Abstract
Background
Estimated glomerular filtration (eGFR) results based on serum creatinine are frequently inaccurate with differences against measured GFR (mGFR) often attributed to unmeasured non-functional factors, such as muscle mass.
Methods
The influence of muscle mass (measured by dual-energy x-ray absorptiometry, DEXA) on eGFR error (eGFR-mGFR) was evaluated using isotopic mGFR (Tc99m DTPA plasma clearance) in 137 kidney transplant recipients. Serum creatinine was measured by isotopic-calibrated enzymatic analysis, converted to eGFR using Chronic Kidney Disease EPIdemiology (CKD-EPI) formula, then unindexed from body surface area.
Findings
Unindexed CKD-EPI eGFR error displayed absent fixed bias but modest proportional bias against reference mGFR. eGFR error correlated with total lean mass by DEXA (r=-0·350, P<0·001) and appendicular skeletal muscle index (ASMI), a proxy for muscularity (r=-0·420, P<0·001). eGFR was falsely reduced by -5·9 ± 1·4 mls/min per 10 kg lean mass. Adipose mass and percentage fat had no effect on error. Muscle-associated error varied with each eGFR formula and influenced all CKD stages. Systemic eGFR error was predicted by ASMI, mGFR, recipient age, and trimethoprim use using multivariable regression. Residual plots demonstrated heteroscedasticity and greater imprecision at higher mGFR levels (P<0·001), from increased variance corresponding to higher absolute values and unreliable prediction by serum creatinine of high mGFR. Serum creatinine correlated with ASMI independent of mGFR level (r = 0·416, P<0·001). The diagnostic test performance of CKD-EPI eGFR to predict CKD stage 3 (by mGFR) was weakest in cachexia (sensitivity 68·4%) and muscularity (specificity 47·4%, positive predictive value 54·5% for the highest ASMI quartile).
Interpretation
Serum creatinine and eGFR are imperfect estimates of true renal function, with systemic errors from muscle mass, tubular secretion, and intrinsic proportional bias; and additional inaccuracy at the extremes of renal function and patient muscularity. Cautious interpretation of eGFR results in the context of body habitus and clinical condition is recommended.
1. Introduction
]. Creatinine is convenient, inexpensive to measure, and widely available. Technical improvements including mitigation of non-creatinine chromogens from the Jaffe reaction, introduction of enzymatic methods and isotopic standardization of creatinine have increased accuracy and measurement precision. Because creatinine is small (113 Da), water-soluble, non-protein bound, and freely filtered across the glomerulus without significant tubular reabsorption, it is the principal endogenous indicator of glomerular filtration rate (GFR) used in clinical practice.
], and scores of eGFR formulae have now been published [
,
,
,
]. The Chronic Kidney Disease EPIdemiology collaboration (CKD-EPI) formula [
] has supplanted the older Modification of Diet in Renal Disease (MDRD) formula [
]. Automated eGFR reports now accompany serum creatinine results at point-of-care and are widely used for detection of chronic kidney disease, patient management, and research. However, eGFR values derived from serum creatinine are frequently inaccurate when compared against measured GFR (mGFR) reference methods. Only 24–38% of eGFR results typically fall within the clinically relevant P10 accuracy standard (proportion absolute percentage error
]. Disease misclassification by eGFR are well-known pitfalls to experienced nephrologists. Examples include false underestimation of GFR in young muscular men (incorrectly labelled as renal failure), or overestimation in frail, sarcopenic women (underestimating CKD with near normal creatinine).
