Role of lean body mass in estimating glomerular filtration rate in Alzheimer disease

Abstract

Background. The association between estimated glomerular filtration rate (eGFR) and progression of Alzheimer disease (AD), as measured by cognitive decline and brain atrophy, has been infrequently studied. Since AD is characterized by sarcopenia and other changes in body composition, which are known to influence GFR, a determination of how lean mass (LM) affects estimation of GFR in AD patients is important.

Methods. Participants were drawn from a prospective longitudinal study of brain ageing and AD in community-dwelling individuals. Control (n = 60) and AD (n = 61) participants were enrolled. Estimated GFR was calculated using the four-variable Modification of Diet in Renal Disease (MDRD), Cockroft–Gault, Macdonald appendicular LM and Taylor LM equations. Association of eGFR with 2-year change in cognitive function and brain volume was assessed.

Results. Individuals with AD demonstrated a paradoxical finding in which lower baseline MDRD eGFR was associated with less cognitive decline (P = 0.04) and brain atrophy (P = 0.02), a phenomenon not observed in non-AD controls. This finding was abolished in the AD patients when either the Macdonald appendicular LM or Taylor LM equations were used. While significant group-by-eGFR interactions were present for cognitive decline (P = 0.006) and brain atrophy (P = 0.001) when the MDRD equation was used, no group-by-eGFR interactions were present when either the Macdonald LM (P = 0.58 and P = 0.10 for cognitive decline and brain atrophy, respectively) or Taylor LM (P = 0.97 and P = 0.55) equations were used.

Conclusions. Accounting for measures of LM in GFR estimation appears to significantly mitigate counterintuitive relationships between measures of AD progression and eGFR as calculated by more traditional measures of renal function. This suggests that consideration of LM in eGFR calculations may be important in patients with sarcopenia, such as the AD population.

Introduction

Recent years have witnessed increased interest in the applicability of measures of estimated glomerular filtration rate (eGFR) to populations with specific diseases or medical conditions, such as cardiovascular disease [1–3] and diabetes [4–7]. The growing appreciation that factors such as metabolic state and body composition influence estimation of renal function [8–15] provides an opportunity to examine how such estimates may be influenced by various disease processes. One such condition of interest is Alzheimer disease (AD). While the clinical hallmark of AD is the progressive decline in cognitive function (most notably memory), accompanied in most cases by corresponding brain atrophy, AD is also characterized by alterations in body composition, including weight loss [16] and sarcopenia [17,18]. Currently, 5.2 million Americans are estimated to have AD; this number is expected to rise dramatically as the population ages [19], making individuals with this disease an increasingly important patient population to study.

Many prior studies have linked eGFR with cognitive function in populations with chronic kidney disease (CKD), demonstrating that lower eGFR is associated with cognitive decline [20–29]. A recent study in individuals without dementia found that eGFR was associated with decline in cognitive performance over a mean of 3.4 years [29]. The study sample consisted of substantial numbers of patients with major comorbidities (e.g. vascular disease, congestive heart failure, etc.) that, despite adjustment for in statistical modeling, could still confound the relationship between renal and cognitive functions. Unfortunately, there is a paucity of data examining the relationship between eGFR and dementia progression specifically in the AD population. While one recent study in AD patients found a modest association between MDRD eGFR and dementia, the analysis was cross-sectional in nature only [30].

We therefore assembled a cohort of mildly demented AD patients who were nearly free of major comorbidities, along with control individuals without dementia, and followed them prospectively for 2 years to determine how various baseline eGFR measures are associated with cognitive decline and brain atrophy. Because there is increasing evidence that the structural and functional brain changes in AD are accompanied by systemic physiologic processes including loss of bone density[31] and lean mass (LM), we were particularly interested in studying how consideration of LM in eGFR calculations might affect this relationship. Accordingly, we assessed eGFR with the Modification of Diet in Renal Disease (MDRD) equation [32], the Cockroft–Gault equation [33] (which actually measures creatinine clearance, but which is still widely used to assess renal function) and two equations which utilize measures of LM, the Macdonald [11] and Taylor [9] equations. The Macdonald equation was specifically developed to utilize appendicular LM as measured by dual-energy X-ray absorptiometry (DXA), while the Taylor equation uses DXA-derived whole-body LM, making them appealing equations to study in the context of disease processes characterized by sarcopenia. As we did not have isotopically determined GFR values, our goal was not to validate which equation performs most accurately in AD patients. Rather, we sought to examine specifically how consideration of different measures of eGFR, some accounting for LM, might affect bedside estimates of GFR compared to more traditional equations, and provide future hypotheses for testing.

