Associations of Chronic Kidney Disease Markers with Cognitive Function: A 12-Year Follow-Up Study

Exposures (kidney damage and kidney dysfunction)

Albuminuria (encompassing macro- and microalbuminuria) was defined by albumin:creatinine ratio≥3.5 mg/mmol (women) or≥2.5 mg/mmol (men) from an early morning random spot urine sample. Kidney dysfunction was indicated by eGFR <60 ml/min/1.73 m² (where eGFR was calculated from serum creatinine according to the Chronic Kidney Disease Epidemiology Collaboration [CKD-EPI] equation [8]). The primary analyses were based on albuminuria/kidney dysfunction characterized at baseline (1999/00; i.e., given this afforded the longest duration of exposure). However, comparisons were also made on the basis of the 2004/05 classification to explore potential differences according to exposure duration (i.e., 12 years versus 7 years).

Since cystatin C (an alternate filtration biomarker that may hold advantages over creatinine for eGFR calculation) was also measured in 1999/00, we repeated the 12-year exposure analyses after redefining kidney dysfunction using two cystatin C-based eGFR equations (one using cystatin C only; the other both cystatin C and serum creatinine combined; available in 95% of the final sample) [9].

Outcome measures (cognitive function)

Two domains of cognitive function comprised the primary outcome measures [10]:

Since CVLT and SDMT scores were based on different scales of measurement (i.e., making it difficult to evaluate differences in memory and processing speed relative to each other), these data were re-expressed as z-scores (i.e., standardized values based on number of standard deviations from the mean).

Premorbid/age-independent cognition was estimated by level of educational attainment in all individuals; however, a formal assessment—the Spot-The-Word (STW) test of verbal ability [13]—was undertaken in a subgroup of 4,037 individuals (scored from 0 to 60 based on the number of ‘real’ words identified from 60 pairs of real versus non-words). This afforded sensitivity analyses of between-group differences adjusted for STW score, rather than education level, in 98% of the final sample.

Individuals 60 years and older were additionally screened for global cognitive impairment using the Mini-Mental State Examination (MMSE; <24 out of 30 indicating cognitive impairment) [14].

Covariates

Age, sex, education level, medical history, medication use, and clinical/behavioral risk factors were derived from a combination of interviewer-administered questionnaires and clinical/ biochemical data. Diabetes was deemed to be present in cases of a self-reported previous diagnosis and/or use of hypoglycemic medication, or if World Health Organization (WHO) oral glucose tolerance testing (OGTT) criteria were met [15]. Pre-diabetes was also classified on the basis of WHO OGTT criteria. Hypertension was defined by blood pressure≥140/90 mmHg (seated rest; mean of at least two recordings) and/or use of antihypertensive medication. Physical activity was quantified using a validated survey [16], based on which individuals were classified as being sufficiently active (i.e., meeting the public health-recommended volume of≥150 min/week moderate-intensity activity, or an equivalent volume at higher intensity), insufficiently active (i.e., <150 min/week moderate-intensity activity or equivalent), or inactive. Due to its previously reported non-linear relationship with cognition [17], alcohol intake was categorized as follows: abstainers; >0 but <10 g/day; ≥10 to <20 g/day; 20 + g/day (where 10 g defines 1 standard drink in Australia). Dyslipidemia was defined by any abnormality of total cholesterol (≥5.5 mmol/L), HDL cholesterol (<1.0 mmol/L), LDL cholesterol (≥3.5 mmol/L), or triglycerides (≥2.0 mmol/L), or use of lipid-lowering medication. Other covariates included body mass index (BMI; standard calculation from height and weight), smoking status (current versus former smoker versus never-smoked), and depressive symptoms (based on a Center for Epidemiologic Studies Short Depression Scale [18] score of≥10). All covariates were assessed at each time point except for depressive symptoms (2011/12 follow-up only).

