Excess body weight is associated with an increased risk of incident CKD, kidney function decline, and development of ESKD in the general population.^123–4 Despite this evident and strong epidemiologic link, the underlying biologic mechanisms still remain unclear. Metabolic factors linked to excess body weight, such as insulin resistance, hypertension, hyperuricemia, and dyslipidemia, may act as causal intermediates in the relationship between overweight/obesity and renal risk, but their relative contributions in this relationship are unknown.^5678–9

Insulin resistance, hypertension, hyperuricemia, and hypercholesterolemia are established metabolic risk factors that have been associated with risk of various diseases, such as cardiovascular disease, CKD, and ESKD.^{5,1011121314151617181920–21} Overweight and obesity are in general associated with a worse profile in these metabolic factors. However, some overweight/obese individuals show a lower burden of metabolic abnormalities, often referred to as overweight/obese metabolically healthy individuals. In contrast, a fraction of individuals present with an unhealthy metabolic profile even among normal weight individuals. Studying how overweight/obesity in combination with the metabolic health status affects the risk of kidney disease is a first step in better understanding the complex relationships between excess body weight, metabolic factors, and kidney disease.

However, studies about the relationship between overweight and obesity, metabolic health, and the risk of ESKD are scarce and results conflicting. In the Reasons for Geographic and Racial Differences in Stroke study, a higher body mass index (BMI) was associated with a lower risk of ESKD in metabolically healthy individuals, and a twofold higher ESKD risk was found in metabolically unhealthy participants irrespective of their weight status.²² Another US cohort study found no association between metabolically healthy obesity and ESKD either, but confirmed the increased risk in metabolically unhealthy individuals.²³

These two studies suggest that the effect of BMI on ESKD might be mediated in part through metabolic factors. Indeed, a recent Mendelian randomization study with UK biobank data provided evidence for a largely causal association between adiposity (measured as BMI) and CKD, and showed that BP and diabetes explained about two-thirds of this causal relationship.²⁴ Moreover, in a previous Austrian population-based study, we have shown that the triglyceride-glucose index (TyG index; a recently proposed indicator of insulin resistance²⁵) was significantly associated with incident ESKD and mediated approximately 40% of the relationship of BMI with ESKD.¹³ However, it is still unclear which overweight- and obesity-related metabolic factors contribute most to the association of BMI with ESKD risk.

Using data of >100,000 participants of the Vorarlberg Health Monitoring and Promotion Program (VHM&PP), the aims of this study were (1) to evaluate the increase in ESKD risk of a metabolically unhealthy profile compared with a healthy metabolic profile; and (2) to quantify the contributions of insulin resistance, hypertension, hyperuricemia, and hypercholesterolemia as mediators in the relationship between BMI and ESKD risk, both individually and in combination.

Methods

Data Source and Study Population

The VHM&PP, conducted by the Agency for Preventive and Social Medicine, is a large population-based risk factor surveillance program in Vorarlberg, the westernmost province of Austria. In the course of the program, which was initiated in 1985, every adult residing in Vorarlberg was invited to participate, and a screening examination was performed by local general practitioners according to a standard protocol. Between January of 1985 and June of 2005, 99,894 female and 85,473 male residents aged older than 18 years (more than half of the adult population of Vorarlberg) were enrolled in the VHM&PP. During the screening examination, height and weight (in light clothing) were measured by medical staff, smoking status was inquired, and a blood draw was taken. Occupational status (blue collar, white collar, or self-employed) was determined and used as a surrogate measure of socioeconomic status. Participants who were retired at baseline were classified according to their former occupation, and individuals not employed outside the home were classified according to their spouse’s or partner’s occupation. A more detailed description of the program and examination procedures is reported elsewhere.^3,26,27

Because an overnight fast was part of the protocol only from 1988 onwards, we excluded 8073 participants (4.4%) who did not have an examination with a blood draw in fasting status. Moreover, as uric acid (UA) was routinely determined only in women above the age of 50 years at screening, but not measured in women at younger ages, we excluded women under 50 years of age. Furthermore, participants were excluded with incomplete exposure, covariate, or mediator variables; with a baseline BMI<20 kg/m² or a baseline age>75 years; and with a follow-up time <2 years, resulting in a final analysis population of 100,269 individuals (75,282 men and 24,987 women). Detailed numbers are provided in the flow chart in Supplemental Figure 1. Of note, the decision to apply these exclusion criteria was made a priori without any data exploration. In order to avoid nonlinear effects due to a J-shaped association between BMI and ESKD when including underweight individuals as shown by Herrington et al.,²⁸ we restricted our analysis to individuals with a BMI of 20 kg/m² and higher. The first 2 years of follow-up were excluded to reduce the possibility of reverse causation.