Materials and methods

Study design and sample recruitment

The data were obtained from a longitudinal study of brain ageing and AD (the ‘Brain Aging Project’) conducted in community-dwelling adults between 2004 and 2009. The primary aim of the 2-year observational study was to assess the relationship of a variety of lifestyle-related factors with AD progression, as measured by change in cognitive function and by brain atrophy. Lifestyle-related factors included cardiorespiratory fitness, physical activity, body composition [e.g. fat mass, lean mass and bone mineral content by DXA] and metabolic measures (e.g. inflammatory status, insulin sensitivity).

Participants in the Brain Aging Project were ≥ 60 years old and were either non-demented [Clinical Dementia Rating [34] (CDR) 0 (n = 60)] or diagnosed with early-stage AD [CDR 0.5, very mild dementia (n = 51) or CDR 1, mild dementia (n = 10)] as detailed below. The non-demented participants were recruited by self-referral in response to media coverage and through word of mouth, while the AD participants were recruited largely by media appeals and through a referral-based memory clinic. Because the Brain Aging Project was focused on assessing predictors of disease progression specifically in individuals with the earliest stages of AD (where therapeutic strategies are likely to have the most impact), and because individuals with more severe disease would have a much-reduced likelihood of full participation in the 2-year study, individuals with moderate (CDR 2) and severe (CDR 3) dementia were excluded.

Other study exclusions included neurologic disease other than AD, diabetes mellitus (DM), recent (< 2 years) history of ischemic heart disease, clinically significant depressive symptoms, use of antipsychotic and investigational medications, and significant sensory impairment or systemic illness that could impair completion of the study. Presence of a medical condition was determined by self-report and through medication reconciliation by the study nurse (i.e. by identification of all current prescriptions and their indications). All demented participants were accompanied by a study partner who was knowledgeable about the participant’s medical history, daily activities and medication use. Only participants who completed MRI, cognitive testing, DXA, and demographic and laboratory assessments were included in the study.

All participants provided informed consent. The study was approved by the institutional review board at the University of Kansas Medical Center and was undertaken in accordance with the principles of the Declarations of Helsinki.

Diagnostic testing

Clinical assessment of dementia: A nurse clinician collected demographic information, an educational and medical history, and current medication list from the participant and a study partner. A neurologist performed a standard physical and neurological examination. All participants were then evaluated using a semi-structured interview of the participant and a study partner to determine the presence or absence of dementia, and its severity if present, using the CDR [34]. Diagnostic criteria for AD require the gradual onset and progression of impairment in memory and in at least one other cognitive and functional domain [35]. These diagnostic methods have an accuracy for AD of 93% [36] and sensitively detect the earliest stages of AD [37] by focusing on intra-individual change rather than comparison with group norms. A global CDR score is derived from individual ratings in each domain with CDR 0 indicating no dementia, CDR 0.5 very mild dementia and CDR 1 mild dementia; participants with moderate (CDR 2) or severe (CDR 3) dementia were not enrolled.

Body composition determination: DXA (Lunar Corp., Madison, WI, USA) was used to determine total body measures of lean mass and fat mass. DXA testing was standardized by having all patients fast, void and dress in a hospital gown. Percent body fat represents the percent of total body mass (determined by DXA) composed of fat (i.e. total fat mass × 100%/total body mass). LM was measured exclusive of fat and bone mineral content, and appendicular LM consisted of a summation of the lean mass of the upper and lower extremities.

