, Introduction
Obesity is associated with an increased risk of various diseases including hypertension, diabetes mellitus (DM), and cardiovascular disease [1,2]. Since these comorbidities have been recognized as risk factors for chronic kidney disease (CKD) [3,4], the prevalence of obesity among patients with CKD has been rapidly increasing, contributing to a high overall prevalence [5]. Accordingly, there has been considerable interest in the associations between obesity and various adverse clinical outcomes in the CKD population.
Although an elevated body mass index (BMI) is associated with increased mortality in the general population [6], the phenomenon termed the ‘obesity paradox’ or ‘reverse epidemiology’—where higher BMI appears to improve survival—has been observed in patients with CKD [7–9]. However, previous studies investigating the association between BMI and initiation of kidney replacement therapy (KRT) in patients with CKD have reported inconsistent results across diverse ethnic groups [4,8–13].
Furthermore, although the majority of previous studies used baseline BMI values for analysis, there is a paucity of consensus regarding the predictability of a single BMI measurement for adverse outcomes in patients with CKD, given the substantial weight fluctuations often experienced in this population [14,15]. To address this limitation, the marginal structural Cox model (Cox-MSM) analysis—a robust statistical approach—enables estimation of the causal effect of a time-varying exposure with regarding time-varying covariates, which may simultaneously function as both confounders and intermediate variables within a potential outcome framework [16]. Despite the advantages of Cox-MSM, no study has yet applied this approach to investigate the association between BMI and risk of KRT initiation in non-dialysis-dependent CKD (NDD-CKD) patients.
Therefore, we aimed to investigate the association between time-updated BMI and initiation of KRT in patients with NDD-CKD in Korea using Cox-MSM to account for fluctuations and interactions of BMI and confounders over time. We hypothesized that a higher time-updated BMI would be associated with increased risk of KRT initiation in patients with NDD-CKD.
Methods
Study design and participants
The KoreaN Cohort Study for Outcomes in Patients With Chronic Kidney Disease (KNOW-CKD; NCT01630486 at http://clinicaltrials.gov) is a large-scale prospective cohort study involving nine tertiary care hospitals in Korea, which enrolled 2,238 patients aged 20–75 years with pre-dialysis CKD stages from G1 to G5 between 2011 and 2016 [17]. CKD stages were defined based on the estimated glomerular filtration rate (eGFR) calculated by the CKD-EPI (Chronic Kidney Disease Epidemiology Collaboration) equation [18]. Patients with a previous history of dialysis, any organ transplantation, heart failure (New York Heart Association class 3 or 4), liver cirrhosis (Child-Pugh class B or C), or cancer; currently pregnant women; patients with single kidney due to trauma or nephrectomy; and patients who were unable or unwilling to provide written consent were excluded from this study. Among the 2,238 patients enrolled, 102 with missing values for baseline demographic, anthropometric, and laboratory data were excluded. Thus, a total of 2,136 patients were included in this study. All patients were followed until March 2023.
This study was approved by the Institutional Review Board of each participating clinical center (Seoul National University Hospital [No. 1104–089-359; May 25, 2011], Seoul National University Bundang Hospital [No. B-1106/129–008; August 24, 2011], Severance Hospital [No. 4–2011-0163; June 2, 2011], Kangbuk Samsung Medical Center [No. 2011–01-076; June 16, 2012], The Catholic University of Korea, Seoul St. Mary’s Hospital [No. KC11OIMI0441; June 30, 2011], Gachon University Gil Hospital [No. GIRBA2553; August 8, 2011], Eulji General Hospital [No. 201105–01; June 10, 2011], Chonnam National University Hospital [No. CNUH-2011-092; July 5, 2011], and Inje University Busan Paik Hospital [No. 11–091; July 26, 2011]) and conducted in accordance with the principles of the Declaration of Helsinki. Written informed consent was obtained from all subjects.
Study outcome
The primary outcome was initiation of KRT including dialysis or kidney transplantation during the follow-up period. All patients were under close observation, and the events were reported by each participating center. All events were adjudicated by a committee. Patients who were lost to follow-up were censored at the date of their final visit.
Study exposure
The main exposure of interest was BMI, calculated as weight (kg) divided by height squared (m2). Patients were divided into five ordinal categories of BMI: <18.5, 18.5–22.9, 23.0–24.9, 25.0–29.9, and ≥30.0 kg/m2. According to the criteria established by the World Health Organization Asia-Pacific [19], the normal BMI category of 18.5–22.9 kg/m2 was determined as the reference group. Furthermore, patients with BMI 23.0–24.9 kg/m2 were classified as overweight and those with BMI ≥25.0 kg/m2 were classified as obese. BMI was measured at baseline, 6 months, and then annually. For time-updated BMI, the most recent measure of each patient was used.