]. Practical difficulties linking accurate muscle mass measurements with renal functional reference methods have hindered research. The introduction of Dual Energy X-Ray Absorptiometry (DEXA) has overcome that obstacle. Originally developed for bone mineral content (BMC) measurement and osteoporosis diagnosis, DEXA also produces reliable and inexpensive measurements of muscle mass with miniscule radiation exposure (about 0·1μGy). The relative attenuation characteristics of two X-ray energy peaks determine the elemental contents of tissues. Mathematical algorithms then assign pixels based on BMC, fat, or lean muscle which are overlayed onto body regions (Fig. 1). Lean muscle mass by DEXA strongly correlates with CT and MRI reference standards (R2 0·86 and 0·96) [
,
], but is much less expensive and time-consuming to perform. DEXA produces total, truncal, appendicular muscle mass measurements, and appendicular skeletal muscle index (ASMI) a relative index of muscularity normalized to height [
] (comparable to BMI scaling for obesity) [
]. ASMI accurately diagnoses sarcopenia and cachexia. In patients with renal failure, lean mass may contribute to bias for MDRD eGFR, however results are conflicted and accurate data sparse [
,
].
Fig. 1Workflow of DEXA whole body composition analysis. Patients are scanned with a constant potential X-ray generator that produced a beam separated into high and low energy regions (A, image compliments of GE Healthcare). The energy discriminating detector uses the differential attenuation characteristics of these beams to determine the elemental content of tissues using mathematical algorithms (B), separating them into bone (black), fat (red), and lean muscular tissue (green) (C, D, E). Illustrated study patients are small with minimal muscular mass (C, BMI 17·3, ASMI 4·7 kg/m2), normal (D, BMI 25·1, ASMI 7·7 kg/m2), and large and muscular (E, BMI 11.4, ASMI 38·6 kg/m2). DEXA images are scaled to patient height.(For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
], and the absolute level of mGFR. Muscle mass reduced the performance of CKD-EPI eGFR to predict CKD stage 3 at the extremes of body habitus.
3. Results
3.1 Clinical body composition and eGFR results
From 164 patients screened, 37 were excluded (no DEXA for geographical or social reasons, n = 17 or non-contemporaneous, n = 15, physical constraints, n = 5 [large size 2, amputations 2, gamma nail 1]). Included recipients (n = 137) were 49·2 ± 14·1 years old, 61·3% male, and 78·8% received a deceased donor transplant (detailed Table S2). Immunosuppression comprised: tacrolimus (92·7%) or cyclosporine (6·6%); mycophenolate (81·8%), azathioprine (8·8%), sirolimus (2·2%), or leflunomide (5·1%); and prednisolone. Recipient’s ethnicity was Caucasian (n = 94), East Asian (n = 23), Indian/South Asian (n = 15), Pacific Islander (n = 4), and Australian Aboriginal (n = 1).
]. A patient’s eGFR was affected by creatinine generation markers (ASMI, body weight, male recipient) and functional clearance indicators (mGFR, optimal kidney-pancreas donor kidney, and lack of chronic tubular damage) using multivariable regression analysis (R2 0·731, Tables 2, S3-S4).
Table 1Key demographic of study population renal functional and body composition parameters by DEXA at 12-months (detailed Table S1). Key: ASMI, appendicular skeletal muscle index; DEXA, dual-energy x-ray absorptiometry; eGFR is estimated GFR.
Table 2Predictors of transplant eGFR. The unindexed eGFR value was affected by both creatinine generation markers (ASMI, body weight, male recipient) versus renal functional clearance indicators (mGFR, optimal SPK kidney, and lack of chronic tubular damage). Multivariable predictors of eGFR using regression analysis, and included isotopic mGFR and ASMI (R2 0·731, df 126, constant 36·18). Key: ASMI, appendicular skeletal muscle index; DEXA, dual-energy x-ray absorptiometry; SPK, simultaneous pancreas kidney transplant (optimal donated kidneys).
3.2 Performance of eGFR against reference mGFR
Fig. 2Inter-relationships between GFR estimations and measurements. Panel A illustrates the non-linear relationship between serum creatinine and unindexed measured GFR. Unmodified eGFR correlated with corrected mGFR (B, Pearson correlation). Difference plots of (indexed) eGFR error against averaged corrected eGFR and mGFR (mls/min/1·73 m2, Bland-Altman method, Panel C) demonstrate absent fixed and proportional biases. The residual plot against mGFR showed heteroscedasticity and wider variances at higher mGFR levels (Panel D). The absolute value of unindexed eGFR error (eGFR–mGFR, mls/min) increased with higher isotopic mGFR (black columns), however conversion to absolute percentage eGFR error ([eGFR–mGFR]/mGFRx100%), abrogated the influence of renal function (gray columns, Panel E, * P<0·05, *** P<0·001 vs <40 mls/min). Corrected GFR error (eGFR-mGFR difference) compared against corrected mGFR (alone) excluded fixed bias, however detected proportional bias with eGFR understimated with greater renal function levels (Panel F).
