Albuminuria, Kidney Function, and Cancer Risk in the Community

Yejin Mok; Shoshana H. Ballew; Yingying Sang; Josef Coresh; Corinne E. Joshu; Elizabeth A. Platz; Kunihiro Matsushita


Am J Epidemiol. 2020;189(9):942-950. 

In This Article


Study Population

We used data from the Atherosclerosis Risk in Communities (ARIC) Study, a community-based cohort of 15,792 individuals aged 45–64 years at visit 1 (1987–1989) from 4 US communities (Forsyth County, North Carolina; Jackson, Mississippi; suburban Minneapolis, Minnesota; and Washington County, Maryland). Thereafter, cohort participants underwent 5 examinations in 1990–1992 (visit 2), 1993–1995 (visit 3), 1996–1998 (visit 4), 2011–2013 (visit 5), and 2016–2017 (visit 6). Of 15,792 participants, 11,656 attended visit 4, at which ACR was first evaluated in ARIC. We excluded participants with missing data for CKD measures (n = 441) or a history of cancer (n = 1,318). Additionally, we excluded the small number of non-White/non-Black participants (n = 28), Black participants from the Minnesota and Maryland centers (n = 31), and those with missing covariates (n = 903), resulting in a final sample of 8,935 participants.

Kidney Disease Measures

Our exposures of interest were eGFR and ACR at baseline. We used the CKD Epidemiology Collaboration equation with serum creatinine and cystatin C for eGFR.[24,25] Serum creatinine was measured with a modified kinetic Jaffe reaction. Creatinine values were calibrated to the Cleveland Clinic Laboratory,[26] then standardized to the isotope-dilution mass spectrometry traceable method.[27] Serum cystatin C was measured in 2008 from stored frozen samples collected at visit 4 by a particle-enhanced immunonephelometric assay (N Latex Cystatin C; Siemens Healthcare Diagnostics, Deerfield, Illinois) with a BNII nephelometer (Siemens Healthcare Diagnostics). Urine albumin was measured by a nephelometric method, and urine creatinine was measured by the Jaffe method using spot samples stored at −70°C immediately after collection.


Age, sex, race, and education were assessed at visit 1. Smoking status, alcohol intake, health insurance status, and medical history were evaluated at visit 4 by trained interviewers using standardized questionnaires. The amount of alcohol consumed (in g/week) was calculated, assuming that 4 oz of wine contains 10.8 g, 12 oz of beer contains 13.2 g, and 1.5 oz of liquor contains 15.1 g of ethanol. Smoking status was categorized as current, former, and never smokers. Pack-years of smoking was calculated as the number of packs per day multiplied by years of smoking among current and former smokers at visit 4. History of cardiovascular disease included prevalent coronary heart disease, stroke, and heart failure. All medications were verified by trained personnel. Family history of cancer was obtained based on questionnaire at visit 2. Participants were also asked how often they have routine physical examination for general checkup and either "at least once a year" and "at least once every five years" were considered to be having routine physical examination for general checkup. Body mass index was calculated as weight (kg) divided by height (m) squared. Hypertension was defined as systolic blood pressure of ≥140 mm Hg, diastolic blood pressure of ≥90 mm Hg, or use of antihypertensive medications. We defined diabetes mellitus as fasting serum glucose level of ≥126 mg/dL, nonfasting glucose level of ≥200 mg/dL, self-reported physician diagnosis of diabetes, or antidiabetic medication use. Total cholesterol was determined using enzymatic methods. High-sensitivity C-reactive protein was measured by the immunoturbidimetric assay on a BNII analyzer (Siemens Healthcare Diagnostics).

Ascertainment of Cancer Incidence and Mortality

The outcomes of interest were overall and site-specific cancer incidence and mortality. Cancer incidences were ascertained in 1987–2012 by linkage to state cancer registries of the states of North Carolina, Mississippi, Minnesota, and Maryland, and supplemented by abstraction of medical records and hospital discharge codes for self-reported cases. Participants who self-reported a diagnosis of cancer on an annual follow-up telephone call (semiannual from 2012) were contacted separately for more information on cancer diagnoses, and medical records pertaining to cancer diagnoses and treatment were collected. High-priority cancers (bladder, breast, colorectal, liver, lung, pancreas, and prostate) not captured by registries were confirmed by an adjudication team through all collected materials. Other types of cancer were confirmed through the cancer registry, abstraction of archived medical records for time intervals not covered by the registry, and death certificate data, but they were not otherwise adjudicated.