Laboratory methods

Urinary albumin was analyzed by rate nephelometry with the Beckman Array (Beckman/Coulter, Fullerton, USA). Urinary creatinine in 1999/00 was derived from the modified kinetic Jaffé reaction (Olympus AU600 autoanalyzer; Olympus Optical Co. Ltd., Tokyo, Japan), whereas at follow-up it was based on the spectrophotometric-Jaffe alkaline picrate method (Roche Modular; Roche Diagnostics, Indianapolis USA). To ensure standardization of eGFR values across the entire cohort, serum creatinine was measured on stored, frozen (–80°C) samples by enzymatic assay calibrated to the Isotope Dilution Mass Spectrometry method (Roche Modular). Cystatin C was measured using an immunoturbidimetric assay (Roche/Hitachi 917, MODULAR P analyzer, Roche Diagnostics).

Plasma glucose, total cholesterol, HDL-cholesterol, and triglycerides were measured enzymatically using an Olympus AU600 analyzer (Olympus Optical). Glycated hemoglobin (HbA1c) was derived from HPLC (Bio-Rad Variant Hemoglobin Testing System).

Statistical analyses

Data were expressed as mean±standard deviation (continuous variables) or as percentages (categorical variables), with unadjusted group comparisons reflecting a t- or chi-square test, respectively. Transformations were applied to achieve normality where appropriate.

Independent associations of baseline (i.e., 1999/00) albuminuria and kidney dysfunction with cognitive function in 2011/12 were analyzed using a series of ANCOVA models with adjustment for specific sets of covariates. The base model (Model 1) featured age, sex, and education. Model 2 additionally adjusted for atherosclerotic cardiovascular disease (i.e., myocardial infarction or stroke), diabetes, and hypertension (i.e., major causes/consequences of CKD known to contribute to cognitive impairment; where atherosclerotic cardiovascular disease was a composite of previous history at baseline and incident events during the 12-year follow-up) [17]. In model 3, lifestyle and other risk factors influencing cognitive function were also included (i.e., BMI, physical activity, smoking, alcohol intake, and dyslipidemia) [17]. Where between-group differences were observed in model 1, the percentage changes in the standardized β from model 1 ⟶ 2, and model 1 ⟶ 3 in turn, were calculated. These values estimated the proportions of the effect of albuminuria/kidney dysfunction explained by the added covariates (i.e., beyond age, sex, and education).

Sensitivity analyses included the following: 1) additional adjustment for the presence of depressive symptoms at 12 years; and 2) replacement of education level with STW score in the fully adjusted model.

Interactions were tested by re-analyzing the fully adjusted model inclusive of an interaction term for each of the following covariates, in turn: age, sex, education level, atherosclerotic cardiovascular disease, diabetes, and hypertension (where p < 0.10 indicated the presence of an interaction effect). For the categorical outcome variable cognitive impairment, multiple logistic regression was applied to determine its associations with each exposure, independent of confounders. All analyses were undertaken using STATA version 14 (STATA, College Station, USA) and p < 0.05 defined statistical significance.

Participant characteristics

Clinical profiles of individuals with and without both albuminuria and kidney dysfunction in 1999/00 are displayed in Table 1. Although each CKD marker was associated with older age, the difference was less marked in the albuminuria group (∼5 years older than controls; mean 53 years) compared with the kidney dysfunction group (∼20 years older than controls; mean 68 years). Each of albuminuria and kidney dysfunction were associated with lower education level, higher prevalence of pre-diabetes/diabetes, hypertension, and myocardial infarction, higher mean values of BMI, systolic blood pressure, total cholesterol, and HbA1c, lower mean HDL cholesterol, and lower alcohol intake. Individuals with albuminuria also reported lower levels of physical activity and were found to have higher diastolic blood pressure (not observed in the kidney dysfunction group). In turn, those with kidney dysfunction tended to have higher LDL cholesterol (not observed in the albuminuria group). Table 1 also displays raw/unadjusted cognitive function data (full analyses described in the sections below). MMSE screening in the subgroup of individuals aged 60 + revealed cognitive impairment in 3.5%.

Albuminuria and cognitive function at 12 years

Individuals with albuminuria in 1999/00 had worse memory 12 years later, indicated by a lower mean CVLT score that persisted across all stages of covariate adjustment (Fig. 2). Relative to the base model inclusive of age, sex, and education, the proportion of this effect explained by the collective covariates of atherosclerotic cardiovascular disease, diabetes, and hypertension was 8% (model 2). When lifestyle and other risk factors were included, the proportion mediated increased to 21%.

Fig. 2.

