The long-term effects of blood urea nitrogen levels on cardiovascular disease and all-cause mortality in diabetes: a prospective cohort study

Patients who participated in the National Health and Nutrition Examination Survey (NHANES) and those included in the 10 consecutive cycle datasets (1999–2000, 2001–2002, 2003–2004, 2005–2006, 2007–2008, 2009–2010, 2011–2012, 2013–2014, 2015–2016, and 2017–2018) were enrolled in the present study. The NHANES includes adults and children from 0 to 80 years of age. Briefly, the National Center for Health Statistics (NCHS) of the Centers for Disease Control and Prevention used complex, multistage, stratified methods to collect a representative sample of the US population. NHANES covers several variables, which are mainly related to the demographic, laboratory, dietary, examination, and important questionnaire data. These questionnaire variables were used by trained professionals during in-home and mobile examination center (MEC) interviews. More detailed information about NHANES is available online (https://​www.​cdc.​gov/​nchs/​nhanes). The NHANES study was approved by the NCHS research ethics review board (protocols 98 − 12, 2005-06, 2011-17, and 2018-01). Patients aged ≥ 18 years who provided written informed consent upon enrollment were included in the study. Those who self-reported as pregnant were excluded (n = 1,305). Diabetes was ascertained according to the following criteria: [1] self-reported physician-diagnosed diabetes or current use of oral antihyperglycemic agents or insulin [2], a glycated hemoglobin A1c (HbA1c) level of ≥ 6.5% [3], a fasting blood glucose level of ≥ 7.0 mmol/L, and/or [4] a postprandial 2-h plasma glucose level of ≥ 11.1 mmol/L [13]. Patients with diabetes [1] with unavailable follow-up information (n = 19) [2], diagnosed with cancer at baseline (n = 1,739), and [3] with missing information on BUN levels (n = 717) were also excluded. Ultimately, 10,507 patients with diabetes were included in the analysis (Supplemental Fig. S1).

BUN levels were measured using the enzymatic conductivity rate method, which is described in detail in the NHANES. The reference range for BUN levels was 3.1–9.5 mmol/L for adults. The outcomes were based on the mortality data of the National Death Index, which is an NCHS-centralized database, from the date of NHANES interview until follow-up (December 31, 2019) [14]. All underlying causes of death in the US population were precisely recorded following the International Statistical Classification of Diseases, 10th Revision (ICD-10). The ICD-10 codes I60–I69, I20–I51, I13, I11, and I00–I09 were used to determine the cause of CVD death, and malignant neoplasm was ascertained using the C00–C97 codes [15]. The follow-up interval was calculated using the difference between the last known date recorded in the mortality file for each individual and the date of the initial examination.

Sociodemographic information on age, marital status (widowed, divorced, married, separated, living with partner, and never married), race/ethnicity (other races, non-Hispanic White, other Hispanics, non-Hispanic Black, and Mexican–American), physical activity (inactive, active, and insufficiently active), smoking history (never, current, and former smokers), educational degree (< high school level, high school level, and > high school level), and alcohol intake (none, heavy, and low-to-moderate drinker) were obtained using standardized interview questionnaires. Household income was measured using the family income-to-poverty ratio (PIR). Patients who smoked ≥ 100 cigarettes, ≥ 100 cigarettes and then stopped smoking, and < 100 cigarettes were considered current, former, and never smokers, respectively [2]. During leisure time, patients who did not perform physical exercise were categorized as “inactive.” Those who engaged in moderate activity with metabolic equivalents of 3–6 at least five times each week or in strenuous exercise with a metabolic equivalent of > 6 at least thrice per week, which met the recommended levels of leisure–time activity, were categorized as “active”; those who were neither inactive nor did not meet the criteria for the active group were categorized as “insufficiently active” [16]. Alcohol status was determined based on the 24-h dietary intake measured during the MEC interview. Men consuming 0.1–27.9 g of alcohol per day and women consuming 0.1–13.9 g of alcohol per day were classified as moderate drinkers, men consuming > 28 g of alcohol per day and women consuming > 14 g of alcohol per day were classified as heavy drinkers, and patients who consumed 0 g of alcohol per day were classified as nondrinkers. Healthy Eating Index 2015 (HEI-2015) was introduced to measure the diet quality [17]. Body mass index (BMI) was subdivided into three categories (≥ 30.0, 25.0–29.9, and < 25.0 kg/m²) [2]. The criteria for diagnosing baseline hypertension were as follows: [1] taking antihypertensive medications or [2] having a systolic blood pressure of ≥ 130 mmHg and/or a diastolic blood pressure of ≥ 80 mmHg [18]. The diagnosis of dyslipidemia was based on the following conditions: [1] high-density lipoprotein levels of < 40 mg/dL in men and < 50 mg/dL in women [2], a total cholesterol (TC) level of ≥ 200 mg/dL [3], a low-density lipoprotein level of ≥ 130 mg/dL [4], a triglyceride level of ≥ 150 mg/dL, and/or [5] taking lipid-lowering medications [19]. Data on diabetes duration and use of diabetes medications, including oral hypoglycemic drugs and insulin, were collected when the patients were interviewed during medical visits. The homeostatic model assessment of insulin resistance (HOMA-IR) was calculated using the formula adopted in previous studies [20]. A 24-h recall interview was performed to collect information on the patient’s daily diet [21]. We calculated the total energy intake (TEI) using the US Department of Agriculture Automated Multiple-Pass Method [22]. The baseline CVD was diagnosed by a professional physician. The baseline levels of biochemical markers, including serum cotinine, serum calcium, serum TC, urine albumin, and serum triglycerides, were strictly analyzed according to the laboratory standards. Further experimental details were also documented in the NHANES. The duration of diabetes and the medications used were obtained through interviews during medical visits.