Neuropsychological assessment: A trained psychometrician administered a standardized psychometric battery to all participants as previously described [17,38]. The battery included the standard measures of verbal and working memory, and executive function including set maintenance and switching and visuospatial cognition. Cognitive performance scores were converted to Z-scores based on the mean and standard deviation of a larger cohort of non-demented subjects. The mean of each participant’s Z-scores served as an index of global cognitive performance. The difference in Z-scores over 2 years represents the primary cognitive variable of global cognitive decline (where negative numbers represent decline). A standard mini-mental status exam (MMSE) was also performed for descriptive purposes.

Neuroimaging: Brain volume is well measured by magnetic resonance imaging (MRI), and brain atrophy is increasingly used as a ‘biomarker’ of AD progression [39,40]. Structural MRI was obtained for all participants using a Siemens 3.0-T Allegra MRI scanner (Siemens Medical Solutions, Erlangen, Germany). High-resolution T1-weighted anatomic images were acquired to provide detailed gross anatomy with high grey–white matter contrast (MP-RAGE; 1 × 1 × 1 mm³ voxels; TR = 2500 ms, TE = 4.38 ms, TI = 1100 ms, FOV 256 × 256 mm² with 18% oversample, flip angle = 8°). Baseline data analysis was performed using the VBM5 toolbox (http://dbm.neuro.uni-jena.de), an extension of the SPM5 algorithms (Wellcome Department of Cognitive Neurology, London, UK) running under MATLAB 7.1 (The MathWorks, Natick, MA, USA) on Linux. Processing for VBM has been detailed elsewhere [41]. Total grey matter (GM), white matter (WM) and cerebrospinal fluid volumes from the VBM5 segmentation were added to compute total intracranial volume (TICV) using tissue maps of each study participant. GM and WM volumes were added and then divided by TICV to compute normalized whole brain volume (% TICV). We then assessed 2-year rates of global brain atrophy using the Structural Image Evaluation using Normalization of Atrophy method [42]. This registration-based method uses images from two time points to assess brain volume changes by estimating the local shifts in brain edges across the entire brain and then converting the edge displacement into a global estimate of percentage brain volume change between the two time points, where negative numbers represent atrophy.

Assessment of kidney function

Four different parent equations to calculate eGFR were used. The four-variable Modification of Diet in Renal Disease (MDRD) equation [32] was employed, as it is endorsed by the National Kidney Foundation KDOQI clinical practice guidelines to estimate GFR. In a sensitivity analysis, the MDRD equation adapted to isotope dilution mass spectrometry (IDMS)-standardized creatinine was used after converting the Jaffe (alkaline picrate) method creatinine to IDMS creatinine using the manufacturer-derived conversion equation. The Cockroft–Gault (CG) equation [33], which measures creatinine clearance and not eGFR, was also employed, as it is still widely used, particularly in the pharmacy literature. The CG equation used absolute mass and was corrected for body surface area (BSA) with the Du Bois method [43]. A correction factor of 0.8 was used to correct for bias in the original MDRD study sample [44]. As with the MDRD equation, IDMS-standardized creatinine was used in a sensitivity analysis. The Macdonald and Taylor eGFR equations, derived specifically from DXA-derived LM values (appendicular [11] and whole body [9], respectively) were also employed using both traditional creatinine and, in a sensitivity analysis, IDMS-standardized creatinine. The two LM equations do not require external standardization for BSA.

The equations used were:

1. Four-variable MDRD equation [32]:

$$\text{eGFR}=186.3\times {(\text{ SCr})}^{-1.154}{(\times \text{ age})}^{-0.203}(\times 0.742\text{if female})(\times 1.210\text{if black})$$

For the version using the IDMS-standardized creatinine, a factor of 175.0 replaced 186.3 [44].