Covariates
In addition to time-varying exposure and clinical outcomes, we considered clinically meaningful time-fixed and time-varying covariates. Time-fixed covariates involved baseline measurements such as age, sex, smoking status, and a history of DM, hypertension, and coronary artery disease, and time-varying covariates contained eGFR, hemoglobin, and systolic blood pressure (BP) (Fig. 1).
Data collection and measurements
Demographic data and medical history were collected by self-report and review of medical records at enrollment. BP was measured at each study visit. Age-adjusted Charlson comorbidity index was calculated at baseline, with additional points assigned for age. Blood samples were obtained after overnight fasting. Serum creatinine was measured at a central laboratory (Lab Genomics) and quantified by the isotope dilution mass spectroscopy–traceable method. During follow-up, laboratory parameters including creatinine and hemoglobin were annually assessed. Metabolic syndrome was defined as the presence of three or more of the following: 1) systolic BP ≥130 mmHg and/or diastolic BP ≥85 mmHg, or taking anti-hypertensive medication; 2) fasting plasma glucose ≥100 mg/dL, or taking anti-DM medication; 3) triglycerides ≥150 mg/dL, or on medication for elevated triglycerides; 4) high-density lipoprotein (HDL) cholesterol <40 mg/dL for male and <50 mg/dL for female, or on medication for reduced HDL cholesterol; and 5) waist circumference ≥90 cm for male and ≥80 cm for female according to the Asia-Pacific criteria [20].
Statistical analysis including marginal structural Cox model
For descriptive statistics across BMI categories, continuous variables are presented as mean ± standard deviation and categorical variables as number (percentage). Incidence rates in each BMI category were calculated per 1,000 person-years. We conducted a Cox-MSM to investigate the association between time-updated BMI and risk of KRT initiation. Cox-MSM is widely used in various fields of medical research, particularly with longitudinal observational data, to estimate the causal effects between clinical treatments and outcomes of interest in the presence of treatment-confounder feedback within a potential outcome framework [16]. For applying Cox-MSM, we first used multinomial logistic regression to obtain the inverse probability of treatment weight (IPTW) from BMI categories in each visit. We then applied the same method to predict the inverse probability of censoring weight (IPCW) at both the current and previous visits. These two models incorporated time-varying covariates and time-varying BMI while adjusting for time-fixed covariates. The final stabilized weight, calculated as the product of the IPTW and the IPCW, was truncated at above the 99.9th and below the 0.1th percentiles to yield consistent and unbiased causal effects of exposure (Fig. 2). Finally, we fitted a Cox proportional hazards (PH) model with stabilized weight to estimate the effects of time-updated BMI through the hazard ratio (HR). To compare with the Cox-MSM results, two Cox PH models were additionally performed: one including only baseline values and another time-dependent Cox model without inverse probability weights. Subgroup analyses, stratified by age (65 years), sex, DM, metabolic syndrome, and baseline eGFR (60 mL/min/1.73 m2), were conducted alongside the primary Cox-MSM analysis. For sensitivity analysis, we performed a competing risk analysis using a cause-specific hazard model, in which death events occurring before the study outcome were considered competing events and censored at the time of the death event. Two-sided p-values <0.05 were considered statistically significant. All statistical analyses were performed using R version 4.4.2 (R Foundation for Statistical Computing).
Assumptions of the marginal structural Cox model
To obtain an unbiased estimation of Cox-MSM, three assumptions of exchangeability, positivity, and correct model specification are necessary [21]. Regarding the exchangeability assumption, we assumed that the selected covariates were sufficient to adjust for both confounding and selection bias and applied the inverse probability weights to Cox-MSM. To assess unmeasured confounders in the association between time-updated BMI and KRT initiation, we performed E-value estimation for each HR, as proposed by VanderWeele and Ding [22]. The lowest E-value above 1 suggests that no unmeasured confounders are needed to nullify the association [23]. The positivity assumption requires that individuals in all BMI categories are classified based on a combination of confounders. In our large-sample study, no departures from the positivity assumption were observed. We assumed that the outcome model is correctly specified as Cox regression and the exposure model as multinomial logistic regression.
Results
Baseline characteristics
The baseline characteristics of the study population according to BMI category are presented in Table 1. In the total study population of 2,136 NDD-CKD patients, the reference group with normal BMI of 18.5–22.9 kg/m2 accounted for 627 patients (29.4%) and the group with BMI of 25.0–29.9 kg/m2 represented the largest proportion (35.1%). Patients with BMI of 25.0–29.9 kg/m2 were older, while those with BMI ≥30.0 kg/m2 were younger than the reference group. The groups with BMI ≥25.0 kg/m2 had a lower proportion of males and a higher proportion of current smokers, compared to the reference group. Furthermore, obese patients with BMI ≥25.0 kg/m2 had higher systolic BP, larger proportions of those with DM and hypertension, and lower eGFR and higher hemoglobin levels, compared to the reference group.