Table 3Performance characteristics of unindexed eGFR against reference test mGFR (both in mls/min). Fixed bias is the mean (±SD, the precision) difference between eGFR and mGFR versus zero. Proportional bias is the regression (beta coefficient±SE) slope by mGFR level. Accuracy denotes percentage eGFR results within ±10% and ±30% of mGFR reference (P10% and P30%, respectively). Key: CG, Cockcroft-Gault; CKD EPI, Chronic Kidney Disease EPIdemiology; and MDRD, Modification of Diet in Renal Disease (eGFR formulae).
3.3 eGFR error correlated with muscle mass
Fig. 3eGFR error was affected by muscular mass. Muscle mass measurements inversely correlated with unindexed eGFR error (eGFR-mGFR) including total lean mass (Panel A) and appendicular skeletal mass index (ASMI, Panel B), resulting in overestimation in cachexic patients and underestimation in muscular recipients. Adipose mass had no effect on eGFR error (Panels C). Key: Pearson correlation coefficients and 95% prediction bands are presented.
Table 4Multivariable predictors of unindexed GFR error. Predictors of eGFR error (eGFR-mGFR, mls/min) by linear regression included muscularity and trimethoprim use, along with the level of renal function. Isotopic mGFR (mls/min) was included as a covariate to adjust for heteroscedasticity at higher GFR levels in (unweighted) model 1 (R2 0·279, constant 42·855). Predictors of GFR error using weighted linear regression analysis (model 2, against reciprocal mGFR as the weighting term, R2 0·206, constant 26·309). Key: ASMI, appendicular skeletal muscle index.
3.4 eGFR muscular error varied by formula and trimethoprim use
Fig. 4Muscle mass error varies by eGFR formula used. Unindexed eGFR error (eGFR–mGFR) was compared against ASMI, a muscularity marker. The linear regression slope (coefficient±SE) and Pearson’s correlation reflect the relative influence of muscle mass on eGFR error. Negative slope coefficients indicate a muscle-dependent underestimation of true mGFR. The error from CKD-EPI eGFR equation was less dependent on muscularity compared with MDRD formula. In contrast, Cockcroft-Gault (which additionally incorporates weight), showed no demonstrable “eGFR” error (derived from creatinine clearance) from muscular mass. Key: ASMI, appendicular skeletal muscle index; CKD-EPI, Chronic Kidney Disease EPIdemiology; MDRD, Modification of Diet in Renal Disease (formulae). Pearson correlation, linear regression coefficient (±SE), and 95% prediction bands are presented.
], acute tubular injury, glomerulosclerosis (n = 133 biopsies), immunosuppression, corticosteroid exposure, transplant variables, and serum albumin had no effect on error (Table S5).
Random eGFR error was approximated after comprehensive multivariable linear regression analysis. Sequential coefficients of determination (R2) were: 0·293 (physical variables including ASMI, n = 15); 0·309 (trimethoprim); 0·825 (function including mGFR, n = 3); 0·872 (transplant-related, n = 16); 0·911 (kidney pathology, n = 12). The residual variance was 8·9% (1–0·911).
3.5 Serum creatinine independently correlated with muscle mass
], and ASMI; and functional markers of mGFR and serum urea (R2 0·823, Tables 5, S9).