1. Cockroft–Gault equation [33]:

$$\text{eGFR}=0.8\times (1.73/\text{BSA})\times [(140-\text{ age})\times \text{ weight}(\times 0.85\text{ if female})]/(\text{SCr}\times 72)$$

1. Macdonald appendicular lean mass (ALM) GFR equation [11]:

$$\text{eGFR}={\text{antilog}}_{10}[1.008+(-1.014\times {\text{ log}}_{10}\text{SCr})+(0.01644\times \text{ ALM})+(0.07108\times \surd \text{weight})$$

1. Taylor LM GFR equation [9]:

$$\text{eGFR}=(2.4\times \text{ LM})-(0.75\times \text{ LM }\times \text{ Scr})$$

For the sensitivity analysis, we converted Jaffe creatinine values to IDMS values using the following manufacturer-supplied equation:

$$\text{IDMS SCr = {[(Jaffe Scr }\times \text{ 88.4)}-11.8340\text{]/1.0355/ 88.4}}$$

Body surface area (BSA) for the CG equation was computed as follows:

BSA = (W^0.425 × H^0.725) × 0.007184, where the weight is in kilograms and the height is in centimetres [43].

Statistical analysis

Group differences in descriptive data and measures of interest (brain atrophy and cognitive decline) were examined using t-tests for continuous variables and chi-square tests for categorical variables. Given known gender effects on lean mass, ANOVA was used to compare lean mass across groups treating gender as a covariate (or ‘nuisance variable’).

Linear regression was used to model the independent association between eGFR, as calculated by the four equations, with 2-year rates of global cognitive decline and brain atrophy. Age, sex and history of hypertension were included as covariates in all models since each of these factors are known to be associated with changes in cognitive function and brain atrophy. As only three individuals with AD were non-white (two African-American and one Native American), and because race is not associated with either brain volume or cognition, this term was not specifically included in modeling. As no individuals had diabetes mellitus, significant coronary artery disease, stroke or congestive heart failure, there was no need to control for these potential covariates. Standardized β-values and change in the coefficient of determination (R²Δ, a measure of change in effect size expressed as a decimal) were assessed to define the relationships between the different estimates eGFR (as yielded by the various equations) and measures of cognitive decline and brain atrophy, after adjusting for age, sex and hypertension.

Additionally, we assessed whether group (AD vs non-AD) by eGFR interactions were present in predicting cognitive decline and brain atrophy in order to assess whether eGFR was similarly predictive of decline in AD vs non-demented groups. Univariate general linear model was used to examine for the presence of group (AD vs non-demented) by eGFR values in predicting cognitive decline and brain atrophy, controlling for age, sex and hypertension. The presence of significant interactions would be suggestive of differences in the applicability of the eGFR measures in AD vs control groups. For all analyses, SPSS version 17 was used. Statistical significance was inferred for a P-value ≤ 0.05.

Results

Participant characteristics

Participant characteristics are shown in Table 1. Neither age nor sex was significantly different between individuals with and without AD. As expected, females were slightly overrepresented. Fully 95.1% of the AD patients, and all of the non-demented controls, were Caucasians. There was higher appendicular and whole-body LM (controlling for sex) in the non-demented individuals. Overall, rates of major comorbidities were low; although nearly half of the individuals with AD had hypertension at baseline, none had diabetes, a history of stroke, congestive heart failure or significant history of recent (< 2 years) coronary artery disease. As expected, the participants with AD had lower baseline global cognitive function and a greater degree of cognitive decline at 2 years than individuals without dementia. Additionally, normalized whole brain volume was lower at baseline and demonstrated increased atrophy at 2 years in the individuals with AD. Individuals with AD had a trend towards a lower eGFR, as calculated by the MDRD equation (P = 0.10), and a significantly lower eGFR when the CG (P = 0.01) and, especially, the two LM-based equations were used (P = 0.002 for Macdonald and P = 0.007 for Taylor).

Table 1.

