Measurement, Estimation, and Correlates of the GFR before and after Bariatric Surgery

Visual Abstract

Abstract

Key Points

• In individuals undergoing bariatric surgery, the magnitude of decline in GFR postbariatric surgery is directly associated with presurgery GFR.

• Reductions in GFR postsurgery were only weakly correlated with weight loss.

• In obese individuals, GFR-estimating equations performed best when deindexed and when applied in people with reduced kidney function.

Background

Important questions remain about how bariatric (i.e., weight loss) surgery affects measured GFR (mGFR) and eGFR as well as what factors influence change in mGFR postsurgery.

Methods

Data were pooled from all seven available studies (dates: 2004–2018) measuring GFR prebariatric and postbariatric surgery using gold standard methods. Change in postsurgery mGFR, factors that could influence change in mGFR, and effects on five GFR-estimating equations were analyzed using standard statistical methods.

Results

The cohort included 105 individuals from the United States and Europe. Sixty-eight percent were female, 97% were White, the mean age was 50 years (range, 24–70), and the mean body mass index was 46±8 kg/m². The mean presurgery mGFR of 107 ml/min (range, 31–215) fell to 92 ml/min (−14%; 95% confidence interval, −21 to −10) postsurgery, with a strong linear relationship existing between presurgery mGFR and % change in GFR (r=−0.51; −0.64 to −0.35). Individuals with presurgery mGFR ≥90, 60 to <90, and<60 ml/min had −25 (−32 to −19), 1 (−11 to 13), and 7 (0 to 14) ml/min mean postoperative changes in mGFR. Change in weight was significantly correlated with change in mGFR (r=0.22; 0.03 to 0.40). After adjusting for sex, changes in mGFR postsurgery were associated with higher presurgical age (−0.6 ml/min per year; −1.1 to −0.2), mGFR (−0.5 ml/min per 1 ml/min; −0.6 to −0.4), and change in systolic BP (−0.3 ml/min per 1 mm Hg; −0.6 to 0.0). All GFR-estimating equations significantly underestimated postsurgical GFR reductions in people with preserved kidney function and had improved bias, precision, and accuracy when deindexed and applied to individuals with mGFR <90 ml/min.

Conclusions

In individuals with obesity undergoing bariatric surgery, the magnitude of postsurgical decline in mGFR was directly associated with presurgery GFR and weakly correlated with weight loss. In addition, GFR-estimating equations' performance improved when deindexed and used in people with reduced kidney function, with the combined creatinine/cystatin C equations having the best overall performance.

Introduction

Obesity is a global public health crisis that affects 40% of the American adult populace and nearly 50% of Americans with CKD stages 3–4.^1,2 Obesity is also a key risk factor for the development and progression of CKD.^3–5 Studying the obesity–kidney link has consequently become a topic of keen interest for the nephrology community,^6–8 especially in light of the rising prevalence of obesity and the growing availability of effective antiobesity treatments.⁹

Although the pathophysiology of obesity-related CKD is still being elucidated, damage from chronic elevations in the GFR (i.e., hyperfiltration) is believed to be one important mechanism.^10,11 Several research studies have directly measured GFR (mGFR) in individuals with obesity using gold standard methods before and after weight loss,^12–24 but insights on the precise relationship between obesity, GFR, and weight loss as well as identification of factors correlated with or that influence change in GFR have been limited by the modest size of these individual studies.

These few studies aside, GFR is usually estimated for diagnostic, treatment, and even research purposes using one of several equations. Owing to factors such as routine indexing for body surface area (BSA) and the effect of excess adiposity on the two major kidney clearance markers serum creatinine and cystatin C, eGFR is less accurate when used in populations with obesity.^25,26 This, in turn, makes it more challenging to optimally manage people with obesity and CKD and study the effects of obesity on GFR.

To address the timely issues above, we pooled and analyzed all available published data in individuals with obesity who had mGFR performed before and after metabolic (i.e., bariatric) surgery to optimize weight loss.

Methods

Pooled Studies

A literature search identified all published studies that performed mGFR in individuals with obesity undergoing weight loss.^12–24 Approval was obtained from each primary author for inclusion of their data in this study. Specific information about each study is available in their published versions.^12–18 All studies had Institutional Review Board approval.

