke survivors.Reference Kersten, Low, Ashburn, George and McLellan16,Reference Olaiya, Cadilhac and Kim17
The improvement in unmet health care needs of stroke survivors over the study period is encouraging and aligns with overall improvements in the quality of stroke care in Canada and other high-income countries over the past decades.Reference Feigin, Forouzanfar and Krishnamurthi18 Despite using DiD analysis, a robust method of treatment effect estimation for policy interventions, we found no association between the implementation of integrated systems of stroke care and improvements in unmet health care needs after stroke. In addition, in the sample restricted to stroke survivors only, the improvement in unmet health care needs of stroke survivors after 2006 was greater in provinces without integrated systems of stroke care compared to the provinces with these systems. This may be explained by underlying temporal trends in unmet health care needs in the general population which, if not accounted for, can produce biased estimates in the DiD analysis.Reference Dimick and Ryan19
Strengths of our study include the use of a nationally representative sample, the ability to account for clustering among provinces, and the use of advanced statistical analytical methods for assessment of a policy intervention. However, some limitations merit discussion. First, the validation of self-reported stroke in CCHS is low (kappa 0.36, 95% CI 0.34–0.39) compared to a case definition of stroke of one hospital admission with a cardiovascular disease diagnosis code; however, misclassification of stroke is unlikely to be related to province of residence, the implementation of integrated system of stroke care, or to time.Reference Muggah, Graves, Bennett and Manuel20 Second, we did not have information on the time between the survey date and that of stroke events. We also did not have any information on stroke type or severity, associated deficits, processes of stroke care, or discharge destination, all of which could potentially influence our primary outcome. We did not have information on nonrespondents which may limit the generalizability of our findings. Third, the unmet health care needs reported by stroke survivors may be due to factors others than stroke, and thus may not be affected by the implementation of integrated stroke systems. However, the higher unmet health care needs in stroke survivors compared to the general population after adjusting for various factors suggest that being a stroke survivor is an independent predictor of unmet health care needs. Fourth, provinces that were classified as not having implemented integrated systems of stroke care may have employed other measures to improve stroke care. In addition, the implementation of integrated systems of stroke care was heterogeneous across provinces, we were unable to use exact dates for implementation of such systems in the regression analysis, and using a 2006 cutoff may have diluted the ability to detect the impact of these systems. We also had limited information on the actual success of the implementation of organized stroke systems in all provinces, and on the proportion of people with stroke who received integrated stroke care. For all of these reasons, our finding of a lack of association between stroke systems of care and reduction in unmet health care needs should not be taken as evidence of a lack of effect of such systems.
In conclusion, we found that stroke survivors have higher unmet health care needs compared to the general population, but that these needs have reduced over time. Further work is required to identify the nature of these unmet health care needs and to determine how to address them.
Disclosures
MKK holds Mid-Career Investigator Awards from the Heart and Stroke Foundation of Canada. MVV holds a Fellowship Award from the Canadian Institutes of Health Research. JF does not have anything to declare.
Statement of Authorship
MVV was responsible for study conception, design of the study, data acquisition, and analysis and interpretation of results. JF was responsible for design of the study, and analysis and interpretation of results. MKK was responsible for study conception, design of the study, and interpretation of results. MVV wrote the manuscript that was critically assessed JF and MKK.
Availability of Data and Supporting Materials Section
Dataset including statistical code can be available from corresponding author.
Supplementary Material
To view supplementary material for this article, please visit https://doi.org/10.1017/cjn.2019.319