Characteristics of the participants, by disease status

Characteristic	AD (n = 61)	Non-AD (n = 60)	P-value
Age, years	74.4 ± 6.5	73.0 ± 7.2	0.29
Female sex, n (%)	38 (62.3)	34 (56.6)	0.59
Caucasian race, n (%)	58 (95.1)	60 (100.0)	0.25
Appendicular lean mass, kga	17.5 ± 4.2	19.0 ± 5.3	0.03
Lean mass, kga	41.6 ± 8.6	44.6 ± 10.6	0.03
Hypertension, n (%)	28 (45.9)	16 (26.6)	0.04
Follow-up duration, years	2.0 (0.25)	2.1 (0.28)	0.26
MMSE score, points	26.1 ± 3.1	29.4 ± 0.8	< 0.001
Change in MMSE score, points	− 4.1 (6.0)	− 0.3 (1.0)	< 0.001
Cognitive function, Z-score	− 1.62 ± 1.05	0.13 ± 0.50	< 0.001
Change in cognitive function, Z-score	− 0.67 ± 0.68	− 0.04 ± 0.36	< 0.001
Whole brain volume, %ICV	67.0 ± 3.3	70.4 ± 3.2	< 0.001
Change in brain volume, %	− 2.92 ± 2.53	− 0.79 ± 1.28	< 0.001
Serum creatinine, mg/dL	0.87 ± 0.21	0.82 ± 0.20	0.13
eGFR, mL/min/1.73 m2
MDRD	81.3 ± 20.0	87.1 ± 17.7	0.10
CG	57.3 ± 14.4	65.0 ± 17.6	0.01
Macdonald	90.6 ± 26.2	108.1 ± 32.3	0.002
Taylor	69.8 ± 19.5	78.9 ± 16.5	0.007

^aANOVA controlling for sex. Other tests are t-tests, showing mean ± 1 standard deviation.

Maximum score on the MMSE is 30 points.

AD, Alzheimer dementia; MMSE, Mini-Mental Status Exam; ICV, intracranial brain volume; eGFR, estimated glomerular filtration rate; MDRD, Modification of Diet in Renal Disease; CG, Cockroft–Gault.

eGFR and 2-year rates of cognitive decline

Associations between baseline eGFR, as calculated with the four different equations, and cognitive decline are shown in Table 2; 44 AD and 56 non-AD individuals had full longitudinal data for cognitive function testing. As stated, models controlled for age and sex but not race, which had low levels of variability. The only comorbidity of substantial frequency was hypertension, which was also included in the modeling.

Table 2.

Association of estimated glomerular filtration rate with cognitive decline and brain atrophy in individuals with and without Alzheimer disease

	Cognitive decline							Brain atrophy
	AD n = 44			Non-AD n = 56			eGFR × group interactiona	AD n = 34			Non-AD n = 47			eGFR × group interactiona
Equation	β	R2Δ	P-value	β	R2Δ	P-value	P-value	β	R2Δ	P-value	β	R2Δ	P-value	P-value
MDRD	− 0.31	0.081	0.04	0.18	0.028	0.22	0.006	− 0.44	0.155	0.02	0.27	0.067	0.08	0.001
CG	− 0.21	0.020	0.31	− 0.06	0.020	0.77	0.02	− 0.40	0.067	0.15	0.35	0.072	0.07	0.005
Macdonald	− 0.16	0.020	0.31	− 0.01	0.000	0.96	0.58	− 0.23	0.037	0.28	0.18	0.025	0.28	0.10
Taylor	− 0.22	0.026	0.25	0.18	0.014	0.39	0.97	− 0.27	0.036	0.29	− 0.01	0.000	0.97	0.55

^aInteraction term represents the group (AD or non-AD) by eGFR interaction for the respective equations in predicting the cognitive decline and brain atrophy.

β-values represent the direction of correlation between eGFR with cognitive decline and brain atrophy.

eGFR, estimated glomerular filtration rate; AD, Alzheimer dementia; Δ, change; MDRD, Modification of Diet in Renal Disease; CG, Cockroft–Gault.

