As with all biochemical measurements, cotinine levels are subject to both measurement error and normal day-to-day variations. In regression analysis these variations lead to regression dilution bias and subsequent underestimation of odds ratios(26,42). To correct for this bias we estimated RDRs by use of repeat samples. The RDR corrected ORs of cotinine were, as expected, notably higher than the corresponding uncorrected values, indicating that the underlying risk associated with cotinine is substantially underestimated (Figure 1).
