Risk assessment of diesel exhaust and lung cancer: combining human and animal studies after adjustment for biases in epidemiological studies

BACKGROUND: Risk assessment requires dose-response data for the evaluation of the relationship between exposure to an environmental stressor and the probability of developing an adverse health effect. Information from human studies is usually limited and additional results from animal studies are often needed for the assessment of risks in humans. Combination of risk estimates requires an assessment and correction of the important biases in the two types of studies. In this paper we aim to illustrate a quantitative approach to combining data from human and animal studies after adjusting for bias in human studies. For our purpose we use the example of the association between exposure to diesel exhaust and occurrence of lung cancer. METHODS: Firstly, we identify and adjust for the main sources of systematic error in selected human studies of the association between occupational exposure to diesel exhaust and occurrence of lung cancer. Evidence from selected animal studies is also accounted for by extrapolating to average ambient, occupational exposure concentrations of diesel exhaust. In a second stage, the bias adjusted effect estimates are combined in a common effect measure through meta-analysis. RESULTS: The random-effects pooled estimate (RR) for exposure to diesel exhaust vs. non-exposure was found 1.37 (95% C.I.: 1.08-1.65) in animal studies and 1.59 (95% C.I.: 1.09-2.10) in human studies, whilst the overall was found equal to 1.49 (95% C.I.: 1.21-1.78) with a greater contribution from human studies. Without bias adjustment in human studies, the pooled effect estimate was 1.59 (95% C.I.: 1.28-1.89). CONCLUSIONS: Adjustment for the main sources of uncertainty produced lower risk estimates showing that ignoring bias leads to risk estimates potentially biased upwards.