Fig. 5Muscle mass increases serum creatinine concentration. Both the ASMI (in kg/m2) and lean muscular mass (in kg) significantly correlated with serum creatinine concentration (Panels A and B), but not with percentage body fat (C). When stratified using unindexed mGFR into CKD stages, serum creatinine still associated with ASMI within each strata of function (A). The plot of serum creatinine against isotopic mGFR when stratified into ASMI quartiles (1st quartile is cachexia, 4th is the most muscular, Panel D) is a family of four curves which are affected by creatinine generation from muscle. eGFR underestimation error was greatest in muscular (ASMI quartile 4) patients with high mGFR levels (E). The negative eGFR error from muscle mass displayed a consistently negatively slope across all CKD stages (panel F). Key: ASMI, appendicular skeletal muscle index; CKD, chronic kidney disease levels were defined as stage 1 (≥90mls/min), 2 (60–89mls/min), 3 (30–59mls/min), and 4 (<30mls/min). Pearson correlation coefficients and 95% prediction bands are presented.
Table 5Multivariable predictors of serum creatinine. Multivariable linear regression analysis of loge serum creatinine concentrations, controlled for renal function using mGFR and serum urea (n = 137, R2 0·823, constant 3·602). Key: ASMI, appendicular skeletal muscle index.
3.6 ASMI and mGFR exert differential effects on serum creatinine
], and serum urea were added (R2=0·823, Tables 6, S11–14). Within each mGFR level, ASMI associated with serum creatinine (R2 0·413–0·805). At high function (GFR≥90 mls/min, n = 34), ASMI was the dominant predictor of serum creatinine and could be estimated from creatinine alone, irrespective of mGFR (R2 0·471, Table S10c).
Table 6Competing multivariable predictors of serum creatinine by CKD stage. Multivariable predictors of serum creatinine (loge µmol/L) using linear regression to assess relative contributions of renal clearance versus muscular input: using isotopic mGFR (per 10mls/min) and ASMI (kg/m2) at different CKD stages. Note that regression coefficient measurement is unit dependent. Key: Coefficient (±SE) is the linear regression slope. Key: ASMI, appendicular skeletal muscle index; CKD, Chronic Kidney Disease.
3.7 Diagnostic test performance of eGFR by ASMI quartile
Table 7Diagnostic performance of eGFR by ASMI quartiles. The test performance of CKD EPI eGFR (unmodified as mls/min/1·73 m2) to detect CKD stage 3 (mGFR<60 mls/min/1·73 m2) by ASMI muscularity quartile (n = 137) Key: PPV and NPV are positive and negative predictive values, respectively. ASMI, appendicular skeletal muscle index where quartile 1 is the least muscular. Results are percentages.
3.8 Surrogate predictors of ASMI and eGFR error
We investigated simple physical markers as potential surrogate predictors for muscularity. ASMI muscularity correlated with body weight (r = 0·825, P<0·001), male (r = 0·480, P<0·001), height (r = 0·537, P<0·001), BMI (r = 0·710, P<0·001), and BSA (r = 0·711, P<0·001), but not with recipient age (Fig. S3, Table S14). Multivariable regression found male recipient and weight, independently predicted ASMI (P<0·001, R2 0·739, Table S15). However, the clinical utility of physical markers to predict error was poor. Unindexed eGFR error correlated with weight (r = 0·230, P = 0·008), male sex (P = 0·028, t-test), and BMI (r = 0·197, P = 0·023, Fig. S4). Multivariable regression found male sex and BMI only weakly predictive of error (R2 0·115, Table S16).
4. Discussion
]. A recent large systematic review of over 70 eGFR formulae found suboptimal P30 accuracy metrics of 60–90% for CKD (10–40% exceeded P30); 40–90% for diabetic nephropathy; and 30–90% for kidney transplantation [
]. Poor agreement, lack of concordance, and unpredictable errors were common which remained unchanged by cystatin C use or IDMS calibration. Misclassification of CKD stage was frequent. The authors controversally concluded that eGFR was an unreliable tool to assess renal function in health and disease [
,
,
]. eGFR inaccuracy incorporates systemic bias and imprecision, which are best considered separately. Our study found substantial systemic eGFR error from measured muscle mass that was proportional to lean muscle mass and ASMI quartile, influential across all CKD stages, and dependent on eGFR formula used for calculation. Modest proportional bias underestimated mGFR at higher function (and vice versa), and tubular secretion increased eGFR error by 12·5%. Increasing imprecision which paralleled greater mGFR levels (with dispersion in difference and residual plots) caused by greater variance with higher absolute values, combined with suboptimal predictive capability of serum creatinine in normally functioning kidneys.
