Comparison of PET Oxygen Extraction Fraction Methods for the Prediction of Stroke Risk

MATERIALS AND METHODS

This study was a retrospective analysis of prospectively collected clinical and PET data from 68 patients with symptomatic atherosclerotic carotid artery occlusion and complete quantitative OEF studies enrolled in the STLCOS (1). All patients had atherosclerotic occlusion of 1 common or internal carotid artery and ipsilateral ischemic symptoms. Patients were followed for the occurrence of stroke or death for the duration of the study. Eighteen healthy volunteers were studied to establish a reference range of OEF values. All studies were performed under institutional review board–approved protocols. Informed consent was obtained from all subjects.

RESULTS

Mean hemispheric (±SD) absolute OEF in the healthy volunteers was 0.41 (±0.09; range, 0.26–0.64). Absolute OEF was not a significant predictor of ipsilateral stroke using the upper limit of the reference range (4 patients categorized with increased OEF, P = 0.21). When the upper 95% confidence limit for absolute OEF (0.44) was used, 33 of the 68 patients were categorized as having increased OEF. Eight of the 9 strokes occurred in these patients (P = 0.0042).

All 3 methods were significant predictors of stroke risk as individual continuous variables. In multivariate analysis comparing all 3 methods, only the count-based method remained in the final model (P = 0.0028). Absolute hemispheric values performed better than ratios of absolute values (P to remove = 0.0043). ROC analysis (Fig. 1) revealed a similar order of performance, with the area under the curve for the absolute OEF (0.769) falling between the count-based ratio method (0.815) and the ratio of absolute hemispheric values (0.737).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1.

ROC curves for all 3 OEF methods. On solid line, ○ indicates hemispheric ratios of count-based OEF; on dotted line, ▿ represents hemispheric ratios of absolute OEF; and on dashed line, □ indicates absolute values of OEF. These curves were generated by progressively increasing threshold for determining whether OEF was increased. At lowest thresholds (far upper right of curve), all patients had increased OEF and all 9 strokes occurred in these patients. At this point in curve, all methods have 100% sensitivity (ability to identify patients with subsequent stroke) but poor specificity (ability to identify patients without subsequent stroke). As threshold becomes higher (moving leftward toward origin), sensitivity falls as specificity improves. The curves are similar; however, there is greater area beneath curve for count-based ratio method, which suggests best performance.

DISCUSSION

All 3 methods are predictive of stroke risk as both continuous and dichotomous variables. Both ratio techniques were previously shown to be significant predictors of risk for ipsilateral stroke as dichotomous variables for this same group of 68 patients (Kaplan–Meier log rank, P = 0.0048 [count-based] and P = 0.025 [absolute ratios]) (4). The failure of hemispheric ratios of absolute OEF as a dichotomous variable to predict stroke in the study of Yamauchi et al. (3) may be caused by several factors, including the threshold used to identify increased OEF and the inclusion of patients with bilateral carotid occlusion. The importance of optimizing the threshold used to identify increased OEF is emphasized by the initial failure of absolute OEF to predict stroke in this study.

The multivariate and ROC analyses of our data support the conclusion of Yamauchi et al. (3) that absolute values of OEF are better predictors of future stroke than hemispheric ratios of absolute OEF values. This study also suggests that the count-based OEF method performs better than absolute values of OEF measured by the technique of Mintun et al. (5). In part, this may be because of the effect of blood volume asymmetry. Blood volume may increase with autoregulatory vasodilation. Hemispheric increases in the count-based OEF image may reflect the effects of both increased oxygen extraction and increased blood volume, because there is no correction for unextracted oxygen remaining in the blood with the count-based method.

CONCLUSION

All 3 methods are predictive of stroke risk in patients with unilateral carotid occlusion. Count-based ratios of OEF perform as well as or better than absolute values or ratios of absolute values of OEF, when OEF is measured using the technique of Mintun et al. (5). The count-based OEF method does not require arterial input data, a measurement of blood volume, or complicated metabolic processing. The count-based method is a practical and accurate technique for identification of patients with increased OEF, particularly in the setting of a multicenter trial.