nium octaoxide, which are insoluble, have little potential to cause renal toxicity but could cause pulmonary toxicity if exposure is by inhalation (ATSDR, 1999b). Insoluble uranium compounds can remain in the pulmonary tissues, especially the pulmonary lymph nodes, for a long time and constitute a localized radiologic hazard. As a general rule, uranium in the intestinal tract is less readily absorbed than uranium from the respiratory tract and results in lower doses per unit intake.

The committee was charged with evaluating the scientific literature on the effects of depleted uranium. As detailed in this report, the evaluation focused on direct experimental and observational evidence in animals and human populations. The committee acknowledges that there is a broader literature on risk assessment of radiologic and chemical toxicants, including uranium.

In general, population-based quantitative risk assessment is used in public health to inform intervention strategies, for example, in setting policy and regulations. Such risk assessment is not intended to estimate risk to any given individual in a population or to determine causality. Rather, it is intended to characterize population-attributable risk broadly to support population-level, not individual-level, decisions.

The current approach to quantitative risk assessment, developed by the National Research Council, consists of four steps: hazard identification, exposure assessment, dose-response modeling, and risk characterization (NRC, 1983, 1994). Of the four steps, the committee emphasized two as most relevant to its charge: hazard identification (that is, Does evidence of toxicity of depleted uranium exist at any level of exposure?) and exposure assessment (that is, What actual levels of exposure were experienced by military personnel serving in the Gulf War?). The committee considered mechanisms of both radiologic and chemical toxicity and a variety of cancer and noncancer outcomes or end points.

Elements of the risk assessment approach—notably dose-response modeling—vary among the cancer and noncancer end points. Cancer and genetic changes are modeled as a mathematical function in which risk increases with increasing exposure or dose. Although considerable controversy remains about the shape of the dose-response curve, especially at low doses, a linear no-threshold model has traditionally been used. This approach has been used for ionizing radiation as a carcinogen; for example, the National Research Council report Health Risks from Exposure to Low Levels of Ionizing Radiation: BEIRVII Phase 2 endorses the use of such a model for radiogenic-cancer risk estimation (NRC, 2006).

A linear no-threshold dose-response model implies that cancer risk increases proportionally with increasing dose and that no “safe” dose exists (that is, every exposure or dose conveys some risk—low doses have low risk and higher doses proportionally higher risk). Such a model continues to be used for population-based quantitative risk assessment in public health, in spite of substantial uncer-