Kidney Function and Risk of Renal Cell Carcinoma

The incidence of renal cell carcinoma (RCC) is projected to increase in the coming decades, particularly in Latin America, Asia, and Africa due to a transition to Western lifestyle (1, 2). RCC survival is highly dependent on the stage at diagnosis. RCC patients with localized disease have 5-year survival rates of 93%, which drops to 75% for regionally spread disease, and further to 16% for metastasized disease (3). The early detection of RCC is challenging, partly because RCC is asymptomatic at an early stage (4) and is often detected incidentally during imaging for unrelated purposes (5).

Several studies have proposed strategies to improve the early detection of RCC, including the use of urine sticks (6), blood biomarkers [e.g., KIM1 (7), AQP1, and PLIN2 (8)],abdominal contrast-enhanced CT or screening programs for high-risk patients according to their comorbidities (9). However, there is currently no widespread screening for RCC in asymptomatic individuals and it is not clear how to identify individuals at a sufficiently high risk to benefit from screening (10).

The relationship between kidney function and incident RCC has not been frequently studied; however, some studies have indicated that individuals with poor kidney function have an increased risk of RCC (11). However, it is not clear whether this association reflects the etiologic role of kidney function (12) and whether markers of kidney function may inform RCC risk assessment. Kidney function was measured as the glomerular filtration rate (eGFR) which is typically calculated from blood creatinine measurements. One study indicated an increased risk of kidney cancer with reduced eGFR, suggesting that poor kidney function may be involved in RCC etiology or present as an early marker of RCC (13). Creatinine is a residual chemical compound produced during energy-producing processes in the muscles. Kidneys filter creatinine from the blood into urine along with other waste products of the body. Creatinine is widely used in clinical practice to diagnose chronic kidney disease (CKD), for prognostication and monitoring of CKD, but is strongly influenced by other factors, including hydration level, body size, diet, physical activity, pregnancy, or the use of some drugs (14). Blood cystatin-C has been suggested as a reliable marker of eGFR, which is less affected by previously mentioned factors (15, 16).

Our study aimed to evaluate the etiologic role of kidney function in RCC by describing the temporal relationship between kidney function and incident RCC. We also evaluated whether kidney function may be useful for predicting RCC risk.

The UK Biobank (UKB) is a general population cohort study that recruited 502,369 participants across the United Kingdom from 2006 to 2010 (17). Detailed descriptions of data collection, including information on lifestyle, environmental, and anthropometric measurements, have been previously described in detail (17). Blood samples were collected from all participants at baseline. We excluded 4,930 participants who were diagnosed with kidney cancer or unknown type of cancer before recruitment. We also excluded participants who lacked measurements of creatinine or cystatin C (n = 34,261, Supplementary Fig. S1). Most of the missing biomarker measurements (80%) occurred because the participant did not donate a blood sample, and the remaining were due to a combination of technical issues. There were no apparent differences between the study participants with biomarker measurements and those without them [with creatinine and cystatin C missing: 54% of females vs. 57% with available measurement, a median body mass index (BMI) of 26.9 without measurement vs. 26.7, a median age of 58.1 without measurement vs. 58.3, no difference was shown in smoking status, hypertension, diabetes, systolic and diastolic blood pressure]. Incident RCC and death information was obtained via individual record linkage to national cancer and mortality registries. We defined the RCC cases following the International Classification of Diseases (ICD-0–3) histology codes: 8050, 8140, 8260, 8270, 8280–8312, 8316–8320, 8340–8344 (18). Data on tobacco exposure history, alcohol intake and clinical information were obtained via questionnaires at the recruitment visit. BMI was calculated from weight and height (weight/height², kg/m²), measured at baseline.

All participants provided written informed consent, and the study protocol was approved by the North West Multicenter Research Ethics Committee of the United Kingdom. This study accessed relevant UKB data under application number 97846.

Creatinine was measured using enzymatic assays [in mmol/L, average within-laboratory coefficients of variation (CV): 1.9%], and high-sensitivity C-reactive protein (CRP) was measured using immunoturbidimetry (in mg/L, average within-laboratory CV: 1.7%), both on a Beckman Coulter AU5800. Cystatin C was measured using latex-enhanced immunoturbidimetry (in mg/L, average within-laboratory CV: 1.4%) on a Siemens ADVIA 1800.

The eGFR was calculated using the 2022 CKD-EPIC creatinine formula, 2012 CKD-EPI cystatin C formula, and 2021 CKD-EPI creatinine and cystatin C combined formula (Table 2 in Inker and colleagues; ref. 19).