Measurements

Each study had one prebariatric surgery visit and at least one postsurgery visit. Three studies included two or more postsurgery visits (one at 6 and 12 months,¹⁴ n=11; one at approximately 1–2 weeks and 1 year,¹³ n=4; and one at 3, 6, 12, and 24 months, n=12¹⁶). Only data points from the 12-month visit were included from these studies because relatively few individuals in the entire cohort had >1 measurement and because we wanted to include one long-term follow-up visit common to all participants. mGFR was performed before and after bariatric surgery using one of several gold standard methods as previously described.^12–18 The period from presurgery mGFR to bariatric surgery and from bariatric surgery to postsurgery mGFR differed between studies because of differing protocols. In 20 individuals, the former period was not measured or reported. Five major serum creatinine–based and/or serum cystatin C–based GFR-estimating equations were used to estimate GFR.^27–29 BSA was calculated using the DuBois and DuBois equation.³⁰ Nonindexed eGFRs (in ml/min) were calculated by multiplying eGFR by (actual BSA [m²]/1.73 m²). mGFR was indexed by multiplying mGFR by (1.73 m²/actual BSA [m²]). Albuminuria and proteinuria were measured by either a spot urine method (mg/g) or 24-hour urine collection (mg/24 hour) depending on the study. Of the 105 serum creatinine samples, 96 used the standardized isotope dilution mass spectrometry–traceable method. For serum cystatin C measurements, 27 of the 72 specimens¹⁵ were traceable to the international standard, nine were not, and for 36 specimens, use of the standard was unknown because the original laboratory no longer exists.³¹ Glomerular hyperfiltration referred to throughout this paper is defined as an increased single-nephron GFR with or without total kidney increased GFR. Specific assays used to measure serum insulin, hemoglobin A1c (HbA1c), glucose, and albuminuria or proteinuria and techniques to measure BP have been previously described.^12–18 All participants had measurements performed in the fasting state except one.¹⁵ Presurgical presence of diabetes was defined by the use of diabetes medications or HbA1c of >6.5% and hypertension was defined by systolic and diastolic pressure of >140 and >90 mm Hg, respectively, or use of antihypertensive medications.

Outcomes

Study outcomes included evaluating the effect of bariatric surgery on mGFR and other kidney-related and metabolic-related parameters, identifying factors associated with change in mGFR, and identifying the most predictive eGFR equation before and after surgery.

Statistical Analysis

Demographics, presurgery and postsurgery patient characteristics, and change (both absolute and percent) in characteristics were summarized. Continuous variables were described by mean (SD) and median interquartile ranges (IQRs). Categorical variables were summarized by frequency (relative frequency by percentage). Paired t tests and Wilcoxon signed-rank tests were used to compare the change in patient characteristics over time as appropriate. One-way ANOVA was performed to compare changes of GFR and weight postsurgery among different presurgery GFR groups. A spaghetti diagram was plotted to show average and individual changes in GFR postsurgery and presurgery. T tests and Fisher exact tests were used in all participants and subsets to compare demographics and presurgery characteristics between participants with and without reductions in GFR postsurgery. Pearson correlation coefficients were calculated between differences from presurgery to postsurgery on GFR, HbA1c, weight, body mass index (BMI), systolic BP, serum insulin, and serum glucose. Spearman correlation coefficient (rho) was used to represent monotonic correlation between presurgery mGFR and % change in mGFR. Univariable linear regression analysis was performed to identify associations between presurgery characteristics and changes in biomarkers with changes in mGFR. Linear regression using stepwise variable selection was used to identify a final parsimonious model after forcing age, sex, and presurgery mGFR and also including presurgery albuminuria and changes in weight, systolic BP, HbA1c, and albuminuria, with significance levels α=0.15 for entry and α=0.05 to stay. Of note, weight loss was not included in the final multivariable model due to high collinearity with other variables. T tests were also used to compare differences from presurgery to postsurgery on GFR between participants with and without diabetes in all participants and subsets. Ninety-five percent confidence intervals (CIs) were reported as appropriate.

Performance of eGFR equations was described by bias (mean [SD]) and precision (median; IQRs) of bias on differences between calculated and mGFR. Accuracy was defined as the percentage of values within ±20% (P_20%) or 30% (P_30%) of mGFR. eGFRs from different equations were compared with mGFR by comparative boxplots at each presurgery and postsurgery time point, by indexing, and stratification by presurgery GFR. eGFR equations were considered to perform well when yielding small absolute mean and median bias and high accuracy. T tests and Wilcoxon signed-rank tests were used to compare the changes in mGFR and difference between the changes in mGFR and each eGFR equation at postsurgery from presurgery.

Results

Of the 13 studies identified as measuring GFR before and after weight loss,^12–24 seven provided data and all involved bariatric surgery.^12–18 The six studies were excluded due to inability to contact study investigators¹⁹ and inability of study investigators to locate study data^20–23 or provide data.²⁴ Supplemental Figure 1 shows a flow chart of the seven included studies.

