Systematic review and meta-analysis of the effect of cognitive impairment on the risk of admission to long-term care after stroke

Search strategy

Three databases were searched (PubMed, PsycInfo and CINAHL). Title, abstract and Pubmed Medical Subject Headings (MeSH) (mapped to similar topic terms in the other databases) terms for stroke, cognitive impairment, dementia, LTC, and discharge outcome were used in developing the search strategy. The Pubmed search strategy is provided as extended data (Table S1¹⁴). Searches were limited to English peer-reviewed journals between January 2000 and June 2018 to obtain the most current research and a feasible number of articles. Search results were exported to Endnote X8 and duplicates were removed. Abstracts were screened by a team of reviewers (AMc, LB, AS), with each abstract screened independently by two reviewers, followed by full text review of potentially relevant articles (JS & ES). Any disagreements were discussed until consensus was reached.

Inclusion and exclusion criteria

Studies that used a longitudinal, cross-sectional and observational design were included. Studies with ischaemic stroke patients only or mixed stroke (ischaemic and haemorrhagic, with or without transient ischaemic attack (TIA)) patients were included. Studies with haemorrhagic stroke patients or TIA patients only were excluded. Studies that assessed any definition of PSCI, including as a continuous or categorical variable, or as mild CI or dementia, were included. Studies that assessed pre-stroke cognition only were excluded. Eligible studies could be hospital-based or population-based and were included if they reported the discharge destination at any time post-stroke.

LTC was defined as a discharge setting outside home where a patient receives professional support from qualified staff for a long period of time. Terminology used for settings varies, and can include terms such as nursing home, residential care and skilled nursing facility. Destinations excluded from this broad definition were home living, home care, inpatient rehabilitation, acute care, transitional care, and delayed discharge. If discharge to LTC was examined in combination with other discharge outcomes (e.g., formal home care), the study was included but these were examined separately in stratified analysis.

Studies published in English in peer-reviewed journals were included. Qualitative studies, commentary or review articles, case reports and conference abstracts were excluded.

Data extraction

Data were extracted by JS using a standard form (Table S2 in extended data¹⁴) and checked by ES. Data items included: study population; setting; definition of PSCI; proportion with PSCI; discharge destination; effect sizes (e.g., odds ratio) and measures of uncertainty (e.g., confidence interval) for the association between PSCI and discharge destination and covariates adjustment.

Quality assessment

The Crowe Critical Appraisal Tool (CCAT) was used for quality assessment¹⁵. Individual studies were scored on a scale of 0 to 5 on eight dimensions; introduction, design, sampling, ethical matters, results, discussion and preliminaries (abstract and title). Quality Assessment was done by (JS) and checked by (ES). With a maximum possible score of 40, three categories of study quality were created; high quality (CCAT score >30, a mean score >3.8/5 across dimensions), medium quality (CCAT score 25–30, or a mean score of 3–3.8/5 across dimensions) and low quality (CCAT score <25, or a mean score <3/5 across sections).

Analysis

The analysis included a narrative synthesis and a quantitative meta-analysis. For the narrative synthesis, study results were tabulated summarising the population, definition of PSCI, discharge destination, covariate adjustment and results.

Where feasible, study estimates were pooled in a random-effects meta-analysis. To facilitate the quantitative pooling of studies, results from studies were converted to Odds Ratios (OR) and 95% Confidence Intervals (CI), where possible, consistent with Cochrane guidelines¹⁶. To convert the mean scores to ORs, two effect size conversion calculators were used. Mean cognitive functioning scores for discharge destinations in individual studies were first converted to Cohen’s d and its confidence intervals¹⁷. ORs and their confidence intervals were subsequently calculated from Cohen’s d¹⁸.

The meta-analysis was divided into two parts: i) analysis of association between PSCI and LTC, and ii) association between post-stroke dementia and LTC. Within the PSCI analysis, studies that defined PSCI as a continuous variable, and as a categorical variable, were pooled separately. The following sensitivity analyses were planned where feasible: how studies measured CI, study quality categories, definition of LTC and whether studies reported adjusted estimates. Heterogeneity was assessed using the I² statistic and interpreted using the Cochrane Review guidelines¹⁶. Data management and analysis were done using the Cochrane Review Manager 5 (RevMan 5) software.

For studies that had more than one measure of PSCI, estimates of the measure best representing the primary focus of this study was selected for the meta-analysis¹⁹, decided based on consensus discussions (JS and ES). If adjusted and unadjusted estimates were reported, the adjusted estimate was included in any analysis. If LTC was compared with more than one discharge destination (e.g. home, rehabilitation), the estimate using home as the reference group was included in the meta-analysis, as this was more similar to other studies in the review.

Included articles

A total of 1,219 unique records were identified and the full texts of 59 articles were reviewed (see Figure 1). The full texts of 30 articles were further examined in detail and 18 articles were selected for qualitative synthesis^20–37. Of these, 12 provided sufficient information for inclusion in meta-analysis.

Figure 1. Flow chart of included studies in systematic review and meta-analysis.