For the MDRD equation, there was no association between eGFR and cognitive decline (P = 0.22) in the non-demented group. However, as shown in Table 2, individuals with AD demonstrated a ‘paradoxical’ inverse association in which lower baseline eGFR was associated with less cognitive decline over 2 years (β = − 0.31, P = 0.04), i.e. lower eGFR was associated with preservation of cognitive function. A group (AD vs non-AD) × eGFR interaction was present (P = 0.006), suggesting that the relationship between eGFR and cognitive decline was significantly different in AD individuals compared to non-demented controls. The results for the CG equation for the non-AD individuals were similar to those of the MDRD equation in that there was no association between baseline eGFR and cognitive decline in the non-AD controls (P = 0.77). While there was not a significant association between eGFR and cognitive decline (β = − 0.21, P = 0.31) when using CG in the AD patients, there again was significant group × eGFR interaction (P = 0.02). When using the Macdonald equation, however, a different pattern emerged. In the AD patients, the relationship between eGFR and cognitive decline observed when using the MDRD equation was abolished (β = − 0.16, P = 0.31) in AD patients and remained unchanged in the non-demented controls (β = − 0.01, P = 0.96). The group × eGFR interaction term was now non-significant (P = 0.58), suggesting no difference in the relationship of Macdonald eGFR and cognitive decline between the AD and non-AD groups. The results using the Taylor equation were similar to those of the Macdonald equation in that the group × eGFR interaction term remained non-significant (P = 0.97).

Overall, incorporation of body composition into GFR estimates (i.e. total mass, appendicular LM or whole-body LM) resulted in a decrease in the R²Δ (expressed as a decimal, representing the change in the effect size of the model after accounting for the covariates of age, sex and hypertension). For example, in the individuals with AD, R²Δ was relatively high for the MDRD equation in predicting cognitive decline (0.081), but much lower for the other three equations (0.020–0.026), demonstrating a general diminution in the counterintuitive relationship between higher eGFR and greater cognitive decline seen with the MDRD equation, suggesting that incorporation of body composition into the estimates of GFR attenuated the paradoxical observation in which lower MDRD eGFR was associated with less dementia progression.

Figure 1 demonstrates the associations between eGFR and cognitive decline for the MDRD (A), Macdonald (B) and Taylor (C) equations. While there was a significant divergence in the cognitive decline–MDRD eGFR correlations between the AD and non-AD individuals (represented by the interaction P-value of 0.006), there was no such divergence between AD and non-AD individuals when the Macdonald and Taylor equations were used.

Fig. 1.

Association between 2-year cognitive decline, as measured by the Z-score, and estimated glomerular filtration rate, as calculated by the MDRD (A), Macdonald appendicular lean mass (B) and Taylor lean mass (C) eGFR equations. When the MDRD equation is used, the association of cognitive decline with eGFR is significantly different between the demented and non-demented individuals (P-value for the interaction = 0.006). However, when the Macdonald and Taylor equations are used, there is no such interaction (P = 0.58 and 0.97, respectively).

In a sensitivity analysis, we used IDMS-standardized creatinine values; in no instance did use of this measure change the results (data not shown).

eGFR and 2-year rates of brain atrophy

In a similar fashion, associations between baseline eGFR and brain atrophy, after accounting for age, sex and hypertension history, are shown in Table 2; 34 AD and 47 non-AD individuals had full longitudinal MRI data to assess brain atrophy. For the MDRD equation, non-demented individuals demonstrated a trend relating lower eGFR with greater cognitive decline (β = 0.27, P = 0.08), as might be expected. However, individuals with AD demonstrated a paradoxical inverse association in which lower baseline eGFR was associated with significantly less brain atrophy (β = − 0.44, P = 0.02). A group × eGFR interaction was present (P = 0.001), suggesting that the relationship between eGFR and brain atrophy was significantly different in AD individuals compared to non-demented controls. The results for the CG equation were broadly similar; again, the group × eGFR interaction was highly significant (P = 0.005). When using the Macdonald equation, the relationship between eGFR and brain atrophy was non-significant in AD patients (β = − 0.23, P = 0.28) and remained unchanged in the non-demented controls (β = 0.18, P = 0.28). The group × eGFR interaction was non-significant (P = 0.10), suggesting no difference in the relationship between eGFR and cognitive decline across groups when eGFR was calculated using the Macdonald equation. The results using the Taylor equation were similar to those of the Macdonald equation, again demonstrating non-significant interaction between groups (P = 0.55). Examination of R²Δ in the AD patients showed a sizable difference between the value for the MDRD equation (0.155) and the Macdonald (0.037) and Taylor (0.036) equations, again suggesting a diminution in the counterintuitive relationship between higher MDRD eGFR and brain atrophy when appendicular and whole-body LM were considered.