Presurgery Description of Population

Of the 105 individuals included, 74 were from the United States, nine were from the England, ten were from France, 12 were from Spain, 68% were female, and 97% were White. The average age was 50 years (range, 24–70 years), weight was 130 kg (SD=28; range, 86–238 kg), BMI was 46 kg/m² (SD=8; range, 33–82 kg/m²), and BSA was 2.3 m² (SD=0.3; range, 1.8–3.2 m²). Presurgery measurements took place 0–446 days before surgery in the 85 patients for whom this information was available. The mean mGFR was 107 ml/min (SD=40; range, 31–215 ml/min), with 24%, 8%, and 4% of participants having GFRs between 60 to <90, 45 to <60, and 30 to <45, respectively. The mean serum creatinine was 1.0 mg/dl (SD=0.4; n=104) and cystatin C was 1.1 mg/L (SD=0.4; n=72). The median urinary excretion of albumin was 16 (IQR, 5–92) and protein 86 (IQR, 30–1100) in mg/g or mg/24 hour. A comprehensive list of presurgery participant characteristics is presented in Supplemental Table 1.

Postsurgery mGFR, Weight, and Other Parameters

Supplemental Table 2 presents participant characteristics at the postsurgery visit, which took place an average of 280 (SD=96) days after surgery. Table 1 presents changes in a variety of clinical and biochemical parameters that occurred over an average 395 days (SD=123; median, 378 days; range, 117–838 days) between presurgery and postsurgery measurements. A mean of 36.0 kg (95% CI, −38.4 to −33.7) was lost (SD=12; range, 7–69 kg), leading to a reduction in average BMI by 12.8 k/m² (95% CI, −13.6 to −12.0).

Table 1.

Changes in clinical and biochemical parameters after versus before bariatric surgery

Variables	Presurgery		Postsurgery		Differencec
	N	Mean (SD)a or Median (25th–75th %ile)b	N	Mean (SD)a or Median (25th–75th %ile)b	N	Mean or Median (95% CI)d
Weight, kga	105	130.0 (28.2)	105	93.9 (23.7)	105	−36.0 (−38.4 to −33.7)
% Changea						−28 (−29 to −26)
BMI, kg/m 2 a	105	46.1 (8.0)	105	33.2 (6.9)	105	−12.8 (−13.6 to −12.0)
% Changea						−28 (−29 to −26)
BSA, m 2 a	105	2.3 (0.3)	105	2.0 (0.3)	105	−0.3 (−0.3 to −0.3)
% Changea						−13 (−14 to −12)
Systolic BP, mm Hga	105	131 (18)	104	125 (17)	104	−6 (−9 to −3)
% Changea						−4 (−6 to −2)
Diastolic BP, mm Hga	105	77 (10)	104	73 (10)	104	−4 (−6 to −2)
% Changea						−4 (−7 to −2)
mGFR, ml/mina	105	107 (40)	105	92 (33)	105	−15 (−21 to −10)
% Changeb						−10 (−15 to −4)
Urinary albumin excretion, mg/g or mg/24 hb	103	16 (5–92)	102	14 (6–46)	100	−1 (−7 to 2)
% Changeb						−6 (−42 to 25)
Urinary protein excretion, mg/g or mg/24 hb	43	86 (30–1100)	44	108 (30–350)	42	0 (−24 to 2)
% Changeb						0 (−24 to 1)
Serum creatinine, mg/dlb	104	0.9 (0.7–1.1)	105	0.8 (0.7–1.0)	104	−0.1 (−0.1 to −0.1)
% Changeb						−11 (−18 to 0)
Serum cystatin C, mg/Lb	72	1.1 (0.9–1.2)	71	1.0 (0.9–1.2)	70	−0.0 (−0.1 to 0.0)
% Changea						−2 (−9 to 8)
Serum glucose, mg/dlb	71	99 (86–120)	71	83 (78–96)	71	−11 (−15 to −7)
% Changea						−14 (−19 to −8)
HbA1c, %b	38	5.8 (5.5–7.8)	38	5.3 (4.9–5.7)	38	−0.7 (−1.0 to −0.3)
% Changeb						−12 (−16 to −5)
Serum insulin, mU/mlb	18	22.4 (19.4–32.0)	18	9.6 (4.7–11.8)	18	−14.4 (−21.3 to −10.0)
% Changeb						−68 (−82 to −45)

BMI, body mass index; BSA, body surface area; CI, confidence interval; HbA1c, hemoglobin A1c; mGFR, measured GFR.

^aReported as mean (SD) for normally distributed variables.

^bReported as median (25th–75th %ile) for non-normally distributed variables.

^cDifferences were calculated using only individuals who had both presurgery and postsurgery data points available.

^dReported as mean (95% confidence interval) for normally distributed variables or median (95% confidence interval) for non-normally distributed variables.

