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 as 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 We identified studies through a literature search and included our own unpublished data.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. (C) 1999 by the National Kidney Foundation, Inc.