Associations of Chronic Kidney Disease With Dementia Before and After TIA and Stroke

Background and Objectives Individuals with chronic kidney disease (CKD) appear to be at increased risk of cognitive impairment, with both vascular and neurodegenerative mechanisms postulated. To explore the vascular hypothesis, we studied the association between CKD and dementia before and after TIA and stroke. Methods In a prospective, population-based cohort study of TIA and stroke (Oxford Vascular Study; 2002–2012), pre-event and new postevent dementia were ascertained through direct patient assessment and follow-up for 5 years, supplemented by review of hospital/primary care records. Associations between pre-event dementia and CKD (defined as an estimated glomerular filtration rate [eGFR] <60 mL/min/1.73 m2) were examined using logistic regression and between postevent dementia and CKD using Cox and competing risk regression models, adjusted for age, sex, education, stroke severity, prior stroke, white matter disease, diabetes mellitus, and dysphasia. Results Among 2,305 patients with TIA/stroke (median [interquartile range] age, 77 [67–84] years, 1,133 [49%] male, 688 [30%] TIA), 1,174 (50.9%) had CKD. CKD was associated with both pre-event (odds ratio [OR] 2.04 [95% confidence interval (CI) 1.52–2.72]; p < 0.001) and postevent dementia (hazard ratio [HR] 2.01 [95% CI 1.65–2.44]; p < 0.001), but these associations attenuated after adjustment for covariates (OR 0.92 [0.65–1.31]; p = 0.65 and HR 1.09 [0.85–1.39]; p = 0.50). The results were similar when a competing risk model was used (subdistribution HR [SHR] 1.74 [1.43–2.12]; p < 0.001, attenuating to 1.01 [0.78–1.33]; p = 0.92 with adjustment). CKD was more strongly associated with late (>1 year) postevent dementia (SHR 2.32 [1.70–3.17]; p < 0.001), particularly after TIA and minor stroke (SHR 3.08 [2.05–4.64]; p < 0.001), but not significantly so after adjustment (SHR 1.53 [0.90–2.60]; p = 0.12). Discussion In patients with TIA and stroke, CKD was not independently associated with either pre- or postevent dementia, suggesting that renal-specific mechanisms are unlikely to play an important role in aetiology.

Chronic kidney disease (CKD) is associated with a significant burden of cognitive impairment that worsens with declining renal function. 1 In the Reasons for Geographic and Racial Differences in Stroke (REGARDS) study, each 10 mL/min/ 1.73 m 2 decrease in estimated glomerular filtration rate (eGFR) <60 mL/min/1.73 m 2 was associated with an 11% increase in prevalence of cognitive dysfunction. 2 Patients on hemodialysis are 3 times more likely to have severe cognitive impairment than age-matched patients not on dialysis with reported prevalence rates of 30%-40%. 3 Both vascular and neurodegenerative hypotheses have been proposed to underlie CKD-related cognitive impairment. 4,5 In support of the vascular hypothesis, there is a high prevalence of cardiovascular risk factors 6 and a strong blood pressure-dependent association with stroke in CKD. 7 Furthermore, impairment of executive function and processing speed are prominent in CKD, consistent with cerebrovascular disease. 8 Although CKD may also augment neurodegeneration through the interplay of hypertension and Alzheimer pathology, 9 or via high concentrations of uremic toxins such as neuroexcitatory guanidine compounds, 10 any association between CKD and dementia might simply be due to increased cerebrovascular disease associated with CKD. For example, silent cerebral infarction is increased in patients with CKD 11 and is associated with cognitive impairment. 12 However, in one previous study, CKD was predictive of all-cause dementia independent of both previous symptomatic cerebrovascular disease and small vessel disease on brain imaging. 13 We therefore aimed to determine the associations between CKD and allcause dementia before and after TIA or stroke, with adjustment for measures of severity of the initial event, vascular risk factors, and other determinants of susceptibility in a large, longitudinal population-based study with standardized assessment of dementia to 5-year follow-up.