Free Statistics software version 1.4 was used to perform statistical analysis, which was based on the R statistical software package (http://​www.​R-project.​org, the R Foundation) [23]. Categorical and normally distributed continuous variables were presented as numbers (n) with percentages (%) and as means with standard deviations, and nonnormally distributed continuous variables were presented as medians with interquartile ranges. The patients were divided into quarters (Q1–Q4) according to BUN values. The difference between the four groups was compared by one-way ANOVA tests (continuous variables with normal distribution and homogeneity of variance), Kruskal-Wallis test (continuous variables with nonnormal distribution), and χ² test (categorical variables). A P-value of < 0.05 was considered significant. Person-time was calculated based on either [1] the interval between the date of NHANES interview and death and [2] the interval between the date of NHANES interview and follow-up (December 31, 2019), whichever came first. Kaplan–Meier analysis and log-rank test were used in the univariate survival analysis. The hazard ratios (HRs) and 95% confidence intervals (CIs) of CVD, cancer, and all-cause mortality were calculated using the multivariate Cox proportional hazards regression models. In Model 1, none of the variables were adjusted. Variables such as gender, race/ethnicity, and age were included in Model 2. Model 3 was fully adjusted for gender, race/ethnicity, age, PIR; educational status; marital status; BMI; HbA1c and serum TC levels; diabetes medication use; diabetes duration; insulin therapy; serum cotinine, calcium, serum triglyceride, urine albumin, magnesium, and vitamin D levels; alcohol intake; multivitamin supplement use; kidney disease; TEI; HEI; HOMA-IR; hypertension; hyperlipidemia; CVD; smoking status; and physical activity.

Subgroup analysis was performed according to age (≤ 60 or > 60 years), BMI level (< 30 or ≥ 30 kg/m²), gender, race/ethnicity (others or non-Hispanic White), alcohol intake status, smoking status, diabetes duration (≤ 3 or > 3 years), physical activity, and HbA1c level (< 7% or ≥ 7%). To explore the potential interaction effects between subgroups, likelihood ratio tests were conducted to assess the corresponding multiplicative interaction terms, and P values were also calculated. Moreover, a series of sensitivity analyses were performed. First, patients who died within 2 years after the baseline examination were excluded to reduce the possibility of reverse causality. Second, considering that dietary protein intake plays a critical role in the BUN levels, the sensitivity analysis was additionally adjusted for protein intake. Third, severe liver disease may decrease the BUN levels; therefore, we excluded patients with liver disease. Fourth, as part of the sensitivity analysis, we also determined the association of BUN with mortality among patients with normal BUN levels (3.1–9.5 mmol/L). Fifth, considering that other potential confounding factors may still exist, the inverse probability of treatment weighting (IPTW) method was used to compare the Q1 and Q4 groups. Covariates used in the model were gender, age, BMI, race/ethnicity, PIR, alcohol intake, HOMA-IR, smoking status, HbA1c level, physical activity, diabetes duration, and kidney disease. We used restricted cubic splines(RCS) with three knots to visualize the dose-response association between BUN and mortality. In addition, generalized additive models were used to explore the potential nonlinear association of BUN levels with mortality. When a nonlinear association was found in the models, a recursive algorithm was used to calculate the inflection points in order to determine the link between BUN and mortality; a two-piecewise linear regression was also performed.