Outcome data were obtained by linking the VHM&PP database with the Austrian Dialysis and Transplant Registry (OEDTR) and the National Mortality Registry. The OEDTR collects data provided by the Austrian dialysis and transplant centers on all patients undergoing chronic RRT (hemodialysis, peritoneal dialysis, kidney transplantation) in Austria since 1964 with an almost complete follow-up.²⁹

All study procedures were performed in accordance with the Declaration of Helsinki and relevant guidelines. Institutional review board approval for the study was obtained from the Ethics Committee of the State of Vorarlberg. Written, informed consent was obtained from all VHM&PP participants, and all patients registered in the OEDTR had signed a declaration of consent to permit their data to be transferred to the registry.

STrengthening the Reporting of Observational studies in Epidemiology guidelines were followed for the preparation of this article.³⁰

Definitions of Exposure, Mediators, and Outcome

BMI was calculated from height and weight records as weight (kg)/height (m)2, and categorized into normal weight (BMI 20 to <25), overweight (BMI 25 to <30), and obesity (BMI≥30) according to the World Health Organization definition. The TyG index was calculated as ln(fasting triglycerides [mg/dl] × fasting blood glucose [mg/dl]/2),²⁵ and mean arterial pressure (MAP) as diastolic BP+1/3×(systolic BP−diastolic BP). The outcome ESKD was defined as initiation of RRT, either dialysis or kidney transplantation. Follow-up began 2 years after the baseline health examination and ended at the diagnosis of ESKD, or at the occurrence of the censoring events death or end of the observation period (December 31, 2019), whichever occurred first.

The four metabolic factors TyG index, MAP, UA, and total cholesterol (TC) were chosen a priori without any data exploration. As shown previously, the TyG index is a simple and clinically useful surrogate marker for insulin resistance, superior to using glucose and triglyceride values separately.²⁵ Its validity is similar to the homeostatic model assessment insulin resistance (HOMA-IR) index.³¹ HOMA-IR requires a direct measurement of insulin which was not available in our cohort. MAP combines information from both systolic and diastolic BP into one variable, avoiding collinearity problems that would arise when using the two variables jointly in the statistical model. Mediators were dichotomized into low/normal versus high using the 75th percentile as the cut-off value. Further, we constructed a metabolic score (MS) as the sum of the z-transformed values of these four factors. This score was then dichotomized at the value 1, resulting in a group of “metabolically healthy” individuals (i.e., MS<1; 67.6% of participants), and a group of “metabolically unhealthy” individuals (i.e., MS≥1; 32.4% of participants), an approach introduced previously.³² Examples of combinations of values in metabolic factors together with the resulting MSs are shown in Supplemental Table 1.

Statistical Analyses

Only exposure, mediator, and covariate data of the first health examination were included in the analysis. ESKD incidence rates per 100,000 person-years were calculated stratified by BMI category and metabolic health status, and hazard ratios (HRs) along with 95% confidence intervals (95% CIs) were calculated from Cox proportional hazards models adjusted for baseline age, sex, smoking status, and socioeconomic status.

For quantifying mediation of the association of BMI with ESKD through the four mediators TyG index, MAP, UA, and TC, we applied the product-of-coefficients method.^3334–35 Of note, because of the observational nature of our data, we prefer the term association instead of effect. In brief, two sets of regression models were fit to the data, one set of models for the mediators, and one model for the outcome. Specifically, the four mediators were regressed on the exposure (BMI) in separate linear models each, and the outcome (ESKD) was regressed on the exposure (BMI) and the four mediators (TyG index, MAP, UA, TC) in Cox proportional hazards models with time from baseline examination as the underlying time variable. Models were adjusted for age, sex, smoking status, and socioeconomic status, without inclusion of interaction terms. Both linearity and proportional hazards assumptions were assessed visually, by plotting penalized splines and scaled Schoenfeld residuals, respectively; no major deviations were detected.