Albuminuria and cognitive function measured at 12 years of follow-up. Data are between-group differences in CVLT and SDMT z-scores measured at the 2011/12 follow-up. All data are estimated marginal means with 95% confidence intervals (error bars). Error bars crossing the zero line indicate no significant difference between individuals with versus without albuminuria (p > 0.05). Models 1 to 3 reflect incremental adjustment for a progressively more covariates; i.e., Model 1 featured age, sex, and education level only; Model 2: Model 1 covariates + atherosclerotic cardiovascular disease, hypertension and diabetes; Model 3: Model 2 covariates + body mass index, physical activity, smoking, alcohol intake, and dyslipidemia. The percentage values indicate the proportion of the effect of albuminuria mediated by these incrementally added covariates (i.e., beyond the Model 1 covariates of age, sex and education). CVLT, California Verbal Learning Test; SDMT, Symbol Digit Modalities Test.

In contrast to memory, worse processing speed in individuals with albuminuria was only borderline significant in the base model, and largely explained by cardiometabolic factors (model 2 proportion mediated: 31%) and lifestyle/other risk factors (model 3 proportion mediated: 47%).

Kidney dysfunction and cognitive function at 12 years

No differences in memory or processing speed at 12 years were identified between individuals with and without kidney dysfunction at baseline (Fig. 3). In the subgroup of 3,867 individuals whose eGFR could be derived from the alternate formulae based on cystatin C alone and cystatin C/creatinine combined, results were similar (data not shown).

Fig. 3.

Kidney dysfunction and cognitive function measured at 12 years of follow-up. Data are between-group differences in CVLT and SDMT z-scores measured at the 2011/12 follow-up. All data are estimated marginal means with 95% confidence intervals (error bars). Error bars crossing the zero line indicate no significant difference between individuals with versus without kidney dysfunction (p > 0.05). Models 1 to 3 reflect incremental adjustment for a progressively more covariates; i.e., Model 1 featured age, sex and education level only; Model 2: Model 1 covariates + atherosclerotic cardiovascular disease, hypertension and diabetes; Model 3: Model 2 covariates + body mass index, physical activity, smoking, alcohol intake, and dyslipidemia. CVLT, California Verbal Learning Test; eGFR, estimated glomerular filtration rate; SDMT, Symbol Digit Modalities Test.

Duration of exposure

Of the primary analytic sample (n = 4,128), 3,333 individuals (81%) participated in the interim (2004/05) CKD screening assessment and had complete data for all model 3 covariates. This in turn enabled comparisons of cognitive function measured in 2011/12 between those with versus without albuminuria/kidney dysfunction from 1999/00 (12-year exposure) and from 2004/05 (7-year exposure). Rates of albuminuria and kidney dysfunction in this subgroup of n = 3,333 in 1999/00 were 3.3% and 1.0%, respectively (i.e., similar to the primary analytic sample). By 2004/05, these rates had increased to 4.0% and 2.3%. Between-group comparisons of CVLT and SDMT scores according to each of the 1999/00 and 2004/05 classifications are displayed in Fig. 4.

Fig. 4.

Between-group differences in cognitive function according to duration of exposure to albuminuria/kidney dysfunction. Between-group differences in CVLT and SDMT z-scores measured at the 2011/12 follow-up are shown in parts A and B, respectively. Data are estimated marginal means with 95% confidence intervals (error bars). Error bars crossing the zero line indicate no significant difference between individuals with versus without albuminuria/kidney dysfunction (p > 0.05). All data reflect adjustment for model 3 covariates (i.e., age, sex, education level, atherosclerotic cardiovascular disease, hypertension, diabetes, body mass index, physical activity, smoking, alcohol intake, and dyslipidemia; n.b. relevant 1999/00 or 2004/05 values were used for time-varying covariates). CKD, chronic kidney disease; CVLT, California Verbal Learning Test; eGFR, estimated glomerular filtration rate; SDMT, Symbol Digit Modalities Test.

The lower mean CVLT score of individuals with albuminuria from 1999/00 was similar to the primary analysis (n.b. although it was of borderline significance [p = 0.068], the magnitude difference was very similar [i.e., ∼–0.2 units lower]). This difference in CVLT score did not extend to the shorter exposure duration based on 2004/05 classification. In turn, no differences in the between-group comparison of CVLT score were observed based on kidney dysfunction defined in 1999/00 versus its classification in 2004/05. Likewise, SDMT differences were not apparent for albuminuria and kidney dysfunction, irrespective of exposure duration.

