Association Between Obesity and Chronic Kidney Disease: Multivariable Mendelian Randomization Analysis and Observational Data From a Bariatric Surgery Cohort

Obesity is a risk factor for chronic kidney disease (CKD), a rising health care burden worldwide that increases morbidity and premature mortality (1,2). Observational studies suggest that obesity may increase CKD, independent of metabolic risk factors such as type 2 diabetes (T2D) and hypertension (3–11). Unlike observational data, Mendelian randomization (MR) is less prone to confounding and can be used to infer causal associations between an exposure and outcome (12). A prior MR study suggested a causal association between increased BMI and CKD but not increased waist-to-hip ratio (a phenotype more strongly associated with metabolic disease) (13). Multivariable MR studies indicate that increased BMI increases CKD, even after adjusting for hypertension and/or T2D, and has been proposed to be an independent causal factor for CKD (14,15). Dysglycemia and elevated fasting glucose (FG) in the non-T2D range have been implicated in glomerular hyperfiltration and associated with CKD in both observational and MR studies (16–19) and may potentially contribute to obesity-associated CKD. Thus, whether obesity is truly an independent causal factor for CKD after adjusting for dysglycemia and hypertension is not known but may inform future strategies to reduce/prevent CKD, including weight loss (WL)–based interventions.

Although WL can potentially reduce/prevent CKD, the optimal amount of WL needed to reduce CKD is unknown and likely depends on the underlying etiology of obesity-associated CKD. Bariatric surgery (BS) is the most efficacious WL intervention and has shown reductions in estimated glomerular filtration rate (eGFR)–based measures of CKD (CKD_eGFR) (20,21) in observational studies and microalbuminuria (MA) in people with T2D in a randomized control trial (22). Remission rates of T2D (defined as hemoglobin A_1c [HbA_1c] <6.5%) after BS plateau after a WL in the 20–25% range (23). WL above this threshold has been associated with an increased likelihood of normalization of glycemia (HbA_1c <6%) and achievement of normotension (24–26).

In this study, we undertook bidirectional MR to assess potential causal associations between BMI and CKD_eGFR (stage III CKD or worse defined by eGFR <60 mL/min/1.73 m²) (27) and the presence of MA (urinary albumin-to-creatinine ratio [UACR] >25 mg/g in women or >17 mg/g in men) in people of European descent using summary statistics from the largest genome-wide association studies (GWAS). Consistent with prior studies, we found suggestive causal associations between BMI and CKD (14,15) and, therefore, undertook multivariable MR analysis to assess to what extent these potential causal associations are independent of increased FG, T2D, and hypertension. Finally, we assessed the association between WL within 1 year of BS and a 50% decline in eGFR (primary outcome) and CKD hospitalization (secondary outcome).

MR analyses were undertaken in participants of European ancestry using summary statistics from the largest published GWAS. The instruments for glycemic parameters include GWAS of fasting glucose undertaken in people without T2D (i.e., variation in FG in the non-T2D range), as well as GWAS in T2D case and control cohorts (28–34) (Table 1). Informed consent and institutional approval were previously obtained by the individual cohort investigators.

For our primary analysis, we undertook bidirectional MR with CKD_eGFR and MA as outcomes and BMI as exposure with subsequent multivariable MR adjusted for hypertension, T2D, and FG. P < 0.05 was considered significant for each analysis. The recently published Strengthening the Reporting of Observational Studies in Epidemiology using MR reporting guideline (35) has been incorporated in this article (Supplementary File 1).

We also undertook additional bidirectional MR analysis assessing the effect of BMI as exposures on UACR and eGFR creatinine and cystatin C–based eGFR measures.

First, the instrument is robustly associated with the exposure; thus, we used single nucleotide polymorphisms (SNPs) that were associated with the exposure at genome-wide significance (12). Second, there is no horizontal pleiotropy, meaning that the instrument does not influence the outcome via another pathway other than the outcome (12). Finally, the instrument is not influenced by any confounders (12). For bidirectional MR, we used inverse variance weighted (IVW) MR (IVWMR) and additional sensitivity analyses, including MR-Egger, weighted median, weighted mode, and leave-one-out analyses.