From the mediator models, we obtained regression coefficients for the association of BMI with the respective mediator (pathways A₁, A₂, A₃, and A₄). From the outcome model, we obtained (1) regression coefficients (on the log[HR] scale) for the association of BMI with ESKD, independent of the mediators (pathway C); and (2) independent associations (on the log[HR] scale) between each of the mediators and ESKD (pathways B₁, B₂, B₃, and B₄). The indirect association of BMI with ESKD mediated through the ith mediator is defined as the product A_i×B_i. The direct association of BMI with ESKD is given by the regression coefficient C. The sum of the four indirect associations (i.e., joint indirect association) and the direct association gives the total association of BMI with ESKD. All of these associations were converted to the HR scale by exponentiation.

In case of a rare outcome (ESKD incidence of 0.5% in our analysis), and under the assumptions (1) of causal relationships as depicted in the directed acyclic graph in Figure 1, in particular that age, sex, smoking status, and socioeconomic status account for the majority of confounding; and (2) that the models are correctly specified, these associations have a direct causal interpretation.^3536–37 We can think of the indirect association between BMI and ESKD through the ith mediator as the effect of BMI that is due to mediation through the ith mediator, and of the direct association of BMI with ESKD as the remaining effect of BMI not explained through the set of mediators. If one mediator (M₁) causally affects another mediator (M₂), the interpretation of an indirect effect has to be amended, in so far as the indirect effect through M₁ captures the exposure effect mediated through M₁ directly, but not the causal descendant M₂, and the indirect effect through M₂ captures the exposure effect through M₂, with and without the causal antecedent M₁. This needs to be considered when interpreting the indirect associations through the TyG index and MAP, because there is evidence for insulin resistance causally affecting BP^11,38,39 (thus the arrow from insulin resistance to hypertension in Figure 1); see also the discussion part of this manuscript. Evidence for other causal relationships between the four mediators is weak (indicated by the absence of arrows in Figure 1),^{40414243–44} and therefore the interpretation of indirect associations through UA and through TC is more straightforward. Exact definitions of direct and indirect associations in terms of counterfactuals (the full technical terms are interventional direct and interventional indirect effects) and more detailed descriptions can be found elsewhere.^35,37

The proportions of the various derived associations relative to the total association between BMI and risk of ESKD were calculated on the log-transformed HR scale, because HRs are additive on this scale. We used bootstrapping with 5000 samples to calculate 95% CIs for all parameters of interest. Stratified analyses by sex and primary underlying kidney disease were performed as well. In addition, a sensitivity analysis without the exclusion of the first 2 years of follow-up was done.

Finally, we compared associations of BMI with ESKD risk (conditioning on the same covariates as listed above), both adjusted and unadjusted for the mediators, also known as the difference method for mediation analysis.^34,45

Statistical analyses were conducted in R, version 3.5.1.⁴⁶ All statistical tests were two-sided at a significance level of 0.05.

Results

Characteristics of the Study Population

The analyzed cohort included 100,269 participants (24.9% female, mean baseline age 46.4 years, SD 14.6) initially free of ESKD, of whom 463 (0.5%) developed ESKD over a mean follow-up of 23.1 years (Table 1). The primary underlying kidney disease was classified as diabetic kidney disease in 24.4% (n=113) of ESKD cases, vascular nephropathy in 32.2% (n=149) of cases, and other disease in 43.4% (n=201) of cases. Mean time from baseline until ESKD was 15.5 (SD 7.8) years. Mean BMI (kg/m²) at baseline was 26.1 (SD 3.8); 41.2% of the participants were overweight, and 13.9% obese. Baseline characteristics stratified by sex and BMI categories are given in Table 1. Women were markedly older than men (mean [SD] 60.1 [6.8] versus 41.8 [13.6]) due to the exclusion of women younger than 50 years. Distribution of age, laboratory parameters, and other covariates differed across BMI categories as expected.