Methods
Consecutive patients with TIA or stroke were prospectively recruited from April 1, 2002, to March 31, 2012, from the Oxford Vascular Study (OXVASC), an ongoing populationbased study of all acute vascular events (including TIA, stroke, acute coronary syndromes, and peripheral vascular events). The study population comprises all 92,728 individuals, irrespective of age, registered with about 100 general practitioners (GPs) in 9 general practices in Oxfordshire, United Kingdom. The methodology of OXVASC was approved by the Oxfordshire Research Ethics Committee. Multiple methods of ascertainment are used to ascertain patients with TIA or stroke, as detailed elsewhere. 14 Briefly, multiple overlapping methods of hot and cold pursuit are used to achieve near-complete ascertainment of all individuals with TIA or stroke and to minimize selection biases in determining dementia risk. 15 These include a daily, rapid access TIA clinic to which participating GPs and the local emergency department (ED) refer all individuals with unhospitalized TIA or stroke, daily searches of ward admissions (medical, cardiology, stroke unit, neurology), ED attendance register and in-hospital bereavement office death records, and monthly searches of death certificates, coroner's reports (for out-of-hospital deaths), GP and hospital diagnostic/discharge codes, and brain/vascular imaging referrals. Written informed consent (or assent from relatives) was obtained for study interview and follow-up, including ongoing review of all primary care/ hospital records and centralized health and death data. eAppendix 1 (links.lww.com/WNL/B712) provides further details on the study population and case ascertainment methodology.
Patients were assessed by a study physician as soon as possible after their TIA/stroke, usually within 1 or 2 days of the event. TIA and stroke were defined using WHO criteria 16 (i.e., patients with infarction on brain imaging but symptoms lasting <24 hours were classed as TIA) with review of all cases by the same senior vascular neurologist (P.M.R.) throughout the study (see eAppendix 1, links.lww.com/WNL/B712, for definitions of events). A detailed clinical history was recorded in all patients by interview using a standardized questionnaire, including education; medical history and vascular risk factors were recorded in all patients, supplemented by primary care records (see eAppendix 1). Premorbid functional status was assessed using modified Rankin Scale and Barthel scores and stroke severity was assessed with the National Institutes of Health Stroke Scale (NIHSS) score. Baseline brain and vascular imaging and other investigations were performed as reported previously. 14,17 CKD was defined as eGFR less than 60 mL/min/1.73 m 2 for 3 or more months as per 2012 Kidney Disease: Improving Global Outcomes (KDIGO) guidelines. 18 eGFR was estimated using the Chronic Kidney Disease Epidemiology Collaboration Equation (CKD-EPI). eGFR was then categorized into 5 groups based on modified CKD classification by the National Kidney Foundation-Kidney Disease Outcomes Quality Initiative: eGFR ≥90, 60-89, 30-59, 15-30, and <15 mL/min/1.73 m 2 . For the purpose of statistical analysis, eGFR ≥60 mL/min/1.73 m 2 was used as the Glossary CI = confidence interval; CKD = chronic kidney disease; DSM-IV = Diagnostic and Statistical Manual of Mental Disorders, 4th edition; ED = emergency department; eGFR = estimated glomerular filtration rate; GP = general practitioner; HR = hazard ratio; IQR = interquartile range; MCI = mild cognitive impairment; MMSE = Mini-Mental State Examination; NIHSS = National Institutes of Health Stroke Scale; OR = odds ratio; OXVASC = Oxford Vascular Study; PICH = primary intracerebral hemorrhage; SHR = subdistribution hazard ratio; SVD = small vessel disease. reference, as there were no poststroke dementia events for eGFR ≥90 mL/min/1.73 m 2 in some of the subgroup analyses. Similarly, the eGFR 15-30 and <15 mL/min/1.73 m 2 groups were combined as the individual numbers within each group were small. For details on the brain imaging, see eAppendix 1, links.lww. com/WNL/B712. Assessments were made blind to clinical data. A qualitative scale was used (Oxford scale) based on the white matter disease severity score (absent, mild, moderate, or severe) of the Blennow scale for CT scans and a modified version of the Fazekas scale for MRI scans. 19 Multiple methods of follow-up were used to reduce attritional biases in identification of dementia. 