IVWMR was analyzed using meta-analysis of the individual Wald ratio for each SNP. By permitting the IVW to have a nonzero intercept, MR-Egger relaxes the assumption of no horizontal pleiotropy and returns an unbiased causal estimate, in the case of horizontal pleiotropy, provided that the horizontal pleiotropic effects are not correlated with the SNP-exposure effects (instrument strength independent of direct effect assumption) (12,36). The median effect of all SNPs in the instrument was used for analysis using weighted median MR, allowing SNPs with a greater effect on the association to be evaluated by weighting the contribution of each SNP by the inverse variance of its association with the outcome (37). Even if only 50% of the SNPs satisfy all three MR assumptions, the method is robust (37). Finally, SNPs were clustered into groups based on similarity of causal effects for weighted mode MR, with the cluster with the largest number of SNPs deriving the causal effect estimate (38). To assess heterogeneity, Cochran Q test was used, while leave-one-out analysis was conducted to assess whether any MR estimate was biased by a single SNP potentially with horizontal pleiotropic effect (12). F statistics were calculated manually for continuous exposures (12,39,40).

Univariable MR was conducted using the TwoSampleMR package in R (RStudio version 1.3.1073 and R version 4.0.3). Linkage disequilibrium pruning was used to select a proxy (r² >0.8) if a SNP was not directly matched from the 1000 Genomes Project. The ggplot2 and metaphor packages in R were used to create plots.

We undertook multivariable IVWMR to assess the effect of BMI on CKD independent of hypertension, T2D, and FG (41). Our data indicate a significant correlation between the effects of BMI-associated SNPs on FG and T2D (Spearman rank correlation ρ = 0.37; S = 527,411; P = 7.9 × 10⁻⁷). The correlation remained significant after removing four SNPs that were genome-wide significant for both traits (Spearman ρ = 0.32; S = 527,412; P = 2.4 × 10⁻⁵) (Supplementary File 5). Given the potential collinearity between the traits, we undertook an analysis adjusted for each glycemic trait individually and in combination with hypertension. Multivariable MR was conducted using the TwoSampleMR, Multivariable MR, and RMultivariable MR packages in R, where the latter two packages assessed heterogeneity via Cochran Q test and strength of the instrument via F statistics (39,41). Plots were generated using plotobject in R.

A total of 456,426 participants from the UK Biobank composed ∼67% of the Genetic Investigation of Anthropometric Traits (GIANT)/UK Biobank GWAS of BMI, 69% of the Diabetes Genetics Replication and Meta-analysis (DIAGRAM)/Genetic Epidemiology Research on Aging (GERA)/UK Biobank GWAS of T2D, and 80% of the Chronic Kidney Disease Genetics (CKDGEN) GWAS of CKD.

We conducted a retrospective cohort study using population-level health care administrative databases in Ontario, Canada. The data sources included the Ontario Bariatric Registry, which provides information on patients assessed at bariatric centers and whether they eventually had surgery. Other data sources included the Discharge Abstracts Database, which includes detailed information on all hospital admissions in Ontario; the Registered Persons Database, with demographic information on all Ontario residents; and the Ontario Laboratory Information System to which all community and most hospital-based laboratories have gradually enrolled since 2006 to contribute laboratory test results. These data sets were linked using unique encoded identifiers and analyzed at ICES. We have received institutional ethics approval for the study.

We identified all people who underwent primary BS between April 2010 and October 2016. Exclusion criteria included residence outside of Ontario, a diagnosis of CKD prior to surgery, eGFR of ≤45 mL/min/1.73 m² at baseline, last eGFR before index date of ≤0.6× baseline eGFR, no weight recorded at either the 6- or 12-month follow-up visit at the bariatric center, death or left Ontario in the first 12 months after surgery, and no eGFR measurements in the year prior to surgery or after surgery. The final sample size was 5,337 (Supplementary File 4).

The secondary outcome was first hospitalization for CKD based on the following hospital diagnostic codes: ICD-9 (4030, 4031, 4039, 4040, 4041, 4049, 585,586, 5888, 5889, 2504) and ICD-10 (E102, E112, E132, E142, I12, I13, N08, N18, N19). Outcome ascertainment started 1 year after surgery and ended 31 March 2020.

Baseline characteristics ascertained for each patient included age, sex, income (defined ecologically based on neighborhood-level median household income, divided into quintiles), rurality of residence (46), and prior history of diabetes (47). In addition, the following comorbidities were ascertained from the Ontario Bariatric Registry: hypertension, dyslipidemia, and vascular disease (coronary artery disease, cerebrovascular disease, or heart failure) (48).