Risk of kidney replacement therapy initiation according to body mass index category
Among a total of 2,136 patients, 723 (33.9%) underwent KRT during the median follow-up of 8.3 years (interquartile range, 6.4–10.2 years). The incidence rates and HRs of KRT initiation according to statistical models are presented in Table 2. In the baseline model, the risk of KRT initiation was not significantly lower in the groups of obese patients with BMI ≥25.0 kg/m2 compared to the reference group. In the time-dependent model, the groups with obese patients showed a gradual decrease in the risk of KRT initiation as BMI increased. Notably, the groups with BMI of 25–29.9 kg/m2 (HR, 0.75; 95% confidence interval [CI], 0.63–0.89) and ≥30.0 kg/m2 (HR, 0.67; 95% CI, 0.51–0.87) exhibited a significantly lower risk of KRT initiation compared to the reference group. A similar pattern was observed among the groups of obese patients in the Cox-MSM, wherein the groups with BMI of 25–29.9 kg/m2 (HR, 0.73; 95% CI, 0.61–0.87) and ≥30.0 kg/m2 (HR, 0.69; 95% CI, 0.53–0.88) exhibited a significantly lower risk of KRT initiation compared to the reference group, and the risk of KRT initiation gradually decreased as BMI increased (Fig. 3). On the other hand, the underweight group with BMI <18.5 kg/m2 did not show a significant association between time-updated BMI and KRT initiation risk (HR, 1.06; 95% CI, 0.65–1.74).
Subgroup analysis
In the subgroup analysis stratified by age of 65 years, both subgroups showed a gradual association between higher time-updated BMI and a lower risk of KRT initiation in obese patients with BMI ≥25.0 kg/m2 (Table 3). Meanwhile, this trend was more prominent in female patients. Specifically, females with BMI ≥30.0 kg/m2 exhibited a significantly lower risk of KRT initiation (HR, 0.59; 95% CI, 0.40–0.88). Furthermore, a similar pattern was more prominently observed among patients with metabolic syndrome and those with a baseline eGFR lower than 60 mL/min/1.73 m2. On the other hand, the underweight group with BMI <18.5 kg/m2 showed no significant association with the risk of KRT initiation compared to the normal BMI group in all subgroup analyses.
Sensitivity analysis
In the competing risk analysis using a cause-specific hazard model, the inverse association between higher time-updated BMI and a lower risk of KRT initiation in obese patients was consistently observed. This analysis even showed numerically lower HRs compared to the primary Cox-MSM for the groups with BMI of 25.0–29.9 kg/m2 (HR, 0.72; 95% CI, 0.59–0.86) and ≥30.0 kg/m2 (HR, 0.67; 95% CI, 0.52–0.87) (Supplementary Table 1, available online).
Discussion
In this longitudinal observational study using the largest prospective cohort study of NDD-CKD patients in Korea, we demonstrated that a higher time-updated BMI was significantly associated with a lower risk of KRT initiation in patients with NDD-CKD, particularly in obese patients with BMI ≥25.0 kg/m2. This result was achieved by applying the Cox-MSM analysis, enhancing the robustness of our findings, particularly in light of the potential for body weight fluctuations in this population. Consequently, our study suggests that clinicians should carefully manage obesity in patients with NDD-CKD to preserve kidney function, while recognizing that an overly aggressive approach to weight reduction may not always be beneficial.
To the best of our knowledge, our study is the first to investigate the association between time-updated BMI and risk of KRT initiation in patients with NDD-CKD using the Cox-MSM analysis. The associations between BMI and various clinical outcomes in the CKD population have been a subject of interest since the discovery of the ‘obesity paradox,’ wherein a decrease in mortality risk was observed with increasing BMI [7–9], in contrast to the general population where an increase in BMI is associated with a higher risk of mortality [6]. However, in terms of the association between BMI and risk of KRT initiation, previous studies have yielded conflicting results in patients with NDD-CKD. Although most studies reported no significant association between BMI and risk of kidney failure [4,11,12], some showed that a higher BMI is significantly associated with an increase [8–10] or a decrease [13] in this risk in patients with CKD. Moreover, given that a significant weight change is associated with a higher risk of KRT initiation [14], we used the Cox-MSM approach to reflect BMI change during the follow-up period. Although obesity defined as BMI ≥25.0 kg/m2 was associated with increased risk of CKD progression in our previous study [24], the present analysis used more detailed BMI categories with a different reference range and different analytical method, which may explain the discrepancies in our findings. Given these persistent discrepancies highlighted by our methodologically distinct findings, a systematic meta-analysis is further warranted.