Although the mGFR fell by a mean of 15 ml/min (95% CI, −21 to −10) to 92 ml/min (−14% overall decline), change in mGFR differed by presurgery GFR. Figure 1 shows the strong linear relationship between presurgery mGFR and % change in mGFR after surgery (r=−0.51; 95% CI, −0.64 to −0.35) where the higher the presurgery mGFR, the greater the % reduction in mGFR postsurgery. The mean postsurgery change in mGFR in people with presurgery mGFR ≥90 (n=68), 60 to <90 (n=25), 45 to <60 (n=8), and 30 to <45 (n=4) ml/min was −25 (95% CI, −32 to −19), 1 (95% CI, −11 to 13), 7 (95% CI, −3 to 18), and 6 (95% CI, −4 to 17) ml/min, respectively, representing an 18% decrease and 3%, 14%, and 17% increase in GFR, respectively. The mean postsurgery change in mGFR in people with presurgery mGFR ≥90 (n=68) and <90 (n=37) was −25 (95% CI, −32 to −19) and 3 (95% CI, −5 to 11) ml/min, respectively, representing an 18% decrease and 7% increase in GFR, respectively (95% CI of difference between changes, 18 to 3). Individuals with a presurgery mGFR <60 (n=12) versus ≥60 (n=93) ml/min had a 7 ml/min increase (0–14) versus an 18 ml/min decrease (95% CI, −24 to −12) in mGFR postsurgery (95% CI between groups in change in GFR, 8–42). The mean weight loss differed between individuals with baseline mGFR ≥90 versus <90 ml/min (38.5 versus 31.5 kg; 95% CI, 2.1 to 11.7), but no statistical difference was noted between individuals with baseline mGFR ≥60 versus <60 ml/min (−36.1 versus −35.6; 95% CI, −7.0 to 8.0).

Figure 1.

Relationship between presurgery GFR and change in GFR after surgery. CI, confidence interval.

Individual trajectories in mGFR presurgery versus postsurgery are presented in Supplemental Figure 2. Of the 105 total patients, 73 (70%) had a reduction in postsurgery mGFR, while 32 (31%) had either no decrease or an increase in GFR. Supplemental Table 3 shows that the cohort whose mGFR fell postsurgery had a higher presurgery mGFR (118 versus 80 ml/min; 95% CI, −53 to −23) and lower serum creatinine (median, 0.8 versus 1.1 mg/dl; 95% CI, 0.1 to 0.4) and lower cystatin C (median, 1.0 versus 1.1 mg/L; 95% CI, 0 to 0.5), were younger (48 versus 54 years; 95% CI, 1 to 11), and had a longer time gap between presurgery mGFR and bariatric surgery (median 110 versus 82; 95% CI, −86 to 4) as compared with the cohort without a postsurgical GFR reduction. However, in the subgroup with a presurgery mGFR >90 ml/min (N=68), no significant differences were observed in the demographic and laboratory measurements between those whose GFR fell postsurgery (N=58, 85%) versus those in whom GFR did not fall (N=10, 15%).

As presented in Table 1, additional findings postsurgery included reductions in systolic (−6 mm Hg; 95% CI, −9 to −3) and diastolic (−4 mm Hg; 95% CI, −6 to −2) mean BP, median serum glucose (−11 mg/dl; IQR, −15 to −7), median HbA1c (−0.7%, IQR, −1.0 to −0.3), and median insulin (−14.4 mU/ml; IQR, −21.3 to −10.0). Serum creatinine (−0.1 mg/dl; IQR, −0.1 to 0) was also slightly lower, but urinary albumin and protein excretion and serum cystatin C were unchanged.

Correlates of Change in mGFR

Change in mGFR was moderately correlated with change in HbA1c (r=0.40; 95% CI, 0.08 to 0.63); weakly correlated with changes in weight (r=0.22; 95% CI, 0.03 to 0.40), BMI (r=0.24; 95% CI, 0.05 to 0.41), and systolic BP (r=0.23; 95% CI, 0.04 to 0.41); and not significantly correlated with changes in serum insulin or glucose. The magnitude of change in mGFR postsurgery versus presurgery was significantly greater with a higher presurgery GFR (−0.4 ml/min drop per 1 ml/min higher GFR; 95% CI, −0.6 to −0.3), more weight loss (−0.5 ml/min per extra 1 kg; 95% CI, −1.0 to 0.1), and larger falls in systolic BP (−0.4 ml/min per 1 mm Hg drop, −0.8 to −0.1) and HbA1c (−7.2 ml/min per 1% drop; 95% CI, −12.8 to −1.6). Study site, presurgery age, sex, race, Hispanic ethnicity, or presence of presurgery diabetes or hypertension were not significantly associated with change in GFR. After fitting a final multiple linear regression model, presurgical age (−0.6 ml/min per year; 95% CI, −1.1 to −0.2), presurgical GFR (−0.5 ml/min per 1 ml/min; 95% CI, −0.6 to −0.4), and change in systolic BP (−0.3 ml/min per 1 mm Hg change; 95% CI, −0.6 to 0.0) were each significantly associated with change in GFR after surgery once adjusting for sex (NS). Change in weight was not selected in the final model likely due to its high multicollinearity with age (r=0.38 [95% CI, 0.20 to 0.53]) and presurgery GFR (r=−0.40 [95% CI, −0.55 to −0.23]). Supplemental Table 4 presents the unadjusted and final multivariable regression models used. There was no statistical difference between individuals with or without diabetes in change in GFR in the total cohort or in subgroups (<90 versus >90 ml/min).