BMI Categories, Metabolic Health Status, and ESKD Incidence

An unhealthy metabolic profile (prevalence in the total population: 32.4%) was associated with an increased ESKD risk (multivariably adjusted HR, 3.57; 95% CI, 2.89 to 4.40), independent of BMI (Table 2). Likewise, ESKD risk across the BMI categories of normal weight, overweight, and obesity strongly depended on the metabolic health status. Overweight and obesity, paired with a metabolically healthy profile, were associated with an increase in ESKD risk compared with the reference category of metabolically healthy normal weight (adjusted HR_{overweight, healthy}, 1.11; 95% CI, 0.78 to 1.60; HR_{obese, healthy}, 1.98; 95% CI, 1.20 to 3.25); however, to a much lesser extent than when paired with a metabolically unhealthy profile (HR_{overweight, unhealthy}, 3.93; 95% CI, 2.91 to 5.29; HR_{obese, unhealthy}, 5.75; 95% CI, 4.17 to 7.93). A metabolically unhealthy profile was associated with a substantial increase in ESKD risk even in the presence of normal weight (HR_{normal weight, unhealthy}, 4.50; 95% CI, 3.20 to 6.31). Similar patterns were also observed for low/normal (<75th percentile) versus high (≥75th percentile) values of the single factors TyG index, MAP, UA, and, attenuated, for TC (Table 2).

The likelihood of a metabolically unhealthy profile increased over the BMI categories (the prevalence of a metabolically unhealthy profile was 16.8% in normal weight individuals, 40.4% in overweight individuals, and 59.5% in obese individuals), and consequently ESKD incidence rates were lowest in normal weight participants, with a clear increase in overweight and obese individuals (12.9; 95% CI, 10.8 to 15.2, ESKD incident cases per 100,000 person-years; versus 21.8; 95% CI, 18.9 to 24.9; versus 39.7; 95% CI, 32.9 to 47.6; Table 2).

Despite the differences in baseline age, patterns were similar in men and in women, although the analysis of women was limited by small ESKD incidence numbers. Detailed information on sex-specific ESKD incidence rates is provided in Supplemental Tables 2 and 3.

Associations of Mediators with BMI, with ESKD Risk, and among Themselves

BMI was strongly associated with all mediators. Adjusting for age, sex, smoking status, and socioeconomic status, a BMI increase of 5 kg/m² was associated with an increase of 0.41 (95% CI, 0.40 to 0.42) SDs in TyG index, 0.37 (95% CI, 0.36 to 0.38) SDs in MAP, 0.32 (95% CI, 0.31 to 0.32) SDs in UA, and 0.12 (95% CI, 0.11 to 0.12) SDs in TC (Figure 2). The four mediators were mildly to moderately correlated among each other, even after adjusting for BMI, age, sex, smoking status, and socioeconomic status (partial Pearson correlation: ρ=0.088 between TyG index and MAP, ρ=0.168 between TyG index and UA, ρ=0.359 between TyG index and TC, ρ=0.099 between MAP and UA, ρ=0.076 between MAP and TC, and ρ=0.133 between UA and TC; all P<0.001). Regarding the independent contributions of the single mediators to the risk of ESKD, a 1-SD increase in TyG index was associated with an increase in the risk of ESKD by 44% (HR, 1.44; 95% CI, 1.30 to 1.59), a 1-SD increase in MAP was associated with a risk increase by 53% (HR, 1.53; 95% CI, 1.39 to 1.67), and a 1-SD increase in UA with a risk increase of 54% (HR, 1.54; 95% CI, 1.39 to 1.70); in contrast, TC was not significantly associated with ESKD risk (HR, 1.07; 95% CI, 0.96 to 1.17, per 1-SD increase) (Figure 2).