We imputed the missing risk-factor data 10 times using predictive mean matching for alcohol consumption per week (31% missing), physical activity score (20%), smoking intensity (16%), smoking status (1%), BMI (<1%), and hypertension (<1%). Age, sex, and center of recruitment were used as predictors. The HRs obtained in each imputed dataset showed no variation, therefore, we used one imputed dataset in all subsequent analyses (at 1 year: HR from 1.257 to 1.260, at 5 years: HR from 1.107 to 1.110)

The relationship between kidney function (eGFR), related biomarkers (creatinine and cystatin-C), and RCC risk was first evaluated using Cox proportional hazard models, but the Schoenfeld residuals suggested time-dependent relative hazards. Therefore, subsequent RCC risk analyses were carried out using flexible parametric survival models that allow hazards to vary as a function of time (20) and provide the additional advantage of readily facilitating absolute risk estimations (21). We performed both minimally adjusted and risk factor–adjusted analyses. The minimally adjusted model used time since recruitment as the time-scale and was adjusted for age at recruitment and CRP level (log-standardized), with stratification by sex. The risk factor–adjusted analysis included additional covariates (22, 23), specifically smoking status (never, former, current), alcohol intake (grams per week), physical activity score (total metabolic equivalent task, minutes per week), use of nonsteroidal anti-inflammatory drugs (NSAIDs, yes or no), history of kidney failure (yes or no), history of kidney stones (yes or no), diabetes (yes or no), hypertension (yes or no), and BMI (continuous).

We modeled the time-dependent relationship between RCC risk and each biomarker using restricted cubic splines and used three internal knots (located at 25%, 50%, and 75% of the follow-up time). Participants were censored at the time of any primary cancer diagnosis (except nonmelanoma skin cancer), death, or the end of the registry follow-up.

To establish a risk prediction model, we first calculated the risk score for all participants based on the Singleton and colleagues’ risk-prediction model (24). This model was originally developed in the European Prospective Investigation into Cancer and Nutrition (EPIC) study and accounts for sex, age, BMI, smoking status, hypertension status, and diastolic and systolic blood pressure. We trained and assessed the performance of four models: the original Singleton model, the Singleton model recalibrated in UKB, the recalibrated Singleton model with CRP added, and finally, the updated model that also included eGFR. We used a resampling-based algorithm with replacements to internally validate the model by dividing the study population into random training (60% of the UKB) and testing sets (remaining 40% of the UKB) 500 times. Participants were censored upon the first primary malignant RCC, death, or end of follow-up (defined as the first date between the end of the cancer or death registry by the center), whichever occurred first. Model training used flexible parametric survival models with follow-up time as a time scale, with the following covariates included in all models: the logit of the Singleton risk score, presence or absence of CRP, and eGFR (Supplementary Data S1, equations 1–3). Knots were placed in accordance with the Singleton and colleagues’ model (model in Supplementary Data S1, equation 2). We compared the Akaike information criterion (AIC) for models in which eGFR was parameterized with natural cubic splines (with one, two, or three degrees of freedom), with or without log transformation, and with or without intercept in the flexible parametric survival model. The lowest AIC was attained by modeling eGFR directly with splines with three degrees of freedom, with an intercept in the flexible parametric survival model. The optimized models, with and without eGFR, were applied to the testing set (40% of the UKB) to calculate the RCC risk for each participant. Model calibration was calculated as the ratio of expected cases to observed cases. Model discrimination was estimated using AUCs (25).

All analyses were performed using R, version 4.1.2. The following specific packages were used: package survival for Cox proportional hazards models (ref. 26; version 2.44–1.1); package rstpm2 for flexible parametric survival models (ref. 27; version 1.6.2); package timeROC for time-dependent AUCs (version 0.4); package mice for imputation (ref. 28; version 3.13); and package ggplot2 for figures (ref. 29; version 3.2.1).

The data are available upon request to UKB at https://www.ukbiobank.ac.uk/. The scripts, used for this study, are available in a Code Ocean capsule: https://codeocean.com/capsule/2588674/tree/v1.