Performance of GFR-Estimating Equations

Figure 2 and Table 2 present the bias, precision, and accuracy for each equation (including race-inclusive and race-free) presurgery and postsurgery and nonindexed to BSA, as compared with mGFR, as well as relative performance rankings. Before surgery, all indexed equations had high bias (range, −38.9 to −21.7 ml/min), reflecting a large underestimation of the true GFR, low precision (range, −37.0 to −17.5 ml/min), and low-moderate accuracy (P_30% range 36.1%–67.3%), with the creatinine-based equations (race-inclusive and race-free) performing best overall (bias, −21.7 and −24.5 ml/min; precision, −17.5 and −20.5 ml/min; accuracy, P_30% 67.3% and 65.4%), and the cystatin C–based equation performing worst. After surgery, bias, precision, and accuracy improved for all equations, with creatinine-based equations again performing best overall and the cystatin C–based equation again performing worst.

Figure 2.

Boxplots of differences between GFR and eGFR by time point or period, eGFR equation, and presence of equation indexing for BSA. (A) Indexed eGFR−mGFR at presurgery, the difference between presurgery and postsurgery, and postsurgery. (B) The same variables except that eGFR is index-free. Equations represented are the following: Creat, race-inclusive creatinine-based CKD-EPI equation²⁷; comb, race-inclusive combined creatinine and cystatin C–based CKD-EPI equation²⁸; CysC, cystatin C–based CKD-EPI equation²⁸; race-free creat, race-free creatinine-based CKD-EPI equation²⁹; race-free comb, race-free combined creatinine and cystatin C–based CKD-EPI equation.²⁹ Index-free GFR estimations were calculated by multiplying the eGFR by (actual BSA [kg/m²]/1.73 m²). BSA, body surface area; CKD-EPI, CKD Epidemiology Collaboration; mGFR, measured GFR.

Table 2.

Performance of eGFR equations versus measured GFR presurgery and postsurgery

eGFR Equations	Indexed Equations					Nonindexed Equations
	N	Bias, Mean (SD)	Precision (IQR)	Accuracy		N	Bias, Mean (SD)	Precision (IQR)	Accuracy
				(P20%)	(P30%)				(P20%)	(P30%)
Presurgery
eGFRCr	104	−24.5 (27.5)2	−20.5 (−37.3 to −8.0)2	37.52	65.42	104	3.8 (25.5)2	3.3 (−9.9 to 16.5)3	70.23	77.93
eGFRCr+CysC	72	−32.9 (31.6)4	−31.5 (−45.4 to −14.5)4	31.93	48.64	72	−5.4 (26.8)3	−2.4 (−19.5 to 9.0)1	72.22	87.51
eGFRCysC	72	−38.9 (35.6)5	−37.0 (−56.4 to −18.5)5	20.85	36.15	72	−14.0 (31.7)5	−10.0 (−32.0 to 5.5)5	54.25	70.85
Race-free eGFRCr	104	−21.7 (28.0)1	−17.5 (−34.0 to −5.0)1	43.31	67.31	104	7.5 (26.1)4	6.9 (−5.5 to 20.8)4	64.44	76.94
Race-free eGFRCr+CysC	72	−30.4 (32.3)3	−28.0 (−42.0 to −11.5)3	31.93	54.23	72	−2.0 (27.4)1	3.0 (−15.5 to 12.5)2	73.61	84.72
Postsurgery
eGFRCr	105	−1.9 (24.7)2	−1.0 (−15.4 to 13.0)1	56.21	83.81	105	13.1 (24.0)4	11.6 (1.7–26.0)3	50.54	69.54
eGFRCr+CysC	71	−16.2 (25.4)4	−13.4 (−33.0 to 2.0)4	52.13	73.24	71	−1.2 (21.6)1	−0.6 (−12.0 to 11.5)1	67.62	85.91
eGFRCysC	71	−25.7 (27.1)5	23.4 (−42.9 to −7.0)5	39.45	56.35	71	−12.8 (23.1)3	−12.0 (−24.2 to 1.3)4	53.53	77.53
Race-free eGFRCr	105	0.9 (25.6)1	3.0 (−13.3 to 17.0)2	55.22	77.12	105	16.5 (25.0)5	15.0 (3.0–29.0)5	45.75	61.95
Race-free eGFRCr+CysC	71	−13.7 (26.0)3	−11.4 (−29.0 to 4.4)3	50.74	74.63	71	1.7 (22.1)2	1.4 (−10.0 to 14.0)2	69.01	81.72

