Figure 2

Example of regression dilution bias in observational epidemiology. The sizes of the boxes are inversely proportional to the amount of statistical information. The HRs are plotted on a natural logarithmic (doubling scale). The black boxes (and the black dotted line) show the association between mismeasured systolic blood pressure and CHD; the red boxes (and the red dotted line) shows the association between systolic blood pressure and CHD if systolic blood pressure was measured without error. This illustrates that the slope of the association is underestimated when an exposure that is subject to random measurement error is related to a disease outcome. SBP, systolic blood pressure.