20 Follow-up interviews were done by trained nurses or study physicians at 1 month, 6 months, 1 year, and 5 years. If clinic follow-up was not possible, patients were assessed at home or via telephone. Cognitive function was tested with the Mini-Mental State Examination (MMSE) 21 and Montreal Cognitive Assessment 22 at face-to-face interviews and with the telephone Montreal Cognitive Assessment and the Telephone Interview for Cognitive Status-modified. 23 Dementia was defined as pre-or postevent according to whether the diagnosis was made before or after the index event, as described previously (see eAppendix 1, links.lww. com/WNL/B712). 15,20,24,25 Briefly, pre-event dementia diagnosis was made using the following information: (1) baseline clinical assessment by study physician and discussion with relatives or other informant; (2) the presence of any dementia diagnosis, and related consultations and investigations, where available, in the primary care record, with hand-searching of the entire record including individual consultations, clinic letters, and hospitalization documentation. The diagnosis of pre-event dementia was made by an experienced consultant physician with subspecialty interest in dementia (S.T.P.) using DSM-IV criteria (eTable 1).
In patients without pre-event dementia, postevent dementia was diagnosed by S.T.P. using the same methodology (i.e., with clinical and cognitive assessment data and handsearching of primary care records to death or 5-year followup). 15,20,24,25 MMSE was done at each follow-up interview and dementia was diagnosed if MMSE was <24 and remained <24 for all subsequent follow-ups with fulfilment of other DSM-IV criteria including impairment of functional status. 15,20,24,25 In patients who had problems that interfered with testing (e.g., dysphasia), incomplete testing (e.g., because of blindness), were followed up by telephone, could not be tested at the study interview (e.g., because of severe deafness) follow-up, or who did not have a follow-up assessment, dementia was diagnosed on the basis of study records where available and hand-searching of primary care, hospital, and death records based on DSM-IV criteria (eTable 1, links.lww. com/WNL/B712). 15,20,24,25 Standard Protocol Approvals, Registrations, and Patient Consents Written informed consent was obtained from all patients (or guardians of patients) participating in the study (consent for research).

Statistical Analysis
Descriptive statistics were used to summarize baseline characteristics stratified by CKD status and according to eGFR categories. Continuous data (SD) or median (interquartile range [IQR]) and categorical data (n [%]) were compared with Mann-Whitney U tests (or Kruskal-Wallis tests if multiple groups) and χ 2 tests, respectively. Age-adjusted p values for difference were calculated using logistic and linear regression where appropriate.
Associations between CKD and pre-event dementia were determined by binary logistic regression to generate odds ratios (ORs) adjusted for (1) age, sex, and education (model 1) and (2) age, sex, education, and other factors previously reported as associated with pre-event dementia, 25 including white matter disease, prior stroke, diabetes mellitus, and also index stroke event severity (NIHSS) and dysphasia occurring after the index stroke (model 2).
All analyses involving postevent dementia were done after exclusion of cases of pre-event dementia. Cumulative incidence of dementia postevent was calculated by Kaplan-Meier methods censoring at death as described previously. 25 Cox regression was used to determine hazard ratios (HRs) for associations between overall 5-year risk of postevent dementia and separately for early (≤1 year) and late (>1 year) postevent dementia (patients with late dementia were excluded from analyses of early dementia and vice versa). Regression analyses were adjusted (1) for age, sex, and education (model 1); (2) for age, sex, education, and other factors reported as associated with dementia after TIA/stroke, 25 including stroke severity (NIHSS), white matter disease, dysphasia, prior stroke, and diabetes mellitus; and (3) with further adjustment for baseline cognitive test score (model 3). Similar analyses were done restricted to TIA and minor stroke (defined as NIHSS <3 as per OXVASC protocol) and major stroke (NIHSS ≥3). 25 We did not examine primary intracerebral hemorrhage (PICH) separately owing to small numbers with this stroke subtype, but we undertook sensitivity analyses for pre-and postevent dementia in which PICH was excluded. We performed further sensitivity analyses excluding recurrent stroke on follow-up in analyses of postevent dementia.