Nonindexed GFR estimations were calculated by multiplying the eGFR by (actual body surface area [kg/m²]/1.73 m²). Bias is calculated as eGFR–measured GFR. Precision is defined as the median of the bias, interquartile range=25th–75th %ile. Accuracy is defined as the percentage of values within ±20% (P_20%) or 30% (P_30%) of the measured GFR. Superscripts represent the relative ranking of each equation (1-best, 5-worst) in the four categories of presurgery versus postsurgery and indexed versus nonindexed equations. eGFR_Cr, eGFR calculated with the race-inclusive creatinine-based CKD Epidemiology Collaboration equation²⁷; eGFR_Cr+CysC, eGFR calculated with the race-inclusive combined creatinine and cystatin C–based CKD Epidemiology Collaboration equation²⁸; eGFR_CysC, eGFR calculated with the cystatin C–based CKD Epidemiology Collaboration equation²⁸; IQR, interquartile range; race-free eGFR_Cr, eGFR calculated with the race-free creatinine-based CKD Epidemiology Collaboration equation²⁹; race-free eGFR_Cr+CysC, eGFR calculated with the race-free combined creatinine and cystatin C–based CKD Epidemiology Collaboration equation.²⁹

When indexing for BSA was removed, presurgery bias (range, −14.0 to 7.5 ml/min), precision (range, −10.0 to 6.9 ml/min), and accuracy (range, 70.8%–87.5%) all improved compared with the indexed state, with both combined creatinine+cystatin C equations now performing best overall (bias, −2.0 and −5.4 ml/min; precision, −2.4 and 3.0 ml/min; accuracy, 84.7% and 87.5%) and the cystatin C equation performing worst. When indexing was removed postsurgery, the combined creatinine+cystatin C–based equations continue to perform best for bias, precision, and accuracy, while creatinine-based equations performed worst.

The pattern described above was not qualitatively different between individuals with less versus more than the median weight loss postsurgery (Supplemental Table 5). When comparing indexed eGFR with indexed mGFR, the results were qualitatively similar to comparing nonindexed eGFR with nonindexed mGFR (Supplemental Table 6). Supplemental Table 7 shows that GFR changes after surgery significantly differ between mGFR and eGFR, regardless of equation used, particularly in people with preserved presurgical mGFR (i.e., ≥90 ml/min), since median mGFR falls dramatically but median eGFR does not. Supplemental Table 8 presents these findings by showing how eGFR equations have observably greater bias and lower precision and accuracy presurgery in people with preserved mGFR.

Discussion

This study, which contains the largest cohort of individuals with mGFR preweight and postweight loss surgery, offers an opportunity to explore several important clinical and scientific questions. The first involves the effect of bariatric surgery on GFR. Most^{12,14–16,19–21} but not all^17,18,23 reports describe obesity as a state of relative glomerular hyperfiltration that is reversed after bariatric surgery. Our study confirms an average reduction in GFR postsurgery versus presurgery. However, the magnitude of the reduction is not consistent, for as the presurgery GFR declines so too does the magnitude of the reduction, with a minimum absolute change observed as the presurgery GFR approaches 60 ml/min. Importantly, reversal of glomerular hyperfiltration through a fall in GFR is expected to provide renoprotection at least in part by reducing mechanical stress on glomerular and proximal tubule cells.¹¹ It is unknown if renoprotection is conferred in people whose GFR does not fall postsurgery. Identifying factors that mediate changes in GFR even after stratifying for presurgery GFR merits further exploration.

Two possible explanations could explain this finding. The first is that the putative mechanism(s) underlying the glomerular hyperfiltration of obesity—tubuloglomerular feedback³² and possibly obesity-related increases in sympathetic activity, circulating insulin, leptin, and angiotensin II—are progressively impaired as GFR falls below approximately 60 ml/min. However, additional contributory effects on vascular reactivity cannot as of yet be excluded.

