Associations with Blood Lead and Urinary Cadmium Concentrations in Relation to Mortality in the US Population: A Causal Survival Analysis with G-Computation

Abstract

Using the parametric g-formula, we estimated the 27-year risk of all-cause and specific causes of mortality under different potential interventions for blood lead (BLLs) and urinary cadmium (UCd) levels. We used data on 14,311 adults aged ≥20 years enrolled in the NHANES-III between 1988 and 1994 and followed up through 31 Dec 31 2015. Time and cause of death were determined from the National Death Index records. We used the parametric g-formula with pooled logistic regression models to estimate the relative and absolute risk of all-cause, cardiovascular, and cancer mortality under different potential threshold interventions for BLLs and UCd concentrations. Median follow-up was 22.5 years. A total of 5167 (36%) participants died by the end of the study, including 1550 from cardiovascular diseases and 1135 from cancer. Increases in BLLs and creatinine-corrected UCd levels from the 5th to the 95th percentiles were associated with risk differences of 4.17% (1.54 to 8.77) and 6.22% (4.51 to 12.00) for all-cause mortality, 1.52% (0.09 to 3.74) and 1.06% (-0.57 to 3.50) for cardiovascular disease mortality, and 1.32% (-0.09 to 3.67) and 0.64% (-0.98 to 2.80) for cancer mortality, respectively. Interventions to reduce historical exposures to lead and cadmium may have prevented premature deaths, especially from cardiovascular disease.

Keywords: NHANES; cancer; cardiovascular; epidemiology; metals.

Figure 1

Adjusted all-cause, cardiovascular disease, and cancer mortality risks according to blood lead (A) and urinary cadmium (B) levels using parametric g-formula with pooled logistic regression models. The solid lines represent the 5th percentile (0.70 μg/dL for lead and 0.04 μg/L for cadmium) and the dashed lines represent the 95th percentile (9.70 μg/dL for lead and 1.63 μg/L for cadmium). Robust 95% confidence intervals (CIs) for each exposure category estimated by bootstrapping (in the pooled logistic regression model) are presented in Table 2.

Figure 2

Adjusted all-cause, cardiovascular disease, and cancer mortality risks according to blood lead (A) and urinary cadmium (B) levels using parametric g-formula with pooled logistic regression models. The solid lines represent the first quartile and the dashed lines represent the fourth quartile. Robust 95% confidence intervals (CIs) for each exposure category estimated by bootstrapping (in the pooled logistic regression model) are presented in Table 3.