Given that death is a competing risk for dementia (i.e., it precludes its occurrence), we performed exploratory analyses using competing risks regressions using cumulative incidence function covariate analysis, similarly adjusted for the above covariates, to study the overall associations of CKD with postevent dementia, and, according to onset, early vs late postevent dementia. 33 From these regressions, we generated subdistribution HRs (SHRs) for comparison as these are thought to be more informative of risk or prognosis whereas standard (cause-specific) HRs are thought to be better indicators of etiology. 34 Results were considered significant at p < 0.05. Statistical analyses were performed using SPSS version 25.0 (SPSS Inc.) and Stata version 13 (Stat Corp.).

Data Availability
Requests for access to data should be submitted for consideration to the OXVASC Study Director (peter.rothwell@ ndcn.ox.ac.uk).

Results
Of 2  There was a significantly greater risk of postevent dementia in patients with CKD compared to those with normal renal function (log-rank p < 0.001; Figure 1), and this was particularly marked in patients whose index event was a TIA or minor stroke (eFigure 1, links.lww. com/WNL/B712). However, although in unadjusted Cox regression analysis, CKD was strongly associated with postevent dementia to 5

Discussion
In a large, prospective, population-based cohort study, CKD was not independently associated with either pre-or postevent dementia in patients with TIA and stroke. Associations were greatly attenuated after adjustment for age, sex, education, and other factors previously shown to be associated with dementia including stroke severity and white matter disease, 25 suggesting that renal-specific neurodegenerative mechanisms are unlikely to play an important role in the relationship between CKD and dementia, at least not in mild to moderate predialysis stages. However, the consistently higher prevalence rates of pre-and postevent dementia in patients with CKD does identify them as a vulnerable group.
Although CKD has been consistently associated with an increased risk of cognitive decline, 2,26,27 we did not find a strong independent association with dementia in our study, even at greater levels of renal dysfunction or when restricted to major stroke events. There are a number of potential explanations for this discordance. First, most of the cognitive studies to date have focused on the relationship between CKD and mild cognitive impairment (MCI) rather than dementia. 28,29 Second, in contrast to our study, where most patients had predialysis CKD and the majority had stage 3 (mild) CKD, previous dementia studies have been mainly dialysisbased. 30,31 Dialysis patients have unique additional risk factors intrinsic to the dialysis procedure including cerebral hypoperfusion and this intradialytic hemodynamic instability causes transient cerebral stunning, leading to cumulative ischemic white matter changes and cognitive changes over time. 32 Third, although we did not see an association between more advanced CKD and dementia, this is in keeping with the literature that suggests that duration of kidney disease or the rate of eGFR decline correlate better with cognitive dysfunction than severity of CKD. 33,34 Mechanisms underlying the pathogenesis of MCI and dementia in CKD are poorly understood. It has been proposed that uremic neurotoxins interacting with neural progenitor cells, the brain vasculature, the glymphatic system, and monoaminergic neurons may play a role. 35 High concentrations of uremic toxins such as tumor necrosis factor can impair Abbreviations: CKD = chronic kidney disease; eGFR = estimated glomerular filtration rate; NIHSS = National Institutes of Health Stroke Scale; PICH = primary intracerebral hemorrhage. Subdistribution hazard ratios for 5-year incidence of postevent dementia accounting for the competing risk of death, according to CKD status and eGFR category, unadjusted and adjusted for age, sex, and education (model 1), and also for stroke severity, prior stroke, white matter disease, diabetes mellitus, and dysphasia (model 2), and model 2 also adjusted for baseline cognitive score (model 3). Values are subdistribution hazard ratio (95% confidence interval).