The second hypothesis focuses on single-nephron GFR, the summation of which defines whole kidney GFR. This hypothesis posits that obesity induces a state of glomerular hyperfiltration that affects all individual nephrons. Although single-nephron GFR remains elevated and does not fall as presurgery GFR declines, the total functioning kidney mass does decline due to ongoing damage and nephron dropout. Progressively few functioning glomeruli inevitably lead to a lower presurgery GFR. Because of ongoing demands to maintain a high total GFR, single nephron hyperfiltration persists. This hypothesis is reminiscent of that seen with kidney functional reserve, which declines as GFR falls, a phenomenon attributed to the remaining nephrons already working at near maximum capacity.³³ Intriguingly, kidney function reserve is thought to occur at least partly through downregulation of the tubuloglomerular feedback mechanism (similar to obesity) and is stimulated by dietary protein consumption. Friedman et al. reported that while protein consumption modestly raised the GFR in individuals with obesity, it was unlikely to fully explain the phenomenon of obesity-related glomerular hyperfiltration.^34,35

A notable finding in this study relates to the specific factors that associate with changes in mGFR after bariatric surgery. Of particular interest was the weak correlation between change in GFR and weight loss. Several previous studies in individuals treated with bariatric/bariatric surgery or antiobesity pharmacotherapy have also reported modest to no correlation between the magnitude of weight loss and mGFR change,^{12,24,36–39} raising the possibility that weight loss is more of an epiphenomenon rather than the primary driving force behind correction of hyperfiltration and other kidney-related benefits. One possible driving factor is glycemia. Change in HbA1c was the strongest correlate of change in GFR that we identified among the 38 individuals with available presurgery and postsurgery HbA1c, similar to what was previously seen with antiobesity medications.³⁸ It is noteworthy that chronic glycemia is a major factor mediating the hyperfiltration of diabetes.⁴⁰ This important topic deserves more study.

Several factors—excess fat, weight loss, and indexing for BSA—are especially important when studying the performance of GFR-estimating equations in individuals with obesity and/or weight reduction. In general, people with obesity typically have more muscle mass, the primary source of creatinine generation, compared with lean persons.⁴¹ This is reflected in higher serum creatinine levels as compared to after weight loss has occurred, as seen in Table 1. Similarly, excess fat is associated with higher circulating cystatin C, a phenomenon also independent of GFR.⁴² The routine standardization of GFR by BSA leads to GFR underestimation in persons with obesity, as others have observed.⁴³ These three factors help explain why indexed eGFR equations so greatly underestimated mGFR at the presurgery visit. Weight loss should improve indexed-eGFR performance as extra muscle and fat mass are lost and BSA converges toward the standard 1.73 m² unit. This is in fact what we observed after bariatric surgery.

Our results argue for deindexing and using the combined creatinine+cystatin C equations when estimating GFR in individuals with obesity. Removing indexing for BSA improved the performance of all equations, with the deindexed combined creatinine+cystatin C equation performing best. In persons with obesity undergoing weight loss, the superiority of the combined eGFR equation and/or the benefit of deindexing has previously been reported.^12,25,44 Interestingly, deindexing is already recommended by the US Food & Drug Administration to avoid mistakes in drug dosing.⁴⁵ However, even the best performing GFR-estimating equations do not achieve the level of accuracy of direct GFR measurement, so this option is preferable in specific circumstances such as clinical research or certain clinical settings. In this study, eGFR performed relatively poorly among people with preserved kidney function (GFR ≥90 ml/min) in predicting mGFR before surgery or changes in mGFR after surgery (Supplemental Tables 7 and 8). Caution is therefore mandated when applying such equations to that population with an alternative strategy being direct measurement when available. Differentiating development of CKD from reversal of hyperfiltration after surgery will require serial estimations of the eGFR over time. We expect that only when a stable weight has been achieved after bariatric surgery does eGFR change over time regain its ability to reflect changes in kidney function and CKD progression. Of note, the robustness of our observations regardless of the amount of weight lost suggests that they are also applicable to nonsurgical weight loss modalities such as antiobesity medications.

Although the study cohort's generalizability is strengthened by its size relative to previous individual studies, multinational composition, and inclusion of a broad range of presurgery GFR, it has several limitations. Only one study in our analysis had repeated GFR measurements at baseline.^15,46 Future analyses would benefit from multiple measurements in every study participant to increase the precision of baseline GFR estimations. The lack of a control arm in any of the studies included in our analysis raises the possibility of regression to the mean as explaining our findings. Although we cannot definitively rule out this possibility, data from a separate randomized weight loss trial in individuals with overweight and obesity who had baseline normal to hyperfiltrating GFR found that participants randomized to the weight loss intervention had a statistically significant reduction in mGFR after 6 months, while those randomized to a control diet had no change in GFR, and certainly no regression to the mean.⁴⁷ This should ameliorate concern related to this issue, although future studies would benefit from a control arm.