Baseline characteristics of the cohort, stratified by the exposure variable, were compared using ANOVA for continuous variables and χ² tests for categorical variables. Multiple imputation, using five imputed data sets, was used to impute the random missing values for baseline characteristics (49): the number of missing values is included in Table 4. Cox proportional hazards models were fit for each primary and secondary outcome, adjusting for the baseline characteristics and baseline eGFR. The Cox proportional hazards model assumptions were verified with weighted Schoenfeld residuals. Modeling with restricted cubic splines (50) indicated a nonlinear relationship between WL and the primary outcome and linear relationship for the secondary outcome. We therefore assessed outcomes by WL category for the primary outcome. For the secondary outcome, we undertook analyses using WL as a continuous variable. For the primary outcome, patients were censored at the last available eGFR result in the Ontario Laboratory Information System. For the secondary outcome, patients were censored on death or the end of follow-up. We also undertook additional Fine-Gray model analyses with death as a competing risk factor for the secondary outcome (51) but not the primary outcome (as everyone included in the analyses had an eGFR measurement and were alive). Analyses were repeated among the subgroup of patients who had a history of diabetes prior to surgery.

All data are included in the article.

Consistent with previous data, MR suggested that increased BMI increases CKD (b = 0.134; P = 1.64 × 10⁻⁴; odds ratio [OR] 1.143 [95% CI 1.066, 1.226]), which remained significant after adjusting for T2D and hypertension (b = 0.124; P = 0.011; OR 1.132 [1.028, 1.246]). However, the effect was not significant after adjusting for hypertension and FG (b = −0.02; P = 0.72; OR 0.98 [0.87, 1.10) (Table 2, Fig. 1, and Supplementary File 2). Mediation analysis indicates hypertension mediated ∼40% and elevated glucose ∼45% of the effect of obesity and in combination, likely entirely mediated the effect of increased BMI on CKD (Fig. 1).

Univariable MR indicated that increased BMI increases MA (b = 0.255; P = 2.14 × 10⁻⁴; OR 1.291 [95% CI 1.128, 1.478]). This finding was not significant after adjustment for hypertension and T2D (b = 0.141; P = 0.117; OR 1.151 [0.965, 1.372), consistent with prior data. Mediation analysis indicated that hypertension mediates ∼11% and T2D ∼48%, with a combined estimate of ∼52% of the effect of increased BMI on MA. Similarly, multivariable MR suggested that elevated FG is a likely contributor to BMI-associated MA. The causal association between increased BMI and MA was not significant after adjustment for hypertension and FG (b = 0.19; P = 0.08; OR 1.21 [0.98, 1.51]) (Table 2, Fig. 2, and Supplementary File 2). In combination, hypertension and increased FG are estimated to mediate ∼20% of the effect of increased BMI on MA (Fig. 2).

We did not find evidence that CKD defined by eGFR <60 mL/min/1.73 m² (b = 0.009; P = 0.495) or presence of MA (b = 0.003; P = 0.835) increased BMI (Supplementary File 3).

Univariable MR indicated that increased BMI decreases UACR (b = −0.031 ± 0.011; P = 0.004) (Table 3). Multivariable MR after adjustment for hypertension and T2D (b = −0.071; P = 3.9 × 10⁻⁷) and hypertension and FG (b = −0.061; P = 0.001) indicated that BMI decreases UACR. Univariable MR indicated that increased BMI reduces cystatin C–based eGFR (b = −0.05; P = 6.8 × 10⁻¹⁶), which remained significant after adjustment for hypertension and T2D (−0.05; P = 1 × 10⁻⁸) and hypertension and FG (b = −0.03; P = 0.03) (Table 3). Univariable MR suggested that increased BMI did not affect creatinine-based eGFR (b = −0.0004; P = 0.8) (Table 3). We did not find MR evidence that creatinine- or cystatin C–based measures of eGFR or UACR impacts BMI (Supplementary File 3).

The total sample size was 5,337. Baseline demographic features are listed in Table 4. Older age, male sex, rural location, presence of T2D, hypertension, hyperlipidemia, and vascular disease were associated with reduced WL at 1 year after BS, as well as more frequent measurement of creatinine and a consequently shorter interval between baseline measurement and surgery. The median follow-up was 4.63 (interquartile range [IQR] 3.35–5.81) years for the primary outcome and 5.53 (IQR 4.23–6.45) years for the secondary outcome.