Global cognitive impairment

Neither albuminuria at the 1999/00 baseline, nor at 5 years (2004/05) significantly predicted cognitive impairment in the base model adjusted for age, sex, and education (OR 1.7 [95% CI 0.6 to 4.5], p = 0.29 and OR 1.3 [95% CI 0.5 to 3.4], p = 0.56, respectively). The association of kidney dysfunction with cognitive impairment could not be investigated due to too few cases with both conditions (n < 5).

Implications of CKD markers for memory and processing speed

Although excess cognitive impairment in CKD is well-known and—given their shared pathophysiology—unsurprising, the potential for CKD to itself be an antecedent of cognitive impairment remains controversial [19]. Our data suggest that CKD manifesting as albuminuria, but not necessarily as kidney dysfunction, predicts at least one domain of cognitive function in memory. However, we hesitate to conclude differential associations between the two CKD markers given that their cognitive test scores did not deviate substantively, and the smaller kidney dysfunction group (n = 39 for eGFR versus n = 142 for albuminuria) obviously afforded less statistical power. Nevertheless, the trend toward a larger effect for albuminuria was suggestive of this marker having stronger predictive capacity for memory performance. Although it is intuitive to attribute this finding to albuminuria having a stronger basis than eGFR in microangiopathy (and therefore representing a better surrogate of the systemic and cerebral microcirculation), this notion clearly requires evidence beyond the current study. Albuminuria is also associated with more aggressive CKD progression than non-albuminuric kidney dysfunction [20]—the latter phenotype being that exhibited by the majority of patients in the low eGFR group (per Fig. 1). Although the pathophysiology of non-albuminuric CKD remains incompletely understood, it has been reported to be more closely associated with aging, hypertension, and vascular disease [21], key covariates with cognitive function that we adjusted for in our models. A lack of independent predictive capacity of eGFR for cognitive function in this context may therefore not necessarily be surprising, even if this notion is contradicted by the REGARDS study (in which eGFR predicted cognitive impairment only in individuals in the lowest category of albumin excretion [6]). In any case, our findings with respect to eGFR cannot be assumed to be generalizable to individuals with albuminuric kidney dysfunction.

An alternative explanation lies with the age difference between the albuminuria and kidney dysfunction groups. Notwithstanding statistical adjustment, we cannot eliminate the prospect that the younger age of the albuminuria group at baseline (mean 52 years versus 68 years for kidney dysfunction) contributed to their larger deficit in memory 12 years later; i.e., since younger subsamples may be less likely to be biased as a result of attrition due to competing risks (e.g., mortality and incident comorbid disease, which obviously affect older individuals to a greater degree). Consistent with this notion, we previously reported in the AusDiab cohort a stronger association between glycemic control and 12-year cognitive function in younger compared with older individuals (i.e.,<60 versus 60 + years) [10].

CKD and cognitive function

The current study aligns with previous work in some, but not all, aspects. Existent prospective studies have produced mixed findings, though the majority have compared cognitive outcomes in groups defined by eGFR exclusively [2, 19]. Studies incorporating markers of kidney damage (i.e., albuminuria or proteinuria) have been more consistent in reporting associations with cognitive function. Indeed, the Rancho Bernardo Study echoed our own data in reporting albuminuria, but not kidney dysfunction, as a predictor of cognitive function measured ∼7 years later (albeit in men only) [22]. Likewise, proteinuria, but not kidney dysfunction, was a borderline significant predictor of incident vascular dementia in the Three-City (3C) Study [23]. Finally, both albuminuria and kidney dysfunction were associated with cognitive impairment at 4 years in REGARDS (though, as mentioned above, the predictive capacity of each marker appeared to depend on the other being normal) [6].