The mechanisms underlying our finding of a negative association between time-updated BMI and risk of KRT initiation, particularly in obese patients with CKD, remain unclear. Although BMI is widely used to assess obesity in clinical practice due to its simplicity, it is an unreliable measure for body composition in patients with CKD since it does not differentiate between fat mass and muscle mass, body fluid, or bone [25]. However, given the established positive correlations between BMI and both fat and lean mass, our findings align with a previous study in which CKD patients with both high lean and high fat mass exhibited the lowest risk of the composite outcome of all-cause mortality and cardiovascular events [26].
In terms of lean mass, high skeletal muscle mass was associated with a reduced risk of CKD progression [27]. Considering the established associations of risk factors such as insulin resistance, endothelial dysfunction, oxidative stress, and inflammation not only with sarcopenia [28], but also with reduced kidney function and increased albuminuria [29], a higher time-updated BMI due to a higher lean mass may have a potential protective effect against the initiation of KRT.
Furthermore, regarding fat mass, the potential mechanisms underlying its protective effects include greater energy reserves to withstand the catabolic stress of CKD and elevated levels of tumor necrosis factor-α receptors that can counteract proinflammatory cytokines [30]. Moreover, the heterogeneity of fat type results in different metabolic effects. Subcutaneous fat functions as a metabolic buffer, protecting other tissues from lipotoxicity by lipid overflow and ectopic fat [31]. In contrast, visceral fat is associated with metabolic syndrome and cardiovascular risk [32]. Considering that patients with higher BMI may have a higher subcutaneous fat mass rather than visceral fat [33], a higher time-updated BMI due to a higher subcutaneous fat in obese patients with NDD-CKD may also have a protective effect against the initiation of KRT.
Moreover, the subcutaneous adipose tissue of the gluteofemoral region is beneficial for metabolic health, while abdominal subcutaneous adipose and the visceral adipose tissue are associated with a metabolically unhealthy phenotype [34]. Thus, not only the type of adipose tissue but also its distribution may impact metabolic health, which in turn may influence subclinical inflammation and insulin resistance, contributing to CKD progression. Therefore, clinicians should consider body composition, including muscle and fat mass, as well as adipose tissue distribution, when assessing BMI and obesity in patients with CKD. Moreover, further research is warranted to determine the optimal body composition for reducing the risk of KRT initiation using detailed information on body composition. Since a multifrequency bioelectrical impedance analysis was conducted to assess body composition in the KNOW-CKD phase II study [17], we plan to perform an in-depth analysis of the impact of specific body compositions on kidney function.
In the subgroup analysis, a significant negative association was observed between time-updated BMI and risk of KRT initiation among females with BMI above 30.0 kg/m2, while no significant association was identified among males. This observation suggests the potential involvement of sex hormones in this disparity. Since estrogens, which can be produced in adipose tissue, reduce mesangial proliferation, stimulate nitric oxide generation in the kidneys, and possess antioxidant properties [35,36], a higher BMI may exert a protective effect against KRT initiation in females. Furthermore, since advanced CKD is associated with protein-energy wasting, which induces inflammatory process and promotes muscle degradation resulting in low BMI [37], a significant inverse association between time-updated BMI and risk of KRT initiation in obese patients with CKD may be prominently observed in the subgroup with baseline eGFR <60 mL/min/1.73 m2.
This study has several limitations. First, owing to the observational nature, residual confounding factors and reverse causation may be present. However, in our analysis of E-values, all of the lower limits were above 1, which suggests no unmeasured confounders (Supplementary Table 2, available online). Second, the number of time-varying covariates was relatively small compared to that of time-fixed covariates. Given that the exposure of our study was annually measured BMI, time-varying covariates also had to be selected from annually measured variables, resulting in a relatively small number compared to time-fixed covariates. Third, as a potential consequence of the limited time-varying covariates, the variability of stabilized weights at each time interval was relatively small. Fourth, this study exclusively included Korean patients; therefore, further research involving other ethnic groups is required to enhance the generalizability of our findings. Fifth, although underweight status has been known to increase the risk of KRT initiation due to the systemic oxidative stress and chronic inflammation caused by sarcopenia and malnutrition [38,39], the number of patients with underweight status in our cohort was relatively small, limiting the statistical power to sufficiently evaluate its effect. Thus, further research focusing on the underweight status is also warranted.
In conclusion, a higher time-updated BMI is associated with a lower risk of KRT initiation in obese patients with NDD-CKD in Korea. Therefore, clinicians should carefully evaluate obesity and consider body composition when managing weight in patients with NDD-CKD. Further research is necessary to investigate the effect of specific body composition on patient outcomes.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation
Supplement table 1
Print