There were 172 events among 5,337 patients. In unadjusted models, WL of 20 to <30% (3.6%), 30 to <40% (2.5%), and ≥40% (3.2%) was associated with a reduction in the primary outcome compared with <20% (5.2%; P = 0.02). After adjustment for covariates, WL was associated with a reduction in the primary outcome, which was significant for the 30 to <40% group (hazard ratio [HR] 0.53 [95% CI 0.32, 0.87]) (Fig. 3) but not the 20 to <30% group (HR 0.72 [0.44, 1.2]) or the ≥40% group (HR 0.73 [0.4, 1.3]). Similar trends were seen in the subgroup of patients with diabetes (n = 2,227 [41.7%]; 20–30% group: HR 0.61 [0.36, 1.05]; 30–40% group: HR 0.4 [0.22, 0.72]; ≥40% group: HR 0.58 [0.28, 1.21]) (Fig. 3).

There were 53 hospitalizations for CKD. In unadjusted models, WL of 20 to <30% (1.3%), 30 to <40% (0.7%), and ≥40% (0.4%) was associated with a reduction in the secondary outcome compared with <20% (2.6%) (P = 0.0005). After adjustment for covariates, compared with the <20% WL category, increased WL was associated with a progressive reduction in CKD hospitalization, which was significant for 30 to <40% (HR 0.37 [95% CI 0.17, 0.83]) and ≥40% (HR 0.24 [0.07, 0.89]) WL but not 20 to <30% WL (HR 0.60 [0.29, 1.23]). Fine-Gray analysis with death as a competing outcome yielded similar estimates (20 to <30% group: HR 0.63 [0.30, 1.3]; 30 to <40% group: HR 0.39 [0.17, 0.89]; ≥40% group: HR 0.26 [0.07, 0.97]) (Fig. 3).

Similar trends were seen in the diabetes subgroup (20 to <30% group: HR 0.71 [95% CI 0.32, 1.56]; 30 to <40% group: HR 0.38 [0.16, 0.91]; ≥40% group: HR 0.33 [0.09, 1.25]). Fine-Gray analysis with death as a competing outcome yielded similar estimates (20 to <30% group: HR 0.74 [0.33, 1.66]; 30 to <40% group: HR 0.40 [0.16, 0.97]; ≥40% group: HR 0.35 [0.09, 1.33]) (Fig. 3).

Assessment of WL as a continuous variable showed that after adjustment for covariates, each 1% increase in WL was associated with a further reduction in hospitalization for CKD (HR 0.94 [95% CI 0.91, 0.98). Fine-Gray analysis with death as a competing outcome yielded the same conclusion (HR 0.94 [0.91, 0.98]). Analysis of the patients with diabetes was similar with each 1% additional WL associated with an HR of 0.94 (0.91, 0.98). Fine-Gray analysis with death as a competing outcome yielded similar estimates in patients with diabetes (HR 0.95 [0.90, 0.99]).

Observational and MR studies have suggested that obesity may be an independent cause of CKD based on an increased risk of CKD after adjusting for T2D and hypertension (3–6,13–15). Findings from our MR analyses suggest that obesity is likely not an independent cause of CKD but, rather, the effect is mediated by hypertension and dysglycemia. Notably, elevated FG in the non-T2D range is likely a significant contributor to both eGFR-based measures of CKD and MA in people with obesity. The mediation analysis findings suggest that increased FG is likely the largest contributor to the creatinine-derived eGFR measure of CKD, while T2D is the biggest contributor to MA. Concordant with MR analysis, WL at thresholds previously associated with normoglycemia and normotension is associated with reduced CKD outcomes after BS. WL of 30 to <40% was associated with significantly reduced primary (50% decline in eGFR) and secondary CKD outcome (hospitalization for CKD), while ≥40% WL was associated with a significantly greater reduction in the secondary outcome. Prior data suggests T2D remission rates plateau with 20–25% WL, but further WL is associated with increased normoglycemia and normotension (23–26,52). The reduction in the primary outcome did not reach statistical significance with ≥40% weight loss, which likely reflects the smaller sample size for this category and hence merits some caution in interpretation. Given the nonlinear relationship between WL and the primary outcome, studies with a larger sample size will likely yield more definitive insights into the relationship between the amount of WL and reduction in eGFR. Subgroup analyses in people with T2D yielded generally similar findings for both the primary and secondary outcome. Whether achieving normoglycemia and normotension via means other than WL/BS also reduces obesity-associated CKD outcomes remains to be determined.