BMI and Risk of ESKD Mediated through TyG Index, MAP, UA, and TC

A 5-kg/m² higher BMI was associated with an increase in ESKD risk by 57% (HR_{total association}, 1.57; 95% CI, 1.38 to 1.77) (Table 3). This association could be explained by the joint contribution of the four mediators TyG index, MAP, UA, and TC (proportion jointly mediated, 99%; 95% CI, 76% to 140%; HR_{joint indirect association}, 1.56; 95% CI, 1.49 to 1.64), and the remaining direct association was not statistically significant (HR_{direct association}, 1.01; 95% CI, 0.88 to 1.14). Regarding the contributions of the single mediators, the pathways via TyG index (proportion mediated, 33%; 95% CI, 22% to 49%), MAP (34%; 95% CI, 24% to 50%), and UA (30%; 95% CI, 21% to 45%) contributed to the total association of BMI with ESKD risk in comparable amounts, whereas the pathway via TC (2%; 95% CI, −1% to 4%) did not significantly contribute (Table 3). The negligible contribution of TC was due to the combination of both a weaker association of BMI with TC and a much weaker, nonsignificant independent contribution of TC to risk of ESKD, compared with the other mediators (Figure 2). Subgroup analyses revealed similar patterns across men and women (Table 4). Stratified analyses revealed distinct patterns of mediation depending on the underlying kidney disease (Supplemental Table 4).

Investigating BMI categories, the total association of BMI with ESKD risk (expressed as HRs) was 1.25 (95% CI, 0.99 to 1.56) for overweight versus normal weight, and increased to 2.35 (95% CI, 1.81 to 3.01) for the obesity group. For both groups, the total association was fully (i.e., nonsignificant direct association) mediated through MAP, TyG index, and UA, but not TC (HR_{direct association}, 0.80; 95% CI, 0.63 to 1.00, for overweight; and 0.89; 95% CI, 0.66 to 1.18, for obesity, respectively) (Table 3).

The finding of this mediation model, namely full mediation of the association of BMI with ESKD by the metabolic factors under consideration, was confirmed in separate analyses using the traditional difference method for mediation analysis (Supplemental Table 5). The difference method does not allow a further decomposition into single components, though. The inclusion of the first 2 years of follow-up did not substantially alter the results (Supplemental Table 6).

Discussion

In this large, observational, population-based cohort study we aimed to quantify the contributions of several different overweight- and obesity-related metabolic factors to the association of BMI with ESKD risk. In a first step, we showed that a metabolically unhealthy profile was associated with a markedly increased ESKD risk, not only in overweight and obese, but also in normal weight individuals. This means that ESKD risk is not only associated with BMI, but, importantly, also with the metabolic health status, and suggests that metabolic factors might be important mediators in the relationship between BMI and ESKD. In a second step, we quantified the independent contributions of each of the four metabolic factors TyG index, MAP, UA, and TC as mediators to the BMI-ESKD relationship. We found that the three mediators TyG index, MAP, and UA were each significantly and independently associated with the risk of ESKD. Each of them independently mediated around one-third of the total association of BMI with ESKD risk, and thus TyG index, MAP, and UA mediate the BMI-ESKD relationship to a high degree. In contrast, the mediating role of TC was negligible. No sex differences were detected, and thus these findings apply equally to men and women.

The association between excess body weight and the development of CKD and ESKD is well supported by global data in millions of individuals of variable baseline risk and age including adolescents.^2,4,47 Our finding of a 57% higher risk of ESKD with each 5-kg/m² higher BMI in a middle-aged low-risk general population confirms the relevance of this preventable risk factor. This observed risk increase is consistent with the strength of association between BMI and ESKD as observed in UK data on >1.4 million people with similar baseline characteristics as in our study.²⁸ This fact becomes even more important considering the growing number of obese individuals worldwide, with the prevalence of obesity having more than doubled since 1980 in >70 countries as reported by the Global Burden of Disease study.⁴⁸

Far less is known about the causal pathways connecting excess body weight and ESKD. Using data of 281,228 genotyped UK Biobank participants in a Mendelian randomization study, Zhu et al. found that the mediators diabetes mellitus and BP in combination explained about two-thirds of the association between BMI and CKD, a portion underestimated using conventional epidemiologic analysis.²⁴ In our study, the two factors TyG index and MAP mediated 67% of the total association of BMI with ESKD, a similar number to the one reported by Zhu et al. despite several important differences in study design and analysis (prospective versus cross-sectional design, differences in baseline age, use of TyG index versus diabetes mellitus, ESKD as the outcome versus a broader CKD definition). Despite these differences, findings of the two studies are similar, indicating robustness and generalizability of the results. A further strength of our study is the addition of the potential mediators UA and TC which were not investigated by Zhu et al.