Notwithstanding these comparisons, it is important to appreciate several methodological factors that distinguish our own study. Firstly, as alluded above, our cohort was ∼1–2 decades younger than those of Rancho Bernardo, 3C, and REGARDS, and was followed for a longer period of time (12 years). Indeed, to our knowledge, no longitudinal studies have examined cognitive outcomes past 7 years of follow-up in CKD. The importance of this was apparent from our results; i.e., no impairment of memory or processing speed was evident for individuals who only showed evidence of CKD from the interim 2004/05 assessment (i.e., 7 years exposure). Of course, while this approach has the advantage of characterizing prolonged exposure effects and minimizing the confounding influence of baseline cognitive impairment, it obviously gives rise to the exclusion of more advanced disease (i.e., due to aforementioned competing risks). Indeed, this was apparent from the more favorable clinical profile of individuals with CKD included the final analytic sample compared with the group with CKD at baseline who had to be excluded (e.g., due to death or study withdrawal before follow-up cognitive function testing).

The other major novel aspect of our study was in its outcome measures. Indeed, we are unaware of any longitudinal studies examining associations between albuminuria and the specific cognitive domains of memory and processing speed (though these are obviously components of tests of global cognition; e.g., MMSE). That we identified a deficit in memory, but not processing speed, in individuals with albuminuria underscores the value of assessing multiple domains. However, confirmation that albuminuria has a greater proclivity to affect memory will require further study since, at this stage, it is unclear whether the finding has a genuine mechanistic basis or simply reflects differences between the CVLT and SDMT in sensitivity and reproducibility.

Clinical relevance

Although the memory impairment suggested by lower CVLT scores in individuals with albuminuria represents a small-to-medium effect according to statistical criteria (Cohen’s d), its clinical relevance is less certain. While it could be considered a form of mild cognitive impairment, the deficit observed in this single domain was obviously not sufficient to manifest in cognitive impairment according to the more commonly applied MMSE-based definition (incorporating multiple cognitive domains). Indeed, with <5% of those aged 60 + years showing evidence of cognitive impairment based on the MMSE, ours was clearly a relatively ‘cognitively healthy’ cohort overall (likely reflecting attrition from what was originally a population-based sample). The prospect that this may have biased our results toward the null should certainly be considered (i.e., a stronger effect for CKD may have been detected with a shift in the distribution of outcome measures toward less favorable values).

Even with the relatively small between-group difference, we can at least be confident that there was an effect on memory of albuminuria per se; i.e., only 21% of the association was contributed by the major antecedents/consequences of CKD known to precipitate cognitive decline (i.e., diabetes, hypertension, and cardiovascular disease) and other relevant risk/lifestyle factors. In adjusting for the covariates that we did, we believe our study addresses some of the potential unmeasured confounding for which previous studies on this topic have been criticized [19]. Attesting to the importance of this was the apparently large contribution of covariates to the difference in processing speed between individuals with and without albuminuria (i.e., 43% from model 1 to model 3).

Relevance to global prevention of dementia

Although some cardiometabolic risk factors for dementia have demonstrated a decline in prevalence in recent decades (e.g., smoking and hypertension), others have failed to improve. The trajectory of CKD, though stable on a global level, still shows an escalation in certain regions and population subgroups [24]. Even if prevalence growth can be stalled or reversed, our ever-increasing lifespan will ensure a burgeoning number of cases worldwide for the foreseeable future. In this context, studies such as ours examining long-term cognitive outcomes in individuals with versus without CKD, and the extent to which these reflect shared cardiometabolic risk factors, are vital in establishing the importance of CKD relative to other modifiable determinants of dementia. In this way, future preventive strategies can be better targeted. The current study suggests that CKD may indeed predispose to worse cognition in the longer term. Although further work is needed to translate these findings more broadly, the implication is that CKD has the potential to be a salient contributor to dementia at a global level.

Study limitations

Cognitive tests were not included in the first wave of testing of the AusDiab cohort. This limitation was mitigated to some extent by the relatively long 12-year follow-up period. Definitions of exposures were based on once-only screening tests, whereas CKD diagnosis usually requires persistence of abnormal ACR and/or low eGFR for at least 3 months [25]. Finally, the battery of cognitive function tests was relatively restricted in encompassing the domains of memory and processing speed only (i.e., beyond global and premorbid function indicated by MMSE and STW, respectively). Future studies would be well-served by the inclusion of tests of other domains known to be affected by cardiometabolic disease (particularly executive function).