There were several limitations of the current study. First, the wavelengths of the SLO were not optimized for oximetry. However, despite this we found statistically significant differences among the stages of disease. Second, we used only one image for analysis, but in the future more images can be acquired to reduce measurement variability. In fact, the standard deviations in our study usually were larger than those in previous reports (though similar to those using SLO
36), which may have reduced the power to discern differences in biomarker variables among the stages of disease. Third, a fixed calibration factor was used to calculate vessel diameters and, thus, did not account for variations in refractive error among subjects. However, subjects with high refractive error (>6 diopters) were excluded from the study. Fourth, we did not account for measurement variations as a function of the cardiac cycle. While Knudtson et al.
53 have found that image quality is more important than the cardiac cycle as a source of measurement variability, Chen et al.
54 have reported changes in vessel diameter during the heart cycle as 3.46% and 4.82% in arteries and veins, respectively. This suggests that the heart cycle may have been a major source of variability in our data over and above the error of measurement. In the future, taking this into account may reduce the variability substantially. Fifth, aging has been shown to affect retinal SO
2 measurements in healthy and diabetic subjects.
28,55 Although control and diabetic subjects were age matched, adjustments for age were taken into account in the models. While changes in the optical properties of the eye due to disease were minimized by the calculation of optical density ratios, future studies are needed for rigorous determination of the effects of alterations in lenticular light transmission on SO
2 measurements. Sixth, our method of calibrating values of SO
2 from values of ODR sometimes led to values exceeding 100%. However, we used the same method other investigators have used and they also obtained SO
2 values above 100%. Finally, the distribution of participants by race and diabetic group was not even (e.g., an underrepresentation of African Americans and Hispanics among No DR and also an underrepresentation of whites among NPDR and PDR). Nevertheless, the observational study design and multivariate linear regression models allow for estimates that appropriately adjust for race.