Hypercholesterolemia did not contribute as a relevant mediator to the BMI-ESKD association due to a weaker association of BMI with TC and a much weaker independent contribution of TC to the risk of ESKD, compared with the other three studied mediators. In line with this finding, Hsu et al. and Kastarinen et al. did not observe an association between TC levels and long-term risk for ESKD over a 25-year period^18,49; and also in a randomized, controlled trial simvastatin plus ezetimibe, although lowering cholesterol levels, did not lead to a reduction in ESKD incidence compared with the placebo group.⁵⁰ A time-window analysis exploring the association between metabolic factors and ESKD showed that the effect of hypercholesterolemia was highest in the first 5 years, diminishing thereafter and lacking after ≥15 years of follow-up.⁵¹ Hypercholesterolemia in the years before ESKD may be caused by progressive CKD and nephrotic proteinuria, for example, in diabetic kidney disease, and may therefore be considered a consequence rather than a cause of progressive kidney disease. This view is supported by a report from the Chronic Renal Insufficiency Cohort study showing that TC levels are not associated with CKD progression over a period of 4 years.⁵²

Insulin resistance has been found to be associated with CKD independently of diabetes.^53,54 Mechanistically, insulin resistance is assumed to cause glomerular hyperfiltration, increased sodium retention, tubular dysfunction, and renal tissue inflammation and fibrosis.^6,8,55 As an indicator for insulin resistance, we used the TyG index, which has been proven to correlate well with the euglycemic-hyperinsulinemic clamp test and HOMA-IR index,³¹ and was recently found to be significantly associated with the risk of ESKD.¹³

Hypertension is another established and important risk factor of ESKD.^19,56 The relationship between body weight and BP and the development of hypertension with excess body weight is firmly supported by epidemiologic and preclinical human and animal experimental evidence.^9,57,58 Among the several different biologic mechanisms between obesity and hypertension, insulin resistance itself is a contributing mediator.^43,59 This has to be considered when interpreting the indirect associations mediated through the TyG index and MAP in our study. Specifically, the component of the association of BMI with ESKD risk that is mediated via the “double-mediator” pathway BMI→TyG index→MAP→ESKD is part of the indirect association mediated through MAP, but not the indirect association through the TyG index. If anything, the implicit involvement of the TyG index in part of the indirect association through MAP renders the role of the TyG index in the relationship between BMI and ESKD even more prominent. Indeed, we previously demonstrated that the TyG index as a single mediator mediated 42% of the BMI-ESKD relationship,¹³ whereas the contribution independent of the other metabolic factors was 33% in this study.

A new finding, which to our knowledge has not been described previously, is the significant mediating contribution of UA on the association of BMI with ESKD. Hyperuricemia is independently associated with new-onset CKD as shown in a large meta-analysis of 13 individual studies, with an increasing effect estimate with longer follow-up.⁵ In contrast, the effect of UA on the progression of CKD remains controversial.^{60616263–64} Randomized, controlled trials showed no beneficial effect of UA-lowering therapy on the progression of CKD in patients with type 1 diabetes and diabetic kidney disease⁶⁵ or advanced CKD stages 3 and 4.⁶⁶ However, these studies did not address UA-lowering therapy in obese patients with hyperuricemia and normal kidney function or early stages of CKD. Therefore, UA could play a role as a causal risk factor for kidney disease onset rather than disease progression. Regarding ESKD, Hsu et al. reported increased serum UA levels to be an independent risk factor for ESKD.¹⁸ However, far less is known about the role of UA in the association between excess body weight and ESKD. Our finding of a roughly equal amount of mediation through UA as compared with the TyG index and MAP emphasizes that besides the established mediators hypertension and insulin resistance (or diabetes), hyperuricemia should be in the focus of future studies investigating the effect of overweight and obesity on kidney function as well.

