The impact of stroke, cognitive function and post-stroke cognitive impairment (PSCI) on healthcare utilisation in Ireland: a cross-sectional nationally representative study

The Irish Longitudinal Study on Ageing (TILDA) involves a nationally representative sample of 8,175 community-dwelling adults aged 50 years and older in the Republic of Ireland [33]. The present study is a cross-sectional analysis of wave 1 TILDA data, collected between 2009 and 2011 [34]. Cognitive assessment data were available for those who completed a health assessment in wave 1 (n = 5,859). Of 133 stroke survivors, 92 completed a cognitive assessment (see Fig. 1). A detailed description of the TILDA methodology is available elsewhere [33, 34]. This study follows the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) Guidelines [35].

The main outcome of this study was self-reported healthcare use in the previous 12 months, including the number of General Practitioner (GP) visits, emergency department visits, outpatient visits,¹ and nights admitted to hospital. This study also explored the use of rehabilitation services,² specifically physiotherapy, occupational therapy and psychology.

Covariates included age, sex, education (none/primary school, secondary school/high school, third level/university), employment status (not working or working) and depression (higher scores on the Center for Epidemiological Studies Depression Scale (CES-D)) [42]. Additional covariates included disability status (no disability or (i) disability in activities of daily living (ADL) necessary for basic functioning, e.g., walking, dressing, or (ii) disability in instrumental activities of daily living (IADL) which allow individuals to live independently, e.g., cooking, managing finances or medications). Questions relating to disability were based on validated scales, namely The Index of Independence in Activities of Daily Living [43], and The Lawton Instrumental Activities of Daily Living Scale [44]. An indicator of socioeconomic status was derived from possession of a medical card/ GP visit card (yes or no). In Ireland, lower income groups often qualify for the medical card, which entitles holders to free or subsidised public healthcare, while the GP visit card permits free GP care for all individuals aged 70 years and older [45].

A combined stroke and cognitive status variable was generated to describe the sample. This variable comprised the following four categories; (i) no stroke, no cognitive impairment, (ii) no stroke, cognitive impairment, (iii) stroke, no cognitive impairment, and (iv) stroke, cognitive impairment. Relationships between the combined exposure variable and categorical covariates (e.g., gender, education, disability status) were explored using chi-square tests, or one-way ANOVA for continuous covariates (e.g., age). As the combined exposure variable included more than two categories, Cramer’s V effect size is reported. Associations between the combined exposure variable and healthcare utilisation (GP visits, emergency visits, outpatient visits and number of nights in hospital) were explored using the Kruskal–Wallis test for non-normally distributed outcomes. Two-tailed tests with an α-level of 0.05 for statistical significance were applied. Chi-square tests were employed to examine utilisation of rehabilitation services, a dichotomous outcome (yes/no).

The continuous MoCA score variable was reverse-scored and converted to a standardized z-score for the regression models, which re-interprets scores in terms of standard deviations from the mean [46]. Multivariate negative binomial regression, reporting incidence-rate ratios (IRR) for effect sizes, was used to model the relationship between stroke, cognitive function and each healthcare utilisation variable separately. This regression technique is suitable for continuous count variables, and is preferred over Poisson regression when data are over-dispersed (variance larger than the mean) [47]. Utilisation of rehabilitation services in the last 12 months (physiotherapy, occupational therapy and psychology) was also explored. These outcomes were combined and modelled as one single variable, which represented use of one or more rehabilitation services. Multivariate logistic regression, reporting odds ratios (OR) for effect sizes, was applied to model use of rehabilitation services. Due to the small numbers utilising rehabilitation services overall, the adjusted p-value was reported using Fisher’s exact test.

In the unadjusted and adjusted multivariate analyses, incidence-rate ratios (IRR), odds ratios (OR) and 95% confidence intervals (CI) are presented. Collinearity was investigated by examining the correlation between independent variables; where two variables were highly correlated (e.g., polypharmacy and medical card), the covariate with the lowest correlation with other independent variables in the model, was selected for inclusion in the regression. A number of covariates, namely healthcare coverage, the presence of one or more cardiovascular conditions and the presence of long-term health problems, had negligible effects on the overall results and were excluded from the final regression models (see Additional File 1). The final adjusted models included demographic variables, medical card status, disability (impairment in ADL/ IADL) and depression. The main effects and the interaction between the two exposure variables (stroke and cognitive impairment) were explored in these models. First, healthcare utilisation rates were compared between stroke and non-stroke participants. Second, the impact of varying degrees of cognitive function on healthcare utilisation was evaluated. Finally, the combined impact of stroke status and cognitive function on health service utilisation was explored using interactions. Stata 14 was used to analyse the data [48].