Table 1 displays the characteristics and results of the 18 included studies, with more detail provided in Table S2 (extended data¹⁴). Study settings were mixed, with rehabilitation centres (k=7) and stroke units (k=7) the most common recruitment settings, and two community-based studies. In total, 15 studies examined overall PSCI, while three examined dementia and one examined dementia and CIND separately. Of the studies examining overall PSCI, 8 examined PSCI as a categorical variable, and 7 as a continuous variable.

The most frequently used measure of cognitive function was the Mini-Mental State Examination (MMSE, k=5), with other measures used including the Repeated Battery for the Assessment of Neurological Status (RBANS)²³, the (MoCA)^24,32, and the cognitive category of the Functional Independence Measure (FIM)^22,25,30,31. Of the three studies that examined dementia, one used DSM-IV criteria²⁰, one used ICD-9²⁷ and one used Vascular Dementia criteria²¹.

Most studies adjusted for covariates such as age or sex. Overall functional impairment was adjusted for in 14 of the 18 studies, using a measure such as the modified Rankin scale (mRS) or Barthel Index^{20,23–25,27–31,33–37}.

Study quality

Two studies were in the low-quality category^22,32 (score <25); nine studies^{25–27,29–31,33–35} were in the medium-quality category (25–30) and seven studies^{20,21,23,24,28,36,37} were in the high-quality category (>30). Low quality (score <3/5) was most frequently observed in the sections on reporting results and analysis (k=5) and reporting ethical matters and approval (k=8), with poor performance defined as a score <3/5. Full quality appraisal scores by dimension are displayed in Figure S1 (extended data¹⁴).

PSCI and admission to LTC

PSCI was reported as significant predictor of LTC in 12 studies^{22,23,26,28–32,34–37}, with three studies failing to find a significant association^21,24,25.

PSCI as categorical predictor

Eight studies reported PSCI as a categorical predictor^{21,25,28,29,31,34–37}, with six^{28,29,31,34–36} reporting a significant association with admission to LTC, and two^21,25 reporting no significant association. One of these two examined CIND separately from dementia²¹, and the other was the only study in the review to focus on severe stroke alone²⁵.

Four studies^29,31,34,35 reporting sufficient information were pooled together in a random-effects meta-analysis. PSCI significantly increased the odds of LTC [OR: 2.36 (95% CI: 1.18, 4.71, p=0.01)] and there was substantial heterogeneity (I²=77%, p=0.004), see Figure 2. Two of the four studies had clear definitions of PSCI, including standard tests and cut-offs^29,35. When these were combined in a sensitivity analysis, there was no significant heterogeneity (I²=0%, p=0.75) and PSCI significantly predicted LTC, [OR: 3.11 (95% CI: 1.79, 5.43, p<0.0001)] (see Figure S2, extended data¹⁴).

Figure 2. Forest plot of PSCI and admission to LTC (studies reporting PSCI as a categorical variable).

One study combined several outcomes in the “poor discharge” category, including death before discharge²⁹. Excluding this study in a sensitivity analysis, PSCI did not significantly predict LTC [OR: 2.09 (95% CI: 0.98 to 4.45, p=0.06)] and heterogeneity was significant (I²=76%, p=0.002), (see Figure S3, extended data¹⁴). A further sensitivity analysis of three studies that reported adjusted estimates^29,31,35 indicated that PSCI significantly predicts LTC [OR: 2.13 (95% CI: 1.02, 4.46, p=0.04)]; (I²=76%, p=0.006) (see Figure S4, extended data¹⁴). All four studies were in the medium quality category, and no sensitivity analysis by quality category was carried out.

PSCI as a continuous predictor of admission to LTC

Seven studies reported PSCI as a continuous predictor of LTC^{22–24,26,30,32,37}. Of these, four studies^22,23,30,31 reported that lower levels of cognitive impairment (or better cognitive function) reduced odds of LTC. One study examined three PSCI measures (MMSE, CDT and CST), adjusting for all simultaneously, reporting that only one (CST) significantly predicted the discharge destination³⁷. One study using the MoCA did not find PSCI to significantly predict discharge destination²⁴, when adjusted for covariates, while another study found stroke patients in home care settings to have worse PSCI than stroke patients in long-term care²⁶. This study was excluded from the meta-analysis as it assessed cognitive deficits (i.e., a higher score represented worse cognition), while other studies assessed cognitive function (a higher score represented better cognition), and pooling estimates was not feasible²⁶. Another study did not report sufficient estimates to facilitate inclusion in a meta-analysis³².

Five studies^{22–24,30,37} were pooled together in a meta-analysis (Figure 3). Low levels of PSCI significantly reduced the odds of LTC [OR: 0.89 (95% CI: 0.84,0.95, p=0.0007)] and heterogeneity was not statistically significant, (I²=48%, p=0.10).

Figure 3. Forest plot of PSCI and admission to LTC (cognition measured as a continuous variable).