Figure 2 demonstrates the associations between eGFR and cognitive decline for the MDRD (A), Macdonald (B) and Taylor (C) equations. While there was a significant divergence in the cognitive decline–MDRD eGFR correlations between the AD and non-AD individuals (represented by the interaction P-value of 0.001), there was no such significant divergence between AD and non-AD individuals when the Macdonald and, especially, the Taylor equations were used.

Fig. 2.

Association between 2-year brain atrophy, measured as percent change in brain volume, and estimated glomerular filtration rate, as calculated by the MDRD (A), Macdonald appendicular lean mass (B) and Taylor lean mass (C) eGFR equations. When the MDRD equation is used, the association of brain atrophy with eGFR is significantly different between the demented and non-demented individuals (P-value for the interaction = 0.001). However, when the Macdonald and Taylor equations are used, there is no such interaction (P = 0.10 and 0.55, respectively).

At no point did use of IDMS-standardized creatinine values demonstrate different patterns.

Discussion

In this study, we examined the relationship of eGFR with functional and structural changes over time in patients with and without AD. To our knowledge, this is the first study to examine measures of renal function and longitudinal dementia progression, using both cognitive testing and brain imaging, in the AD population specifically. Our results suggest that the relationships of eGFR with cognitive decline and with brain atrophy are dependent upon consideration of lean mass in the estimation of renal function.

When the MDRD equation was used, we found distinctly counterintuitive associations between eGFR and cognitive decline and brain atrophy in AD patients such that individuals with higher eGFR had greater declines in the outcome measures. This association was strongest in the AD patients with the MDRD equation, which does not consider mass, and was attenuated when the Macdonald and Taylor equations were used, which were specifically designed to incorporate different measures of LM (as measured by DXA) in the calculation of eGFR. Significant interactions were observed between groups (i.e. when comparing the observed relationships in AD individuals to those of non-AD individuals), which demonstrates that accounting for LM by either equation attenuated the apparent association of higher MDRD eGFR with greater cognitive decline and brain atrophy in AD. This suggests that LM may account for the ‘paradoxical’ relationship between MDRD eGFR and markers of disease progression in AD. These observations, along with others suggesting that dementia and AD are associated with sarcopenia [17,18], highlight the potential importance of accounting for LM in AD. Not doing so, such as when using the MDRD equation, would lead to the paradoxical conclusion that higher eGFR in individuals with AD is associated with greater cognitive decline and brain atrophy. This implausible conclusion, however, would not be made when accounting for LM to estimate GFR.

Several investigators have demonstrated that cognitive decline is associated with decreased eGFR at baseline in non-demented individuals [20–29]. For example, Khatri et al. [28] recently showed that CKD was associated with cognitive decline in a large population-based survey. Buchman et al. [29] reported similar findings in a prospective cohort of individuals without AD but with decreased eGFR (mean eGFR 59.0 mL/min/1.73 m² by MDRD) and substantial prevalences of major comorbidities (e.g. 11.4–13.5% prevalence for each of stroke, myocardial infarction and diabetes), for which control using statistical modeling was attempted by the investigators. Large population-based studies such as the REGARDS (23 405 persons studied) [23] and NHANES III (4849 persons) [22] have reported similar findings in a cross-sectional fashion, and in the longitudinal Health ABC study, baseline eGFR was associated with cognitive decline in 3075 individuals at follow-up several years later [20]. Thus, the association between decreased kidney function and cognitive impairment and decline is well established in cohorts of non-demented individuals with varying frequencies of major public-health comorbidities, even after adjustment for these factors.