The TILDA dataset is a rich source of population health information, underpinned by a robust methodology. A major strength of this study is the large sample of nationally representative older adults. This study utilised the entire wave 1 cohort in order to compare healthcare utilisation between stroke and non-stroke participants, to establish the independent effects of stroke and cognitive function on healthcare utilisation. Cognitive function was measured using the MoCA, which is suitable for cognitively intact populations, and is more sensitive to the detection of mild cognitive impairment than the Mini-Mental State Examination (MMSE) [36]. However, it is important to note that cognitive screening tests are not diagnostic tools; rather, these assessments provide an indication of cognitive difficulties that require further investigation through comprehensive neuropsychological assessment. Hence, screening tests using standard cut-off values may fail to identify every individual with cognitive impairment. For this reason, the continuous MoCA score (converted to a standardised score) was utilised in the regression models, which increased the power of these analyses.

This study had several limitations. The number of stroke participants with a completed cognitive assessment was small (n = 92); hence, it was not appropriate to conduct multivariate regression analysis on the stroke sample specifically. In an effort to overcome this issue, the entire wave 1 sample was included in the regression models, and the relationship between stroke status and cognitive score was explored using an interaction test. However, the analysis may have lacked statistical power to identify interactions, due to the small number of participants in the stroke and cognitive impairment groups. In wave 1, less than half of participants with stroke (n = 51, 39%) had experienced stroke within the previous 3 years, and healthcare utilisation rates may differ significantly among those with less recent stroke. Limitations pertaining to the public TILDA dataset were also evident. Speech and language therapy was not available in this dataset for analysis, though the numbers using this service were extremely small (n = 19/8504) [66], and are unlikely to have changed the presented results. Additionally, the dataset did not include information on the reason for hospital re-admission or Emergency Department presentation, which would have influenced length of hospital stay and patient outcomes. It was not possible to adjust for a number of potential confounders; for example, this study did not explore the impact of non-cardiovascular comorbidities and chronic health conditions, which could increase healthcare utilisation in older populations [67]. Healthcare coverage was selected as a proxy for socioeconomic status in the regression analysis. Given that most of the Irish population over age 70 possess a medical card or GP visit card, a more specific measure of socioeconomic status would have been preferable.

Other limitations related to the generalisability of the findings. Firstly, healthcare utilisation in TILDA was self-reported, which raises the question of recall accuracy, especially among participants with cognitive impairment. Likewise, cognitive impairment can affect insight and reduce an individual’s awareness of their own health status, suggesting that self-reported diagnosis may be less accurate among those affected by cognitive decline. Selection bias is also a concern [68], in that cognitive data were only available from participants who took part in the TILDA health assessments, and these respondents tended to be younger and more able-bodied, with higher levels of education than the overall TILDA sample [69]. Thirdly, misclassification bias during data collection is also a possibility [68]; participants who experienced minor strokes or transient ischaemic attacks for which medical attention was never sought, may not have been classified appropriately (non-differential misclassification). Finally, nursing home residents and people with known or suspected dementia were excluded from wave 1 of TILDA [34]. Given these different potential sources of bias, the results of the present study may underestimate the prevalence of cognitive impairment. In addition, the utilisation rates reported in this study may underestimate health service use in more severe cases of stroke and cognitive impairment.

It was not possible to look at data in more recent waves of the public TILDA datasets as many of the required variables (e.g., healthcare utilisation, stroke status or cognitive status) were either not available, or were recoded as categorical variables, making comparison problematic. However, the results of this study are still relevant. Post-acute stroke care in Ireland has not changed significantly over the last decade [70], and access to rehabilitation continues to be an enduring problem for stroke survivors living in the community [29, 31, 64]. This study provides an indication of healthcare utilisation in stroke, and has identified poor cognitive function as a potential barrier to accessing outpatient services. These findings have important implications for the future planning of healthcare services for stroke survivors and individuals with cognitive impairment.