Our final analysis included 463,178 participants from the UKB who were followed up for 5,696,963 person-years (median, 12.6 years; SD, 1.90 years; Table 1). During the follow-up, 1,447 study participants were diagnosed with RCC, including 65 within 1 year of follow-up, 472 within 5 years, and 1,148 within 10 years. Females represented 54% of our cohort, and the median age at blood draw was 58 years for people without a cancer diagnosis and 62 years for participants with a cancer diagnosis during the follow-up. The median eGFR calculated from creatinine and cystatin was 88.9 mL/minute/1.73 m² (IQR, 20.9) for participants with RCC diagnosis and 95.9 mL/minute/1.73 m² (IQR, 19.6) for non-cases.

In the initial risk analysis, which focused on evaluating the potential importance of kidney function in RCC etiology, we restricted the studied cohort to the participants without diagnosis of any cancer (except nonmelanoma skin cancer) before recruitment and censored the cohort at any first cancer, end of follow-up, or death, whichever occurred first, this analysis included 442,119 participants from UKB who were followed for 5,222,905 person-years (Supplementary table S1). In the minimally adjusted risk analysis, eGFR was associated with RCC risk throughout the follow-up period, regardless of whether eGFR was calculated using creatinine, CRP, or both markers (dotted lines in Fig. 1). Accounting for RCC risk factors attenuated the association between eGFR and RCC risk, but lower eGFR remained associated with higher RCC risk within the last 5 years leading up to diagnosis, (solid lines in Fig. 1). The association of HR for lower eGFR with RCC risk gradually attenuated with increasing time from blood draw to diagnosis, and there was no risk increase among participants diagnosed more than 5 years after recruitment. For instance, the risk factor–adjusted HR (HR_sd) estimates for one SD decrease in eGFR measured with both creatinine and cystatin-C (eGFR_creat-cystC) were 1.24 (95% CI, 1.19–1.30) when measured 1 year prior to diagnosis, 1.12 (95% CI, 1.07–1.17) 5 years before diagnosis, and 1.05 (95% CI, 1.00–1.09) 10 years before diagnosis. The corresponding HR_sd estimates for eGFR measured with either creatinine (eGFR_creat) or cystatin-C (eGFR_cystC) were similar and followed the same pattern by time from blood collection to diagnosis.

The risk prediction part focused on evaluating whether eGFR could improve upon a published risk model. We restricted the study cohort to participants without a diagnosis of RCC before recruitment and censored the cohort at RCC diagnosis, end of follow-up, or death, whichever occurred first. This analysis included 463,178 participants from the UKB, who were followed up for 5,696,963 person-years (Table 1). As the association of eGFR with RCC was time-dependent on follow-up time (Fig. 1), we evaluated calibration and discrimination for models predicting the 1-, 5- and 10-year risk of RCC (Fig. 2). We initially assessed the performance of the published Singleton and colleagues’ model trained in EPIC and found that it discriminated significantly well (1 year: AUC = 0.69, 95% CI, 0.63–0.79; 10 years: AUC = 0.70, 95% CI, 0.69–0.72), but predicted slightly more cases than observed [1-year expected/observed (E/O): 1.27, 95% CI, 1.01–2.01; 10-year E/O: 1.07 (95% CI, 0.96–1.14)]. After recalibrating the Singleton and colleagues’ model in the UKB and incorporating CRP (Supplementary Data S1, equation 2), the E/O ratio slightly improved [1 year E/O: 0.97 (95% CI, 0.66–1.58); 10 years E/O: 1.02 (95% CI, 0.90–1.10); Fig. 2].

We subsequently incorporated eGFR_creat-cystC and CRP into the risk model – the updated model (Supplementary Data S1, equation 3), which was well-calibrated across the follow-up period [1-year E/O: 0.89 (95% CI, 0.65–1.47); 10-year E/O: 1.08 (95% CI, 0.93–1.14)]. Model discrimination (AUC) was 0.73 (95% CI, 0.67–0.84) for 1-year RCC risk, 0.75 (95% CI, 0.71–0.77) for 5-year risk and 0.72 (95% CI, 0.70–0.74) for 10-year risk. The difference in AUCs, at 1 year, between the published model and the updated model with eGFR at 0.04 (95% CI for difference in AUCs, –0.013–0.104).

To illustrate the impact on RCC risk for different eGFR levels (estimated using both creatinine and cystatin-C), we calculated the absolute 1-year risk trajectories as a function of age across risk factor combinations and eGFR categories (Fig. 3), with risk factor stratifications by sex (Fig. 3A), BMI (Fig. 3B), hypertension (Fig. 3C), and smoking status (Fig. 3D). The corresponding model parameters are presented in Supplementary Table S2.