Despite having pooled all available data on mGFR in bariatric surgery patients, the study is still relatively modest in size and heterogenous in participant composition. Therefore, it will be important to confirm our exploratory observations in future analyses.

The individual pooled studies used differing protocols and techniques to measure GFR and other parameters and did contain missing data. However, variability introduced by these differences was minimized by the fact that patients served as their own controls. The overwhelming White and Western makeup of the participants limit generalizability to other races and ethnicities and testing of the race-free eGFR equations. Because bariatric surgery was the model used, caution should be used when applying it to other weight loss modalities such as antiobesity pharmacotherapy.

Future studies with longer postsurgery follow-up that included assessment of biomarkers to help tease out the pathophysiology would add scientific value, as would analyses that could define the effect of medications on changes in mGFR. An ongoing kidney-related randomized clinical trial comparing bariatric surgery with best medical care on progression of diabetic CKD (Randomized Study Comparing Metabolic Surgery With Intensive Medical Therapy to Treat Diabetic Kidney Disease, OBESE-DKD, NCT04626323) will help shed light on the potential differential effects of medical-related versus surgical-related weight loss strategies. Finally, although not all the creatinine and cystatin C measurements available in our pooled analysis were traceable to the international standard, the overall results involving nonindexed GFR estimations are qualitatively similar to findings from a cohort that used the traceable method.¹⁵ The OBESE-DKD trial, which will also be simultaneously measuring and estimating GFR using international standards, will be helpful in confirming our findings.

In conclusion, among individuals undergoing bariatric surgery for weight loss, the magnitude of postsurgical mGFR decline is directly associated with presurgery mGFR and weakly correlated with magnitude of weight loss. In addition, GFR-estimating equations' performance improved when deindexed and used in people with reduced kidney function, with the combined creatinine/cystatin C equations having the best overall performance. These findings have implications for evaluating, treating, and studying kidney disease in individuals with obesity.

Supplementary Material

jasn-37-110-s001.pdf ^{(1.4MB, pdf)}

jasn-37-110-s002.pdf ^{(516.4KB, pdf)}

Disclosures

Disclosure forms, as provided by each author, are available with the online version of the article at http://links.lww.com/JSN/F332.

Author Contributions

Conceptualization: Allon N. Friedman.

Data curation: Alex R. Chang, Ling Ling Chuah, Guillaume A. Favre, Allon N. Friedman, Caroline Grangeon-Chapon, Carel W. le Roux, John C. Lieske, Enrique Morales, Esteban Porrini.

Formal analysis: Avry Chagnac, Allon N. Friedman, Katie A. Lane, Yang Li.

Investigation: Allon N. Friedman.

Methodology: Avry Chagnac, Allon N. Friedman, Katie A. Lane, Yang Li.

Project administration: Allon N. Friedman.

Resources: Allon N. Friedman.

Software: Yang Li.

Supervision: Allon N. Friedman, Yang Li.

Validation: Katie A. Lane, Yang Li.

Writing – original draft: Allon N. Friedman.

Writing – review & editing: Avry Chagnac, Alex R. Chang, Ling Ling Chuah, Guillaume A. Favre, Allon N. Friedman, Caroline Grangeon-Chapon, Katie A. Lane, Carel W. le Roux, Yang Li, John C. Lieske, Enrique Morales, Esteban Porrini.

Funding

None.

Data Availability Statements

Only previously published data were used for this study, and the published citation is included. All citations are included in the manuscript.

Supplemental Material

This article contains the following supplemental material online at http://links.lww.com/JSN/F333.

Supplemental Table 1. Presurgery descriptive statistics.

Supplemental Table 2. Postsurgery descriptive statistics.

Supplemental Table 3. Comparison of presurgery patient characteristics in cohorts that did or did not have a reduction in GFR postsurgery.

Supplemental Table 4. Unadjusted and multivariable linear regression models of change in GFR.

Supplemental Table 5. Performance of eGFR equations versus mGFR in participants stratified by > or < median (33.8 kg) weight loss.

Supplemental Table 6. Performance of eGFR equations compared with mGFR and indexed GFR.

Supplemental Table 7. Changes in mGFR versus eGFR postsurgery when stratified by presurgery mGFR.

Supplemental Table 8. Performance of eGFR equations as stratified by presurgery mGFR.

Supplemental Figure 1. Flow chart with study dates and inclusion and exclusion criteria.

Supplemental Figure 2. Individual changes in GFR between presurgery and postsurgery.