Our study suggests that in individuals with overweight and obesity focused interventions, either pharmaceutic or behavioral, on the risk factors insulin resistance, hypertension, and hyperuricemia might carry the potential to strongly decrease the renal risk. In this regard, sodium-glucose cotransporter 2 inhibitors, which have recently been shown to significantly slow the progression of CKD,⁶⁷ may be a promising therapeutic option, because they lower BP and serum UA, improve insulin sensitivity, and decrease body weight and obesity-related glomerular hyperfiltration.^68,69

The outstanding importance of maintaining a healthy metabolic profile, not only in overweight and obese but also in normal weight individuals, is supported by our analysis comparing the association of metabolic health with ESKD risk, both within BMI categories and overall. Within all categories, including normal weight, a metabolically unhealthy profile increased ESKD risk markedly (by a factor of 3 or more), whereas the risk increase across BMI categories within categories of metabolic health, in particular in metabolically unhealthy individuals, was comparatively moderate (2.0- and 1.3-times higher risk for obese versus normal weight individuals in the metabolically healthy and unhealthy groups, respectively). In a recent US cohort study, metabolically unhealthy obesity, but not metabolically healthy obesity, was associated with an increased risk of ESKD over a median follow-up of 5 years compared with healthy normal weight.²³ This is similar to our finding, and although in our study also metabolically healthy obesity was associated (albeit to a lesser degree) with ESKD, the null finding in the US cohort study is probably due to insufficient statistical power (only six ESKD cases in metabolically healthy obese individuals). Further, the two studies differ regarding study population characteristics, length of follow-up, and definition of metabolic health. In the absence of metabolic abnormalities, a long follow-up time might be necessary to adequately capture associations between metabolically healthy overweight/obesity and ESKD. However, metabolically healthy overweight and obesity may represent a temporary intermediate low-risk state and may later be burdened by a worsening metabolic risk factor profile, which eliminates the primary lower risk.

Some limitations of our study warrant mention. First, although the sample size is large, the statistical power and the precision of the estimates is limited by the number of outcomes (n=463). ESKD is a rare event; however, our observed cumulative ESKD incidence of about 0.5% in a middle-aged low-risk general population is in line with the estimated age-related risk of RRT in Europe and the 0.4% ESKD incidence in a large meta-analysis including 39 general population cohorts.^4,70 Second, excess body weight was determined by BMI only. Information on waist circumference or waist-to-hip ratio, which allow a better characterization of body fat distribution,^71,72 was not available in the VHM&PP cohort. Third, the TyG index is only an indicator of insulin resistance; however, no direct measurement of insulin was available. Fourth, data on baseline kidney function, physical activity, other lifestyle factors, pharmacologic therapy, and other potential confounders were not available, which might introduce confounding bias to our results. However, the long time period of 15.5 years between baseline and ESKD and the exclusion of the first 2 years of follow-up in our analysis make it unlikely that a significant number of participants already had a relevant kidney disease at baseline. Lastly, our results are on the basis of a middle-aged, low-risk, population-based White cohort, and therefore it is uncertain how far findings can be generalized to other ages and ethnic groups or to patients with higher comorbidity burden.

Strengths of our study include the length of follow-up, the accurate assessment of the outcome ESKD using the OEDTR, and the application of a state-of-the-art mediation analysis tool, allowing a mathematically consistent decomposition of the total association of BMI with ESKD risk not only into a direct and indirect association, but also further splitting up the joint indirect association into independent contributions of the single mediators. This is not possible with traditional methods for mediation analysis, such as the difference method.^34,45

In conclusion, we found that the total association of excess body weight with the risk of ESKD is to a high degree mediated through insulin resistance, hypertension, and hyperuricemia, with a similar contribution of around one-third for each of these three factors. In contrast, hypercholesterolemia was not a relevant mediator. Apart from promoting body weight reduction to reach normal weight, focused intervention on the three targets insulin resistance, hypertension, and hyperuricemia not only in overweight and obese but also in normal weight individuals might lower renal risk. This should ideally be tested in controlled trials in the future.

Disclosures

K. Lhotta reports Honoraria: Amgen, AstraZeneca, Baxter, Bayer, Boehringer-Ingelheim, Vifor. E. Zitt reports Honoraria: Boehringer-Ingelheim, Otsuka Pharmaceutical GmbH, Vifor Pharma; Advisory or Leadership Role: AMGEN, AstraZeneca; and Other Interests or Relationships: Board Member of the Austrian Society of Nephrology, Member of the European Renal Association, Member of the European Renal Association Chronic Kidney Disease-Mineral and Bone Disorder Working Group, and Medical Research Director Agency for Preventive and Social Medicine. All remaining authors have nothing to disclose.

Funding

None.