Our MR analyses suggests that increased BMI increases eGFR estimated by cystatin C measurement, even after adjustment for hypertension, increased FG, and/or T2D. However, previous work suggests that increased BMI, diabetes, and inflammation can underestimate eGFR based on cystatin C measurement compared with direct measures of eGFR, thus precluding definitive conclusions from these data (53). Consistent with prior data (14), we also report that increased BMI decreases UACR but with no effect on eGFR. We did not have individual-level data on participants, including medication use. Greater use of renoprotective medications in people with a higher BMI may plausibly explain these findings. Illustrating this, recent MR analyses suggest that increased BMI lowers apolipoprotein B likely because of the confounding effects of cholesterol-lowering medication, an effect not seen when analyzing people not taking statins (54).

The strengths of this study include MR analyses with the largest available GWAS in populations of European descent along with observational data from patients undergoing BS. However, this study has several limitations. We used a creatinine-based diagnosis of CKD, which is an indirect measure of renal function and can be affected by muscle mass. Similarly, cystatin C–based measures of eGFR can be underestimated with increased BMI, diabetes, and inflammation (53). The retrospective observational nature of the BS cohort with potential uncaptured confounders is a major limitation; however, these limitations are less likely with MR analyses. There was a >50% sample overlap between MR population cohorts, which can overestimate the effect size when weak instrument bias is present, although this effect is attenuated by the strength of the instruments (40). The MR analyses was undertaken in European populations and may not translate to other ethnic groups. The greater availability of genetic data from other ethnicities will enable similar analyses in these cohorts. Although we do not have data on ethnicity in our BS cohort, previous data suggests that ∼20% of patients undergoing BS are of non-European ancestry (55). Lack of individual-level data, including medications, is a limitation of our analyses. We also do not have details on the type of BS performed, MA/proteinuria, and longer-term WL.

In summary, the MR analyses suggest that obesity is likely not an independent causal factor for CKD, with its deleterious renal effects mediated by dysglycemia and hypertension. These data underscore the likely causal role of hyperglycemia below the T2D threshold to obesity-associated CKD. WL at or above thresholds known to improve/remit these cardiometabolic parameters are associated with reduced CKD outcomes after BS, and these findings await confirmation with well-powered prospective studies. Whether achieving normoglycemia and normotension in the absence of WL reduces CKD also awaits independent confirmation.

Acknowledgments. The authors thank Lei Sun, PhD (Dalla Lan School of Public Health and Department of Statistical Sciences, University of Toronto), for invaluable guidance in statistical analysis and interpretation. The Ontario Bariatric Registry (OBR) collects baseline, treatment, and postoperative data of patients undergoing bariatric treatment at Bariatric Centres of Excellence across Ontario. The OBR is managed by the Centre for Surgical Invention and Innovation in collaboration with the Population Health Research Institute and the Ministry of Health and Long-Term Care.

Funding. S.D. receives funding from the Canadian Institute for Health Information, the Heart and Stroke Foundation of Canada, Diabetes Canada, and the Banting & Best Diabetes Centre (The DH Gales Family Charitable Foundation new investigator award and a Reuben and Helene Dennis scholar in diabetes research). This study was supported by ICES (formerly the Institute for Clinical Evaluative Sciences), which is funded by an annual grant from the Ontario Ministry of Health.

ICES is an independent, nonprofit research institute whose legal status under Ontario’s health information privacy law allows it to collect and analyze health care and demographic data, without consent, for health system evaluation and improvement. Parts of this material are based on data and/or information compiled and provided by the Canadian Institute for Health Information and the Ontario Ministry of Health. The analyses, conclusions, opinions, and statements expressed herein are solely those of the authors and do not reflect those of the funding or data sources; no endorsement is intended or should be inferred.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. A.N., R.K., Y.G., A.M., B.R.S., A.D.P., and S.D. designed the study. A.N., R.K., A.M., and S.D. wrote the manuscript. S.D., A.D.P., A.N., R.K., Y.G., B.R.S., D.R., M.A., B.L., D.Z.I.C., and M.E.F. reviewed the data and edited the manuscript. S.D. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.