For instance, a male aged 60 years (with no hypertension, an average BMI and a never smoker) and an eGFR of 95 mL/minute/1.73 m² (i.e., average eGFR > 60 mL/minute/1.73 m²) had a 0.01% 1-year RCC risk (95% CI, 0.009–0.016) compared with 0.05% (95% CI, 0.03%–0.09%) for a male, with similar clinical parameters, but with eGFR 49 mL/minute/1.73 m² (average eGFR below 60 mL/minute/1.73 m²; Fig. 3A). The corresponding 1-year risks for females, with similar clinical parameters, were estimated at 0.006% for eGFR of 95 mL/minute/1.73 m² and 0.03% for eGFR of 49 mL/minute/1.73 m² (Fig. 3A). Corresponding risk trajectories for the 5-years RCC risk models for females are provided in Supplementary Figs. S2 and S3.

We subsequently evaluated the 1-year RCC risk as a function of age, separated for males and females (Fig. 4) and by combining the risk factors of RCC using the same model as above. A female or a male with normal eGFR (eGFR > 60, not smoker, not hypertensive, BMI < 25) had a 1-year risk of 0.01% (95% CI, 0.00%–0.01%) at 60 years of age. A female aged 60 years with decreased kidney function (eGFR ≤ 60) and hypertension, BMI above 30 kg/m² who is currently smoking had 1-year RCC risk of 0.06% (95% CI, 0.04%–0.10%), and a male with the same profile had a 1-year RCC risk of 0.12% (95% CI, 0.07%–0.20%). The corresponding 5-year RCC risk trajectories as functions of eGFR are shown in Supplementary Fig. S4.

Supplementary Figure S5 shows the effect of incorporating eGFR_creat-cystC measurements for individual-level risk assessment in the recalibrated Singleton and colleagues’ model.

In this study, we described the relationship between eGFR, a marker of kidney function, and incident RCC. We confirmed that low eGFR was associated with higher RCC risk when measured within 5 years of diagnosis, but other risk factors explained the association with RCC risk beyond 5-years of follow-up. We found that eGFR measurements provided a small improvement in RCC risk prediction compared with that of a published model when predicting 1-year risk.

The association between poor kidney function and RCC risk has been previously reported (11, 12, 13, 23, 30, 31), but questions remain as to whether this association reflects a causal effect on RCC development (12). Recently, Lees and colleagues studied the association between kidney function measured using eGFR in the UKB and 10 different cancer sites. For renal tract cancer (C64-C67), they observed a 40% risk increase for study participants with low eGFR (<60 mL/minute/1.73 m²) based on 3,549 renal tract cancer cases (31). Park and colleagues studied more than 10,000 kidney cancer cases in the Korean population and reported an 18% increased risk in individuals with eGFR ≤30 mL/minute/1.73 m² compared with people with normal eGFR (13). Other studies have reported a 2- to 3-fold increase in the risk of RCC in individuals with CKD (23). Our study confirmed the relationship between low eGFR and increased RCC risk, but describes a clear dependency of the relationship on the timing of eGFR measurement relative to RCC diagnosis. Specifically, we found that the inverse relation between eGFR and RCC risk was strongest close to RCC diagnosis and gradually attenuated with a longer time between eGFR assessment and RCC diagnosis. After accounting for the key RCC risk factors, the association between eGFR and RCC was not evident after 5 years of follow-up.

However, the mechanism by which kidney function causes RCC development remains unclear. Some have speculated that alterations in the immune system in patients with chronic kidney disease may lead to immunodepression and a higher prevalence of kidney infections, which may influence the risk of RCC (12, 32). Other proposed mechanisms include hypoxia-induced fibrosis which is central to the impaired renal function (33). However, considering our epidemiologic observations and the lack of an established and distinct mechanism, our study does not provide clear support for the hypothesis that poor kidney function influences RCC development in a causal manner. Rather, the observations from our study are consistent with early preclinical malignant changes in the kidneys that influence kidney function several years prior to diagnosis, thus causing the observed risk associations through reverse causality.