Results from the four studies that reported adjusted effect sizes^22,23,30,37 showed that heterogeneity reduced (I²=18%, p=0.30) and lower levels of PSCI remained significantly associated with reduced odds of LTC [OR: 0.91 (95% CI: 0.87, 0.95, p<0.01)] (see Figure S5, extended data¹⁴).

When estimates from the four studies^22,24,30,37 that defined LTC as a combination of nursing home/long term care and other destinations, were pooled together, low levels of PSCI significantly reduced odds of LTC [OR: 0.88 (95% CI, 0.79, 0.98, p=0.02)] and heterogeneity was moderate to substantial (I²=61%, p=0.05), see Figure S6 (extended data)¹⁶.

Three studies that were in the high study quality category^23,24,37 were pooled together. Low levels of PSCI did not significantly reduce odds of LTC [OR: 0.77 (95% CI: 0.57,1.05, p=0.10)] and heterogeneity was substantial (I²=69%, p=0.04) (see Figure S7, extended data¹⁴). Although all 5 studies^{22–24,30,37} used different cognitive assessment tools, each included a clear definition of PSCI, and a sensitivity analysis by PSCI definition was not conducted.

Dementia and admission to LTC

Four studies examined the relationship between dementia and LTC and they all found dementia to significantly increase the odds of LTC after stroke^20,21,27,33. Three studies^20,27,33 were pooled in a meta-analysis and dementia statistically significantly increased odds of LTC [OR (CI: 2.58 (95% CI, 1.38, 4.32, p=0.003)] and heterogeneity was substantial though not significant (I²=60%, p=0.80), see Figure 4.

Figure 4. Forest plot of post-stroke dementia and admission to LTC.

When two studies that were in the medium quality category^27,33 were pooled together in a sensitivity analysis, heterogeneity was increased (I²=68%, p=0.08) and dementia did not significantly predict LTC [OR (CI 95%): 2.70 (0.91, 7.97, p=0.07)] (see Figure S8, extended data¹⁴). All three studies had clear definitions of PSCI and LTC, and adjusted for key covariates, and sensitivity analyses were thus not carried out with respect to these factors.

Study quality

Studies had some consistency in the definition of CI with studies using the same cognitive screening tool reporting the same PSCI cut-off scores. However, some studies did not provide clear definitions of CI and the measures used for cognitive screening. Two studies described CI as cognitive deficits^31,34 and one study did not specify the cognitive screening tools used³⁴.

Few studies reported how they statistically handled missing data and some studies did not provide comprehensive information on ethical approval which affected the quality of the evidence. In some studies, the definition of LTC was combined with other destinations and stroke outcomes including death, contaminating the definition of LTC.

Strengths and limitations of the review

To our knowledge, this is the first systematic review and meta-analysis to examine the effects of CI and dementia post-stroke on admission to LTC. The review was conducted according to the PRISMA guidelines for conducting systematic reviews. Rigorous quality appraisal was undertaken using the CCAT. Sensitivity analyses were carried out in relation to definitions of CI and nursing home admissions, covariate adjustment, and study quality.

The meta-analysis was limited by the number of studies that were eligible for pooling. PSCI measured as a categorical and continuous variable had to be analysed separately. A number of studies with negative results did not report parameter estimates, and could not be meta-analysed, introducing a potential bias towards positive results in the meta-analysis. Only studies in high-income countries were identified, limiting the global generalizability of results.

Implications and conclusion

Previous studies have reported that LTC is resource-intensive care for stroke patients³. The findings from this review can inform population-based planning of social care services for stroke, and highlight the potential benefits of interventions that aim to prevent or delay CI post-stroke, particularly dementia, in reducing the costs associated with post-stroke LTC.

Underlying data

All data underlying the results are available as part of the article and no additional source data are required.

Extended data

Figshare: Extended Data.docx. https://doi.org/10.6084/m9.figshare.12291005.v1¹⁴

This projection contains the following extended data:

Table S1: Sample Search Strategy for Systematic Review

Table S2: Full Details of Included Studies

Figure S1: Quality Appraisal Results

Figure S2: Sensitivity analysis using definition of PSCI (CI as categorical predictor)

Figure S3: Sensitivity analysis using the definition of long-term care (PSCI as categorical predictor

Figure S4: Sensitivity analysis using whether studies reported adjusted estimates (PSCI as a continuous predictor).

Figure S5: Sensitivity analysis using whether studies reported adjusted estimates (PSCI as a continuous predictor).

Figure S6: Sensitivity analysis using how studies defined long-term care (PSCI as a continuous predictor).

Figure S7: Sensitivity analysis using study quality categories (PSCI as a continuous predictor).

Figure S8: Sensitivity analysis using study quality categories (dementia as a predictor of LTC).

Reporting guidelines

Figshare: PRISMA checklist for ‘Systematic review and meta-analysis of the effect of cognitive impairment on the risk of admission to long-term care after stroke’ https://doi.org/10.6084/m9.figshare.12291005.v1¹⁴

Data are available under the terms of the Creative Commons Zero "No rights reserved" data waiver (CC0 1.0 Public domain dedication).