Few investigators have studied the relationship between eGFR and cognitive function in AD patients. A recent analysis did demonstrate a modestly significant adjusted association between eGFR, calculated using the MDRD equation, and cognitive function in a cross-sectional fashion [30]. To our knowledge, the present investigation is fundamentally different from all previous studies. We specifically sought to investigate longitudinal changes in both functional (cognitive) and structural (brain volume) markers of dementia utilizing a relatively ‘pure’ cohort of individuals with AD in which the frequency of major comorbidities was very low. Further, our cohort had predominantly preserved eGFR (that is, mean eGFR > 60 mL/min/1.73 m²), at least as measured by raw serum creatinines and by the MDRD and LM-based equations. Utilizing this cohort permitted us to decrease the possibility that residual confounding (such as the presence of clinically significant vascular disease) was present between diseases that affect both renal and cognitive functions. These results indicate that consideration of LM may be a key factor in accurately estimating renal function in AD. Whether eGFR is independently associated with morbidity and mortality in AD should be examined in future studies, but care should be taken in applying an appropriate measure of GFR. Whether the Macdonald or Taylor equation in particular, or indeed any other equation, is associated with these clinical outcomes requires further work.

There are several important limitations of our study. First, the number of participants was relatively small, reducing our power to detect associations, if they existed. We were, however, able to detect the strikingly counterintuitive relationship between MDRD eGFR and both brain volume and cognitive function. While the degree of statistical significance of our findings was, for the most part, modest, the consistent patterns we observed when LM was invoked (via the Macdonald and Taylor equations) for examining the association of eGFR with both cognitive decline and brain atrophy lend strength to our conclusions. Second, we relied on medical histories provided by a caregiver (typically a spouse or adult child). Although it is possible that all medical comorbidities may not have been recalled by the study partner, the participants underwent extensive evaluation involving a complex history taken in the presence of a trained research coordinator upon enrolled, as well as medication reconciliation (a process by which indications for medications are specifically investigated); it seems unlikely that major medical conditions went undetected. We also had only a single measurement of creatinine on which to base estimates of GFR. While serial values would have been preferable, significant decline in renal function over 2 years in individuals with relatively intact GFR (in the absence of poorly treated comorbidities) seems unlikely to have occurred widely, and our study was mainly designed to elucidate broad trends that may be suitable for future investigation. Another potential limitation is that nearly all of the participants were Caucasian, reducing our ability to generalize our findings to other racial groups, although there is no reason to posit that this phenomenon would not be present in individuals of other races.

Additionally, we did not have isotopically determined GFR values. However, our goal was not to validate which equation performs most accurately in AD patients, but rather to examine how consideration of LM might affect bedside estimates of GFR and, more profoundly, how LM may be intimately related to the AD disease process of itself. While recognizing the impracticality of routine assessments of LM to assist in GFR estimation, the present efforts were undertaken to generate hypotheses about the role that LM assessment in some form might play in assessing renal function, as well as to provide preliminary evidence that more traditional or widely used measures of eGFR (like the MDRD equation) might be inadequate in diseases characterized by sarcopenia. Overall, these limitations are probably counterbalanced by the relatively healthy and homogenous nature of the participants with relatively intact levels of kidney function (a design feature of the study, which was intended to allow for examination of the relationship of eGFR with outcomes of interest unconfounded by other comorbidities), as well as the fact that LM was assessed by DXA (rather than with anthropometric equations) and that we used two eGFR equations derived specifically using DXA-derived LM values.

In summary, consideration of LM alters the relationship between eGFR and changes in brain volume and cognitive function in patients with AD. We cannot at this time determine whether appendicular or whole-body LM is the better measure for estimation of GFR, but the broad pattern of concordant findings between the Macdonald and Taylor equations suggest that consideration of LM is important in this population. Our findings, while preliminary, add to the growing body of evidence that AD involves a systemic process characterized by alterations in body composition, particularly in LM, and that use of the MDRD equation for estimating GFR may be suboptimal. Much further work investigating both the most appropriate method of GFR estimation in individuals with reduced LM as well as the potential role of loss of LM in AD must be undertaken.