The lack of a clear role for kidney function in RCC etiology does not preclude the possibility of the potential use of eGFR in risk prediction. Several previous studies have developed risk models for RCC (24, 34). In addition to age and sex, these models are mainly based on modifiable risk factors such as smoking history, BMI, and blood pressure (34). Singleton and colleagues established an RCC risk model in the EPIC study in 2021 (24); however, despite this model performing well in the training sample, it was not able to identify individuals at very high risk in the general population of EPIC participants. Several other biomarkers have been studied to improve risk prediction and early detection of RCC; however, few studies have been conducted using prediagnostic samples. For instance, Scelo and colleagues highlighted KIM-1 as a potential candidate for RCC detection 5 years before diagnosis in a case–control study (n cases = 190) in EPIC (7). In this study, we found that combining eGFR and CRP measurements with the Singleton risk model provided a small improvement in short-term (1-year) risk discrimination (AUC of 0.73 for integrated model, versus AUC of 0.69 for Singleton and colleagues). We further evaluated the model-predicted risks for different risk factor combinations/scenarios and found that individuals who accumulate multiple risk increasing factors have substantial RCC risk. For instance, a 60-year-old male with an eGFR below 60 mL/minute/1.73 m² who was also obese, hypertense, and actively smoking had a 2.4-fold higher 1-year RCC risk (0.12% vs. 0.05%) risk compared with a 60-year-old male without these risk factors (Fig. 4). It is important to understand that these risk estimates reflect the extremes of the UKB participants, but it is possible that individuals with such extreme risk profiles may benefit from additional interventions or surveillance.

Our study has several strengths, first, we used a cohort of participants longitudinally followed for more than 10 years with baseline blood-based kidney function measures for 463,178 study participants. As RCC is a moderately common cancer, the large cohort sample allowed us to obtain enough cases to perform internal model training and testing, as well as the opportunity to evaluate the temporal relationship between eGFR and RCC risk. However, as is common for population cohorts, the UKB cohort is known to be a healthy cohort (35), which may limit the external validity of our prediction models and decrease the number of individuals with CKD. The UKB did not provided any information on cancer stage or grade for the moment. Finally, in clinical settings, kidney function was also assessed using urine albumin measurements, and 75% of the UKB sample lacked urine markers. Further analyses and risk prediction models should be performed including participant ethnicity, urine markers, in a population with high prevalence of poor kidney function. It would also be helpful to evaluate whether the association between eGFR with RCC risk depends on the tumor size at diagnosis. This information was not available for our study. In addition, conducting a longitudinal study with individual repeated measures of kidney function markers prior to cancer diagnosis would be compelling.

Our study confirms that individuals with poor kidney function have an elevated risk of renal cancer for up to 5 years following eGFR measurement. However, these observations do not provide support for a causal role of kidney function in RCC etiology. The extent to which eGFR measurements can inform RCC risk discrimination was limited overall, but eGFR may be of potential use to identify individuals in the extremes of the risk distribution.

R.M. Martin reports grants from Cancer Research UK during the conduct of the study. D.C. Muller reports grants from Cancer Research UK during the conduct of the study, grants from Cancer Research UK, and grants from NIH/NCI outside the submitted work. No disclosures were reported by the other authors.

Where authors are identified as personnel of the International Agency for Research on Cancer/World Health Organization, the authors alone are responsible for the views expressed in this article and they do not necessarily represent the decisions, policy, or views of the International Agency for Research on Cancer/World Health Organization. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR, or the Department of Health and Social Care.

K. Alcala: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. H. Zahed: Validation, writing–review and editing. R. Cortez Cardoso Penha: Data curation, formal analysis. N. Alcala: Methodology, writing–review and editing. H.A. Robbins: Methodology, writing–review and editing. K. Smith-Byrne: Writing–review and editing. R.M. Martin: Funding acquisition, writing–review and editing. D.C. Muller: Writing–review and editing. P. Brennan: Funding acquisition, writing–review and editing. M. Johansson: Resources, supervision, investigation, writing–original draft, project administration, writing–review and editing.

This research was supported by Cancer Research UK 25 (C18281/A29019) programme grant (The Integrative Epidemiology Programme). R.M. Martin is a National Institute for Health Research senior investigator (NIHR202411). R.M. Martin is supported by a Cancer Research UK 25 (C18281/A29019) programme grant (the Integrative Cancer Epidemiology Programme). R.M. Martin is also supported by the NIHR Bristol Biomedical Research Centre that is funded by the NIHR (BRC-1215–20011) and is a partnership between University Hospitals Bristol and Weston NHS Foundation Trust and the University of Bristol. R.M. Martin is affiliated with the Medical Research Council Integrative Epidemiology Unit at the University of Bristol which is supported by the Medical Research Council (MC_UU_00011/1, MC_UU_00011/3, MC_UU_00011/6, and MC_UU_00011/4) and the University of Bristol.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.