Humans are exposed to a multitude of environmental toxicants daily, and there is a great interest in developing statistical methods for assessing the effects of chemical mixtures on various health outcomes. One difficulty is that multiple chemicals in the mixture can be subject to left-censoring due to varying limits of detection. Methods have been proposed to handle a single biomarker with limit of detection, including a nonstandard application of the Cox model, by considering the biomarker measure as the “event time” and treating the limit of detection as a right-censored time after a reversal of scale (Dinse et.al. (2014)). The hazard ratio in such a Cox model reflects the relationship between disease status (predictor) and the biomarker value (outcome). We extend this method to handle multiple correlated biomarkers subject to limits of detection, through a newly proposed multivariate Cox model. We apply the proposed method to a subset of Sister Study participants whose cadmium, arsenic, lead, mercury, iron, manganese and selenium were available from toenail clippings, to understand the effect of the metal mixture on breast cancer incidence.