synapse function and memory. 36 The glymphatic clearance of waste products occurs primarily during sleep and CKD is associated with sleep disorders. 37 In addition, high levels of the circulating phosphaturic hormone fibroblast growth factor 23 (FGF-23), often found in CKD, have been associated with incident dementia. 38 However, the absence of an independent association between CKD and dementia in our patients with TIA/stroke questions any renal-specific mechanisms over and above the effects of CKD on cerebrovascular disease itself. For example, our associations were stronger for late (>1 year) postevent dementia, particularly in the TIA and minor stroke subgroup, in whom new vascular dementia is often related to recurrent strokes or progressive cerebral small vessel disease (SVD). 39 Severe SVD has been implicated as the most important mechanism in late poststroke dementia 40 and is strongly associated with CKD, especially at younger ages. 41 In addition to the large study size, population-based design, and comprehensive adjustment for potential confounders, and detailed consideration of various biases in detection of dementia, [26][27][28][29] a further strength of our study is the inclusion of a competing risk analysis to account for the risk of death-the importance of which is being increasingly recognized in survival analyses of CKD given the high baseline mortality rate. 42,43 The results from the competing risk analysis were qualitatively and quantitatively similar to those from the Cox (cause-specific) proportional hazards model with the exception of the associations between advanced CKD (eGFR <30 mL/min/1.73 m 2 ) and postevent dementia, which decreased. This likely reflects the stronger association between advanced CKD and the competing event (i.e., death) resulting in lower SHRs than expected, 44 and is consistent with the findings in a previous study relating dialysis to risk for developing dementia. 43 Our study has a number of limitations. First, TIA/stroke and dementia diagnoses were adjudicated by 2 investigators rather than by a consensus panel, but this ensured consistent diagnosis over the 15-year time period of this study. Second, pre-event dementia was retrospectively diagnosed, but the use of multiple sources should have minimized misclassification. Third, we did not study the relationship between CKD and specific subtypes of dementia because clinical classification is challenging in older patients and mixed pathology is the most common finding in neuropathologic studies. 45 Moreover, most dementia occurring after stroke is coded as being of "unspecified" cause in routine practice. 35 Fourth, we were unable to measure urine albumin excretion, which some studies suggest is more strongly and independently associated with cognitive dysfunction than eGFR. 46,47 Fifth, as most of the CKD population had only mild to moderate disease, we may not have captured independent associations with dementia that may exist in later stages or in those who are dialysis-dependent. In addition, this was a highly select population enriched with vascular risk factors and therefore, the results may not be entirely generalizable to less multimorbid CKD. Sixth, as GFR was estimated from serum creatinine, this can lead to an overestimation of GFR in those with low muscle mass. 48 Because muscle mass loss is also associated with cognitive decline, 49 this may have underestimated the effect size of any association between CKD and cognitive decline. Our study should ideally be replicated using GFR estimated from cystatin-C molecule independent of muscle mass, which may increase precision of eGFR equations and has also been shown to colocalize with β-amyloid in the brain. 50 Dementia is more common in patients with CKD than in the general population but predialysis CKD itself appears not to be independently associated with either pre-or poststroke dementia, except possibly with late-onset dementia in those with minor stroke events. Patients with CKD appear to have a clustering of risk factors associated with dementia including prestroke factors (advanced age, diabetes, and atrial fibrillation), stroke factors (greater event severity, dysphasia, and disability), and lower brain reserve (low education, premorbid dependency, and leukoaraiosis) that likely mediate much of the unadjusted relationship between CKD and dementia. Further studies are needed to determine if there are additional unique mechanisms or pathways leading to late-onset dementia in those with CKD and minor stroke events.