]. Lean mass and ASMI proportionally influence serum creatinine concentrations across all CKD stages.
Interestingly, eGFR error associated with ASMI varied according to the formula used for its calculation. The reduced error for CKD-EPI compared with MDRD, could be explained by better muscular estimation from age, sex, and African race variables. CKD-EPI is actually two distinct formulae for male or female subjects, with separate exponents for input demographic predictors. In contrast, Cockcroft-Gault formula (which additionally inputs weight) eliminated muscle mass error (weight strongly correlated with ASMI, r = 0·825, P<0·001). Discrepancy of eGFR error from muscularity reflects an inadequate compensation of muscular creatinine generation by each formula’s variables.
,
,
,
], which possess a greater dynamic range and ability to alter function. The CKD-EPI formula adjusts the serum creatinine exponent above an inflexion point at 62 and 80 µmol/L (female and male, respectively). Poorly-functioning kidneys operate fewer individual nephrons at maximal capacity (hyperfiltration), an display reduced physiological reserve and capability to upscale GFR. Additional imprecision from cachexia and muscularity were amplified at the extremes of renal function (Fig. 5E).
,
,
]. However, the residual variance after comprehensive multivariable modeling of 47 demographic, muscular, functional, and clinical inputs was only 8·9%. Published within-subject CV values are 4·5% to 8·0% for mGFR (healthy individuals and CKD, respectively); 4·4% (0·7% analytical variation) for serum creatinine; 5·3% for CKD-EPI [
]; and 2·0% for DEXA muscle measurements [
]. Another explanation for eGFR-mGFR differences is methodological mismatch. Both eGFR and mGFR reflect “true GFR”, but calculate results over differing time spans using alternative methodologies. Hour-to-hour GFR acutely varies with hydration, protein loading, and blood pressure, and can precipitously drop to minimal levels with severe hypotension (creatinine and eGFR are initially unchanged but deteriorate after hours or days). Isotopic plasma clearance only calculates GFR during the linear clearance phase following bolus equilibration (i.e. 1 or 2 h) [
]. In contrast, eGFR integrates serum creatinine levels over several days to deduce renal function.
]. Kidney transplant recipients avoid the intrinsic collinearity between muscularity and body size (large, muscular individuals with bigger kidneys produce greater GFR). Routine trimethoprim blockade of tubular secretion without altering mGFR converts serum creatinine into an “ideal” filtration marker [
]. Fixed and proportional biases, precision, and accuracy were analysed separately for clarity. Weakness are relatively small numbers of CKD4 patients (n = 6) and trimethoprim use in a transplant cohort, which limits extrapolation of error estimates to the general population, which is likely greater due to variability in inhibited tubular secretion.
]. Panels of multiple markers (e.g. low-molecular-weight proteins or metabolites) or novel non-renal serum markers of muscularity may help [
,
]. Until then, clinicians should carefully interpret eGFR results with observed muscularity, reserve accurate mGFR or 24-hour creatinine clearance for selected divergent cases, and use old-fashioned clinical judgment for pateints at the extremes of body habitus.
Acknowledgments
Funding
Not applicable.
Author’s contributions
All authors participated in manuscript writing and revision: SMG originated the idea of comparing muscular mass to isotopic GFR, within a collaborative project on body composition developed by GJE; SG, LFJN, and BJN undertook primary data acquisition; BJN was responsible for research design and data analysis.
Data sharing statement
Extensive summary data and analysis are presented within the supplemental material. These contain 23 highly-detailed tables of de-identified summated clinical data with their univariable and multivariable statistical analyses to allow open scientific scrutiny. Federal privacy laws and local institutional ethics forbid the placement of confidential individual patient information onto any public data-sharing website or allow for its unauthorised sharing. Specific questions of clinical science can be directed to the corresponding author.
Disclosures
None.
