Does blood flow accurately predict thrombosis or failure of hemodialysis synthetic grafts? A meta-analysis

A number of studies have reported that a single low blood flow (Qa) measurement in synthetic hemodialysis grafts predicts thrombosis or failure. In a meta-analysis of these studies, we computed receiver operating characteristic (ROC) curves that evaluated the predictive accuracy of a Qa measurement. The ROC curves plotted sensitivity versus false-positive rate for predicting thrombosis or failure at different Qa thresholds. A perfect predictor has an area under the curve (AUC) of 1.0, whereas a predictor with no discriminative ability has an AUC of 0.5. We identified studies through a literature search and included our own unpublished data. A random-effects model was used to combine the ROC curves from different studies. Of 19 identified studies, 12 were suitable for computing binormal ROC curves (6 predicted thrombosis; 6 predicted failure). The studies measured Qa and then observed outcome during periods of 1.5 to more than 6 months. The combined AUCs from these studies indicate Qa was a relatively poor predictor, with 0.70 +/- 0. 04 (range, 0.61 to 0.84) for thrombosis and 0.76 +/- 0.07 (range, 0. 62 to 0.90) for failure. The wide range of AUCs also shows there was much heterogeneity between studies. We conclude that a single Qa measurement does not appear to have enough accuracy to be a clinically useful predictor of graft thrombosis or failure. Serial Qa measurements and identification of factors that caused heterogeneity between studies may be needed to achieve sufficient accuracy.