If a participant had >1 type of incident cancer during follow-up, the earliest date of cancer incidence was chosen for analysis of the overall cancer. Cancer mortality included all individuals at risk for cancer at visit 4 in whom cancer was identified as the underlying cause of death.

Statistical Analysis

Baseline characteristics of the study population were summarized according to 4 categories of eGFR (in mL/minute/1.73 m2: <45, 45–59, 60–89, and ≥90) and ACR (in mg/g: <10, 10–29, 30–299, and ≥300) according to the international clinical guideline,[28] previous literature,[13,29] and the distributions in our study.

In the survival analysis, we performed the analyses for overall cancer as well as overall cancer with the exclusion of prostate cancer (overall nonprostate cancer), because screening for prostate cancer with prostate-specific antigen has been controversial and guidelines changed several times during ARIC follow-up.[30–32] Also, the concentration of prostate-specific antigen is reportedly influenced by kidney function.[33,34] Age/sex/race-standardized rate of cancer incidence and mortality were calculated across categories of eGFR and ACR using the age, sex, and race distribution in the US population.[35] To visualize the full-spectrum association between both kidney disease measures and cancer risk, hazard ratios of a linear spline model of eGFR and ACR with 3 knots (45, 60, and 90 mL/minute/1.73 m2 for eGFR, with 95 mL/minute/1.73 m2 as a referent, and 10, 30, and 300 mg/g for ACR, with 5 mg/g as a referent) were plotted using Cox proportional hazards models. We also quantified cancer risk by cross-categories of eGFR and ACR (eGFR of ≥90 mL/minute/1.73 m2 and ACR of <10 mg/g as a referent). We implemented several models to evaluate the impact of potential confounders: Model 1 adjusted for age, sex, and race-centers; and model 2 further accounted for body mass index, hypertension, diabetes, total cholesterol, statin use, aspirin use, high-sensitivity C-reactive protein, history of cardiovascular disease, family history of cancer, alcohol amount (g/week), the combination of smoking status and pack-years of smoking (never, former (<25 and ≥25 pack-years), and current (<25 and ≥25 pack-years)), and kidney disease measures (ACR was accounted for in the analyses for eGFR and vice versa).

We conducted several sensitivity analyses. First, we adjusted for health insurance (yes vs. no) and routine physical examination for general checkup (yes vs. no) as a measurement of the likelihood of cancer-screening opportunity to address access to medical care. Also, we restricted to those who had cancer-screening opportunity. Second, we assessed subgroups by age (<65 vs. ≥65 years), sex, and race (White vs. Black). Interactions were tested by likelihood ratio test comparing models with and without product terms of interest. Third, we censored overall or site-specific cancer cases within the first 3 years, to reduce the possibility of reverse causation between kidney function/damage and cancer risk. Fourth, for lung cancer, the most prevalent cancer in our study, we conducted stratified analysis by smoking status (never vs. former vs. current smokers). Fifth, we conducted competing-risk analysis with noncancer mortality as a competing end point of cancer risk using Fine and Gray subdistribution hazard regression.[36] Finally, we also quantified the risk of cancer by cystatin C–based eGFR.

For site-specific cancers, we analyzed only cancer sites with ≥50 cases and presented hazard ratios (as a logarithmic scale) in a forest plot. Due to sparse data in some categories, we modeled the kidney disease measures as continuous (per 15-mL/minute/1.73 m2 decrement in eGFR below 90 mL/minute/1.73 m2 and per 8-fold increment in ACR with log-transformed ACR) variables for the site-specific analyses as well as subgroup analysis. All analyses were performed using Stata, version 14 (StataCorp LP, College Station, Texas). All statistical tests were 2-sided and statistical significance was determined as P < 0.05.